OpenMotion, Under Apache 2.0, Builds Video in Code From a Reference and Hits 10,000 GitHub Stars
OpenMotion is an open studio that builds videos in code, with no video generators involved. You paste a link to a reference video and upload a brand kit. The agent breaks the video down into shots, transitions and rhythm, then rebuilds it in HTML, CSS and JavaScript. The project is licensed under Apache 2.0, works with 22 coding agents and has collected 10,000 stars on GitHub. The result stays editable and does not turn into a throwaway file.
AI-processed from TG: @cgevent (Метаверсище и ИИще); edited by Hamidun News
Typical neural video generation produces a finished file. Changing a logo, a color or the ending in that file almost always means generating the video again. The OpenDesign team (the nexu-io organization on GitHub) takes a different approach. Its project OpenMotion builds motion graphics in code, so any detail can be changed afterward. The project is open, licensed under Apache 2.0, and the post's headline cites a milestone of 10,000 GitHub stars.
How it works
The workflow is simple. You find a video whose style you want to copy and paste a link to it. Then you upload a brand kit: logo, colors, fonts. From there, an agent takes over. It breaks the reference down into shots, transitions and rhythm, generates any missing assets, and rebuilds the video as ordinary HTML, CSS and JavaScript code.
Say a marketer needs a promo video for a new product, paced like a competitor's video they liked. Before, that meant either hiring an editor or making a dozen attempts with a video generator. Here, the reference becomes a project you can open and edit.
The studio includes a timeline. You can rearrange clips and change products or endings with prompts. The finished video exports to MP4 for YouTube, Shorts, Reels and TikTok. The same project can be exported in several formats without rebuilding it from scratch.
What's in the project
OpenMotion positions itself as an open alternative to Claude Motion and CapCut. The repository includes a library of more than 300 Motion.md guides. These are text instructions that tell the agent how to build particular animations and techniques. The project supports 22 coding agents, including Claude Code, Codex, DSH and WorkBuddy. Running it requires Node.js version 22.18 or newer.
The "video as code" approach works because coding agents are already good at writing and rewriting text-based projects. When a video is described in HTML, CSS and JavaScript, you can store it in Git, compare versions, and swap in different texts and products. The same build can be run across a hundred variants, for example for different markets or products.
What to keep in mind
The project description does not make clear how accurately the agent reproduces complex references. The source includes no comparative measurements against Claude Motion or CapCut. GitHub stars show community interest, not the quality of the output. In addition, code suits graphics, text, transitions and product interfaces well. It will not replace live footage or photorealistic scenes: for those, the agent adds assets, and that is generation again.
There is also a legal question. If you use someone else's video as a reference, the team should check what exactly is being reproduced: rhythm and structure, or recognizable elements of another brand. That is a matter of company policy, not of the tool.
What to do in practice
For developers who already work with Claude Code or Codex, the barrier to entry is low: the right version of Node.js and the repository are enough. Product and marketing teams should run one or two typical videos through it, such as a release announcement or a short ad, and compare the time and revisions against their usual editing. For founders of small products, the project is interesting as a way to make videos in a consistent style without a studio. The Apache 2.0 license allows using the code in commercial projects and adapting it to your needs.
The source notes that this is already the second solution of this kind in two days: the market is clearly shifting toward video that is assembled rather than generated. An open implementation lets you test the idea without subscriptions, and you can evaluate it today.
- Nov 18, 2026Willison Releases Three External Sandboxes So Agent-Run Code Stays Off Your Laptop
- Nov 18, 2026Dreamina Launches Motion Studio: Motion Graphics Written as Code From a Prompt
- Nov 18, 2026Nvidia's data-center revenue jumps 110% to $107.6 billion: the AI pause didn't hurt
- Nov 18, 2026AI Finds Unknown V8 Vulnerability for $4.24 in 13 Minutes, After Failing for $350 in February
Key Questions Answered
How it works?
The workflow is simple. You find a video whose style you want to copy and paste a link to it. Then you upload a brand kit: logo, colors, fonts. From there, an agent takes over. It breaks the reference down into shots, transitions and rhythm, generates any missing assets, and rebuilds the video as ordinary HTML, CSS and JavaScript code. Say a marketer needs a promo video for a new product, paced like a competitor's video they liked.
What's in the project?
OpenMotion positions itself as an open alternative to Claude Motion and CapCut. The repository includes a library of more than 300 Motion.md guides. These are text instructions that tell the agent how to build particular animations and techniques. The project supports 22 coding agents, including Claude Code, Codex, DSH and WorkBuddy. Running it requires Node.js version 22.18 or newer. The "video as code" approach works because coding agents are already good at writing and rewriting text-based projects.
What to keep in mind?
The project description does not make clear how accurately the agent reproduces complex references. The source includes no comparative measurements against Claude Motion or CapCut. GitHub stars show community interest, not the quality of the output. In addition, code suits graphics, text, transitions and product interfaces well. It will not replace live footage or photorealistic scenes: for those, the agent adds assets, and that is generation again. There is also a legal question.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.