Routes audio plus image across five RunComfy models to make talking-head videos. The real value is classification: it picks OmniHuman for clean portrait-plus-voiceover jobs, HappyHorse when you only have a script and need in-pass audio generation, Seedance v2 Pro for cinematic multi-modal composition, and Wan 2-7 or 2-2 for scene control or stylized characters. Each route ships the documented prompting pattern and the exact runcomfy CLI invoke. Use this when you want lip-sync without manually comparing model cards, or when you're building UGC tools, dubbed demos, or virtual presenter pipelines and need the agent to handle model selection. The alternatives it targets (HeyGen, Synthesia) charge per seat; this runs on your own RunComfy credits.
npx skills add https://github.com/agentspace-so/runcomfy-agent-skills --skill ai-avatar-video