There is so much to like about the way you’ve enabled search on the collection! I’m curious if you are considering how to combine the visual dimensions with key words (e.g., a search for “woman” in objects that are blue and have concentrated light) and more generally how you interpret the user’s intent when more than one dimension/facet is chosen. Is it always the intersection of the selected values, meaning there will be fewer results, or are there times when the results should be the union of search terms?
As for the search results themselves, showing large images is a good choice. I would like to see the number of results because that may influence how I refine my search. Have you also thought about ways to present degrees of similarity between & among images based on the user’s search?