Summary
Routes your talking-head and lip-sync jobs to the right RunComfy model without you memorizing four different CLIs. Feed it a portrait and audio file, it picks OmniHuman for clean UGC voiceovers. Hand it a script with no audio, it reaches for HappyHorse 1.0 with in-pass speech generation. Ask for cinematic composition with reference subjects, it queues Seedance v2 Pro. The skill knows when Wan 2-7's audio_url field matters (full scene control, open weights) and when it's overkill. Useful if you're building dubbed demos, virtual presenters, or character dialog and don't want to re-read four model cards every time your input format shifts.
Install
npx skills add https://github.com/doany-ai/skills --skill ai-avatar-video
