Visual Jev

The win is in the execution. Software asks a vision–language model for a decision about an image rather than for prose, and usually asks several questions at once.

Description read from the page itself, because the listing it was found in gave none.