ai-avatar-video

Use the OmniHuman route: run `runcomfy run bytedance/omnihuman/api` with an `image_url` for the portrait and an `audio_url` for the voiceover. The model generates a video where the subject speaks or gestures naturally to the audio, with no prompt required.

OmniHuman is the default for portrait plus audio file jobs. Use Wan 2-7 with `audio_url` for full scene control, Wan 2-2 Animate for stylized characters, HappyHorse 1.0 when you only have a written script, and Seedance v2 Pro for cinematic multi-reference shots.

Yes, HappyHorse 1.0 generates speech in-pass from the prompt. Quote the spoken line exactly with `says clearly: "..."` inside the prompt, and optionally describe the audio tone separately, such as a friendly tone with a neutral accent.

Yes, you can reuse the same portrait image with a different audio file per language to produce dubbed clips with the same identity. Clean voiceover audio without a music bed gives the best mouth sync quality.

HappyHorse regenerates audio on each call, so it cannot lock lip-sync to a specific MP3 file. For precise sync to a pre-recorded track, use OmniHuman or Wan 2-7 with the `audio_url` field instead.

Exit code 77 means you are not signed in or the token was rejected. Run `runcomfy login` to authenticate, or set the RUNCOMFY_TOKEN environment variable in CI and container environments.