OpenMontage turns an AI coding assistant such as Claude Code, Cursor, Copilot, Windsurf, or Codex into a video production pipeline. A request in plain language (an explainer, a trailer, a documentary montage) is carried through research, scripting, asset generation, editing, and final rendering by the assistant itself.
There is no code orchestrator. The assistant reads a YAML pipeline manifest that lists stages, tools, and success gates, then a Markdown director skill for each stage, and calls Python tools for the actual work. Every pipeline runs the same flow (research, proposal, script, scene plan, assets, edit, compose), checkpoints state as JSON with a decision log and cost snapshot, and pauses for human approval at creative decision points.
Production does not depend on paid generation APIs. With no keys, Piper narrates, archival and stock footage comes from Archive.org, NASA, Wikimedia Commons, Pexels, and Unsplash, and Remotion, HyperFrames, or FFmpeg composes the result. A documentary montage pipeline builds a CLIP-searchable corpus from open footage and edits real motion clips into a timeline. Adding provider keys unlocks cloud video, image, voice, and music models, and a scored selector ranks providers on task fit, quality, control, reliability, cost, latency, and continuity.
A run can also start from a reference video (YouTube, Shorts, Reels, TikTok, or a local clip): the agent analyses its transcript, pacing, scenes, and style and proposes differentiated concepts with cost estimates. A local Backlot storyboard shows each stage as it runs, holds asset generation on a scene-by-scene approval sheet, and replays finished runs.
| Pipeline | Produces |
|---|---|
| Animated Explainer | Researched explainer with narration, visuals, and music |
| Animation | Motion graphics and kinetic typography |
| Avatar Spokesperson | Avatar-driven presenter videos |
| Cinematic | Trailers, teasers, and mood edits |
| Clip Factory | Ranked short clips from one long source |
| Documentary Montage | Thematic montage from open footage archives |
| Hybrid | Source footage plus generated support visuals |
| Localization & Dub | Subtitled, dubbed, and translated video |
| Podcast Repurpose | Podcast highlights as video |
| Screen Demo | Software screen recordings and walkthroughs |
| Talking Head | Footage-led speaker videos |
Features
- Agent-driven pipelines: YAML manifests and Markdown director skills tell the coding assistant how to run each production stage
- Live web research: the agent searches YouTube, Reddit, news, and academic sources and cites them in a research brief before scripting
- Reference-video starts: analyses a video you like and proposes original concepts that keep its pacing and structure, with cost estimates
- Zero-key path: Piper narration, open archives and stock footage, and Remotion, HyperFrames, or FFmpeg rendering work without paid APIs
- Provider selection: 20+ video, 15+ image, and 10+ speech providers, cloud and local GPU, ranked by a scored selector
- Budget governance: cost estimates before execution, spend caps, and per-action approval thresholds
- Quality gates: pre-compose validation and post-render self-review with ffprobe, frame extraction, and audio analysis
- Backlot storyboard: a local live board that tracks stages, costs, and assets, gates generation on approval, and replays runs
- Post-production tools: stitching, trimming, mixing, color grading, subtitles, upscaling, background removal, lip sync, and WhisperX transcription
