You know what this deserves a GIF

And just to prove I am not gen spamming:

Manual tidy up 🫩

I think it is something with my config but I always get weird results with having Hires enabled by default, so I normally do it after the original image is generated. I pretty much always have to touch up eyes in Photoshop though or every character would look straight out of Oshi no Ko 😅
Will take a look at Face Detailer but every Bbox model appears to be blocked in my region at least through Stability Matrix UI 🤔
Ah, I see! Now I understand. Yeah, the reason I asked was because Inference is built on ComfyUI, and without FaceDetailer, ComfyUI typically doesn't produce good looking faces. So I wondered if it might've been due to your use of a refiner, or the high number of steps, or simply the checkpoint model you used. XD