TastePrompt

Much has been discussed about Taste with the advent of Generative AI, TastePrompt builds a visual taste profile from your images and lets others prompt with it speeding up the creative process.

Comment

Inspiration

The discourse around AI slop got me thinking about why generated content feels hollow. The model doesn't know your taste. tastePrompt tries to fix that by extracting structured aesthetic signal from a curated image collection and synthesizing it into a persistent fingerprint that Claude can actually reason from.

What it does

Run a Tumblr archive through Claude vision and it extracts structured aesthetic signal from each image, then synthesizes it into a taste fingerprint. You can query that fingerprint for personalized recommendations across any domain, or load two profiles and blend them to find what two people would both love or where their tastes diverge. Users can use to prompt the profiles fingerprint to help with creative decisions.

How we built it

React frontend with no backend, all API calls client side. Claude vision extracts structured JSON from each image in a batch script, a synthesis pass generates the fingerprint, and profiles are static JSON bundled at build time. Streaming responses via the Anthropic SDK, cover art pulled from OMDB, iTunes, and Open Library.

Built With

Share this project:

Updates