AIToolScan

OpenMontage — The Open-Source, Agentic Video Production System

OpenMontage (AGPL-3.0, by calesthio / Calesthio AI Labs) turns your AI coding assistant into a full video production studio. It is not another single-clip text-to-video model. It is an end-to-end production pipeline that runs the same structured workflow a real production team follows — research, proposal, script, scene planning, asset generation, editing, and final composition — but automated by the agent you already use. The pitch is that you get a finished, quality-gated video, not a 5-second clip you have to edit yourself.

The defining architectural choice is agent-first: there is no proprietary code orchestrator. Your AI coding assistant IS the orchestrator — Python provides the tools and persistence, while all creative decisions, review criteria, and quality standards live in readable Markdown skill files and YAML pipeline manifests that you can inspect and customize. It ships dedicated instruction files for Claude Code (CLAUDE.md), Cursor (CURSOR.md), GitHub Copilot, Codex, and Windsurf, all pointing to a shared AGENT_GUIDE.md agent contract. As of the current build it is trending as #1 Repository of the Day on GitHub with roughly 66k stars.

What is in the box

• 12 production pipelines: Animated Explainer, Animation, Avatar Spokesperson, Cinematic, Clip Factory, Documentary Montage, Hybrid, Localization & Dub, Podcast Repurpose, Screen Demo, and Talking Head — each a complete idea-to-video workflow.
• 100+ production tools: video generation, image creation, text-to-speech, music, audio mixing/dubbing, subtitle generation, upscaling, background/face restoration, color grading, scene detection, and transcription.
• 60+ provider integrations: cloud APIs, local GPU models, free stock libraries, and open archives, all behind one scored selection layer. Swapping providers is free — no vendor lock-in.
• 700+ agent skill and knowledge files: pipeline directors, creative technique guides, quality checklists, and deep technology knowledge packs that teach the agent how to use every tool like an expert.
• Reference-driven creation: paste a YouTube/Reel/TikTok clip you like and the agent analyzes its transcript, pacing, and style, then returns 2–3 differentiated concepts with an honest tool path and cost estimate before producing anything.

Real footage, not just “animate an image”

Most “free AI video” stacks quietly mean “Ken Burns over a still image.” OpenMontage can do that, but its Documentary Montage pipeline builds a CLIP-indexed, semantically searchable corpus from Archive.org, NASA, Wikimedia Commons, and optional free-key sources like Pexels, Unsplash, and Pixabay — then cuts together actual motion footage into a finished timeline. Web research is a first-class stage: before writing a single line of script, the agent runs 15–25+ live web searches across YouTube, Reddit, Hacker News, and academic sources and cites everything in a research brief, so videos are grounded in real, current information rather than hallucination.

Provider menu (behind one scored selector)

• Video (20+): Kling (via fal.ai + official), Google Veo 3.1, Runway Gen-4, Seedance, MiniMax/H3, Grok Imagine, Higgsfield, HeyGen gateway — and free local GPU options WAN 2.1/2.2, Hunyuan, CogVideo, LTX-Video (via ComfyUI or Modal).
• Image (15+): FLUX, Google Imagen 4, GPT Image 2, Seedream, Nano Banana, Recraft — plus free local Stable Diffusion.
• TTS (10+): ElevenLabs, OpenAI, Google (700+ voices), Azure, and free offline Piper. Music via Suno AI and ElevenLabs Music/SFX.
• Composition engines:Remotion (React/Node.js programmatic video) and HyperFrames (HTML/CSS/GSAP motion graphics) for local rendering, with FFmpeg as the always-free assembly backbone. The runtime is chosen at proposal time and locked through the run.

Quality gates and budget governance

• Delivery-promise enforcement blocks “slideshow-looking” renders; pre-compose validation catches broken plans before wasting GPU time; and a mandatory post-render self-review (ffprobe + frame extraction + audio analysis + subtitle checks) must pass before you ever see the output.
• Budget governance: estimate cost before execution, reserve budget before the call, reconcile actual spend after, and pause for per-action approval above a threshold (default $0.50) with a total cap (default $10). The repo’s own showcase videos list real costs of roughly $1–5 each.
• Backlot living storyboard: a local board that fills in as the pipeline runs, turning asset generation into a scene-by-scene approval gate so you sign off on visuals before the render, with a full scrubbable replay of every completed run.

Quick start

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

Prerequisites: Python 3.10+, FFmpeg, Node.js 18+, and an AI coding assistant (Claude Code, Cursor, Copilot, Windsurf, or Codex). Every API key in .env is optional — with zero keys you still get Piper narration, free archival footage, Remotion/HyperFrames rendering, FFmpeg post-production, and auto-generated word-level subtitles. Add keys (Fal, Kling, ElevenLabs, Runway, Suno, etc.) to unlock premium cloud generation, or make install-gpu to run free local video models on your own GPU.

Honest limits

• It requires an AI coding assistant to run. The agent is the orchestrator; this is a knowledge-and-tool layer, not a standalone GUI app. If you want a one-click consumer editor, this is not that.
• Very large and young. 700+ skill files and 100+ tools is a lot of surface area; the project is built nights-and-weekends on an AGPL-3.0 license, so some pipelines and providers are less battle-tested than others.
• Cloud generation costs money. The zero-key path makes real videos, but premium cinematic quality leans on paid provider APIs and a GPU for the free local models. Budget for it.
• AGPL-3.0 is a strong copyleft license. If you build a network service on top, AGPL source obligations apply — review it before commercial use.
• Local-LLM operation is not shipped yet. Full offline pipelines via Ollama/LM Studio are listed as coming soon; today you generally need a cloud LLM for the assistant itself.

Links

GitHub: https://github.com/calesthio/OpenMontage · Website: https://openmontage.video · Provider/pricing guide: docs/PROVIDERS.md in the repo · Community: GitHub Discussions · Build updates: @calesthioailabs on X · License: GNU AGPL-3.0