How we index images for RAG | Hacker News

Hacker News discussed eager media ingestion for indexing images in RAG systems, so visual content can be searched and retrieved more effectively.

The thread compared upfront image processing with thumbnail and size pre-generation as practical ways to make retrieval pipelines cheaper and faster.

A key concern was that newer LLMs may extract different or additional context from the same image, so indexed outputs may need reprocessing or multi-model combination.

Participants also noted that usage patterns can help prioritize which assets to process first, reducing cost while improving coverage.