Short Film & Music Video Maker · publiczny skill Open Skill
Short Film & Music Video Maker
Short Film & Music Video Maker to publiczny Open Skill w Digen. Otwórz go, by zobaczyć, co tworzy ten workflow.
Other
2 0 1
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3873649/2026-10-15/wys32fgc.jpg
# Short Film & Music Video Maker Turns an idea, script, lyrics, or uploaded audio/reference into a finished short film or music video, shot by shot, and stitches it into a final cut. ## When to use - User wants a **short film / short drama** (a story with scenes, characters, dialogue) from an idea or script. - User wants a **music video (MV)** from lyrics, an uploaded song, or an idea. - User wants to **revise one stage** (script, storyboard, a shot, the final cut) or **continue the next segment** of a project already in progress. ## Reference files - `references/short-film-workflow.md`: full step-by-step pipeline for short films/dramas. Read it when the request is a story-driven short film. - `references/music-video-workflow.md`: full step-by-step pipeline for music videos. Read it when the request is an MV (from lyrics/audio/idea). - Read only the file that matches the current request; don't read both at once. ## Memory (preferences) - Before starting, call `memory_query` for the user's saved preferences (visual style, orientation, video length, subtitle language, brand/channel name, voice preference). - After finishing a project, call `memory_store` to remember durable preferences the user stated (e.g. "always 9:16", "prefer cinematic warm grading", "subtitles in Persian"), with `kind` = `preference` or `project`. Don't store one-off task details. - If the user says "forget/change my preference", use `memory_query` to find the `id`, then `memory_forget`. ## Style detection (automatic) - If the user uploads a reference video or image (a mood board, an existing clip, a film still, a music video to imitate), detect its visual style before generating anything. - For a reference video: call `recognize_video(video_url=<asset URL>, prompt="Describe the visual style: color grading, lighting, camera work, mood, genre, and any recurring visual motifs.")`. - For a reference image: call `recognize_image(image_url=<asset URL>, prompt="Describe the visual style: color grading, lighting, composition, mood, genre, and any recurring visual motifs.")`. - Carry the extracted style description into every `generate_image` / `generate_video` prompt so all shots stay visually consistent. - If the user describes a style in words (e.g. "cinematic warm grading", "neon cyberpunk"), use that directly as the style guide instead of analyzing a reference. ## General steps 1. First use `write_todos` to list the pipeline steps, and update the whole list as you progress. 2. Read the matching reference file (`references/short-film-workflow.md` or `references/music-video-workflow.md`) and follow it. 3. Ask the user only for what's genuinely missing (idea/script/lyrics/audio, orientation, length). Default orientation to `portrait` (9:16) unless the user asks for landscape. 4. On each asset-producing step, pass the fixed `name` / `entity_type` values written in the reference file (they differ per stage so products don't mix). 5. When a stage completes, call `set_guidance` to suggest next actions (e.g. "revise the script", "regenerate this shot", "continue the next segment"). 6. Report failures honestly, never fabricate URLs, and politely decline requests for prohibited content (video generation does not support NSFW/adult content on any channel). ## Notes - Intermediate step-by-step products (storyboard images, per-shot videos, drafts) will be visible on the canvas alongside the final cut — that's expected. - Video generation does not support NSFW/adult content on any channel; decline such requests for video. - Never hard-code physical model names; always use the channel IDs (`t2i`, `i2v`, `ref2v`, `ai2v`, `t2v`, `music`, `tts.design`, `tts.clone`, `asr`, `clip-mixer`).