Block out a scene's camera and character positions before generating AI video.
Direct Claude live to build a storyboard while watching it happen in the browser.
Export a motion and camera reference clip to feed into a video generation model.
Save reusable sets and shots as a project for future scenes.
| dennisonbertram/emulsion | adguardteam/recovery | adm1nsys/safari-ai-extension | |
|---|---|---|---|
| Stars | 8 | 8 | 8 |
| Language | JavaScript | JavaScript | JavaScript |
| Last pushed | — | 2018-03-16 | — |
| Maintenance | — | Dormant | — |
| Setup difficulty | easy | easy | hard |
| Complexity | 3/5 | 2/5 | 2/5 |
| Audience | vibe coder | general | developer |
Figures from each repo's GitHub metadata at analysis time.
Just needs Python 3 and a browser, Node is only required for the optional headless export helpers.
Emulsion is a tool for planning out shots for AI generated video, built as a skill for Claude Code. It gives Claude a small browser based 3D app where scenes can be blocked out using simple shapes, like boxes, spheres, and blocky stand in people, along with a movable camera. The idea is that you describe a scene in plain language to Claude in your terminal, for example asking for a funeral procession on a winding road shot with an old style camera, and Claude builds that scene live in the app while narrating what it is doing. You can watch it happen, and you can also reach in and drag things yourself, since every change either of you makes shows up for the other. Objects can be given keyframes so they move over time, and the camera can be keyframed too so it pans, zooms, or tracks a subject like a real shot would. Once a shot is blocked out, you can export the camera's movement as a short video clip. That clip is meant to be fed into an AI video generation model as a motion and camera reference, alongside separate reference images and a written prompt describing the scene beat by beat, so the AI model has a clearer sense of the intended camera work rather than guessing from text alone. Under the hood it is a simple setup with no build step: a three.js app running in the browser, and a small Python server that keeps everyone's changes in sync by having the agent send full scene updates and the browser poll for them. Projects, including shots, sets of scenery, and generated assets, are saved as plain JSON files. This project is aimed at people experimenting with AI filmmaking who already use Claude Code, particularly those trying to get more deliberate camera and blocking control out of AI video generators rather than relying purely on text prompts. It is released under the MIT license.
A Claude Code skill with a 3D browser app for blocking out camera shots to guide AI video generation.
Mainly JavaScript. The stack also includes JavaScript, three.js, Python.
Licensed under MIT, free for anyone to use and modify.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly vibe coder.
This repo across BitVibe Labs
Verify against the repo before relying on details.