Your meme folder, now searchable by vibes

What it does

MemeMeow is a natural-language search engine for meme collections. You describe a scene or emotion and it returns matching reaction images using embedding models, eliminating the need to remember filenames or tags. The core is a Streamlit app, but the project also ships a public API and a community pack system so the library can grow beyond its initial seed data.

The interesting bit

The project treats meme curation as infrastructure. It defines a three-tier manifest format—community group, repository, and individual meme library—that lets users import, sync, and share packs without downloading every image upfront. There is also an ecosystem of third-party clients—an iOS Shortcut, a browser userscript, and even an IME plugin—suggesting the maintainers see meme retrieval as an OS-level utility.

Key highlights

• Uses embedding models to match natural-language queries against image descriptions or filenames.
• Supports VLM auto-labeling to avoid manually tagging hundreds of images.
• Offers a public API at api.zvv.quest/search with simple q and n parameters.
• Community resource packs use a manifest hierarchy for incremental updates and lazy image loading.
• Ecosystem includes web UI, iOS Shortcut, Tampermonkey script, and QQ bot integrations.

Caveats

• The README warns of a Streamlit-specific memory leak: refreshing the page after a code update regenerates ImageData and leaks memory; this only triggers during development, but it is a visible rough edge.
• The default dataset is a specific collection of Zhang Weiwei memes sourced from Zhihu, so out-of-the-box utility depends on your interest in that corpus.
• The hosted web demo does not support importing custom resource packs; you need a local deployment or the API for full flexibility.

Verdict

If you curate reaction images for a community or just want to stop scrolling through a Downloads folder full of unnamed JPEGs, this is a pragmatic, if niche, search layer. Everyone else can safely ignore it unless they happen to need a ready-made Zhang Weiwei meme API.