magic-hour-music-video
Installation
SKILL.md
Magic Hour music video
Start with the user's track, selected phrase and desired feeling. Accept “make a video for this song”; choose a coherent direction and technical settings within the existing budget. Reuse supplied audio and references. Do not replace the song with generated narration or make a singer when the brief asks for an instrumental visualizer.
Choose the actual route
| Desired result | Route | Important distinction |
|---|---|---|
| Complete song-driven scene assembly with optional singing shots | Web Music Video Generator | Its multi-scene/lyric/reference workflow is not exposed as an equivalent endpoint in the current public MCP/OpenAPI |
| A short visual guided by supplied audio and an optional starting image | audio_to_video_create_video |
Audio interval and visual prompt; no exposed beat map, lyric timestamps or lip-sync toggle |
| Evolving stylized visuals with an audio-sync camera effect | animation_create_video |
Art style, camera effect, transition speed and output fps; not identity-locked performance |
| A portrait performs the supplied audio | Talking Photo | Review singing against the actual sung audio; spoken examples do not validate it |
| Existing performer matches new audio | Lip Sync | Preserve source action; inspect mouth alignment and the full final phrase |
| Exact cuts at chosen musical moments | Accepted clips plus an available compositor | Cut to the supplied track; do not claim a generated clip has deterministic beat alignment |
For the web Music Video Generator, prepare the selected audio, reference images, aspect, creative direction and whether visible singing is wanted. If the agent cannot operate that surface, provide those ready-to-use inputs and the direct link, then continue from the downloaded result. Do not fabricate an MCP endpoint or silently call a different route equivalent.