ImageCluster

ImageCluster is a local web application for analyzing digital image collections with visual embeddings, CLIP-like multimodal models, UMAP/t-SNE projection maps, optional clustering, and semantic text search. The goal of ImageCluster is to give non-technical users a practical interface for exploring image collections and comparing different embedding models on those collections. The project is inspired by Lev Manovich's ImagePlot software - https://github.com/culturevis/imageplot - and adapts that tradition of visual cultural analytics to current embedding-based image analysis. Images are read from the local img/ folder and output files are written under output/. Features: Scan a local image folder. Generate image embeddings with OpenCLIP, CLIP, SigLIP, SigLIP 2, MetaCLIP, MobileCLIP and other supported models. Create interactive UMAP or t-SNE projection maps. Explore the projection as points or image thumbnails. Compare how different embedding models organize the same image collection. Run natural-language semantic search over image embeddings. View ranked search results with thumbnails and image detail previews. Optionally run K-Means clustering. Export PNG charts, projection TSV files, search TSV files, sessions and debug reports.