Tag
Kapa.ai describes their approach to indexing images for RAG by using a cheap vision model to generate text descriptions at indexing time, avoiding query-time vision costs, resulting in better answers with minimal per-query overhead.