cluster them by image embedding, then use an expectation embedding as a protype of each cluster to generate a folder name (e.g. w CLIP interrogator)
check metadata, might already have prompt embedding?
- plugin to auto generate prompt tags (for e.g.lightroom smart queries) from exif
- if can recover prompt, separately cluster on prompt
clip-text-similarityand imageclip-image-similarity