ScaleMuse Skills

Video playbooks for AI agents

Free, public Skills that plug into agents like Codex; the final video renders on your own machine.

A Skill is a working manual written for an agent: how to pick an angle, write the script, break it into shots, when to spend credits, and how to judge the result. Skills are free to download, edit and share. Paired with the ScaleMuse Runtime installed locally, the agent creates the project, lays out the timeline and renders an MP4 on your computer; only model-backed steps such as speech synthesis or image and video generation are billed in credits through your API key.

Catalog 1.0.0 · 5 Skills · source MontageAI/scalemuse-video (main, aa1d549) View on GitHub

Three steps to start

  1. 1

    Install the ScaleMuse Runtime

    The Runtime is the local program that owns project files, timelines, jobs and rendering, and the package already carries every official Skill. On an Apple Silicon Mac, download the trial package and drag "ScaleMuse Runtime" into Applications; other platforms run it from source, see the repository.

    Download the Runtime Source & docs
  2. 2

    Run setup codex once from Codex

    Send the sentence below to Codex as is. The Runtime installs its bundled official Skills into the Codex Skill directory and creates the scalemuse command; no plugin, no MCP configuration. After updating the Runtime, send the same sentence again to sync.

    What to send to Codex
    请运行 "/Applications/ScaleMuse Runtime.app/Contents/MacOS/scalemuse" setup codex

    Codex usually picks up newly installed Skills automatically; restart Codex if they do not appear.

  3. 3

    Just ask for the video you want

    On the next turn, describe the video. The entry Skill $scalemuse-video routes the request to the matching vertical Skill, and the agent follows the manual: confirm constraints, research and write, create the local project, lay out the timeline, render the MP4. You can also name a Skill explicitly with $. Before speech synthesis or image/video generation, the Runtime states the expected credits and then charges through your API key.

    Just ask
    Make me a 45-second, 9:16 explainer about AI chips.
    Name a Skill
    Use $finance-tech-video to turn "AI chip supply and demand" into a 45-second explainer.

    Create the API key in your profile and set it as the SCALEMUSE_API_KEY environment variable for the Runtime. Skills only call public Runtime commands and never see the plaintext key.

To install a single Skill on its own, or try it in another agent: every card carries a GitHub install prompt and the Skill is available on the next turn; creating projects and rendering still needs the local Runtime.

Skill catalog

Name, version, status, capability requirements and install URL on every card come from the public catalog file in the repository, so they match exactly what Codex installs.

  • PreviewPreview: installable and usable, degrades to what the Runtime can currently do and reports what is missing instead of pretending to finish.
  • ExperimentalExperimental: the method can be read and iterated on, but end-to-end output is not promised yet; the awaited capabilities are listed on the card.

ScaleMuse Video

EntryPreviewv0.1.0
$scalemuse-video

Entry point: just say "make me a video"

Catches any request to make a video with ScaleMuse: checks the Runtime first, then routes to the finance, talking-head, livestream or ecommerce Skill by content. When none fits, it asks only for the inputs that decide the output, then creates, compiles, validates and renders the project through the CLI without pretending to have done more.

What the agent reads

Create, continue, validate, or render videos with the local ScaleMuse Runtime. Use whenever a user asks to make a video with ScaleMuse, including a generic request such as “帮我做一条视频”; route specialized finance, talking-head, livestream, and ecommerce work to the matching installed workflow.

This is the exact text Codex uses to decide whether the Skill applies. Kept verbatim.

The SKILL.md and reference files the agent actually reads, verbatim.
Start from
An ideaConfirmed scriptLocal mediaExisting project
Outputs
MP4 videoScaleMuse project file (.smvideo)
Required capabilities Provided by the local Runtime; no credits.
Local projectTimelineLocal render
Optional capabilities Missing ones degrade gracefully and are reported; model-backed ones go through the ScaleMuse API and cost credits.
Asset importMedia inspectionSpeech recognitionSpeech synthesisImage generationVideo generation
Runtime
>=0.1.0 <0.2.0
Codex install prompt
Install this Skill from https://github.com/MontageAI/scalemuse-video/tree/main/skills/scalemuse-video.
Sample prompt
Use $scalemuse-video to help me make a video with the local ScaleMuse Runtime.

Finance & Tech Video

Previewv0.1.0
$finance-tech-video

Source-backed finance and technology explainers

Start from a topic, a set of sources or a confirmed script; build a claim table before prose so every fact is traceable. Create the local project, lay out the timeline, re-time it to the measured narration after TTS, and render a vertical or horizontal cut on your machine.

What the agent reads

Create source-backed finance or technology explainer videos from a topic, sources, confirmed script, or existing ScaleMuse project. Use for research-led business, market, company, AI, or technology videos; do not use for ordinary talking-head cleanup, livestream clipping, or physical-product UGC ads.

This is the exact text Codex uses to decide whether the Skill applies. Kept verbatim.

The SKILL.md and reference files the agent actually reads, verbatim.
Start from
An ideaSourcesConfirmed scriptExisting project
Outputs
MP4 videoScaleMuse project file (.smvideo)
Required capabilities Provided by the local Runtime; no credits.
Local projectTimelineLocal render
Optional capabilities Missing ones degrade gracefully and are reported; model-backed ones go through the ScaleMuse API and cost credits.
Asset importAsset inspectionSpeech synthesisImage generationVideo generation
Runtime
>=0.1.0 <0.2.0
Codex install prompt
Install this Skill from https://github.com/MontageAI/scalemuse-video/tree/main/skills/finance-tech-video.
Sample prompt
Use $finance-tech-video to turn my topic into a source-backed local video project.

Ecommerce UGC Remake

Experimentalv0.1.0
$ecommerce-ugc-remake

Structural remake of ecommerce UGC ads

Break a reference ad into hook, problem, demonstration, evidence and CTA beats, then replace every shot and claim with your own product facts and authorized assets, generating only what is truly missing. Never clones a third party's face, voice, copy or protected footage. Waiting on reference analysis and asset capabilities.

What the agent reads

Rebuild the advertising structure and shot rhythm of a reference video for a user's physical product using authorized product facts and assets. Use for ecommerce UGC ads; do not use for software promos or copying a third party's identity, voice, script, or protected footage.

This is the exact text Codex uses to decide whether the Skill applies. Kept verbatim.

The SKILL.md and reference files the agent actually reads, verbatim.
Start from
Reference videoLocal mediaExisting project
Outputs
MP4 videoScaleMuse project file (.smvideo)
Required capabilities Provided by the local Runtime; no credits.
Local projectTimelineLocal render
Optional capabilities Missing ones degrade gracefully and are reported; model-backed ones go through the ScaleMuse API and cost credits.
Asset importReference analysisSpeech synthesisImage generationVideo generation
Runtime
>=0.1.0 <0.2.0
Codex install prompt
Install this Skill from https://github.com/MontageAI/scalemuse-video/tree/main/skills/ecommerce-ugc-remake.
Sample prompt
Use $ecommerce-ugc-remake to rebuild this ad structure for my physical product.

Livestream Clips

Experimentalv0.1.0
$livestream-clips

Clip discovery and batch editing for long streams and podcasts

Transcribe and segment hours of livestream or podcast, score candidates on hook strength, completeness, energy and independence, deduplicate, then derive one child project per approved range for cleanup and batch rendering. Waiting on long-media analysis.

What the agent reads

Find, score, deduplicate, and edit self-contained short clips from a long livestream or video podcast. Use when the main decision is which moments to extract; do not use for polishing one already-selected talking-head take.

This is the exact text Codex uses to decide whether the Skill applies. Kept verbatim.

The SKILL.md and reference files the agent actually reads, verbatim.
Start from
Local mediaExisting project
Outputs
MP4 videoScaleMuse project file (.smvideo)
Required capabilities Provided by the local Runtime; no credits.
Local projectTimelineLocal render
Optional capabilities Missing ones degrade gracefully and are reported; model-backed ones go through the ScaleMuse API and cost credits.
Asset importSpeech recognitionWord alignmentTopic segmentation
Runtime
>=0.1.0 <0.2.0
Codex install prompt
Install this Skill from https://github.com/MontageAI/scalemuse-video/tree/main/skills/livestream-clips.
Sample prompt
Use $livestream-clips to find and edit the strongest self-contained moments.

Talking-head Editor

Experimentalv0.1.0
$talking-head-editor

Non-destructive cleanup of single-presenter A-roll

Keep the source recording and timecodes, use word-level transcript timing to find silences, fillers and repeated takes, cut them non-destructively after confirmation, then add captions, reframing and a speech-safe mix. Waiting on word alignment and source-time mapping.

What the agent reads

Non-destructively clean and enhance speech-led A-roll using transcript timings, edit decisions, captions, reframing, overlays, and audio mixing. Use for a single-presenter recorded video; do not use for selecting highlights from a long livestream or generating a research-led faceless explainer.

This is the exact text Codex uses to decide whether the Skill applies. Kept verbatim.

The SKILL.md and reference files the agent actually reads, verbatim.
Start from
Local mediaExisting project
Outputs
MP4 videoScaleMuse project file (.smvideo)
Required capabilities Provided by the local Runtime; no credits.
Local projectTimelineLocal render
Optional capabilities Missing ones degrade gracefully and are reported; model-backed ones go through the ScaleMuse API and cost credits.
Asset importMedia inspectionSpeech recognitionWord alignment
Runtime
>=0.1.0 <0.2.0
Codex install prompt
Install this Skill from https://github.com/MontageAI/scalemuse-video/tree/main/skills/talking-head-editor.
Sample prompt
Use $talking-head-editor to clean up my recorded talking-head video.

FAQ

Do Skills cost anything?

No. Skills are public method files anyone can download, edit and share, and the Runtime is free to install. Credits are charged at standard rates only when the agent calls ScaleMuse model capabilities (speech synthesis, speech recognition, image or video generation) through the Runtime, and the expected cost is stated before submission.

Do I have to use Codex?

Skills and the Runtime follow the MCP standard, so agents such as Codex, Claude Code and WorkBuddy can connect to the same entrypoint; the current install guide and testing focus on Codex.

Where is the video rendered? Is my footage uploaded?

On your own computer. Project, timeline and rendering all happen locally and local footage is not uploaded; only model capabilities you explicitly approve send the necessary inputs to ScaleMuse, and that step states the expected credits first.

FE 15293fe-20260903021855 | BE ...