facefork

Upload a photo, ask “what if…”, every answer becomes a branch.

An infinite canvas for playing with your own face: each prompt (“what if I had a beard?”) grows a new node under the photo you asked it of. Select two photos — even of different people — and combine them. Star the keepers, compare any node with its parent, export the tree, or turn it into something postable (before/after, polaroid, magazine cover, a looping clip, …).

Everything runs in your browser on your own OpenRouter key. Photos stay in the browser until you ask for an edit; then the prompt and the reference photos go directly to OpenRouter and the model provider you chose. facefork.com is a static site — its server never receives your photos or your key, and a Content-Security-Policy header forbids the page from connecting anywhere but OpenRouter. No analytics, no backend.

Run it

npm install
npm run dev        # http://localhost:5173

Paste an OpenRouter key when the app asks (or under ⚙). Pick any image-editing model OpenRouter offers; the default is Nano Banana 2 at about $0.07 per image.

How it is built

• React + React Flow for the canvas, dagre for layout, Vite to build.
• Generation through OpenRouter’s image API, called directly from the browser.
• Your tree lives in IndexedDB; ⇩ Export writes a .facefork file (zip of JPEGs + graph).
• Share formats are rendered on a canvas in the browser; clips are encoded with WebCodecs.
• Fully static: npm run build → dist/, deployable to any static host (it runs on Cloudflare Workers static assets; see wrangler.jsonc).

The bundled demo tree was generated once with make demo (needs OPEN_ROUTER_KEY in .env); “Sam” is an AI-generated portrait, the other sitter is Leonardo’s.

License

0BSD — do whatever you want with it; no attribution required. The demo portrait of “Sam” is AI-generated; the Mona Lisa is public domain.