CLI tool for semantic image search, locally without using third party APIs.
Powered by node-mlx, a machine learning framework for Node.js.
sisi.mp4
GPU support:
- Macs with Apple Silicon
CPU support:
- x64 Macs
- x64/arm64 Linux
(No support for Windows yet, but I might try to make MLX work on it in future)
For platforms without GPU support, the index command will be much slower, and will take many hours indexing tens of thousands of images. The index is only built for new and modified files, so once your have done the initial building, updating index for new images will be much easier.
Install:
npm install -g @frost-beta/sisiCLI:
━━━ Semantic Image Search CLI - 0.0.1-dev ━━━━━━━━━━━━━━━━
$ sisi <command>
━━━ General commands ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
sisi index <target>
Build or update index for images under target directory.
sisi list-index
List the directories in the index.
sisi remove-index <target>
Remove index for all items under target directory.
sisi search [--in #0] [--max #0] [--print] <query>
Search the query string from indexed images.Build index for ~/Pictures/:
sisi index ~/Pictures/Search pictures from all indexed images:
sisi search 'cat jumping'Search from the ~/Pictures/ directory:
sisi search cat --in ~/Pictures/Search images with image:
sisi search https://images.pexels.com/photos/45201/kitty-cat-kitten-pet-45201.jpegIt works with local files too:
sisi search file:///Users/Your/Pictures/cat.jpgThe index is built by computing the embeddings of images using the CLIP model, and then stored in a binary JSON file.
Searching the images is computing cosine similarities between the query string and the indexed embeddings. There is no database involved here, everytime you do a search the computation is done for all the embeddings stored, which is very fast even when you have tens of thousands of pictures.
The JavaScript implementation of the CLIP model is in a separate module: frost-beta/clip.
MIT