diffusion-ui — live page mirror

About

This is an experimental Chrome extension that mirrors your active browser tab in real time and restyles the page from a prompt, using live image diffusion.

It grabs your active tab a couple of times a second, runs each frame through a realtime img2img model, and shows the result in a side panel next to the real page. Scroll or type on the page and the mirror keeps repainting it as Claymation, Lego, Van Gogh, Manga, a wooly Wool/Felt world, or any of 35+ other prompts you can pick or write yourself.

It's a bit laggy and expensive to run, so it's not meant for production. But it's a fun way to experiment with image diffusion models, and a glimpse of where this kind of thing might go.

Expect roughly 2 frames a second.

How it works

It runs on the FLUX.2 [klein] model on fal.ai.

- Capture: it screenshots your active tab with Chrome's tabs.captureVisibleTab, about twice a second.
- Diffuse: each frame goes to fal's realtime img2img model (via @fal-ai/client) with your prompt.
- Render: the result streams into a side panel next to the page, following whatever tab you're on.
Everything runs client-side with your own fal API key. There's no backend of mine in the loop. Each frame goes straight from your browser to your fal account.

Getting started
1. Make a fal account at https://fal.ai and add credits (generation is pay-as-you-go).
2. Create an API key at https://fal.ai/dashboard/keys.
3. Open the side panel, paste your key, pick a style, and watch your tab repaint.

Also bundled
A few self-contained playgrounds for poking at the same engine without capturing any tabs: a drawing pad, a small walkable 3D scene, and some mock pages. They diffuse only their own page.
Follow fal's terms of service and privacy policy.