video-prompting
Video prompting
Turn the user's idea into a clear, model-native video prompt. This skill is for prompt writing and prompt revision across Codex, Claude Code, and other compatible agents. It has no preferred generation platform and does not depend on Higgsfield.
Route by model and task
Identify the exact model and task before drafting:
- For Wan 3.0, read references/wan-3.md.
- For Seedance 2.0, read references/seedance-2.md.
- For Seedance 2.5, read references/seedance-2-5.md.
Do not transfer reference tokens, shot formatting, parameter names, or assumed limits between these models. If the user says only “Wan” or “Seedance,” use the version they name or ask which one they have selected. If they do not know, state a reasonable version assumption and still give them a useful draft.
Determine only the choices that materially change the prompt: text-to-video, image-to-video, reference-to-video, editing, or extension; target duration and aspect ratio; the role of each supplied reference; action and camera movement; and whether dialogue, sound effects, or music are wanted. Ask about missing essentials briefly. Otherwise make minimal assumptions and label them.