AI Chinese Whispers

Upload a photo. It gets described by a vision model, redrawn from that description alone by an image model, then described again — ten times over. Each generation only ever sees the words, never the picture. Watch the drift.

1 · Connection

Calls go straight from this page to openrouter.ai — no server, no proxy. Your key never leaves the browser except in the Authorization header.

2 · Models

Each cycle = 1 caption + 1 generation.
Per cycle—
Estimated run—
Actually spent$0.0000
Estimates use OpenRouter list prices; Actually spent is the real billed usage.cost returned per call.

3 · Mode

A mode is just a different prompt for the captioner — and the caption is the next image's prompt, so that one swap steers the whole chain. It is re-applied every cycle, so the effect compounds: ten cycles of Funnier have been pushed toward comedy ten times.
Edit the prompt for this run
Edits are picked up when you press Run and last only while this tab is open — switching modes and back keeps them, reloading does not. Copy anything you want to keep.

4 · Source image

Drop a photo here or click to browse
JPEG / PNG / WebP / HEIC — resized in-browser before upload
—