OpenMontage: turning a coding assistant into a video studio

OpenMontage is an open-source video production system with one unusual design decision: it contains no orchestrator. The intelligence that runs a production is the AI coding assistant you already have open, following the project's instructions.

The repository, published by the developer calesthio, showed 60,813 stars and 7,729 forks on its GitHub page on September 22, 2026. A writeup on dev.to dates the repository's creation to March 29, 2026 and notes it carries no tagged releases, describing a rolling main branch rather than a versioned product. It is licensed AGPL-3.0, which matters if you plan to build a commercial service on top of it.

That absence of an orchestrator is the whole architecture. Python supplies the tools and the state files, and the coding agent supplies the judgement. The repository documents 12 production pipelines and more than 100 tools, and its agent guide states that every video request must go through a named pipeline, with ad hoc scripts and bypassing the checkpoint loop disallowed by design. Work moves through seven stages: research, proposal, script, scene plan, assets, edit, compose. Each stage pairs a YAML manifest in the pipeline definitions with a Markdown director skill that teaches the agent how to use the tools correctly.

The pipeline list is broader than the usual clip machine. Animated explainer, cinematic, documentary montage, clip factory, talking head, avatar spokesperson, screen demo, localization and dub, podcast repurpose and hybrid all ship as named formats. You can also hand it a YouTube video, a Short, a Reel or a local clip and get back a production plan with concepts, a tool path, cost estimates and a sample before it spends anything on the full run.

The cost story is the reason to read past the star count. The repository's zero-key path narrates with Piper, an offline text-to-speech engine, takes free footage and images from Archive.org, NASA, Wikimedia Commons, Pexels and Pixabay, and renders through Remotion and FFmpeg. That produces a complete video with voice, cut and captions for nothing beyond your own compute. Add credentials and the optional providers come online, including Kling, Google Veo 3.1, Seedance, Higgsfield, FLUX, GPT Image 2 and ElevenLabs. For anyone with a GPU, it also documents a local path for Wan 2.1, HunyuanVideo, LTX-Video and CogVideo at no per-render cost.

Getting started is short. You need Python 3.10 or later, FFmpeg, Node.js 18 or later, and one supported agent: Claude Code, Cursor, GitHub Copilot, Codex or Windsurf. Clone the repository, run make setup, which installs the Python packages, the Remotion Node packages and Piper in one command, then open the folder in your agent and state what you want in plain language. The documented examples include a 60-second animated explainer about how neural networks learn and a 75-second documentary montage on city life in the rain that uses real footage only.

Two pieces of the design are worth knowing before you trust it with a client job. A local storyboard server called Backlot fills itself in as the pipeline runs and acts as an approval gate before rendering, which is how the agent stops for a human at creative decision points. And the project runs its own quality checks: validation before composition plus a mandatory post-render self-review that probes the file with ffprobe, extracts frames, analyses audio levels, checks subtitles and verifies the delivery promises.

What it is not is a replacement for a generation model. OpenMontage is the production layer above clip generators: it writes the script, decides whether to call a generator or pull archival footage, narrates, scores, cuts and renders. Reviews of the output note the gap between technically finished and polished widens on the free stack, and that the archival-footage pipelines hold up better than the ones demanding a GPU.

A creator's practical read is narrow and useful. If your prompts are producing ten-second fragments and the assembly is what eats your week, this is a system built for exactly that layer, and it costs nothing to test on an existing repository.

Clone github.com/calesthio/OpenMontage, run make setup, and open the folder in your coding agent. The zero-key demo path renders without any provider credentials, so the first video costs nothing but time.