What problem does it solve? Choosing the wrong image-to-video model wastes generation credits and iterations. This Skill classifies the user's intent — portrait animation, custom-voiceover lip-sync, or multi-modal composition — and routes the request to the matching model in the RunComfy catalog with the correct schema and prompting patterns. ## Core Features & Use Cases - Intent-based model routing: Selects HappyHorse 1.0 I2V for general animation with native audio, Wan 2.7 with audio_url for lip-sync to a custom voiceover, or Seedance 2.0 Pro for multi-modal input combining image, reference video, and reference audio. - Documented schemas and prompting patterns: Each route includes required fields, defaults, constraints (duration, resolution, aspect ratio, file size), and prompt-writing guidance. - CLI execution: Invokes runcomfy run <vendor>/<model> with a JSON input body and downloads results to a local output directory. - Use Case: A marketer needs the same spokesperson image lip-synced to voiceovers in three languages. The Skill routes to Wan 2.7, reuses one prompt with a locked seed, and swaps audio_url per language. ## Quick Start Ask the agent to animate an image by providing a source image URL and describing the motion, for example: animate this portrait photo with a gentle camera orbit and soft lighting using the image-to-video skill.
