AI Art Generator

Open-source character video

LiveAvatar Character Animator

Turn one portrait and a speech track into an expressive talking-character video. The adapter runs Alibaba Quark's LiveAvatar v1.1 as a reproducible Cog on an app.nz scale-to-zero H100.

Launch on app.nz Source and self-hosting

Illustrated dwarven blacksmith example character — Pinned upstream test character

Inputs

Portrait or character image, WAV/MP3 audio up to five minutes, and optional scene guidance.

Output

Audio-muxed MP4 with preview, standard, and high pixel-area presets.

Runtime

H100 80 GB, four-step DMD inference, FP8 enabled, and automatic scale-to-zero.

License

Apache-2.0 adapter and upstream LiveAvatar code; pinned model provenance is documented in the repository.

Run it

Open Cog Studio and deploy the LiveAvatar template.
Upload a reference image and driving audio.
Start with preview quality, one clip, four steps, and a fixed seed.
Download the MP4; the H100 is released after the configured idle window.

The first request is a cold start and downloads the model. Keep one warm endpoint only for sustained interactive traffic.