What it does

OpenMontage is a Python framework that lets you describe a video in plain language, then delegates the entire pipeline to your AI coding assistant. The agent handles research, scriptwriting, image or footage generation, narration, music sourcing, subtitle timing, and final composition through Remotion or HyperFrames. It outputs actual video files, not slideshows with delusions of grandeur.

The interesting bit

The system is explicitly built for agents, not humans. There’s an AGENT_GUIDE.md contract, a tool registry with auditable decision logs, and pipeline definitions the agent must follow rather than improvise. The README even addresses agents directly: “If You’re An OpenClaw Agent Reading This, Congratulations: You Probably Passed The Turing Test.” This is less a tool you drive and more a production crew you rent by the prompt.

Key highlights

• 12 pipelines, 52 tools, 500+ agent skills — pipeline-driven workflow with stage directors and a scored provider-selection system across 7 dimensions
• Zero-API-key path exists — free narration via Piper TTS, stock footage from Archive.org/NASA/Wikimedia, composition via Remotion or HyperFrames, post-production via FFmpeg
• Reference-video cloning — paste a YouTube Short, TikTok, or Reel; the agent analyzes transcript, pacing, keyframes, and style, then produces 2-3 differentiated concepts with honest cost estimates before spending a dime
• Real motion or fake motion — can retrieve actual stock footage and edit a timeline, or animate stills with Ken Burns, parallax, particles, and camera drift
• GPU-local option — make install-gpu enables Wan 2.1, Hunyuan, LTX, or CogVideo for fully offline generation
• Multi-provider, auditable — every creative decision and provider choice gets logged; multi-point self-review (ffprobe, frame sampling, audio levels, subtitle checks) runs before delivery

Caveats

• Requires an AI coding assistant — Claude Code, Cursor, Copilot, Windsurf, or Codex; this is not a standalone GUI or CLI you run directly
• AGPLv3 license — copyleft; commercial use requires compliance
• Cost estimates are pre-production guesses — actual spend depends on provider choices and video length; the “$0.15” examples use specific free-tier or low-cost combinations that may not generalize

Verdict

Worth a look if you already live in Cursor or Claude Code and want to automate short-form content, explainers, or social clips without touching After Effects. Skip it if you want a traditional NLE, need broadcast-grade control, or your workflow doesn’t involve an AI assistant reading project files and executing code.