About this project

Diffusion Studio is an open-source video editor designed to work with AI coding agents such as Claude Code, Codex, Cursor, Copilot, and Gemini CLI. The core concept is bidirectional editing: changes made on the visual canvas update the underlying code, and code changes redraw the canvas. Projects are authored as JSX using SolidJS modules, where each element carries an ID that enables write-back from canvas interactions to the corresponding code. The desktop app registers an MCP server with the user's agent, providing tools for media inspection, timeline editing, and rendering. A CLI (dapi) exposes the same capabilities as shell commands, including probing media metadata, extracting frames, generating filmstrips and waveforms, transcribing audio, and querying multimodal models about video content. Every command returns a single JSON object suitable for piping and programmatic use. Supported workflows include cutting clips, removing filler words, adding subtitles, applying color correction and filters, creating motion graphics (explainers, promos, title sequences), generating assets (images, video, voiceover) declared in code, clipping highlights for social media, and video understanding tasks like summarization and scene search with timestamps. The project is structured as a monorepo with packages for the headless runtime (koota ECS, no DOM dependency), JSX authoring API, asset management with content hashing, offline encoding via mediabunny, and a SolidJS universal renderer reconciler. A headless mode is available for automated pipelines. Available for macOS (Apple Silicon) under MPL-2.0, with brand assets excluded from the license.