Poster Studio

Your team's AI creative agency, following the traditional agency flow:

1. Client brands. Add a client: details, about text, website, Facebook and Instagram (URL and/or screenshots), brochures (PDF), old ads, logo. The planning team reads everything and runs a brand & competitor study with live web research. Brands are stored, so the team can keep many clients.
2. Create poster. Pick a brand, write the brief, choose the creative depth (1 = straight and clear … 10 = double-meaning, award-level concept), choose one or more positioning styles (Fear, Lust, Greed, Pride, Belonging, Nostalgia…), then assemble the creative team: Copywriter, Creative/Art Director, Visualizer and Photographer. Each has a detailed profile (experience, approach, signature, influences, what they avoid, a sample) that is given to the AI in full. Every run creates two creatives: 1 · graphic, 2 · realistic photo (set in CREATIVE_STYLES in lib/craft.ts).
3. Two ideas. The team develops 2 different ideas (insight, headline, subhead, body copy, CTA, tagline, visual concept, double meaning), then an Executive Creative Director reviews them, sharpens the copy to the chosen depth and notes what changed. Everything is editable.
4. Two creative briefs. The art director lays out each ad, then the Creative Director writes a full creative brief per ad in the house format (THE BRIEF, strategic insight, central creative idea with territories and the chosen direction, creative philosophy, art direction, copywriting, channel execution, creative test, final deliverable). That brief is the final image prompt. Format and channel (e.g. LinkedIn, 4:5 at 1080 × 1350) are part of it. The alternative mode "Picture only + studio typesetting" asks for a text-free picture and typesets the copy in the app.
5. Posters + post copy. Each creative comes with a ready-to-publish social caption (post copy). Download all saves the 2 finished posters and 2 .txt caption files. Images are generated with OpenAI's GPT Image models (default gpt-image-1-mini), capped at $0.04 per image. By default the copy and logo are typeset by the app with real fonts (crisp, editable, no AI lettering). Adjust position, colour and font, then download.

Experts and positioning styles are plain lists in lib/experts.ts and lib/appeals.ts.

Add your OpenAI API key

1. Go to https://platform.openai.com/api-keys, click Create new secret key, name it "Poster Studio", copy it (starts with sk-). You only see it once.
2. Add credit at https://platform.openai.com/settings/organization/billing (e.g. $10) and, to be safe, set a monthly budget limit under Settings → Limits.
3. Image models may need a verified organization: Settings → Organization → Verify organization.
4. In Vercel: open the project → Settings → Environment Variables → add OPENAI_API_KEY = your key (all environments) → Save.
5. Deployments → latest deployment → ⋯ → Redeploy. Environment changes only apply after a redeploy.

Never put the key in the code or share it in chat; it lives only in Vercel.

Image cost limit

Each image is limited to $0.04 by default (MAX_IMAGE_COST_USD). The server enforces it: if a quality would cost more, it automatically uses the best quality that fits. The poster screen shows the model, quality and estimated cost before you shoot.

Approximate prices with the default gpt-image-1-mini:

Format	Low	Medium	High
Square 1024×1024	$0.005	$0.011	$0.036 ✅
Portrait / landscape	$0.006	$0.015 ✅	$0.052 ❌ over budget

✅ = what "Best within budget" picks. So square posters come out in high quality and portrait/landscape in medium. For a different trade-off, change MAX_IMAGE_COST_USD (e.g. 0.06 lets portrait use high quality) or OPENAI_IMAGE_MODEL. Prices are third-party estimates of OpenAI's rates; check your OpenAI usage page after the first few posters. Text steps (study, ideas, prompts) cost roughly $0.01–0.05 per run.

Deploy on Vercel

1. Import this repo at https://vercel.com/new.
2. Environment variables:
   • OPENAI_API_KEY: your OpenAI key (required)
   • APP_PASSWORD: shared team password (recommended)
3. Share brands with the team (optional): by default brands are saved in each person's browser. To share one brand list with everyone, go to your Vercel project → Storage → Create / Connect Database → Redis (or Upstash for Redis; the free tier is enough) and connect it to the project. Vercel adds REDIS_URL (or KV_REST_API_URL / KV_REST_API_TOKEN for Upstash) automatically. Redeploy. Brands already saved in a browser can then be moved to the team with the Move them to the team button on the Brands page.
4. Deploy.

Notes

• The brand study uses OpenAI's web search tool. If your model/account doesn't support it, the app falls back to the model's own knowledge plus your material, and says so on the study.
• The study and high-quality images can take 1–3 minutes. If you see timeouts on the Vercel Hobby plan, make sure Fluid compute is enabled (Project → Settings → Functions), or use Medium quality.
• Facebook and Instagram usually block automated reading, so screenshots of the page/profile and a few posts give the best results.
• PDFs up to 3 MB each (serverless upload limit).

Run locally

cp .env.example .env.local   # fill in OPENAI_API_KEY
npm install
npm run dev                  # http://localhost:3000

Without Redis variables, brands are saved in the browser (IndexedDB).

Configuration

Variable	Required	Default	Purpose
OPENAI_API_KEY	yes	–	OpenAI API key
APP_PASSWORD	recommended	none	Shared team password
REDIS_URL, or KV_REST_API_URL + KV_REST_API_TOKEN	to share brands	–	Redis or Upstash Redis (added by Vercel); otherwise brands stay in each browser
OPENAI_TEXT_MODEL	no	gpt-4.1	Model for study, ideas and prompts
OPENAI_IMAGE_MODEL	no	gpt-image-1-mini	Image model (gpt-image-1.5, … also priced in lib/imageCost.ts)
MAX_IMAGE_COST_USD	no	0.04	Maximum estimated cost per image
OPENAI_BASE_URL	no	https://api.openai.com/v1	API base URL