Script-to-Video Drama Maker · Open Skill 公開スキル
Script-to-Video Drama Maker
Also use to redo or tweak one stage of an ongoing script-to-video project
Story
3 0 3
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4224455/2026-10-17/en84ek6c.png
--- allowed-tools: - generate_image - generate_video - generate_tts - generate_music - mix_clips - probe_media - create_document - update_document - get_document - transcribe_audio --- # Script-to-Video Drama Maker ## When to use - The user provides a text script (short drama, AI drama, horror, thriller, romance, etc.) and wants it turned into a full-length video. - The user emphasizes that the video length must match the script (e.g. "5-minute script → ~300 seconds"), not a fixed short clip. - The user wants facial expressions to follow the script's emotional beats, character-consistent faces, character voices, background music, and a final stitched cut. ## Reference files - `references/pipeline.md`: the full per-scene pipeline, timing math, and `entity_type` conventions. Read it before starting a full build. ## Steps 1. Get the script. If the user pasted text, use it directly. If they uploaded a document, call `get_document(asset_id=...)`. If they uploaded a video/audio of a script, use `transcribe_audio` (speech) and/or `recognize_video` (visuals) to extract it. 2. Read `references/pipeline.md` for the full workflow before generating anything. 3. Use `write_todos` to list the stages (script → scenes → timing → characters/expressions → backgrounds → clips → voiceover → music → stitch → verify) and update it as you progress. 4. Break the script into scenes; for each scene record the setting, characters present, dialogue lines (speaker + exact text), and the dominant emotion/expression. 5. Compute per-scene target durations so the total matches the script length (see the timing math in `references/pipeline.md`). Never cram the whole script into one short clip. 6. Generate character sheets and per-scene expression images, then scene backgrounds. 7. Generate per-scene video clips with durations matching the computed timing. 8. Generate voiceover per dialogue line and one background music track. 9. Stitch all clips with `mix_clips`, adding voiceover, music, and burned-in subtitles. 10. Verify the total duration with `probe_media` and report the result to the user. ## Notes - The whole point is **duration accuracy**: a 5-minute script must produce ~300 seconds of video. Compute timing from dialogue word count + action beats, split long scenes into multiple clips (each `generate_video` clip is short, typically ≤ ~10s), and stitch them. - Facial expressions must follow the script: generate a per-scene expression image from the character sheet (via `i2i`) and use it as the start frame so the face animates with the right emotion. - Use `orientation='portrait'` (9:16) for short-drama / vertical platforms; use `landscape` (16:9) if the user asks for cinematic widescreen. - Video generation does not support NSFW/adult content on any channel — politely decline such requests. - Report failures honestly and never fabricate URLs. Intermediate stage products will appear on the canvas; that's expected. - When a stage completes, call `set_guidance` to suggest next actions (e.g. "redo the horror scene's expression", "make the total longer", "add subtitles").