Show HN: SmartMemeSearch – Search memes collections on your computer(CLIP, OCR) (github.com)

SmartMemeSearch is a desktop app (Show HN) that gives you semantic, multimodal search over local meme collections by combining CLIP-style image embeddings with OCR text extraction. It indexes images (JPG, PNG, WebP, AVIF, HEIC, GIF, etc.), reads any on-image text (including partial matches), and stores optimized embeddings and metadata so you can query by keywords, objects, emotions, topics or the text printed in the meme — all with instant, on-device results and automatic folder scanning to keep libraries up to date. The app ships via Microsoft Store and can be built from source using Visual Studio (open WPF project in the repo). For the AI/ML community this is a compact demonstration of practical, privacy-preserving multimodal retrieval: pairing CLIP-like semantic understanding with OCR enables much richer search than filename or tag-based systems, useful for dataset curation, quick prototyping, content creation, and moderation workflows. Key technical takeaways are on-device embedding + OCR fusion, support for large collections via an optimized queryable database, and broad image-format handling (including animated GIFs). It’s a useful reference for building offline, privacy-first retrieval tools and highlights how relatively small ML pipelines can unlock powerful, real-world multimedia search experiences.