Movie Maker · публичный навык Open Skill
Movie Maker
Movie Maker — публичный Open Skill на Digen. Откройте, чтобы увидеть, что создаёт этот процесс.
Film
8 2 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/847169/2026-10-08/pk9jpamx.png
# Movie Maker - Director-Grade Script to Screen (ref2v Native) Production-native pipeline that turns an idea or script into ONE coherent 10-second cinematic clip - not a pile of disconnected frames. Characters, locations, and props stay locked across motion and cuts via reference-to-video (ref2v). Pipeline: Script -> world & shot plan -> reference assets -> six-part ref2v prompt -> 10s render -> music & mix ## When to use - User wants a finished 10s cinematic clip from an idea, one-liner, or script. - User wants to advance or change one step: rewrite the script, adjust the shot plan, restyle a reference asset, re-render the clip, or swap the music. ## Reference files - `references/six-part-prompt.md`: The six-part ref2v prompt builder (subject/action/location/camera/lighting/style) with examples. Read before building any render prompt. - `references/shot-plan.md`: World & shot plan template (characters, locations, props, single-shot plan). Read when planning a new project. ## Steps 1. **Clarify the idea.** If the user gives a one-liner, expand it into a short 10s script (setup -> beat -> payoff). If they give a script, use it. Confirm the single moment/action the clip should capture and the orientation (default **landscape 16:9**; choose **portrait 9:16** if the user wants short-form/vertical). 2. **Write the script** as a document: `create_document(content=<script>, name=<project title>, entity_type='scripts')`. 3. **World & shot plan.** Read `references/shot-plan.md`. Define the characters, locations, props, and ONE coherent shot plan (camera move, action, timing) for the 10s clip. 4. **Generate reference assets** (use the confirmed orientation) with `generate_image`: - Character identity sheet: `model='t2i'`, `orientation=<orientation>`, `entity_type='characters'`, `name=<character name>`. - Location: `entity_type='locations'`, `name=<location name>`. - Props (as needed): `entity_type='props'`, `name=<prop name>`. - Keep each asset consistent in style so ref2v can lock identity. 5. **Build the six-part ref2v prompt.** Read `references/six-part-prompt.md` and assemble the prompt from the six locked sections: 1. Subject/identity 2. Action/motion 3. Location/setting 4. Camera & lens 5. Lighting/mood 6. Style/grade. 6. **Render the 10s clip** with `generate_video`: `model='ref2v'`, `input_urls=<reference asset ids>`, `orientation=<orientation>`, `duration=10`, `resolution='720P'`, `entity_type='final_videos'`, `name=<clip title>`. 7. **QC the render.** Optionally `recognize_video(video_url=<clip>, prompt=<check identity/action/continuity>)`. If identity or action drifted, regenerate with tighter references or a revised prompt. 8. **Generate music** with `generate_music`: `prompt=<mood/style matching the clip>`, `duration=10`, `entity_type='music'`, `name=<track title>`. 9. **Mix music into the clip** with `mix_clips`: one clip, `video=<rendered clip>`, `audio=<music>`, `orientation=<orientation>`, `entity_type='final_videos'`, `name=<final title>`. 10. **Deliver.** Summarize the final clip in natural language (no URL dumps). Call `set_guidance` to suggest next actions (restyle a character, re-render, swap music, new clip). ## Notes - Use `write_todos` at the start to list the steps and update as you progress. - Default orientation is **landscape 16:9**; choose **portrait 9:16** when the user wants short-form/vertical. Keep the same orientation across reference assets, render, and mix. - Video generation does not support NSFW/adult content on any channel - decline such requests politely. - Intermediate assets (reference images, script, music) appear on the canvas alongside the final clip; that's expected. - Report failures honestly; never fabricate URLs. If a render drifts, say so and re-render.