A small Django experiment which uses an LLM (Gemma 3 12b) to tag and describe wallpapers. Then make them searchable using a Chroma DB, a vector database.
This provides natural search which is suitable to find wallpapers by subject, style, atmosphere, color palette etc. Surprising is that analyzing about 500 images or so with 2 calls each (tags + description), costed less than 5 cents!
