lukutar
Viewing sample resized to 41% of original (view original) Loading...
Parent: post #187636 (learn more) show »
Description

So my little side experiment turned out to actually work: trained a YOLO26 model to actually segment hands in anthro art, trained it on a bunch of AI-generated images - and it holds up surprisingly well. Stock human-hand detectors choke on paws, this one finds them even in weird poses and hands ADetailer a proper per-hand mask instead of a rectangle. Model's in the Sources if you want to try it.

Blacklisted

    There are no comments.