OpenMontage is an open-source, agentic video production system that turns an AI coding assistant into a full video studio: it researches a topic, writes a script, gathers or generates footage, adds voice and music, and cuts the finished file. It is meant for people who already work inside an agentic tool such as Claude, Cursor or Copilot and would rather drive a production from prompts than from a timeline — solo creators, channel operators who publish on a schedule, and developers wiring content into an automated pipeline. The project is written in Python, licensed under AGPL-3.0, and has collected roughly 58,000 stars and 7,300 forks since it appeared in March 2026.
What it does
The repository packages 12 production pipelines, more than 100 tools, and over 700 agent skill and production-knowledge files. Those skill files are the heart of the idea: instead of asking a model to improvise, the project hands the agent written craft knowledge about how video is actually made, and the pipelines walk it through the stages a real production runs.
A single run can cover the whole chain:
- research on the subject of the video
- a written script and shot plan
- footage, including real stock clips rather than animated stills
- narration, music and the final cut
The demonstration the project leans on is a science-fiction trailer that was scripted, shot and scored entirely by agents, at a reported provider cost of $1.33. That is the only cost figure the project publishes, and it covers that one short film — not a general price for any video you might make.
How it works
OpenMontage does not ship a model of its own. Your coding assistant is the runtime: it reads the skills, pipeline definitions and tool wrappers in the repository and executes them step by step, the same way it would work through any other codebase. What you install is the knowledge and the plumbing, not the intelligence.
The generation itself is delegated outward. The project's topics name ElevenLabs for speech, FLUX and Stable Diffusion for images, and OpenAI among the providers, with ffmpeg doing the encoding and assembly and Remotion available for programmatic rendering. A docs/PROVIDERS.md file covers which services can be plugged in, and an AGENT_GUIDE.md tells the agent how to behave inside the repository. Because every stage is a file you can open, a pipeline is editable: change the skill text or the tool, and the next run follows the new instructions.
Getting started
The README opens with a Quick Start section and a "paste a video you already love" path for people who want to start from a reference rather than a blank prompt, plus a page of sample prompts to try. Beyond cloning the repository you will need two things: an agentic coding assistant to drive it, and API keys for whichever providers you choose from the providers document — the footage, the voices and the images all come from paid services, so running OpenMontage means paying per call on top of whatever your assistant costs.
The project also runs a site at openmontage.video and a YouTube channel that documents the build, and it keeps a PR review guide for contributors. It was still being pushed to in September 2026.
When to use it / when not
It fits well if you already pay for an agentic coding tool and want that same subscription to produce finished video; if you want the workflow to live in a repository you can read, fork and version rather than inside a closed app; or if you want something more structured than one-shot text-to-video, with research, scripting and editing as separate reviewable steps.
It fits badly in a few clear cases. There is no graphical timeline here — if you want to drag clips, this is the wrong tool, and you should expect to work from a terminal and a prompt. AGPL-3.0 is a strong copyleft licence with a network clause, which matters if you intend to build a hosted commercial product on top. The project is also young, only months old, and moving quickly, so pipelines and provider integrations are likely to shift under you. And for work where every frame must be signed off, an agent that generates its own shots will still need a human in the loop for each cut.
Anyone who publishes video regularly and is already comfortable living in an AI coding assistant should look at this seriously, if only to see what a production pipeline looks like when it is written down as files instead of hidden in an editor's muscle memory. The 700-plus skill files are worth reading on their own, even if you never run a full pipeline, because they are an unusually explicit attempt to encode filmmaking craft in a form an agent can follow. Treat the $1.33 trailer as a demonstration rather than a budget, budget for provider fees, check the AGPL against your plans, and go in expecting a fast-moving repository rather than a finished product.