← LeaderboardThe Skill Leaderboard

scenario-kinetic-film

Use when making a launch film, game teaser, trailer, product promo or brand film from a brief in kinetic typography: huge type on every beat, whip pans, 3D letters, key art cut in, beat sync to generated music and sound effects, a 1080p or 4K render with motion blur. Not for a song the user already has (scenario-kinetic-music-video).

Published by scenario-labs ↗scenario-labs/skills

Try your first task

A source-grounded starter task has not been verified for this skill yet.

SKILL.md

name: scenario-kinetic-filmdescription: "Use when making a launch film, game teaser, trailer, product promo or brand film from a brief in kinetic typography: huge type on every beat, whip pans, 3D letters, key art cut in, beat sync to generated music and sound effects, a 1080p or 4K render with motion blur. Not for a song the user already has (scenario-kinetic-music-video)."license: MIT

Kinetic Launch Film

Overview

A film here is a spec: scene blocks placed on the beat grid of its music, plus a palette, a caption track and a logo. The engine turns it into one web page where every frame is a pure function of time, so a renderer can seek anywhere and average sub-frames into real motion blur. Nothing in the motion is generated: every word is the word you wrote, and one palette change rebrands the whole film.

  • Code makes all the motion, from 14 blocks: kinetic type, a 3D title, key-art cut-ins, an end card.
  • Scenario makes what code cannot: a music bed at the exact tempo the grid needs, a kit of one-shot sound effects, and any key art, product shot or character the Picture block cuts in.
  • finish.py passes a film only with 40% still frames or less, no still stretch over 0.6 s, and a median motion of 2.5 px/frame or more before the end card. A film that parks fails.

Needs: Node 20+, ffmpeg and ffprobe, uv (or Python 3.12+), Google Chrome or Playwright's Chromium. new_film.sh puts the locked Playwright (package.json, package-lock.json), the pinned Python packages (requirements.txt) and the fonts (get_fonts.sh) in the film folder.

Connection and the generation loop (recommend, model_schema_get, dry_run, jobs_wait): the scenario skill. A song the user brings, with lyrics on its sung onsets: scenario-kinetic-music-video. 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.

Quick reference

$K is this skill's folder. Run everything from the film folder.

bash $K/scripts/new_film.sh my-film --example starter   # or kaiju-postal, tour; --force finishes or refreshes a folder
cd my-film && python3 -m http.server 5671                # preview: space plays, arrows step one beat
node $K/scripts/snap.mjs stills 0.5 b4 b12 b20           # half-size stills at beats (or seconds), from a second shell
.venv/bin/python $K/scripts/beat_grid.py audio/music.mp3 --bpm 172 --window 40
.venv/bin/python $K/scripts/mix.py .                     # out/mix.wav from the film's own cue list
.venv/bin/python $K/scripts/render.py . --res 1080       # out/video_1080.mp4, workers sized to memory
.venv/bin/python $K/scripts/finish.py . --video out/video_1080.mp4 --audio out/mix.wav

Then render and finish again at --res 4k with --video out/video_4k.mp4. Outputs are replaced, so the check run and the master share out/final.

Rules:

  • Everything is in beats: blocks, shots, captions, cues. B(k) is k beats in seconds.
  • One idea per beat or half beat: one or two words, filling the frame width.
  • No still stretch over 0.6 s before the end card.
  • Transitions are moves (whipIn/whipOut, a zoom through a letter, a collapse to a point), never fades.
  • Flip the ground (bg, accent, paper) on phrase changes. Palette colors are 6-digit hex or rgb().
  • A pure function of t: no Math.random, Date or state kept between frames, since motion blur renders sub-frames out of order.
  • On-screen numbers come from the film ({lines}, {frames}, {seconds}, the HUD's frame count and timecode) or the user, never invented.
  • 4K masters stay at H.264 Level 5.2 or lower and under 1 GB.
AssetCapabilityInputs to look for in model_schema_get
Music bedtxt2audiotempo, instrumental switch, duration
Sound kittxt2audioduration, negative prompt
Key art and productstxt2img, img2imgreference images, 16:9, 2K or more

The steps, each with its reference:

  1. Scaffold the film folder with new_film.sh (the first command above). Every later step runs in it, and the downloads go in its audio/ and art/.
  2. Brief to beat sheet: method.md.
  3. Music, sound kit and pictures on Scenario: sound.md.
  4. spec.js: blocks.md, with the motion rules in craft.md.
  5. Preview and stills: scrub, snap the beats, fix, repeat.
  6. A 1080p check render until finish.py prints ALL CHECKS PASS, then the 4K master: render.md.

Worked example: a 24-second game teaser

Kaiju Postal (spec.js) is a teaser for a fictional game about a courier delivering mail in a city of giant monsters. Kinetic type runs on a 172 BPM drum and bass track, two key arts cut in on the drop, a door-knock gag rides the beat cuts, and the title lands on the last hit.

  1. Scope and folder. The scenario skill connects and picks the team and project. new_film.sh kaiju-postal scaffolds the film folder from the starter; every later command runs in it.

  2. Beat sheet (free). 68 beats at 172 BPM (23.7 s):

    • 0-8 type: a charge, then THE CITY / HAS NEW / VERY BIG NEIGHBORS. / THEY STOMP.;
    • 8-16 key art 1 on the drop; 16-24 type ending in a zoom through the D of DELIVER.;
    • 24-32 key art 2; 32-43 a fly-through (DODGE THE BIG FEET.); 43-51 the 3D title YOU ARE THE POST.;
    • 51-57 the knock gag; 57-68 the title card.

  3. Music at an exact tempo. recommend:

    { "capability": "txt2audio", "prompt": "Instrumental drum and bass for a playful video game trailer at a fixed 172 BPM, a tense intro, then a huge drop, punchy kick on every downbeat, no vocals", "duration": 40, "limit": 4 }

    Pick a model whose schema has a tempo input and an instrumental switch, read the input names off model_schema_get, and price two takes:

    { "model_id": "<id from recommend>", "dry_run": true, "parameters": { "prompt": "<the prompt pattern in sound.md>", "instrumental": true, "duration": 40, "bpm": 172, "numOutputs": 2 } }

    After the user's go (unattended, every step: run only inside a budget the task states, else report the prices and stop): run, jobs_wait, asset_download, curl -L -o. beat_grid.py --bpm 172 on each take; keep the clearest grid and the strongest drop. Take B held 172.0 BPM with a drop at 10.74 s, so the film starts 8 beats earlier (music: { offset: 7.95, stopBeat: 58 }): the drop lands on the first key art and the music stops dead on the title.

  4. The sound kit. The 12 core sounds of sound.md (14 runs with the _1/_2 variants), plus the film's own jokes (stomp, roar, bell, knock): a dry run each and one total. After the user's go: one run each, "wait": false, one jobs_wait.

  5. Key art. Two pictures, each priced before the user's go, the second only once the first exists: it passes the first's asset id as a reference so the courier and the city match. The scooter came back with a maker's logo, painted out before use.

  6. spec.js, an excerpt:

    bpm: 172, beats: 68,
palette: { bg: '#0B1020', paper: '#F3EEE3', accent: '#FFC21A', hot: '#FF8A1F', ink: '#14120E' },
music: { file: 'audio/music.mp3', offset: 7.95, stopBeat: 58, gain: -7 },
sounds: { stomp: -1, roar: -5, bell: -7, knock: -1, hit: 1 },
cues: [[5.5, 'stomp'], [8.25, 'roar', .9], [24.6, 'bell']],
scenes: [
  [Words, 0, 8, { charge: true, shots: [
    { at: 1, style: 'punch3d', text: 'THE CITY', fx: 'punch' },
    { at: 2, style: 'slam', text: 'HAS NEW', ground: 'accent' },
    { at: 3, style: 'stack', text: ['VERY BIG', 'NEIGHBORS.'], ground: 'paper', in: 'whipX', fx: 'none' },
    { at: 5, style: 'split', text: ['THEY', 'STOMP.'], in: 'whipY', fx: 'punch', big: true } ] }],
  [Picture, 8, 16, { src: 'art/key1.png', weather: 'rain', words: [[1, 'MONSTERS'], [3, 'NEXT DOOR.', 'accent']], cutIns: [[5, 1.65, [.62, .2], 'stomp']] }],
  ...

  7. Mix and check render. mix.py lists any hit quieter than the music: raise it in spec.sounds. The first finish.py failed on two pictures that held 1.5 s and 1.2 s; the Picture block now pushes, pans, punches in and cuts in on the beat. Its rain carries each picture between words; under snow, dust or no weather, give every beat a word or a cut-in. The teaser passed with 24% still frames, no hold over 0.6 s and 12.5 cuts per 10 s.

  8. Master. render.py . --res 4k and finish.py: a 4K master under 1 GB, a 1080p version and a 9:16 version (the film centered on black), each checked.

Common mistakes

  • Parking between hits. A film can land every cut on the beat and still read as slides: a word or a picture that sits still for a beat is a hold. Keep something moving every frame.
  • Picture prompts that backfire. "A calm darker band for the title" came back as a gray band: describe the composition and darken with tint. "No watermark" beside a reference image was blocked: say "no text, no logos". Vehicles and clothes come back with makers' logos: inspect at 100%.
  • Trusting a generated track's tempo. Fit the stretch the film uses (--window) and give the intended tempo (--bpm): without it a 172 BPM track reads as 86.
  • Measuring in custom code. Measure once, while visible (a hidden element measures 0), and use relPos, not offsetLeft, inside a transformed parent.
  • A nested film sharing state. The Rewind block builds a second film from the same spec: never store per-film state on shared objects.
  • Editing the engine in the film folder. new_film.sh --force replaces it. Custom blocks go in spec.js or a custom.js it imports.
  • Changing render.mjs, browser.mjs or post.py without render.md. Its trap list is why they look the way they do.
  • More 4K workers than render.py plans. Six filled 36 GB and stalled the machine.

Files

scenario-kinetic-film by scenario-labs · The Skill Leaderboard