Downloads the model once (kept in the browser's cache for next time) and runs it on your GPU through WebGPU. Prompts come from a precomputed gallery: the 11 GB text encoder does not fit in a browser.
This page runs the model on your own GPU through the fastwan engine. Start it on this computer, allowing this site:
fastwan
Chrome and Edge may ask to let this site reach devices on your local network: allow it. Safari blocks pages on https from reaching 127.0.0.1, so use Chrome, Edge or Firefox.
This engine runs without the text encoder, so it can only make its precomputed prompts: pick one above. Add more with fastwan encode.
fastwan encode
3 denoising steps, 16 fps. Ctrl/⌘ + Enter generates. Clips stay on your machine: the engine runs on your GPU, and this page only talks to it.Clips stay on your machine: the model runs on your GPU inside this tab.
The engine is running the tiny synthetic model (--synthetic), so clips are coloured noise. Start it with --model-dir for real video.
--synthetic
--model-dir
Finished clips appear here. The engine keeps the last eight, so you can reopen them.