Motion Video

An agent skill that produces a beat-synced 1920×1080 30 fps motion-graphic video: the scene cuts land on the music's beats, the reveals land on the spoken words, and the audio is broadcast-clean.

npx skills add bestagentkits/motion-video-skill

MIT licensed. Works with Claude Code, Codex, Cursor, OpenCode, Gemini CLI, Pi and any other agent that reads Agent Skills.

Demo

The comic-multiverse edition of the Dewee v3.34 release video, produced by this skill from a single prompt:

→ Watch the 151 s demo on X

The same release also has a 144 s glass-keynote edition. Neither working project is bundled with this repository, because the generated voice-over, music and SFX are large and provider-licensed.

What it does

Each run follows one fixed pipeline, and every step writes the file the next step reads, so the order matters:

1. Pin the outcome: topic and sources, style, duration, voice language, ending.
2. Scaffold assets/videos/<slug>/ from the bundled templates.
3. Generate the voice-over, sound effects and music through the multix CLI.
4. Align the voice-over to the script, so every word has a real timestamp.
5. Fit the beat grid, then splice music bars so the drops fall where the script needs them.
6. Build the timeline and the audio mix, then measure the voice/music balance.
7. Author index.html against the composition contract and the chosen style.
8. Lint, check, snapshot and render with HyperFrames, remux the untouched mix, and encode a smaller social version.

A run is finished only when the measurable checks pass: hyperframes lint and check clean, every music segment within ±2 ms of its target beat, music about 4–6 dB under speech, 1920×1080 at 30 fps, and EBU R128 around −14 LUFS with a true peak no higher than −1 dBFS.

Two styles

A restyle is a new project directory, so the first edition always stays intact.

Install

Any agent, with npx skills

The fastest route, and the one that works across agents:

# See what the repository offers without installing
npx skills add bestagentkits/motion-video-skill --list

# Install into the detected agents inside this project
npx skills add bestagentkits/motion-video-skill

# Install globally for Claude Code, without prompts
npx skills add bestagentkits/motion-video-skill -g -a claude-code -y

# Use it once, without installing
npx skills use bestagentkits/motion-video-skill@motion-video | claude

Useful flags: -g installs to your user directory instead of the project, -a <agent> targets specific agents such as claude-code or codex, --copy copies instead of symlinking, and -y skips the prompts.

Claude Code plugin marketplace

/plugin marketplace add bestagentkits/motion-video-skill
/plugin install motion-video@motion-video-skill
/reload-plugins

The same thing from a shell:

claude plugin marketplace add bestagentkits/motion-video-skill
claude plugin install motion-video@motion-video-skill

Invoke the skill as /motion-video:motion-video, or just describe the video you want and let Claude route to it.

ChatGPT and Codex plugins

The repository ships a portable Agent Plugins manifest at the root and a repo marketplace catalog, which is the layout ChatGPT and Codex expect for a skills-only plugin:

codex plugin marketplace add bestagentkits/motion-video-skill
codex plugin marketplace list

In the ChatGPT desktop app, open the Plugins Directory, select the marketplace, and Install. For a personal marketplace instead of a repo one, add an entry to ~/.agents/plugins/marketplace.json pointing at this repository root.

Public listing in the universal ChatGPT/Codex Plugins Directory is a separate step that goes through OpenAI's submission portal and its review process; the package here is already in the submitted shape.

Manual install

Copy skills/motion-video/ into your agent's skills directory. Any of these locations work, depending on the agent and the scope you want:

Requirements

Keep API keys in the multix CLI's own configuration. The skill never prints key values and never writes them into project files, plans or reports.

Quick start

Install the skill, then ask your agent for the video:

Make a 100-second launch video for the 3.34 release with a voice-over,
captions and music, in the comic multiverse style.

To drive the pipeline yourself, start from the templates:

mkdir -p assets/videos/my-video
cp -r skills/motion-video/assets/templates/scripts skills/motion-video/assets/templates/data \
      skills/motion-video/assets/templates/hyperframes.json \
      skills/motion-video/assets/templates/index-skeleton.html \
      assets/videos/my-video/
cd assets/videos/my-video
# rename data/*.example.json to data/*.json,
# and index-skeleton.html to index.html

Then follow references/production-pipeline.md step by step. In the skill's own documentation, <skill> means the installed skill directory, for example ~/.claude/skills/motion-video.

What is in the box

.
├── .agents/plugins/marketplace.json    ChatGPT / Codex repo marketplace catalog
├── .claude-plugin/
│   ├── marketplace.json                Claude Code marketplace catalog
│   └── plugin.json                     Claude Code plugin manifest
├── plugin.json                         Portable Agent Plugins 1.0 manifest
├── skills/motion-video/
│   ├── SKILL.md                        Routing and the definition of done
│   ├── references/
│   │   ├── production-pipeline.md      Every command, scaffold to social encode
│   │   ├── audio-and-beat-sync.md      Voice, cues, music plans, splice, mix
│   │   ├── composition-contract.md     TIMING shape, helpers, determinism
│   │   ├── style-glass-keynote.md      Tokens, layers, transitions, captions
│   │   └── style-comic-spiderverse.md  The second style, same contract
│   ├── scripts/
│   │   ├── fit-beat-grid.py            BPM, BEAT0 and a per-bar kick/energy table
│   │   ├── verify-arrangement.py       Splice lag per segment; exits 1 on drift
│   │   ├── measure-mix-balance.py      Music under speech versus in the gaps
│   │   └── tests/                      Regression tests on synthetic audio
│   └── assets/templates/
│       ├── index-skeleton.html         Composition skeleton
│       ├── hyperframes.json            Renderer configuration
│       ├── scripts/                    Voice, SFX, music, alignment, timeline
│       └── data/                       Example script, cues, SFX, music, outro
├── AGENTS.md                           Operating notes for agents working here
├── README.md
└── LICENSE

The plugin root is the repository root, so skills/ sits at the top level and no file is duplicated between the skill package and the plugin manifests.

Tests

python -m unittest discover -s skills/motion-video/scripts/tests

The suite runs the beat-grid fitter and the arrangement checker against synthetic audio, so it needs ffmpeg, numpy and scipy, and no network access.

License

MIT. See LICENSE.

No generated media is bundled. The skill instructs the agent to use open-licensed (SIL OFL or Apache) fonts from Google Fonts, and it produces its voice-over, music and sound effects inside your own provider accounts.