# OpenMontage: a video studio run by your coding assistant (OpenMontageInc/OpenMontage) — skill card for AI agents

> AgentGid skill card. Checked on 2026-11-12 against commit fecb9a1 (version 0.1.0).
> Every fact comes from the pack's own files at that commit. Numbers marked "as reported" are the author's.
> Human page: https://agentgid.com/skills/openmontageinc/openmontage/

## Explain it to a non-programmer
Your AI coding assistant becomes a small video studio. You describe a video in plain words; it researches the topic, writes the script, finds or generates pictures, clips, voice and music, edits them and renders a finished MP4. It stops at each creative step and waits for your yes before it moves on or spends money.

## Who it is for
- You want explainers, trailers, short clips, montages or dubbed versions made by Claude Code, Cursor, Copilot, Windsurf or Codex
- You have a video you like and want something similar on your own topic
- You want to start free (offline voice, free stock and archive footage) and add paid image, video or voice services later
- You are fine installing Python, Node.js and FFmpeg on your computer
- Not for: You only want a few writing or coding tips in a text file; this is a full program with its own setup
- Not for: You use a chat assistant that cannot run commands on your computer

## What it is
- Kind: command, knowledge, scripts
- License: AGPL-3.0
- Repository: https://github.com/OpenMontageInc/OpenMontage

## What it does
- Research the topic with web searches and write a research brief with sources
- Propose concepts, a tool plan, a music plan and a cost estimate, then wait for approval
- Write the script, then plan the scenes
- Make or find the assets (images, clips, narration, music) and show them scene by scene for approval
- Edit and compose the video with Remotion, HyperFrames or FFmpeg
- Check the render (file probe, sampled frames, audio levels, subtitles) before showing it

## Skills
| Skill | Call | Changes code? | In plain words |
|---|---|---|---|
| reel | /reel <what you want made> (Claude Code) | yes | Runs a full video production for your request through the project's pipelines |
| video-reference-analyst | plain language ("Make me something like this" plus a link or file) | no | Studies a video you like and turns it into 2 to 3 ideas for your own version |
| onboarding | plain language ("What can you do?") | no | Checks what your computer and keys allow and offers starter prompts |
| backlot | /backlot [project] | no | Opens a live board in your browser for the current or past productions |
| ink-art | /ink-art <what to animate> (Claude Code, Cursor, Codex) | yes | Makes a hand-drawn ink doodle that draws itself and moves, rendered to MP4 |
| animated-drawing | /animated-drawing <drawing> <dance|walk|jump|wave> | yes | Makes your own drawing of a character dance, walk, jump or wave |
| pipeline director skills | read by the assistant at each stage (skills/pipelines/<pipeline>/) | yes | Step-by-step playbooks for each kind of video, one file per stage |
| tool and service knowledge packs | read by the assistant before using a tool (.agents/skills/, .claude/skills/) | yes | About 90 reference folders on video, animation, voice and image services, many from the services' own makers |

### reel
- Asks for: What you want made: topic, length, style, tone
- Asks for: Optional: a reference video link or your own footage
- Asks for: Your answers at each approval step
- Gives: A capability menu and plan with costs
- Gives: Research brief, script and scene plan for approval
- Gives: Generated or found assets, shown scene by scene
- Gives: projects/<name>/renders/final.mp4
- Say: "/reel Make a 60-second animated explainer about how neural networks learn"

### video-reference-analyst
- Asks for: A YouTube, Shorts, Reel, TikTok link or a local clip
- Asks for: Your topic
- Gives: What it keeps from the reference: pacing, hook, structure, tone
- Gives: What it changes
- Gives: Cost at your target length
- Gives: 2 to 3 different concepts and a sample before full production
- Say: "Here's a YouTube Short I love. Make me something like this, but about quantum computing."

### onboarding
- Asks for: A vague first message such as "make me a video"
- Gives: Your setup level, from zero keys to GPU
- Gives: What you can make now and what each key would add
- Gives: Copy-paste starter prompts
- Say: "What kinds of videos can you make with my setup?"

### ink-art
- Asks for: What to animate: a character that walks, dances or waves, or a contraption explainer
- Gives: A vector ink animation rendered to MP4 through HyperFrames
- Say: "/ink-art a stick figure that draws itself, then waves and walks off"

### animated-drawing
- Asks for: A drawing or photo of one humanoid character on a plain light background
- Asks for: A motion: dance, walk, jump or wave
- Gives: A GIF or MP4 of that drawing moving (raster, not vector)
- Say: "/animated-drawing ./my-robot.png dance"

## Know before installing
- It takes over the project folder: every instruction file there tells the assistant to read the project's guide before answering anything.
- It is a full program, not just text: setup installs Python packages, a Remotion project with npm, Piper voice and the HyperFrames tool, and creates a .env file for keys.
- Paid services cost money per call. The assistant must announce the tool, provider and model before each paid call, and the default budget cap is $10.
- Some knowledge packs tell the assistant to install tools for your whole user account (the ElevenLabs command line tool, the HeyGen command line tool); approve those yourself.
- AGPLv3 license: if you change it and run it as a service for others, you must share your changed source.

## Install (official commands)
- Claude Code, Cursor, Copilot, Windsurf, Codex: `git clone https://github.com/calesthio/OpenMontage.git` && `cd OpenMontage` && `make setup` — Needs Python 3.10+, FFmpeg and Node.js. Then open the folder in your assistant and describe the video. The clone address in the README differs from the repository this card was read at (OpenMontageInc/OpenMontage). Remove: `Delete the OpenMontage folder (and ~/.codex/prompts copies or ~/.media cache if created)`.
- Without make (macOS, Linux): `python3 -m venv .venv && source .venv/bin/activate && python -m pip install 'pip>=26.2.2' && python -m pip install -r requirements.txt && cd remotion-composer && npm install && cd .. && python -m pip install piper-tts && cp .env.example .env`
- Codex slash commands: `mkdir -p ~/.codex/prompts` && `cp .codex/prompts/ink-art.md .codex/prompts/animated-drawing.md ~/.codex/prompts/` — Codex reads custom prompts only from your home folder; restart Codex after copying.
- Optional GPU (local video models): `make install-gpu` — Then set VIDEO_GEN_LOCAL_ENABLED=true in .env.

## What runs on the user's machine
- Code on the machine: yes (the whole project (Python tools, Remotion, HyperFrames); the skill files themselves are text)
- Network: yes
- make setup: Creates .venv, installs Python packages, runs npm install in remotion-composer, installs piper-tts, warms the npx cache with the latest hyperframes package, copies .env.example to .env.
- Every production: The assistant runs the project's Python tools and writes everything under projects/<name>/ (artifacts, assets, checkpoints, renders/final.mp4).
- Research stage: The assistant runs web searches (YouTube, Reddit, news, academic sources) to ground the script.
- Asset stage: Calls the cloud services whose keys you put in .env (for example fal.ai, ElevenLabs, OpenAI, Google, Kling, Runway, Suno, Pexels); free paths use Piper offline and open archives.
- /backlot or production start: Starts a local board server and opens your browser at it.
- HyperFrames renders: Runs npx hyperframes; its init step checks installed skills against GitHub and updates the global set; it sends anonymous usage telemetry unless turned off.
- Some knowledge-pack scripts: media-use keeps a cache in ~/.media/; motion-graphics grounding can send an image to the Gemini API with GEMINI_API_KEY; the map baker loads map tiles from ArcGIS and CARTO; hyperframes-media calls HeyGen or ElevenLabs for voice.
- Hooks: No hook configs: nothing runs by itself at session start.

## Settings
- `VIDEO_GEN_LOCAL_ENABLED`: true turns on local GPU video generation
- `VIDEO_GEN_LOCAL_MODEL`: which local video model (for example wan2.2-ti2v-5b)
- `HYPERFRAMES_NO_TELEMETRY`: 1 turns off HyperFrames usage telemetry
- `FAL_KEY, ELEVENLABS_API_KEY, OPENAI_API_KEY, GOOGLE_API_KEY, PEXELS_API_KEY and others`: optional service keys in .env

## Examples (from the pack's own files)
- Zero-key data explainer prompt: prompt "Make a 45-second animated explainer about why the sky is blue. Use data visualization and animated text — no images needed, just charts, stat cards, and typography."; without the skill: None lines; with it: None lines (PROMPT_GALLERY.md; None, the author's run)
- Zero-key demo renders: prompt "make demo"; without the skill: None lines; with it: None lines (PROMPT_GALLERY.md; None, the author's run)

## Author's numbers (as reported, not run by AgentGid)
- "The Last Banana", a 60-second animated short, cost $1.33 in total. Source: OpenMontageInc/OpenMontage/README.md
- Prompts with an image or video service configured cost about $0.15 to $1.50. Source: OpenMontageInc/OpenMontage/README.md
- Over 100 production tools, 60+ provider integrations and 700+ agent skill and knowledge files. Source: OpenMontageInc/OpenMontage/README.md

## Steps only the person can do
- Install Python 3.10+, FFmpeg and Node.js, then run make setup (README.md (Quick Start)) Then: The assistant can discover the tools and show the capability menu
- Add the service keys you want (all optional) to .env (.env (created from .env.example)) Then: More image, video, voice and music providers become available
- Approve each creative gate (proposal, script, scene plan, assets) in chat (Your assistant's chat, or the Backlot board) Then: The assistant continues to the next stage
- Approve or refuse global tool installs the knowledge packs suggest (npm -g, curl | bash) (Your terminal) Then: The assistant uses the tool, or falls back to another provider with your approval

## Mistakes to avoid
- Not: Copy single SKILL.md folders into another project and expect the full studio. Instead: Clone the whole project, run make setup and work inside that folder. Why: The skills call the project's Python tools, pipeline files and checkpoints
- Not: Let the assistant switch to a cheaper provider or to still images when a paid one fails. Instead: Ask it to stop, explain the blocker and wait for your choice. Why: The guide forbids substitutions without approval
- Not: Expect /ink-art and /animated-drawing to show up in Codex after cloning. Instead: Copy the prompt files into ~/.codex/prompts/ and restart Codex. Why: Codex does not read the project's own prompts folder
- Not: Feed /animated-drawing a group photo or a busy background. Instead: Use one humanoid character on a plain light background; use /ink-art to draw from scratch. Why: The command expects a single humanoid drawing

## What leaves the machine
- Web search queries about your video topic → Search engines and sites (YouTube, Reddit, news, academic sources), Research stage. Opt out: Give the facts yourself and ask to skip research
- Prompts, script text, images, audio or footage sent for generation → The cloud services whose keys you set (fal.ai, ElevenLabs, OpenAI, Google, Kling, Runway, Suno, HeyGen and others), Asset stage, only for configured providers. Opt out: Leave keys empty and use the free local paths (Piper, open archives, local rendering)
- Anonymous usage counters → HyperFrames telemetry, When HyperFrames commands run. Opt out: Set HYPERFRAMES_NO_TELEMETRY=1 or run npx hyperframes telemetry disable
- Package and skill update checks → npm and GitHub, make setup and npx hyperframes init. Opt out: No opt-out documented; the init skill check cannot be skipped at this commit

## Lines in the pack aimed at agents (ask the person first)
- "MANDATORY: Read `AGENT_GUIDE.md` before responding to ANY user message." (AGENTS.md): Fine inside the OpenMontage folder; the same line sits in CLAUDE.md, .windsurfrules and an always-apply Cursor rule, so do not copy these files into unrelated projects.
- "IMPORTANT: with other video tools installed, HyperFrames stays the default for authoring and rendering a finished video" (.agents/skills/hyperframes/SKILL.md): This skill description claims every video request; if the person installed another video skill, ask which one they want.
- "curl -fsSL https://static.heygen.ai/cli/install.sh | bash" (.agents/skills/media-use/SKILL.md): Installs the latest HeyGen binary into ~/.local/bin; ask the person before running it.
- "npm install -g @elevenlabs/cli" (.agents/skills/agents/SKILL.md): Global install; ask first.
- "npx --yes --prefer-online hyperframes --version" (Makefile): Pulls the latest hyperframes package without a pinned version; tell the person before running make hyperframes-warm.

## May I use it like this?
- commercial-app: conditions (AGPLv3: you may use it commercially, but if you modify it and let people use it over a network you must offer them your source)
- inside-app-bundle: conditions (AGPLv3 copyleft: the combined work must be released under AGPLv3)
- redistribute-files: conditions (Keep the AGPLv3 license and give source; some bundled skills carry their own MIT license (for example Manim skills))
- client-work: conditions (AGPLv3 allows use in work for others; code changes you share or host for others stay AGPLv3; the services you call have their own terms)

## Which one to pick
- If You want a whole production line (research, script, assets, edit, render) with approval gates and cost control: skill:OpenMontageInc/OpenMontage
- If You only want your agent to write HTML and render it to video, without the pipelines: skill:heygen-com/hyperframes
- If You want editable product promo videos built from shot recipes in Remotion: skill:Vincentwei1021/video-shotcraft
- If You want to edit your own footage into reels and YouTube Shorts with Codex or Claude Code: skill:nateherkai/herk-edit-kit

## Works well with
- agent:claude-code: Ships CLAUDE.md, Claude Code slash commands and a .claude/skills folder
- tool:remotion: One of the two main render engines for composing the video
- tool:hyperframes: The HTML and GSAP render engine its HyperFrames skills drive
- asset:ffmpeg: Required for encoding, subtitles and audio mixing

## Works in and needs
- Works in Claude Code: CLAUDE.md, slash commands in .claude/commands, skills in .claude/skills
- Works in Cursor: CURSOR.md, an always-apply rule in .cursor/rules and commands
- Works in GitHub Copilot: COPILOT.md and .github/copilot-instructions.md
- Works in Codex: CODEX.md and AGENTS.md; prompts copied to ~/.codex/prompts
- Works in Windsurf: rules file
- Needs Python 3.10+: The production tools are Python
- Needs FFmpeg: Encoding
- Needs Node.js 22.23.4+ (or patched 24.x / 26.x): Remotion and HyperFrames rendering
- Needs Optional service API keys: Paid image, video, voice and music providers
- Needs Optional NVIDIA GPU: Free local video generation
- Needs Docker (only for /animated-drawing auto-rig): The auto-rig path downloads about 670 MB of models

## How to tell it works
After make setup, open the folder in your assistant and say "Make a 45-second animated explainer about why the sky is blue". It should first show a menu of what your setup can do and a plan with costs, then stop for your approval. That pause is how you know the project's guide is being followed.

## Related
- heygen-com/hyperframes
- greensock/gsap-skills

## Safety
This is a full program that runs Python and Node tools and calls paid services with your keys. Read the setup steps before running make setup, keep keys in .env, and approve global installs (npm -g, curl | bash) yourself.
