yurijmikhalevich/rclip
Semantic photo search for the command line
What it solves
rclip solves the problem of searching through large local photo libraries without needing to manually tag images or rely on cloud-based services. It allows users to find specific photos using natural language descriptions or example images while keeping all data private and offline.
How it works
The tool uses the OpenCLIP ViT-B/32 AI model to create a semantic index of images. When a user enters a query, rclip compares the semantic embedding of the text or image query against the indexed features of the local photo library to find the most similar matches. It supports incremental indexing, meaning it only processes new or changed images on subsequent runs to save time.
Who it’s for
It is designed for command-line users on Linux, macOS (Apple Silicon), and Windows who want a private, fast, and powerful way to search their local image collections.
Highlights
- Multi-modal Search: Supports natural-language queries, reverse image-to-image search, and combined queries (mixing text and images).
- Arithmetic Queries: Allows weighting and subtracting terms (e.g.,
"2:golden retriever" + "./pool.jpg" - fruit). - Privacy-First: Operates entirely offline with no cloud uploads.
- Broad Format Support: Indexes standard formats like JPG, PNG, and WebP, with experimental support for RAW files (ARW, CR2, DNG).
- Terminal Integration: Provides inline image previews for compatible terminals like iTerm2 and WezTerm.