Add exclusion/Boolean operators to search

[DumfriesDik]

I'm really enjoying Lap so far — I've just started indexing my photo collection and already find the natural-language search very useful.

One feature I'd love to see is support for Boolean or exclusion operators in search.

For example:

bird flying over field -sky

to find images matching "bird flying over field" while excluding images containing sky.

I've also tried natural-language phrases such as "bird flying over a field with no sky visible", but the search still seems to match strongly on the individual terms rather than treating "no sky" as an exclusion.

Even simple support for -term, or perhaps NOT, would make it much easier to narrow down large sets of results.

Thanks for developing Lap — I'm enjoying it!

[julyx10]

Thanks for the detailed feedback and for indexing your collection with Lap, glad the semantic search has been useful!

Boolean/exclusion support is a great idea, and it's something we can build, though it's worth explaining why it doesn't already work the way you'd expect. CLIP's text encoder is known to handle negation poorly — phrases like "with no sky visible" tend to still activate on "sky" rather than suppress it, since the model wasn't trained to understand negation semantically. So natural-language negation isn't something we can fix by prompt tweaking alone.

The more reliable path is explicit syntax handled at the application layer: parsing something like bird flying over field -sky into a positive query (ranked via CLIP as usual) plus an excluded term, then filtering or penalizing results that score too high against the excluded term's embedding separately. This is a "soft" exclusion (similarity-based, not exact), but it should meaningfully narrow large result sets the way you're describing.

I don't have this scheduled for an immediate release, but I've noted it as a search enhancement to pursue — starting with simple -term syntax support. Appreciate you writing it up with a concrete example, it makes it much easier to prioritize. Feel free to open a GitHub issue if one doesn't exist yet, so it stays tracked.