Turn AI coding assistants into video production studios
Paste a link to a YouTube Short you like and say, "Make a quantum mechanics video with this vibe," and the AI handles everything from scriptwriting to asset search, editing, and final rendering. OpenMontage is an open-source system that converts AI coding assistants into video production studios. It supports a method that searches for actual video clips from free archives and places them on the timeline, rather than using tricks to make still images appear to move.
Reference-based production flow
Instead of struggling with prompts on a blank screen, you can use existing videos as reference material. The system analyzes the rhythm, hook, structure, and tone of the video, then proposes two to three differentiated concepts with changed topics and visual treatments. It also displays estimated costs and examples of what can be created with currently available tools. This process is not a list of speculative prompts but a concrete production plan that maintains the strengths of the reference video while transitioning to a new topic.
Backlot board and real-time approval gates
It is difficult to grasp production progress from chat text alone, but the backlot board is a local board that fills up automatically while the pipeline runs. The script is displayed on the scenario page, scene cards glow while assets are being generated, and all provider decisions and expenditures are recorded on the wall. In particular, during the asset generation stage, a contact sheet for each scene is displayed, serving as a gate to approve visual assets before rendering. Production is paused until the user responds via chat, preventing issues that are hard to detect in post-production.
Real video production with free tools
You can create videos with actual motion using free open sources like Piper TTS, Archive.org, NASA, and Wikimedia Commons, even without paid API keys. The documentary montage pipeline builds a CLIP-searchable corpus, searches for actual motion clips from free stock videos and archives, and edits them onto the timeline. This is not a trick to make still images appear to move, but a process of intentionally editing actual video clips to render a finished work. Remotion and HyperFrames are automatically selected based on data-driven explanatory videos and motion graphics-centric briefs to use the optimal rendering engine.
Installation environment and requirements
An AI coding assistant capable of reading files and executing code, such as Claude Code, Cursor, Copilot, Windsurf, or Codex, is required. After setting up the environment with the make setup command, open the project folder in the AI assistant and describe the video you want. API keys are optional, and adding more keys allows you to utilize various paid and free providers such as FLUX, Veo, Kling, and ElevenLabs. To use local GPU models, run make install-gpu and add the relevant settings to the .env file.
Verification system to maintain production quality
All production processes follow a structured flow of research, proposal, script, scene planning, assets, editing, and compositing. Each stage has a markdown-format director skill file that precisely guides the AI on how to execute that stage. Multi-point self-reviews, including ffprobe verification, frame sampling, audio level analysis, and subtitle checks, are performed before and after rendering. All provider selections are scored across seven dimensions to make decisions, and this entire decision process is recorded in an auditable log. Budget management features prevent unexpected costs by setting cost estimates before execution, spending caps, and approval thresholds per action.
