what is eidoverse-video fr?

Eidoverse Video is a toolkit that lets an AI coding agent produce finished videos, similar to how a small film studio works. Instead of a human operating editing software directly, a person and an AI agent collaborate: the conversation between them acts like a writers' room, the agent plays the role of filmmaker, and this toolkit provides the sets, cameras, and equipment. The project describes itself as a prealpha release, version 0.01, extracted from an internal production pipeline that has already been used to make hundreds of videos. Everything renders in real time using Deno, a JavaScript runtime, together with WebGPU and the three.js graphics library, keeping almost all the work on the graphics card rather than the processor. Coding agents like Claude Code, Codex, or OpenCode can read the project's own instructions file and understand how to operate the studio without additional setup beyond installing Deno and ffmpeg, a tool for encoding video and audio. Inside the toolkit are building blocks for many parts of a video. Characters use physics based walking, running, climbing, and other movement, driven by an inverse kinematics system that keeps feet planted on uneven ground, plus lipsync that turns spoken audio into mouth movements. There are tools for building creatures and machines out of reusable parts, terrain and grass for outdoor scenes, cloth and water simulations, particle effects like fire, smoke, and sparks, and over thirty visual post processing effects such as film grain, glitch, and neon looks. A separate asset pipeline can search and pull in existing 3D models, images, and environment lighting from external libraries. The main branch runs directly on your own machine with Deno and ffmpeg installed locally, no containers involved. A second branch, called auto, wraps the same toolkit in a Docker container so it can run as an automated subagent inside a larger pipeline, taking in a brief and returning a finished video file. The render engine also checks its own output for common mistakes, such as floating props or frozen faces, before handing back the result.