scenario-labs/skills

scenario-seedance-music-video

Use when turning a song, track, or audio master into a finished music video with Scenario and Seedance: planning shots against beats and sections, transcribing lyrics, generating clips that cut to music, keeping the shots' own sound while the song stays the…

View source

How to use

  1. Copy the installation command and review the source package before installing it in your own agent environment.
  2. Check the source document for required tools, credentials and supported hosts.
  3. Describe your input and desired deliverable; review the output against the source material.
npx skills add https://github.com/scenario-labs/skills --skill scenario-seedance-music-video

Before you start

  • Installation does not include a model subscription, API allowance or permission to access your files.
  • Repository stars and installation counts describe activity, not an independent quality or security test.
Original skill document

Rendered from the source repository. Headings, examples, code, tables, links, and referenced images are preserved.

Scenario Seedance Music Video

Overview

Five steps: song, lyrics, story, frames, video. The supplied master is the only score, and no shot writes its own. scripts/build.py lays the master over the cut once, at the end, and proves the file untouched by hash. scripts/song.py reads the master so cuts land on its structure. The scripts need ffmpeg and ffprobe on PATH, and numpy for song.py.

Connection and the core generation loop: see the scenario skill in this repo. The Seedance parameter contract and conditioning traps: see the scenario-seedance skill. If a sibling skill named here is missing from your available skills, load it with skill_get (Scenario MCP) or ask the user to install it (https://github.com/scenario-labs/skills#install); unattended, proceed from tool schemas and flag the gap.

The two soundtracks

Music written inside a shot restarts in a new key at every cut, so the score comes from outside the video model: the supplied master. What Seedance makes is the sound bolted to the picture. The choice sets generateAudio on every shot, so make it before generating:

  • Song alone: generateAudio: false everywhere. build.py stream-copies the master whenever MP4

allows, so the delivered soundtrack is the supplied file, bit for bit.

  • Song over the shots' own sound: generateAudio: true, every prompt naming the sound the scene

itself makes (heels on wet asphalt, room tone) after "diegetic sound only:" and nothing else, plus "sound": 0.2 in the edit file. Name no instrument or genre, however well it fits the track, and no exclusion such as "no music": an instrument invites a score, one run failed on an output-audio content check for an added "distant guitar", and the exclusion has failed that check too. build.py cuts each clip's audio to its slot, mixes it under the master at that gain, and refuses a mix that would clip. The delivery is then one AAC encode, trading the bit-for-bit guarantee for the sound; the master file stays hash-checked.

Quick reference

StepCallNotes
1. Songpython3 scripts/song.py master.mp3 -o song.jsonhash, duration, loudness, tempo, sections, cut candidates
2. Lyricsrecommend for transcription, model_runsupplied lyrics win; sung-vocal transcription is a draft
3. Storyone page, shown to the user firstthe cheapest place to be wrong; name what must not drift
4. Framesimage model at the delivery aspect ratioreference stills, one per look to hold
5. ShotsSeedance model_run, audio per the choice abovedry_run first; wait=false; jobs_wait with pending_job_ids
6. Cutpython3 scripts/build.py edit.json out.mp4one ffmpeg pass: conform, concatenate, lay the master, verify

Ask once before starting: team and project, track clearance, aspect ratio and length, sound under the song or not, what must and must not appear, spend ceiling. With no one to answer, take the song alone, and write the story down rather than waiting to show it. Then run without stopping.

Worked example: one verse, three shots

  1. python3 scripts/song.py master.mp3 -o song.json. Sections are shot boundaries, cut candidates

are cut points; listen before trusting them.

  1. Upload the master: multipart upload_asset, then upload_asset_complete (see the scenario

skill). recommend a transcription model (the need in the user's own words as prompt), model_schema_get, model_run, jobs_wait. The transcript is a text asset among the job's asset ids: asset_get returns it whole in metadata.preview when hasFullPreview is true (at authoring time the Whisper-based tool wrote SRT-style timed blocks), else asset_download fetches the file. Instrumental track: say so and move on.

  1. Write one page: what happens, where, how it turns across the sections; name the closing image.

Show it to the user before spending anything.

  1. Generate reference stills with an image model (via recommend) at the delivery aspect ratio, one

per look. Look at them: they set identity and palette downstream. Holding one character across several: see the scenario-consistency skill.

  1. Per shot, decide the conditioning: opening state matters, pass a first-frame image; only

identity and world matter, pass referenceImages (see references/shots.md). Generate each shot one or two seconds longer than its slot for a trim handle. To land an action on a beat inside a shot, timecode it from the clip's start: the beat's master time minus the shot's at, plus any head trim planned as in (an unplanned trim moves every beat). Hold one camera move and one location across the timecodes; timecoded verbs survive generation where mood words do not.

  1. Write edit.json and run python3 scripts/build.py edit.json out.mp4:
json
{
  "master": "master.mp3",
  "fps": 24,
  "width": 1280,
  "height": 720,
  "sound": 0.2,
  "shots": [
    { "clip": "clips/01.mp4", "at": 0.0 },
    { "clip": "clips/02.mp4", "at": 12.5, "in": 1.0 }
  ]
}

Each shot runs until the next starts and the last to the master's end, so gaps are impossible, and every at snaps to the nearest frame. in is an optional head trim. sound is the gain on the clips' own audio: 0.1 to 0.3 under a mastered track, up to 1.0 when the clips run quiet (build.py takes 0 to 4); drop the line for the song alone. The build fails loudly on a clip too short, a mix that would clip, or delivered audio that is not the master.

Common mistakes

  • Letting a shot score itself: its score restarts at every cut.
  • Turning sound on before listening to every clip: generateAudio is one switch over the whole

track, so a clip that leaked music goes back with generateAudio: false rather than an exclusion in words.

  • Trimming, normalizing, fading, or re-encoding the master yourself: build.py copies it or carries

it at unity, and hash-checks the file either way.

  • Trusting a requested aspect ratio: stills and clips land near it, not on it, and build.py pads the

difference; crop to the delivery ratio first.

  • Prompting an opening state in reference mode: it will not appear; pass a first-frame image.
  • Judging a clip from a sparse contact sheet: a continuous camera move looks like a hard cut;

measure first (see references/shots.md).

  • Trusting the beat grid: sections and cut candidates are suggestions; check them against what you

hear.

  • Calling it done without watching the delivery with sound, then muted.
from this repository

More skills

All skills
scenario-labs
Community

scenario-blender-expert

Use when doing any 3D work in Blender through Python or an MCP bridge, headless or in a live session: modeling, sculpting a character or head, retopology, UV unwrapping, baking, texturing, rigging, animating, previs, geometry nodes, lighting and rendering, grease pencil, or finishing an AI-generated mesh into a production asset. Also before exporting or handing off an asset, to check it is ready for a game engine, a render or a 3D print (floating parts, z-fighting, inside-out normals, unapplied

1h installs
3
GitHub stars
4K
Updated
Nov 22
scenario-labs
Community

scenario-blender-sculpting

Use when sculpting in Blender through Python: sculpt a head, face or bust, a stylized or cartoon character, a creature or a realistic portrait; block out from primitives; cut planes, a jaw or gonial angle, cheekbones; fix a jaw or face that looks like a ball or balloon; stylized hair clumps; eyelids; an expression or smirk; plan remesh or multires resolution; reproduce Clay Strips, Crease, Draw Sharp, Scrape, Grab or mask moves without a tablet; final clay renders; or when a sculpt operator cras

1h installs
3
GitHub stars
4K
Updated
Nov 22
scenario-labs
Community

scenario-blender-texturing-shading

Use when texturing, shading or look-developing in Blender: building a PBR material from a texture set, displacement, procedural materials, edge wear, dirt, grime, color variation, decals, box or triplanar mapping on meshes without UVs, stylized character skin (subsurface, pores), eyes and cornea, hand-painted stylized textures, texture painting or painting masks headless. Also when a material looks plastic, CG, waxy or blown out, color spaces are wrong, or EEVEE and Cycles disagree.

1h installs
3
GitHub stars
4K
Updated
Nov 22
scenario-labs
Community

scenario

Use when connecting an AI agent to Scenario (scenario.com) through MCP, or when a task involves generating images, video, 3D, audio, sprites, textures, or game assets. Also when picking a Scenario model, running a LoRA, refining a generation prompt, uploading reference images, waiting on generation jobs, checking credits or quota, hitting Scenario auth, scope, or Forbidden errors, or setting up mcp.scenario.com in Claude Code, Cursor, VSCode, or another agent.

1h installs
2
GitHub stars
4K
Updated
Nov 22