neonwatty/meme-search

Uses semantic search powered by Postgres pgvector to index meme content via local AI vision-language models (Florence-2, SmolVLM, Moondream2), extracting both text and visual features without external API calls. Features include bulk description generation, manual metadata editing, tag-based filtering, directory organization across multiple paths, and drag-and-drop web upload with automatic duplicate handling. Built with Ruby on Rails backend and includes configurable model selection from 200M to 2B parameters for resource-constrained environments.

640 stars. Actively maintained with 1 commit in the last 30 days.