Trace.moe is a visual search engine that identifies anime scenes from screenshots by returning the source anime, episode number, and exact timestamp.
The tool solves the problem of finding where an anime screenshot originates by using content-based image retrieval to match uploaded images against a database of anime video frames. It analyzes image features and generates vectors that enable fast similarity matching across its indexed anime content.
The project is designed for anime fans and developers who need to identify scenes from screenshots, and for anyone building bots or applications that require anime scene lookup. It provides multiple integration points including a web interface, a REST API for programmatic access, a Telegram bot, and a browser extension for convenient image submission. The system is modular and can be self-hosted using Docker Compose, with support for Windows via WSL2. Self-hosting requires organizing video files by AniList ID in a designated directory structure, and the system automatically indexes new video files placed in the configured path.
The project maintains active development across its component repositories with automated Docker image builds for both the web server and API server. The codebase includes tooling for image feature analysis and vector generation, and provides documentation through an API reference site and past presentation slides explaining the system architecture.
