OpenMontage: Agentic Video Production From Your Editor
calesthio/OpenMontageOpenMontage is an open-source video production system you drive through an AI coding assistant. You describe the video in plain language and the agent handles research, scripting, asset generation, editing, and final composition with Remotion, showing per-asset cost and approval gates along the way.
Reach for OpenMontage if you already work in an AI coding agent and want to script AI videos from your editor with visible cost gates before each render. Skip it if you want a one-click GUI generator — this expects you to install Python, FFmpeg, and Node, wire up provider API keys, and drive it from a coding assistant.
The problem it solves
Making a short AI video today means stitching together a pile of separate tools: one for the script, another for images, another for voiceover, another for captions, and an editor to assemble it all. Costs are opaque until the bill arrives, and there's no single place to approve the script or the visuals before you've spent money rendering the whole thing.
What is it?
OpenMontage is an open-source (AGPL-3.0) agentic video production system written in Python. The README describes 12 pipelines, 52 tools, and 500+ agent skills that turn an AI coding assistant into a video studio. It can generate from a text prompt or a reference video — pulling free stock footage, generating images and motion clips, adding narration and captions, and composing the result with Remotion.
Why it's getting attention
The README carries a badge claiming '#1 Repository of the Day on GitHub Trending' and pitches OpenMontage as the first open-source agentic video production system. The specific angle — reproducible AI videos built and cost-tracked from inside your coding agent, with example productions the README says cost as little as a few cents — is unusual enough to draw attention.
Key features
- ✓Agent-driven pipeline: research, script, asset generation, editing, and composition from a plain-language prompt
- ✓Can start from a reference video (YouTube, Short, Reel, TikTok, or local clip) and rebuild it on a new topic
- ✓Final composition through Remotion, with support for FLUX images and motion clips from providers like Veo and Kling
- ✓Narration via text-to-speech and word-level captions (the README mentions WhisperX)
- ✓A 'Backlot' board (`python -m backlot open`) that shows live progress, per-asset cost, and approval gates
- ✓Works with Claude Code, Cursor, Copilot, Windsurf, or Codex — any assistant that can read files and run code
Best use cases
- •Producing short explainer or trailer videos from a written prompt without hand-editing
- •Recreating the pacing and structure of a reference video on a different topic
- •Building cost-tracked AI videos where you approve script and visuals before rendering
- •Turning still images into animated shorts with camera motion and particle effects
How to install / try
The README lists prerequisites: Python 3.10+, FFmpeg (`brew install ffmpeg` / `sudo apt install ffmpeg`), Node.js 18+, and an AI coding assistant. You also supply API keys for the providers you want (image, video, and voice generation). The exact clone-and-setup steps live in the repo's Quick Start section.
How to use
You drive OpenMontage from your AI coding assistant in natural language — describe the video, or paste a reference clip, and the agent proposes concepts, an honest tool path, and cost estimates before producing. The Backlot board, opened with `python -m backlot open`, shows the storyboard as an approval gate so you sign off on visuals before the final render.
Strengths
- ✓Ties the whole video pipeline — script to render — into one agent-driven workflow
- ✓Shows per-asset cost and holds approval gates, so you see spend before committing to a render
- ✓Composes with Remotion and can pull free stock footage, not just animate a few stills
- ✓Works with the coding assistant you already use rather than a separate app
Limitations & risks
- △Heavy setup: Python, FFmpeg, Node, a coding assistant, and provider API keys before you make anything
- △Output cost and quality depend on paid third-party providers (the README's cheap examples still spent real money on some of them)
- △The impressive demo videos and the '12 pipelines / 52 tools / 500+ skills' figures are self-reported in the README, not independently verified
- △AGPL-3.0 copyleft, which can complicate commercial or embedded use
- △It's a young project driven through coding agents, so results will vary with your setup and model access
Alternatives
Who should try it — and who should skip
Developers who already live in an AI coding assistant and want to generate videos from their editor with cost visibility and approval steps. It suits people comfortable installing Python, FFmpeg, and Node and managing provider keys. Anyone wanting a polished GUI or a no-setup, one-click video generator should look elsewhere.
Frequently asked questions
The project is open source under AGPL-3.0. Running it isn't free in practice, since it calls paid image, video, and voice providers that you pay for per use.
You drive it through an AI coding assistant — the README lists Claude Code, Cursor, Copilot, Windsurf, and Codex — describing what you want in plain language while the agent runs the pipeline.
The README says it can do both: animate stills, or build a corpus from free stock footage and open archives, retrieve motion clips, and edit them into a rendered timeline.
The README lists example productions ranging from about $0.02 to $1.33 each, depending on which providers and asset types were used. Your cost will vary with duration and provider choices.