OpenMontage

🗒️ calesthio/OpenMontage — an open-source, agentic video production system. There is no code orchestrator: your AI coding assistant is the orchestrator. It reads YAML pipeline manifests and Markdown “director skills”, calls Python tools, self-reviews its own output, and pauses for your approval at every creative gate. Works with Claude Code, Cursor, Copilot, Windsurf and Codex. AGPLv3.

The pitch that matters: most “free AI video” stacks quietly mean animate still images. OpenMontage does that too, but it can also cut a finished video from real motion footage pulled from Archive.org, NASA and Wikimedia Commons — semantically ranked and edited on a proper timeline.

Links

Description

🧩 What it does:

• 12 production pipelines — animated explainer, animation, avatar spokesperson, cinematic, clip factory, documentary montage, hybrid, localization & dub, podcast repurpose, screen demo, talking head, character animation.
• 100+ registered tools / 60+ provider integrations — video and image generation, TTS, music, mixing, subtitles, enhancement, analysis — all behind one scored selection layer.
• 700+ agent skill and production-knowledge files — pipeline directors, creative techniques, quality checklists and per-technology knowledge packs.
• Reference-driven creation — paste a YouTube video, Short, Reel, TikTok or local clip; the agent analyses transcript, pacing, scenes, keyframes and style, then returns 2–3 differentiated concepts with an honest tool path and cost estimate before generating anything.
• Live web research as a first-class stage — 15–25+ searches across YouTube, Reddit, Hacker News, news and academic sources, cited in a structured research brief, before a word of script is written.
• Backlot living storyboard — a local board that fills itself in as the pipeline runs: stages light up, the script lands as a screenplay page, scene cards shimmer while assets generate, and every provider decision and dollar is on the wall. ▶ REPLAY RUN scrubs a finished production end to end.
• Storyboard as a real approval gate — asset generation pauses on a scene-by-scene contact sheet (takes, prompts, per-asset cost, quality scores), so you approve visuals before the render.
• No vendor lock-in — every capability has a local/open alternative next to the premium API.

Download or use

git clone https://github.com/calesthio/OpenMontage.git
cd OpenMontage
make setup

Then open the project in your coding agent and just ask:

"Make a 60-second animated explainer about how neural networks learn"
"Make a 75-second documentary montage about city life in the rain. Use real footage only, no narration, elegiac tone, with music."

Prerequisites: Python 3.10+, FFmpeg, Node.js 18+, and a coding agent. On Windows, if npm install fails with ERR_INVALID_ARG_TYPE, use npx --yes npm install.

Backlot:

python -m backlot open                  # the library — every project on disk
python -m backlot open <project-id>     # one production's live board
python scripts/backlot_simulate_run.py  # watch a simulated run

Reasoning for

This is the first video stack I’ve seen that treats production like engineering with gates, not like prompt roulette — which is exactly the property that makes it usable inside an agent harness (cf. Harness Engineering, Agentic Coding). It also fills the video blind spot from a different angle than the tools I already track (see Extending Claude Code — Tools for Its Blind Spots):

• Video Use edits footage you already have.
• Claude Video lets the agent watch a video.
• OpenMontage produces a video from nothing but a topic — research, script, assets, edit, render.

The zero-key path is the interesting one for experimentation: HyperFrames and Remotion both ship as composition engines, Piper TTS narrates offline, and Archive.org / NASA / Wikimedia supply real footage. That is a complete video without a single paid API call.

What you get with zero API keys

Capability	Free tool	What it does
Narration	Piper TTS	Offline text-to-speech
Open footage	Archive.org · NASA · Wikimedia Commons	Archival and documentary texture
Extra stock	Pexels · Unsplash · Pixabay	Free stock (free developer keys)
Composition (React)	Remotion	Spring-animated scenes, stat cards, word-level captions, TalkingHead
Composition (HTML/GSAP)	HyperFrames	Kinetic typography, promos, SVG character rigs
Post-production	FFmpeg	Encoding, subtitle burn-in, mixing, grading
Subtitles	Built-in	Word-level timing

OpenMontage picks between Remotion and HyperFrames at proposal time and locks it as render_runtime — Remotion for data-driven explainers, HyperFrames for motion-graphics-heavy briefs.

How it works

Every pipeline follows the same flow:

research -> proposal -> script -> scene_plan -> assets -> edit -> compose

Each stage has a dedicated director skill — a Markdown instruction file teaching the agent how to execute that stage. Python provides only tools and persistence; all creative logic, review criteria and quality standards live in readable YAML + Markdown you can inspect and change.

Three-layer knowledge architecture

Layer 1: tools/ + pipeline_defs/     "What exists" — executable capabilities + orchestration
Layer 2: skills/                     "How to use it" — OpenMontage conventions and quality bars
Layer 3: .agents/skills/             "How it works" — external technology knowledge packs

Each tool declares which Layer 3 skills it depends on — a clean example of Progressive Disclosure applied to a tool registry.

Production governance

• Human approval gates are enforced, not suggested — proposal, script, scene plan, assets and publish all pause for sign-off; the checkpoint writer rejects a “completed” gated stage without recorded approval, and superseded checkpoints are archived so the audit trail survives revisions.
• Pre-compose validation — blocks the render when the delivery promise is violated (a “motion-led” video that is 80% stills), slideshow risk is critical, or the renderer family is missing.
• Post-render self-review — ffprobe validation, frames extracted at 4 positions to catch black frames and broken overlays, audio level analysis for silence/clipping, delivery-promise and subtitle checks. Fail the review and the video is never presented.
• Slideshow risk scoring — 6 dimensions (repetition, decorative visuals, weak motion, shot intent, typography overreliance, unsupported cinematic claims) to prevent “animated PowerPoint”.
• Scored provider selection — 7 dimensions: task fit (30%), output quality (20%), control (15%), reliability (15%), cost efficiency (10%), latency (5%), continuity (5%). Winner and alternatives are logged.
• Decision audit trail — every provider, style, music, voice, renderer and fallback choice logged with alternatives, confidence and reasoning.
• Budget controls — estimate → reserve → reconcile, modes observe / warn / cap, per-action approval above a threshold (default $0.50),totalcapdefault$10.

Alternatives considered

• Video Use — agent-driven editor for footage you already shot; reads transcripts for word-level cuts. Complementary, not competing.
• HyperFrames / Remotion — composition engines only; OpenMontage uses both as its renderers rather than replacing them.
• Single-clip prompt tools (Runway, Veo, Kling standalone) — one clip from one prompt; OpenMontage orchestrates them as providers inside a full pipeline.

⚠️ Licence check: AGPLv3, not MIT — meaningful if you ever want to build a hosted service on top of it.

Resources

• 🔗 Repo: github.com/calesthio/OpenMontage
• 🔗 Providers guide (pricing + free tiers) · Backlot README
• 🔗 AGENT_GUIDE.md — the agent contract · PROJECT_CONTEXT.md
• 📖 Related: Video Use · Claude Video · HyperFrames · Remotion · ElevenLabs · Extending Claude Code — Tools for Its Blind Spots

---

Template: tool