Group near-duplicates

Not exact copies: near-duplicates. Where the dedupe hero finds byte-identical files and deletes them to reclaim disk, this scenario finds images that merely look alike (burst frames, export variants, slight edits, repeat screenshots) and groups them by similarity so you can compare and keep the best. Ten loose look-alikes sit in a jumbled pile; a similarity sweep crosses them, then each group flies together into a labelled box that fades in with a match % and a suggested green "Best" keeper. Nothing is deleted. It rests on four tidy groups, then replays.