# three-ai AdsMetri build story — spoolcast dev log 7 devlog about building AdsMetri across Codex, Manus, and Claude, and why role clarity mattered more than raw AI output. Type: video guidebook Published: 2026-05-13 [Original page]() [Browse this group]() Format: long Duration (seconds): 773 Series: Spoolcast Dev Log ## Core message AdsMetri looked like one product, but it was really three different kinds of work: deciding what the product should be, proving what the live Meta data was actually doing, and making the codebase behave correctly. Codex, Manus, and Claude only became useful together when those roles stopped blurring. ## Production writing: author: Claude; role: screenplay, shot-list, scene prompts audio: voice: Puck; tts: Google Cloud TTS; beatCount: 78 render: engine: Remotion ## Style wojak-gpt2 Modern wojak / doomer-chad meme comic style, built natively on GPT Image 2 (text-to-image + image-to-image). Sister library to wojak-comic (which was built on nano-banana-2). Use this when the session's preferred_model is the GPT Image 2 family. Modern wojak / doomer-chad meme comic style, built natively on GPT Image 2 (text-to-image + image-to-image). Sister library to wojak-comic (which was built on nano-banana-2). Use this when the session's preferred_model is the GPT Image 2 family. builder: the main hooded figure — mid-register wojak (not fully doomer, not fully chad). Alias for the style anchor; serves double-duty as both the style anchor and the builder character reference. ai-figure: a wojak-universe AI figure that is slowly breaking down from overuse. Full-body standing. EMOTIONAL REGISTER: flat-affect, NO displayed emotion, NO smile, NO frown — a dead expressionless stare. Depression is the absence of emotion, not the display of sadness. FACE LINEWORK: brows, nose, mouth drawn in the classic rough / jagged / slightly-disjointed wojak linework (uneven strokes, hand-drawn imperfections). EYES: replaced with hollow loading-spinner icons (dashed rings) set in dark-shadowed sockets. CRACKS: visible hairline cracks spider across the face and forehead as if the outer shell is fracturing from overuse. At least one crack wide enough to reveal a glimpse of AI circuitry underneath — thin circuit-trace lines, a small exposed chip pattern. Fractures read as structural fatigue, not injury. FOREHEAD: a crooked rectangular ink-stamp reading 'AI' in bold capital letters, partially crossed by a crack. HOODIE: pale cream / off-white hoodie, hood down, noticeably lighter than the weary narrator's gray hoodie. POSTURE: slight slouch, arms hanging at sides. PLAIN background, no environment, no other characters, no other text. Single clear readable subject. meme-chad: the exact meme-Chad identity used by earlier Spoolcast devlogs: clean-shaven confident Chad with signature yellow pointed mohawk-spike hair, strong jawline, pronounced cheekbones, smug half-smile, muscular arms, and a red ripped-sleeve OUCH! shirt. He is comedic contrast by default: confident, slightly too pleased with himself, and visually brighter than the weary builder. Do not turn him into a bearded worker, lumberjack, security guard, producer, stagehand, engineer, inspector, or generic muscular man. ## Scenes ### Cold Open - "One Product, Three Problems" So I decided to build an analytics dashboard for my e-commerce store, called AdsMetri. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. A small e-commerce storefront and ad dashboard connected by one simple glowing line, with tired wojak holding a laptop. Keep it simple, one frame, no dense UI labels. Render exactly this text on the frame, hand-lettered in the session style: "ADSMETRI". ### Cold Open - "One Product, Three Problems" AdsMetri started as a Meta ads budget manager. Which sounds simple. And that was the first warning sign. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Small budget button glowing on a dark desk, with a tired wojak staring at it like it is suspiciously calm. No text except maybe `BUDGET`. Render exactly this text on the frame, hand-lettered in the session style: "BUDGET". ### Cold Open - "One Product, Three Problems" Because a budget manager is not just a button that says, "spend less money," although honestly, that would be a strong product category. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: A 1:1 wojak2-style mobile frame: tired wojak at a control desk facing two large arcade buttons. The red button is labeled SPEND MORE and the green button is labeled SPEND LESS in big simple readable words. The joke must be instantly clear. No panels, no extra text, no dense UI. Render exactly this text on the frame, hand-lettered in the session style: "SPEND MORE" | "SPEND LESS". ### Cold Open - "One Product, Three Problems" The app had to connect to Meta, sync campaigns, queue budget changes, run those changes at the right time, show account-specific activity, and not quietly mix one ad account into another. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. One control panel with cables going to five simple props: account key, campaign cards, queue tray, clock, account folders. Single scene, no labels needed. ### Cold Open - "One Product, Three Problems" Then the product grew. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Small budget button expanding into a larger dashboard room, still one frame. ### Cold Open - "One Product, Three Problems" It became a dashboard for reading ad performance: lifecycle charts, funnels, ROAS, CPA, CPC, spend trends, BEROAS, and daily Meta P and L. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Dashboard lab with abstract chart instruments, funnel tube, money gauge, campaign stack. No dense labels. ### Cold Open - "One Product, Three Problems" In plain English: it stopped being only "change the budget" and became "tell me what is actually working." Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Split not as panels: foreground hand moves from a budget knob to a truth gauge on the same console. Render exactly this text on the frame, hand-lettered in the session style: "WHAT IS WORKING?". ### Cold Open - "One Product, Three Problems" That is a much scarier sentence. Because once software starts telling you what is working, the real problem is not whether the chart renders. The real problem is whether the chart is telling the truth. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Beautiful chart on monitor casting a long suspicious shadow. Wojak leans in. ### Why Three AIs That is why this build ended up spread across three AI harnesses: Claude, Manus, and Codex. Not because three sounds more impressive than one. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Three separate workbenches around one unfinished product cube, not a panel grid. Different accents: orange, green, blue. ### Why Three AIs Three is also how many browser tabs I open before forgetting why I am alive. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Wojak buried behind three browser windows, existentially tired. ### Why Three AIs The real reason was that AdsMetri was not one problem. It was three problems pretending to be one. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. One product cube with three tangled cords underneath: design, live data, code. Cords are color-coded, no labels unless needed. Render exactly this text on the frame, hand-lettered in the session style: "ONE PRODUCT / THREE PROBLEMS". ### Why Three AIs One problem was product and interface clarity: what should this dashboard show, how should it feel, what should the user understand first, and what should not be crammed into the UI like a tax form with a growth mindset. That became Claude's strongest lane. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Design studio desk with sparse mockup cards, color swatches, funnel sketch, orange accent. No dense text. ### Why Three AIs One problem was live reality: OAuth, Meta API behavior, token permissions, deployment, database state, dummy data, tests, and whether the product actually worked against real accounts instead of only in a pleasant local fantasy. That became Manus's strongest lane. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Green-accent operator at a live test bench with keycard, cloud gate, database cylinder, check light. ### Why Three AIs And one problem was the codebase itself: account state, analyzer snapshots, scheduling logic, pending changes, chart behavior, and the bugs where the UI looked fine while the underlying logic was holding a small meeting without you. That became Codex's strongest lane. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Blue-accent debugger in a code tunnel untangling state cables behind a polished dashboard. ### Why Three AIs The mistake was treating those tools like interchangeable assistants. The useful version was treating them like specialized harnesses with handoffs. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. First three identical generic robot stamps rejected; then three distinct tools on a handoff table. Single scene, no panels. ### Spoolcast Context Quick context: this devlog was made inside Spoolcast, my AI video pipeline. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Small production desk with source notes flowing into script/audio/visual/render icons. ### Spoolcast Context Spoolcast can start from source material like chats, transcripts, repo notes, research, or story ideas, then help turn that into a script, shot list, visuals, narration, render, and final files. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Conveyor belt of abstract source objects turning into a video reel. Keep labels minimal or none. ### Spoolcast Context I still direct it with feedback and approvals. Which, based on this episode, also includes telling the system to follow the protocol that it wrote down for itself. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: A 1:1 wojak2-style mobile frame: tired wojak holding an absurdly overstuffed clipboard labeled only with simple abstract checkboxes, while three vague helper stations wait in the background. The clipboard must be the obvious focal prop. No readable dense text, no panels. ### Spoolcast Context Software automation is powerful. It is also occasionally a clipboard with confidence. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Confident clipboard on a tiny podium, wojak unimpressed. ### Spoolcast Context For this devlog, I am keeping the story source-backed. I am using the pasted Codex, Manus, and Claude perspectives from the chat. If something is uncertain, I will say it is uncertain instead of sanding it into a cleaner story. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Source notes pinned under a glass pane labeled only `SOURCE`, with red uncertainty flags. ### Spoolcast Context Because the whole point of this video is not "AI made a product." The point is what actually shipped, what broke, and what multi-AI work taught us after the novelty wore off. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Product dashboard, broken bug net, and handoff baton on a workbench. ### Product Foundation The first version of AdsMetri was operational. Connect a Meta ad account. Sync campaigns. Show performance metrics. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Account key plugs into a campaign stack and simple metric monitor. ### Product Foundation Queue budget changes. Execute those changes later. Keep an activity log so the user can see what happened. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Budget cards in a queue tray moving toward a clock and logbook. ### Product Foundation That sounds like product plumbing, because it is. But plumbing is one of those things people only call boring when it works. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Dashboard pipes under the floor; one tiny leak dripping into a calendar. ### Product Foundation Manus reported building much of that foundation: app scaffold, database tables, Meta OAuth, token management, campaign sync, budget queues, scheduler behavior, dummy data, legal pages, deployment setup, and a test suite reported as seventy-three tests. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Manus-green foundation scene: database blocks and connection pipes under an app dashboard. No checklist. ### Product Foundation I am saying "reported" because this script is using the pasted summaries as source. I am not independently re-auditing Manus's entire history here. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Magnifying glass over a source page with one uncertainty stamp. No dense text. ### Product Foundation But the role is clear. Manus was closest to foundation plus live verification. It was the harness dealing with the product as something that had to connect to the outside world. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Manus character connecting product cube to an outside-world signal tower. ### Manus Hard Parts The best example was campaign visibility. Some campaigns were not syncing. The exact count is inconsistent across the pasted notes, so I am not going to invent a clean number. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Campaign cards behind a partly closed permission gate; some visible, some blocked. ### Manus Hard Parts The important part is the root cause: Meta API token permissions. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Keycard denied at a gate while campaign cards sit behind it. ### Manus Hard Parts The sync logic can look suspicious. The database can look suspicious. The frontend can look suspicious. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Three suspect props under a desk lamp: code cable, database cylinder, UI screen. No role grid. ### Manus Hard Parts And then the answer is: the token was not allowed to see the campaign. That is not a bug you solve by staring harder at a React component. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Wojak staring intensely at a React-like component while locked door glows behind him. ### Manus Hard Parts That is why live verification mattered. The product had to be checked against what the real platform allowed, not just what the code hoped was true. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Manus character checking real-platform gate while code blueprint sits nearby. ### Manus Hard Parts Manus also hit timezone work. The scheduler originally had an 11 PM Bangkok assumption, but Meta Ads works on account calendar-day boundaries. Manus reported refactoring that to a midnight-to-12:30 window in the ad account timezone. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Clock room with Bangkok clock crossed out softly, account-time clock centered. Minimal text: `ACCOUNT TIME`. ### Manus Hard Parts A clock can be perfectly accurate and still be wrong for the product. Very inspiring. Very irritating. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Perfect clock proudly pointing at the wrong calendar day. ### Manus Hard Parts Manus also called out numeric precision from the Meta API. Tiny floating-point weirdness sounds harmless until it becomes a dashboard number and someone asks why finance looks haunted. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Tiny decimal dust under magnifying glass growing into worried money dashboard ghost-shadow. Keep no readable numbers. ### Manus Hard Parts So Manus's lane was not just "build features." It was: connect to reality, check reality, and stop reality from entering the product wearing a fake mustache. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Manus at checkpoint removing fake mustache from incoming data packet. ### Codex Repo-Level Work But live verification does not automatically make the app coherent. You can prove the data exists and still display it in the wrong account, the wrong chart, or the wrong moment. That was the Codex lane. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Verified data packet entering a maze of app screens, Codex waits at maze entrance. ### Codex Repo-Level Work Codex was the repo-level implementation and debugging harness. In plain English: it worked where the product's moving pieces had to agree inside the actual code. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Codex-blue mechanic tightening bolts between dashboard, scheduler, analyzer, account selector. ### Codex Repo-Level Work Codex's pasted summary covered dashboard, analyzer, settings, budget flows, campaign sync, account scoping, pending change display, recent activity scoping, multi-account state bugs, analyzer lifecycle behavior, chart interactions, right-edge labels, visibility controls, fulfillment economics, timezone fallback cleanup, account-aware scheduling, and compact-theme implementation after design direction was set. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Codebase map with many abstract nodes converging into one product screen. No readable list. ### Codex Repo-Level Work That list is long because state bugs are social creatures. They rarely arrive alone. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. One bug-shaped warning light pulls several tangled cables behind it. Avoid literal creature if possible; use warning icons/cable knots. ### Codex Repo-Level Work The selected Meta account had to match the dashboard. The dashboard had to match pending changes. Pending changes had to match the scheduler. The analyzer had to match the selected account and available snapshots. Activity logs had to be scoped to the right account. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Domino-like chain of app modules lighting up in correct order, centered on selected account token. ### Codex Repo-Level Work A rendering bug is visible. A state bug hides behind a clean interface and waits for you to trust it. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Polished dashboard smiling politely while tangled cables hide behind wall. ### Codex Repo-Level Work Wrong data does not always look broken. Sometimes it looks professional. It wears a nice font and ruins your week. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Beautiful chart wearing a tiny tie, wojak suspicious. Minimal/no text. ### Codex Hard Parts Codex's hard bugs were often not where they first appeared. A chart issue could be a snapshot coverage issue. A dashboard issue could be an account-filtering issue. A sync issue could be cached database state disagreeing with live Meta state. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Debugging elevator moving between layers: chart surface, snapshot room, account filter, cache basement. ### Codex Hard Parts A timezone issue could be a fallback assumption hiding in one part of the UI. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Tiny hidden clock under a UI panel causing shadows on scheduler. ### Codex Hard Parts This is where the product shifted from "does the screen work?" to "does the screen mean what it says?" That is a much stricter test. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Dashboard passes visual check, then faces a truth scale. Render exactly this text on the frame, hand-lettered in the session style: "MEAN WHAT IT SAYS?". ### Codex Hard Parts Codex also worked on fulfillment economics: fulfillment costs, BEROAS, daily Meta P and L, and break-even thinking. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Profit/workbench with cost weights, revenue gauge, break-even line as abstract balance. ### Codex Hard Parts That changed the product question from "how much did we spend?" to "did this actually make money?" Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Spend meter rotating into profit meter. Render exactly this text on the frame, hand-lettered in the session style: "DID THIS MAKE MONEY?". ### Codex Hard Parts Once money math enters the dashboard, believable is not good enough. You need boringly correct. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Dull accountant stamp approving a dashboard while exciting chart fireworks are unplugged. ### Codex Hard Parts The best financial UI is a little boring. Exciting finance dashboards are how people end up in meetings with legal pads and no joy. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Overly dramatic finance dashboard being quietly replaced by calm plain chart; bored meeting room in background. ### Claude Design And Spec But correct behavior still does not make a product feel usable. The user has to understand it. That was Claude's lane. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Correct but confusing dashboard transforms into clearer layout sketch. ### Claude Design And Spec Claude was strongest as the design and specification harness. Claude made Compact theme HTML mockups for the dashboard, analyzer campaigns, analyzer ad sets, analyzer ads, and settings. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Design studio with multiple mockup boards as abstract cards, orange accent, no dense text. ### Claude Design And Spec It iterated on tooltip shape, funnel layout, sidebar structure, P and L color rules, spend-by-campaign callouts, background colors, outlines, and negative space. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Designer adjusting one mock dashboard with swatches and funnel prop. Avoid list boards. ### Claude Design And Spec In plain English: Claude helped decide what the product should look like and how the interface should communicate the work. Design is not decoration here. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. UI mockup acting like a map/road sign for a confused user. ### Claude Design And Spec If AdsMetri is telling someone where ad performance is changing, the UI has to make comparison easy. If it is showing profit safety or fulfillment break-even, the UI has to make risk legible. If it is handling multiple accounts, the UI has to make the current account obvious. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Three UI magnifiers over comparison, risk gauge, account badge on one dashboard. ### Claude Design And Spec Nobody should have to perform archaeology in a dropdown. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Wojak with tiny archaeologist brush excavating a dropdown menu. ### Claude Design And Spec Claude also wrote a handoff package for Codex: design tokens, infrastructure notes, component guidance, preserve rules, and mockup copies. That turned design output into implementation instructions. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Orange design folder passed to blue code workbench. No role grid. ### Claude Design And Spec When that worked, it was a handoff. When it did not, it was just two assistants confidently standing on different maps. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Two wojak assistants holding different maps pointing in opposite directions. ### Claude Hard Parts Claude's self-report also included mistakes, which is useful because mistakes are where the workflow becomes real. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Design studio with red correction pins on mockups. ### Claude Hard Parts It admitted inventing structures instead of reading the repo closely enough. One example was designing lifecycle stages as grouped tiles when the real app used a flatter grid of per-asset cards. Another was adding profit scatter quadrants that did not exist in the actual component. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Imagined fancy mockup dissolving against a simpler real component blueprint. ### Claude Hard Parts The user caught both. That is not a footnote. That is part of the system. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Human review stamp landing on the design table. ### Claude Hard Parts AI design work can be fast, but if it is not anchored to the real code and product state, it can become very polished fiction. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Beautiful fake UI poster floating over a dark code basement, tether snapped. ### Claude Hard Parts Claude also shipped one isolated analyzer chart improvement: a compact tooltip and right-edge labels using a custom SVG renderer with collision handling. It verified that with typecheck and build. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: A 1:1 wojak2-style mobile frame focused on a large analytics line chart on a monitor. Show a compact tooltip box beside a point, several right-edge line labels, and dashed leader lines avoiding collisions like traffic lanes. A tired builder/wojak can be small in foreground, but the chart improvement must be clearly visible. No readable tiny text, no panels, no desk-only scene. ### Claude Hard Parts But Claude could not visually verify the running app because the environment needed OAuth-related variables it did not have. So the honest version is: the code passed, but the eyes were not invited. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Build checkmark outside a locked preview window, pair of cartoon eyes waiting outside. ### Synthesis And Shipping This is why three AIs were useful, and also why three AIs were dangerous. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Three colored lanes feeding one product; lane dividers are clear but slightly fragile. ### Synthesis And Shipping Claude could help decide what good should look like. Manus could help prove what the live product and platform were actually doing. Codex could make the implementation hold together inside the codebase. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: A 1:1 wojak2-style mobile frame with three distinct work lanes feeding one product: left lane has a simple design/spec sketch icon, center lane has a live verification check screen, right lane has a code editor/build box. A tired wojak watches the three lanes connect into one finished dashboard. Use icons and abstract blocks, no readable dense text, no panels. ### Synthesis And Shipping But if the boundaries were blurry, the workflow created new failure modes. One model could design against an imagined structure. Another could verify live behavior without owning the code path. Another could make a valid repo change against an outdated design direction. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Three lanes crossing into a tangle around the product cube. ### Synthesis And Shipping All three outputs could be individually reasonable and still combine into a product that was wrong. That is the multi-AI lesson. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Three clean puzzle pieces forming a visibly wrong shape. ### Synthesis And Shipping Basically, teamwork. But with more JSON. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Teamwork poster half-covered by curly braces and config blocks. Minimal text: `{ }`. ### Synthesis And Shipping By the end of the pasted history, AdsMetri had become much broader than a budget manager: Meta OAuth, campaign sync, budget queues, account-time scheduling, dashboard metrics, expandable campaign structures, analyzer snapshots, lifecycle behavior, funnel views, multi-account scoping, activity scoping, fulfillment economics, BEROAS, daily Meta P and L, compact UI direction, brand cleanup, tests, legal pages, OAuth redirect fixes, and deployment support. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Finished product dashboard room with abstract modules orbiting it. No list text. ### Synthesis And Shipping That is not one feature. That is a small city of features pretending to be a dashboard. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Dashboard as a tiny city skyline of widgets. ### Synthesis And Shipping The hard parts were mostly between the features. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Dark gaps between bright feature blocks reveal cables, clocks, permissions, account tokens. ### Synthesis And Shipping The app had to stop mixing accounts. It had to stop assuming the wrong timezone. It had to stop treating a permission issue like a sync bug. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Three quick symbolic corrections in one scene: account folders separate, clock corrected, gate key accepted. ### Synthesis And Shipping It had to stop showing plausible charts without enough verification underneath. It had to stop letting design drift away from the real component structure. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Chart bolted to verification foundation; design mockup tethered to repo blueprint. ### Synthesis And Shipping The demo is fast. The product is where the receipts show up. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Flashy demo curtain opens to a receipt printer dumping proof tickets. ### Ending So the lesson from AdsMetri is not "three AIs built a product." That is too clean. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Clean marketing poster crumpled beside messy real build table. ### Ending The better lesson is: AdsMetri only became manageable when the work separated into three real lanes. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Three clear lanes into product dashboard, not panels, not role grid. ### Ending Claude for design and spec. Manus for foundation and live verification. Codex for repo-level implementation and debugging. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Three simple tools placed around product: sketch pen, verification key, debugger wrench. ### Ending Those roles did not remove coordination work. They made the coordination work visible. And once it was visible, it could be managed. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Invisible tangled wires become lit, labeled-by-color pathways under glass. No text labels. ### Ending That is less magical than the usual AI story. It is also more useful. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Sparkly magic wand set aside; practical toolbox opens. ### Ending AdsMetri is what AI-assisted product building actually looked like here: fast progress, false leads, permission problems, timezone traps, chart paranoia, design handoffs, tests, deployment chores, and a lot of state management. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Montage-style single room with artifacts on shelves: key gate, clock, chart shadow, handoff folder, test badge, deployment box. One focal shelf wall. ### Ending Not a clean montage. More like a real build. Except sometimes the debugging assistant opens a pull request, and everyone pretends that is normal now. Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: Wojak2 / wojak-gpt2 style: tired expressive wojak faces, flat shaded comic lighting, rough hand-drawn linework, muted colors, readable staging. One focal idea only. Do not create a multi-panel comic page, role grid, triptych, storyboard, or dense poster. Avoid dense readable text, speech bubbles, checklist boards, and tiny dashboard labels. Do not reuse the same desk setup unless this exact visual asks for it. Tired wojak at final dashboard while a tiny pull-request envelope slides under the door. ### Ending Model: gpt-image-2-text-to-image ## Scene prompt Flat-shaded cartoon illustration in the modern wojak-comic style, matching the provided wojak-gpt2 reference when one is supplied. Keep classic wojak meme-comic anatomy and emotional register: pale gray or muted skin, weary hollow eyes, rough hand-drawn facial linework, simple expressive faces, slouched doomer posture unless the prompt explicitly asks for a confident chad contrast. Avoid glossy anime, heroic superhero posters, polished corporate illustration, semi-realistic concept art, cinematic fantasy art, manga styling, and generic webcomic polish. Use muted colors, soft cel shading, and simple readable staging. Backgrounds may show desks, monitors, paperwork, and props, but they should stay secondary to the wojak figure and the beat. Composition is a single readable comic-style frame with one focal subject, not a comic page. Do not split the image into panels, grids, storyboards, or several mini-scenes unless the scene prompt explicitly requests a before/after or A/B comparison. Avoid dense readable text inside the generated image; screens, papers, and boards should use abstract marks unless exact on-screen text is explicitly declared. Visual digestibility guard: render ONE single-scene illustration with one clear focal idea. Do NOT create a multi-panel comic page, triptych, grid, storyboard, timeline strip, collage, or several mini-scenes in one image unless the scene prompt explicitly asks for a before/after or A/B comparison. Avoid text-heavy posters, dense checklists, whiteboards, document pages, speech bubbles, and dashboards full of tiny readable labels. Use zero extra legible text beyond the declared on_screen_text. Any screens, papers, sticky notes, code editors, charts, or dashboards should show simple abstract marks, blocks, icons, or unreadable texture unless declared text is the whole point of the shot. The viewer should understand the image in about two seconds while narration carries the details. Scene: A 1:1 wojak2-style mobile ending frame: the AdsMetri analytics dashboard is the clear focal point, shown as a dark compact UI with three colored lanes and simple charts; the tired builder is small or secondary, quietly watching the dashboard. No extra text, no comic panels, no dense labels. ## Transcript So I decided to build an analytics dashboard for my e-commerce store, called AdsMetri. AdsMetri started as a Meta ads budget manager. Which sounds simple. And that was the first warning sign. Because a budget manager is not just a button that says, "spend less money," although honestly, that would be a strong product category. The app had to connect to Meta, sync campaigns, queue budget changes, run those changes at the right time, show account-specific activity, and not quietly mix one ad account into another. Then the product grew. It became a dashboard for reading ad performance: lifecycle charts, funnels, ROAS, CPA, CPC, spend trends, BEROAS, and daily Meta P and L. In plain English: it stopped being only "change the budget" and became "tell me what is actually working." That is a much scarier sentence. Because once software starts telling you what is working, the real problem is not whether the chart renders. The real problem is whether the chart is telling the truth. That is why this build ended up spread across three AI harnesses: Claude, Manus, and Codex. Not because three sounds more impressive than one. Three is also how many browser tabs I open before forgetting why I am alive. The real reason was that AdsMetri was not one problem. It was three problems pretending to be one. One problem was product and interface clarity: what should this dashboard show, how should it feel, what should the user understand first, and what should not be crammed into the UI like a tax form with a growth mindset. That became Claude's strongest lane. One problem was live reality: OAuth, Meta API behavior, token permissions, deployment, database state, dummy data, tests, and whether the product actually worked against real accounts instead of only in a pleasant local fantasy. That became Manus's strongest lane. And one problem was the codebase itself: account state, analyzer snapshots, scheduling logic, pending changes, chart behavior, and the bugs where the UI looked fine while the underlying logic was holding a small meeting without you. That became Codex's strongest lane. The mistake was treating those tools like interchangeable assistants. The useful version was treating them like specialized harnesses with handoffs. Quick context: this devlog was made inside Spoolcast, my AI video pipeline. Spoolcast can start from source material like chats, transcripts, repo notes, research, or story ideas, then help turn that into a script, shot list, visuals, narration, render, and final files. I still direct it with feedback and approvals. Which, based on this episode, also includes telling the system to follow the protocol that it wrote down for itself. Software automation is powerful. It is also occasionally a clipboard with confidence. For this devlog, I am keeping the story source-backed. I am using the pasted Codex, Manus, and Claude perspectives from the chat. If something is uncertain, I will say it is uncertain instead of sanding it into a cleaner story. Because the whole point of this video is not "AI made a product." The point is what actually shipped, what broke, and what multi-AI work taught us after the novelty wore off. The first version of AdsMetri was operational. Connect a Meta ad account. Sync campaigns. Show performance metrics. Queue budget changes. Execute those changes later. Keep an activity log so the user can see what happened. That sounds like product plumbing, because it is. But plumbing is one of those things people only call boring when it works. Manus reported building much of that foundation: app scaffold, database tables, Meta OAuth, token management, campaign sync, budget queues, scheduler behavior, dummy data, legal pages, deployment setup, and a test suite reported as seventy-three tests. I am saying "reported" because this script is using the pasted summaries as source. I am not independently re-auditing Manus's entire history here. But the role is clear. Manus was closest to foundation plus live verification. It was the harness dealing with the product as something that had to connect to the outside world. The best example was campaign visibility. Some campaigns were not syncing. The exact count is inconsistent across the pasted notes, so I am not going to invent a clean number. The important part is the root cause: Meta API token permissions. The sync logic can look suspicious. The database can look suspicious. The frontend can look suspicious. And then the answer is: the token was not allowed to see the campaign. That is not a bug you solve by staring harder at a React component. That is why live verification mattered. The product had to be checked against what the real platform allowed, not just what the code hoped was true. Manus also hit timezone work. The scheduler originally had an 11 PM Bangkok assumption, but Meta Ads works on account calendar-day boundaries. Manus reported refactoring that to a midnight-to-12:30 window in the ad account timezone. A clock can be perfectly accurate and still be wrong for the product. Very inspiring. Very irritating. Manus also called out numeric precision from the Meta API. Tiny floating-point weirdness sounds harmless until it becomes a dashboard number and someone asks why finance looks haunted. So Manus's lane was not just "build features." It was: connect to reality, check reality, and stop reality from entering the product wearing a fake mustache. But live verification does not automatically make the app coherent. You can prove the data exists and still display it in the wrong account, the wrong chart, or the wrong moment. That was the Codex lane. Codex was the repo-level implementation and debugging harness. In plain English: it worked where the product's moving pieces had to agree inside the actual code. Codex's pasted summary covered dashboard, analyzer, settings, budget flows, campaign sync, account scoping, pending change display, recent activity scoping, multi-account state bugs, analyzer lifecycle behavior, chart interactions, right-edge labels, visibility controls, fulfillment economics, timezone fallback cleanup, account-aware scheduling, and compact-theme implementation after design direction was set. That list is long because state bugs are social creatures. They rarely arrive alone. The selected Meta account had to match the dashboard. The dashboard had to match pending changes. Pending changes had to match the scheduler. The analyzer had to match the selected account and available snapshots. Activity logs had to be scoped to the right account. A rendering bug is visible. A state bug hides behind a clean interface and waits for you to trust it. Wrong data does not always look broken. Sometimes it looks professional. It wears a nice font and ruins your week. Codex's hard bugs were often not where they first appeared. A chart issue could be a snapshot coverage issue. A dashboard issue could be an account-filtering issue. A sync issue could be cached database state disagreeing with live Meta state. A timezone issue could be a fallback assumption hiding in one part of the UI. This is where the product shifted from "does the screen work?" to "does the screen mean what it says?" That is a much stricter test. Codex also worked on fulfillment economics: fulfillment costs, BEROAS, daily Meta P and L, and break-even thinking. That changed the product question from "how much did we spend?" to "did this actually make money?" Once money math enters the dashboard, believable is not good enough. You need boringly correct. The best financial UI is a little boring. Exciting finance dashboards are how people end up in meetings with legal pads and no joy. But correct behavior still does not make a product feel usable. The user has to understand it. That was Claude's lane. Claude was strongest as the design and specification harness. Claude made Compact theme HTML mockups for the dashboard, analyzer campaigns, analyzer ad sets, analyzer ads, and settings. It iterated on tooltip shape, funnel layout, sidebar structure, P and L color rules, spend-by-campaign callouts, background colors, outlines, and negative space. In plain English: Claude helped decide what the product should look like and how the interface should communicate the work. Design is not decoration here. If AdsMetri is telling someone where ad performance is changing, the UI has to make comparison easy. If it is showing profit safety or fulfillment break-even, the UI has to make risk legible. If it is handling multiple accounts, the UI has to make the current account obvious. Nobody should have to perform archaeology in a dropdown. Claude also wrote a handoff package for Codex: design tokens, infrastructure notes, component guidance, preserve rules, and mockup copies. That turned design output into implementation instructions. When that worked, it was a handoff. When it did not, it was just two assistants confidently standing on different maps. Claude's self-report also included mistakes, which is useful because mistakes are where the workflow becomes real. It admitted inventing structures instead of reading the repo closely enough. One example was designing lifecycle stages as grouped tiles when the real app used a flatter grid of per-asset cards. Another was adding profit scatter quadrants that did not exist in the actual component. The user caught both. That is not a footnote. That is part of the system. AI design work can be fast, but if it is not anchored to the real code and product state, it can become very polished fiction. Claude also shipped one isolated analyzer chart improvement: a compact tooltip and right-edge labels using a custom SVG renderer with collision handling. It verified that with typecheck and build. But Claude could not visually verify the running app because the environment needed OAuth-related variables it did not have. So the honest version is: the code passed, but the eyes were not invited. This is why three AIs were useful, and also why three AIs were dangerous. Claude could help decide what good should look like. Manus could help prove what the live product and platform were actually doing. Codex could make the implementation hold together inside the codebase. But if the boundaries were blurry, the workflow created new failure modes. One model could design against an imagined structure. Another could verify live behavior without owning the code path. Another could make a valid repo change against an outdated design direction. All three outputs could be individually reasonable and still combine into a product that was wrong. That is the multi-AI lesson. Basically, teamwork. But with more JSON. By the end of the pasted history, AdsMetri had become much broader than a budget manager: Meta OAuth, campaign sync, budget queues, account-time scheduling, dashboard metrics, expandable campaign structures, analyzer snapshots, lifecycle behavior, funnel views, multi-account scoping, activity scoping, fulfillment economics, BEROAS, daily Meta P and L, compact UI direction, brand cleanup, tests, legal pages, OAuth redirect fixes, and deployment support. That is not one feature. That is a small city of features pretending to be a dashboard. The hard parts were mostly between the features. The app had to stop mixing accounts. It had to stop assuming the wrong timezone. It had to stop treating a permission issue like a sync bug. It had to stop showing plausible charts without enough verification underneath. It had to stop letting design drift away from the real component structure. The demo is fast. The product is where the receipts show up. So the lesson from AdsMetri is not "three AIs built a product." That is too clean. The better lesson is: AdsMetri only became manageable when the work separated into three real lanes. Claude for design and spec. Manus for foundation and live verification. Codex for repo-level implementation and debugging. Those roles did not remove coordination work. They made the coordination work visible. And once it was visible, it could be managed. That is less magical than the usual AI story. It is also more useful. AdsMetri is what AI-assisted product building actually looked like here: fast progress, false leads, permission problems, timezone traps, chart paranoia, design handoffs, tests, deployment chores, and a lot of state management. Not a clean montage. More like a real build. Except sometimes the debugging assistant opens a pull request, and everyone pretends that is normal now. ## Related tracker entries [three-ai AdsMetri build story — spoolcast dev-log #7]() Status and dates describe the published record, not a recent test of an external service. Missing details are not inferred. Public links do not imply an open-source licence or permission to reuse material. [Directory]()