Videogen

Turn a topic or source into a validated, editable Remotion project, then render and inspect the final MP4. Prefer autonomous execution when the user asks for a finished video. Use review checkpoints only when the user asks to approve the script, storyboard, or assets first.

Core constraints

1. Treat this as a two dimensional layered video generator. Do not promise lip sync, complex character acting, three dimensional scenes, or cinematic generated footage unless another suitable video model is explicitly available and requested.
2. Infer missing creative defaults instead of asking routine questions. Default to 30 seconds, 1080 by 1920, 30 fps, simplified Chinese, concise narration, and the paper-collage style. Treat any duration named by the user as a hard target.
3. Verify factual claims before rendering factual content. Keep a source list in the project notes when browsing was needed.
4. Never clone a real person's voice without clear authorization. The bundled default reference is authorized for distribution with this skill and for use as its default narration reference. Also support a custom reference only when the user supplies it or confirms authorization.
5. Check licenses for fonts, music, generated assets, reference images, and voice models before commercial use.
6. Preserve user supplied wording when the user provides an approved script. Only split it into scenes and captions unless editing was requested.

Load references selectively

1. Read references/video-schema.md before creating or editing video.json.
2. Read references/style-presets.md when selecting visual direction or writing image prompts.
3. Read references/quality-gates.md before rendering and final delivery.
4. For Remotion implementation details, use the installed remotion-best-practices skill when available. All animation must be frame driven. Do not use CSS animations or transitions.