Search photos by describing them
Search your photos by describing them — "children on a beach", "snow at night" — and Portman2 ranks the grid by how well each photo matches, best first.
Type the description into the search box, and the last row of the suggestions offers Photos that look like "…". Pick it to rank the grid you were on.
When it is offered
It needs the keyword model: download and switch on Keywords under Machine learning and let the library be analysed. The row is offered for any description of two or more words, and for a single word that is not already the name of a tag, album or date.
The results
A bar over the grid reads Looks like: …, with the place it was asked from, such as an album or a year. The status line counts the photos that match. Photos that the model has not looked at yet are not in the ranking; the analyse pass reaches every photo in time.
Click the ✕ on the bar to go back to where you asked from.
This is a ranking, not a filter: it orders the photos by how well they match, rather than keeping only the ones with a word.