OpenMontage turns an AI coding assistant into a video production pipeline. You describe the video you want in plain language inside Claude Code, Cursor, Copilot, Windsurf, or Codex, and the agent carries out research, scripting, asset generation, editing, and final composition, pausing for your approval at creative decision points.
There is no code orchestrator: the agent is the orchestrator. It reads a YAML pipeline manifest that defines stages, tools, review criteria, and success gates, then reads a Markdown director skill for each stage that describes how to execute it. Python tools provide generation, editing, analysis, and persistence, and every stage checkpoints state to JSON with a decision log and cost snapshot so runs can resume.
Every pipeline follows the same flow: research → proposal → script → scene plan → assets → edit → compose. Before a render, a validation gate checks the delivery promise and slideshow risk; after the render, a self-review runs ffprobe validation, frame sampling, audio-level analysis, and subtitle checks before the result is presented. Provider selection scores each option across seven dimensions (task fit, output quality, control, reliability, cost efficiency, latency, continuity) and records the reasoning.
You can also start from a reference video (YouTube, Short, Reel, TikTok, or local clip). The agent analyzes its transcript, pacing, scenes, keyframes, and style, then returns two or three differentiated concepts with a tool path, cost estimate, and sample before full production. Backlot, a local board fed by the pipeline's project files, shows stage progress, script, scene cards, provider decisions, and spend, and gates the storyboard on your approval.
| Pipeline | Produces |
|---|---|
| Animated Explainer | Research-backed explainer with narration, visuals, and music |
| Animation | Motion graphics, kinetic typography, animated sequences |
| Avatar Spokesperson | Avatar-driven presenter videos |
| Cinematic | Trailer, teaser, and mood-driven edits |
| Clip Factory | Ranked short-form clips from one long source |
| Documentary Montage | Thematic cut from a CLIP-indexed corpus of free stock and archive footage |
| Hybrid | Source footage plus AI-generated support visuals |
| Localization & Dub | Subtitles, dubbing, and translation of existing video |
| Podcast Repurpose | Podcast highlights as video |
| Screen Demo | Software screen recordings and walkthroughs |
| Talking Head | Footage-led speaker videos |
Features
- Agent-first architecture: YAML manifests and Markdown skills hold the orchestration logic; Python supplies tools and persistence
- Web research stage: the agent runs searches across YouTube, Reddit, Hacker News, news sites, and academic sources and cites them in a research brief
- Real-footage documentary path: builds edited videos from Pexels, Archive.org, NASA, Wikimedia, and Unsplash footage without paid video models
- Provider coverage: 60+ integrations spanning cloud APIs, local models, stock libraries, and open archives, including Kling, Runway, Veo, Seedance, and Gemini video generation
- Scored provider selection: ranks providers across seven dimensions and logs alternatives, confidence, and reasoning
- Quality gates: pre-compose validation and a mandatory post-render self-review with ffprobe, frame extraction, and audio analysis
- Budget governance: cost estimation before execution, spend caps, and per-action approval thresholds
- Backlot board: a local live storyboard and library with scene-by-scene approval and run replay
