Meme Search
Self-hosted AI-powered meme search engine that uses vision-language models to auto-caption and semantically search your local meme library.
Meme Search is a self-hostable meme search engine that turns your local meme folder into a fully searchable library. It runs vision-language models (Florence-2, SmolVLM, Moondream2) locally to auto-generate descriptions and text embeddings for every image, then lets you find the perfect meme using either keyword or semantic vector search.
Unlike cloud meme libraries such as Imgflip, Giphy, and Tenor, Meme Search keeps your entire meme collection on your own hardware. Drag-and-drop or paste images through the web UI, organize them across multiple directory paths, tag them, and search by meaning ("distracted boyfriend", "this is fine dog", "success kid") rather than filename. A bundled Python image-to-text generator handles captioning end-to-end, or you can point it at any OpenAI-compatible vision API while keeping embeddings and search local.
Built on Ruby on Rails, PostgreSQL with the pgvector extension, and Python, the entire stack ships in a single Docker Compose file for one-command deployment. Other features include bulk description generation, directory rescan, dark mode, and a fast keyword + vector hybrid search powered by Postgres.