Ok, some img2img refinement might fix that. But when I do "bigger" edits, something Kontext is really made for, it gets the overall anatomy wrong. Heads are too big, the torso is too small.