Open Montage is a free, open-source repo that gives your AI coding agent a complete video production studio. Twelve pipelines, 52 tools, and most finished videos cost under two dollars to produce. Setup takes ten minutes.
Producing video for your business usually means one of three things. Hire a freelancer and wait. Learn editing software you do not have time to master. Or keep pushing the project back.
Open Montage changes the equation. Describe the video you need in plain language. Your agent reads the description, selects the right pipeline, plans every production step, and renders the file. It runs its own quality check before it hands it back.
The system is free. You own everything it produces. The first run can cost less than a coffee.
The Tool: What Open Montage Actually Is
Open Montage is the first open-source agentic video production system. It does not wrap one video generation model. It connects your agent to 52 tools across five categories: video generation with 14 providers, image generation with 10 providers, text-to-speech with 4 providers, audio and music tools, and a full post-production stack using FFmpeg, Remotion, and related free tools.
Your agent gets access to all of it through a three-layer knowledge system. Layer one holds the tool definitions and pipeline specs. Layer two holds skills files describing how to execute each step. Layer three holds 52 technology knowledge packs, one per tool, so the agent understands the capabilities and limits of each provider before it chooses one.
The system works with Claude Code, Cursor, GitHub Copilot, Codex, and Windsurf. Each assistant gets a dedicated instruction file. All instruction files point to a shared AGENT_GUIDE.md. The workflow stays consistent regardless of which agent you use.
| Detail | Value |
|---|---|
| Pipelines | 12 |
| Tools | 52 |
| TTS providers | 4 (ElevenLabs, Google TTS, OpenAI TTS, Piper local) |
| TTS voices | 700+ via Google TTS alone |
| Budget default | $10 per run, configurable |
| License | GNU AGPLv3 |
| Cost range | $0.15 to $1.33, measured production runs |
Setup: Three Steps
Requirements before you start: Python 3.10 or higher, FFmpeg, and Node.js 18 or higher. You also need at least one AI coding assistant installed.
Clone the repo and run the automated setup:
Copy this.
git clone https://github.com/brokenthumbsmedia/openmontage cd openmontage make setup
If make is not available on your system, run the manual path:
Copy this.
pip install -r requirements.txt npm install cp .env.example .env
Install Piper TTS if you want a fully local, free text-to-speech option:
Copy this.
pip install piper-tts
Open .env and add API keys for any cloud providers you want to use. You need zero API keys to get started. Piper TTS, FFmpeg, and Remotion all run locally at no cost. Cloud providers like ElevenLabs, Google TTS, and the video generation services each need their own key, added only if you choose to use them.
Done.
How to Use It: From Description to Finished File
Open your AI coding assistant inside the cloned repo directory. The agent reads the instruction file and loads the full tool library and pipeline definitions.
Describe your video in plain language. "A 90-second animated explainer showing how our client onboarding process works, professional tone, no live footage." The agent reads the description, selects the matching pipeline, and builds a production plan: shot list, tool assignments, order of operations, and estimated cost.
Review the plan. Approve it as written or request changes before any rendering starts. The default budget cap is $10 per run. You set per-action approval thresholds so the agent asks before spending above your floor.
Once you approve, the agent executes each step in sequence. It does not ask for input at every stage unless the plan requires a decision point.
When rendering completes, the post-render review runs automatically. The agent audits the output file and hands it back with the review report attached.
The 12 Pipelines
| Pipeline | Best For |
|---|---|
| Animated Explainer | Process walkthroughs and product demos with motion graphics, no footage required |
| Animation | Stylized short-form content and character-driven storytelling |
| Avatar Spokesperson | Talking-head videos without a camera or recording session |
| Cinematic | High-production-value brand films, 21:9 aspect ratio output |
| Clip Factory | Batch-cutting long recordings into short clips for distribution |
| Documentary Montage | Long-form content using Archive.org, NASA, or Wikimedia Commons footage |
| Hybrid | Mixed live footage and generated visuals in a single output |
| Localization and Dub | Language-swapped versions of existing videos with re-voiced audio |
| Podcast Repurpose | Audio-to-video conversion with captions, b-roll, and branding |
| Screen Demo | Software walkthroughs with annotations and cursor tracking |
| Talking Head | Recorded founder or team video with polished output |
One additional specialized pipeline is documented in the repo. The confirmed total is 12.
Use-Case Pairing: Which Pipeline for Which Job
If you need to explain a service or internal process and have no footage, use Animated Explainer. The agent generates every visual from your description alone.
If you have a recorded call, webinar, or training session sitting unused, use Clip Factory. Your agent segments the recording and isolates the strongest moments. You define clip length and target platform. The agent handles the cuts.
If you sell into multiple languages, use Localization and Dub. Upload the original video and specify the target language. The agent translates, re-voices the audio using your chosen TTS provider, and syncs it to the existing timeline. You do not re-record anything.
If you want to appear on camera but cannot book studio time, use Avatar Spokesperson. Feed the agent your script. The pipeline manages generation and outputs a polished result.
Podcast Repurpose turns any audio recording into a captioned social video. Documentary Montage pulls stock footage from Archive.org, NASA, and Wikimedia Commons. Both pipelines include built-in source access at zero additional licensing cost.
The output profiles cover the platforms you actually publish to: YouTube at 1080p, 4K, and Shorts format; Instagram Reels and Feed; TikTok; LinkedIn; and Cinematic 21:9 for brand use.
What the Agent Quality Controls Actually Enforce
The system runs two validation passes, not one.
Before the agent begins composing, a pre-compose validation checks the entire plan. Broken tool dependencies, missing files, or a budget that would exceed your cap all block the run. The agent does not start rendering until the plan clears every check.
After rendering completes, the post-render review runs ffprobe against the output file, extracts frame samples to verify visual consistency, and runs audio analysis to check sync and levels. The agent scores the result. Anything below the quality threshold gets flagged before it reaches you.
Provider selection is not random. A scored ranking system evaluates each available tool across seven dimensions and assigns the highest-ranked provider for each task. If your preferred provider is unavailable or over budget, the system selects the next-best option automatically.
This is different from a tool that runs one model and returns whatever comes out. The agent is choosing, validating, and reviewing at every stage.
What Does This Cost Per Video?
Every run starts with a $10 default budget cap you can adjust per project. The measured cost examples give you the real range: $0.15 for a Ghibli-style animation, $0.69 for a product ad, and $1.33 for a 60-second animated short. These are recorded production costs from actual runs, not projections. For teams producing ten videos a month at the $1.33 average, total monthly production cost stays under $15. Running local-only tools, meaning Piper TTS, FFmpeg, and Remotion, brings variable cost to zero for any project that does not require cloud generation.
Honest Limits
Video duration scales with compute time and cost. Longer videos cost proportionally more and take longer to render. Test your chosen pipeline on a short version first before committing to a full-length run.
Local video generation models require a GPU. If you run entirely on cloud providers, this requirement does not apply. If you want local generation at no API cost, the hardware requirement is real and non-negotiable.
The license is GNU AGPLv3. If you distribute software that incorporates Open Montage, your distributed software must also be open-source under the same license. Using Open Montage inside your own business to produce internal or marketing content is not affected by this clause.
The repo is under active development. Pipeline specs and tool counts will change between commits. Pin a specific commit if you need a stable configuration across a repeatable production workflow.
Close: One Action, Ten Minutes
Clone the repo. Run make setup. Open your agent inside the project folder.
Describe a video you have been putting off. A process walkthrough. A product demo. A repurposed podcast episode that has been sitting on your hard drive for three months.
If you do not like the result, you spent ten minutes. If it works, you just cut a production task that would have taken hours and cost you more than a dollar and change. The next video is ready whenever you are.