Music Video Source Sync and Performance Control · publiczny skill Open Skill
Music Video Source Sync and Performance Control
Music Video Source Sync and Performance Control to publiczny Open Skill w Digen. Otwórz go, by zobaczyć, co tworzy ten workflow.
Music
16 3 7
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/1856918/2026-08-20/8vspvr38.png
# Music Video Source Sync and Performance Control ## When to use - The user wants a music video / MV created from uploaded source files (audio, images, videos, reference files) and wants it to strictly follow that material. - The user emphasizes that uploaded audio is the official soundtrack, or that uploaded images/videos are identity/style references that must be preserved. - The user wants on-screen performers to lip-sync lyrics, move to the beat, and express emotions matching the song. - The user wants to check or redo any part of such a music video (a clip, the sync, the performance, the final mix). ## Core principles (mandatory) 1. **Every uploaded file is mandatory creative input.** Do not ignore, replace, reinterpret, or override any uploaded audio, image, video, or reference file unless the user explicitly asks. Give each file a real role in the final result. 2. **Uploaded audio is the sole soundtrack.** Never replace it with AI-generated music, never change its genre, never invent a different track. The video must follow the song's real rhythm, tempo, structure, energy, mood, transitions, beat changes, and emotional tone. 3. **Uploaded images/videos are visual truth.** Preserve recognizable identity, look, style, clothing, objects, setting, and creative direction. Do not substitute random stock visuals, invented characters, or mismatched styles. 4. **Performance must match the music.** Lip sync, body language, gestures, expressions, head movement, dancing, walking, crowd reaction, and camera timing all follow the beat, flow, and emotion of the song. If the music is hip hop, the visual style/pacing/movement/energy must feel like hip hop — never turn it into house, EDM, pop, rock, or any unrelated genre unless the user explicitly requests that change. 5. **Facial expressions follow the music's emotion.** When mood changes, expression, body language, and performance intensity change with it. Keep the performance believable, rhythmically precise, and emotionally aligned. ## Steps ### 1. Inventory and analyze every uploaded source file - List all uploaded files in this session. For each one, determine its intended role: - **Audio file** → the official soundtrack. Call `transcribe_audio(audio_url=<uploaded audio>, cache=true)` to get lyrics, language, and timestamped segments (SRT) so you know structure and exact line timing. - **Image files** → identity/style references (characters, wardrobe, setting, props, look). Call `recognize_image(image_url=<each image>)` to extract the details that must be preserved (face, hairstyle, outfit, colors, setting, objects, mood). - **Video files** → visual/motion references (performance style, choreography, cinematography, energy). Call `recognize_video(video_url=<each video>)` to understand the performance and look to emulate. - If no audio file was uploaded, ask the user to upload the exact track before generating — do not invent or generate a soundtrack. - If a file has no meaningful role, ask the user how to use it rather than silently dropping it. ### 2. Build the music-synced shot plan - From the transcription and timestamps, map the song's real structure: intro / verse / pre-chorus / chorus / bridge / breakdown / outro, with start times and energy/mood for each section. Note beat changes and tempo shifts. - Assign each song section one or more shots. For each shot define: start time, duration, visual content, which character appears, what they are doing, the emotion, camera movement, and how movement/editing lands on the beat. - The shot order and pacing must follow the song structure — do not reorder or invent sections. ### 3. Prepare key visual frames from the uploaded references (only if needed) - If a shot needs a concrete starting frame, use the uploaded images directly as input to image-to-image: `generate_image(model='i2i', input_urls=[<uploaded reference images>], prompt=<shot description preserving identity>, orientation=<as needed>, resolution='1080P', name='