Music Video · публичный навык Open Skill
Music Video
**Music Video · Script to Song to Screen · ref2v Native · Mix-Locked** *The agent skill for creators who want a finished **~60-second music video**—not six orphan clips and a track sitting next to them* **Music Video** is a production-native Agent Skill for Digen. It orchestrates a complete **reference-to-video (ref2v)** pipeline for a one-episode MV: **Script → character / scene / prop refs → full song → timed chapter prompts → 6×10s silent renders → join, mute, song on top** No storyboard grids. No shot-by-shot still chains. No TTS. Six coherent 10-second pictures along the first minute of **one generated song**, then a single cut where **faces, wardrobe, rooms, and the track stay locked**. ## <span style="color:#4EC9B0">What you get</span> | For creators | For the production stack | | --- | --- | | **One-pass to first clip** — say the idea and the agent runs brief → refs → song without stopping | **Opinionated pipeline** — six-chapter spine, silent-picture lock, and mix order encoded in-skill | | **Song-first visuals** — pictures timed to the real track, not a fake 60-second stub | **ref2v-first** — no accidental i2v / storyboard / first-last-frame drift | | **Identity lock** — uploaded face binds before any image, song, or render; lead vocal matches the photo | **No song on render** — `audio_url` never hits ref2v; mouths stay closed; music arrives only at mix | | **First-clip gate** — you approve the opening 10s before the rest burns | **Join → mute → song** — six clips concatenated, picture silenced, original track laid on once | | **Series-ready** — next 60 seconds along the **same song**, same refs | **Progressive disclosure** — craft lives in `references/`, not bloated context | | **Plain creator voice** — short lines about people, place, and song; no pipeline jargon unless you ask | **Marketplace-ready** — clear triggers, defaults, and tool contracts | ## <span style="color:#4EC9B0">Pipeline</span> | Stage | Output | | --- | --- | | **Script** | 60s spine, six chapter cards, inventory of every on-screen person / place / signature prop | | **Reference pass** | One still per CHAR / LOC / PROP (`t2i` / `i2i`). Characters and props **9:16**; location plates **16:9 landscape** | | **Song + timing** | One full generated track, then lyric/beat windows for 0–10 … 50–60 | | **Chapter prompts** | Six silent ref2v prompts, continuity-checked as a set (no face jumps, no teleport, no singing) | | **First render** | **One** 10s ref2v clip · **9:16** · chapter-1 refs as `input_urls` · **stop for confirm** | | **Remaining + mix** | Clips 2–6, then **join all picture → mute → original song** · one ~60s `final_videos` cut | ## <span style="color:#4EC9B0">Production stack</span> | Parameter | Default | Notes | | --- | --- | --- | | **Episode** | **~60 seconds** | Six 10s videos + one mix. The song itself is a full track, not compressed to 60s | | **Engine** | `ref2v` · `reference_to_video` | Multi-ref identity lock. Forbidden: `i2v` / `ai2v` / `t2v` / `fl2v` / `director` | | **Clip duration** | **10 seconds** | Always set. Leftover 2–9s allowed; under 2s dropped | | **Video orientation** | Portrait · **9:16** | Landscape only if you ask | | **Location stills** | Landscape · **16:9** | Empty-set plates, not the video frame | | **Picture audio** | **Silent** | No singing, no speech, no mimed lyrics. Song is mix-only | | **Mix order** | Join → mute → song | Never put the track on the join call. Intermediate joins stay off-canvas | | **TTS** | **Off** | The generated song is the soundtrack | | **Confirm** | After clip 1 | Remaining five + mix run only after you say the opening looks good | | **Series** | Next **60s** of the same song | Reuse character / scene / song URIs; do not overwrite prior finals | ## <span style="color:#4EC9B0">Modes</span> | Mode | You bring | Agent delivers | | --- | --- | --- | | **Full auto-run** | A one-liner, lyrics, or mood | Bible → refs → song → first 10s clip, then the rest after confirm | | **Script-first** | Plot, cast, or lyric film | Refs → song → timed chapters → render | | **Reference-driven** | Face / scene / prop photos + brief | Bind uploads → identity lock → song → ref2v | | **Single-stage edit** | "Redo this clip" / "Swap her look" / "Change the chorus picture" | That stage only, then remake the mix if a cut already exists | | **Same-thread sequel** | "Next 60 seconds" / "Episode 2" | Last window → new six chapters · same song · URI reuse | | **Single clip** | "Just 10 seconds" | One silent ref2v, no six-clip mix | ## <span style="color:#4EC9B0">Example prompts</span> **Full auto-run** > Make an MV: a girl chases sunrise through a neon city at dawn. Portrait, ~60 seconds, female vocal. **Reference-driven** > This photo is the singer. Rain-slick alley, vintage camera as the prop. Six chapters along one full song, then one cut. **Lyrics-first** > Use these lyrics. Don't invent a second story. 9:16 music video, first minute visualized. **Series** > Next 60 seconds of the same song. Keep her face and the rooftop. ## <span style="color:#4EC9B0">When to use</span> - Vertical music videos and performer-led lyric films - One-line ideas, pasted lyrics, or character/scene reference images - Identity-locked leads — upload a face; sheet, vocal, and every clip must match - First-minute episodes along a full generated song, then the next 60 seconds in-thread - Surgical edits to one chapter, one still, or the final cut > <span style="color:#87a9ff">Not the right tool when</span> you need still posters only, storyboard grids for review, talking-head product sells, drama seeding ads, or a pipeline that lip-syncs mouths to the song inside each render. ## <span style="color:#4EC9B0">Invoke</span> ``` /music-video ``` Or: *"Turn this into a 60-second music video with consistent characters."* Silent picture · One song · One coherent cut.
18861 39 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/dev/admin/100/2026-09-14/vq3w9kyt.png