Anonymous
8/24/2025, 2:59:28 AM
No.106363546
I don't know how SDXL works or if it has some max intelligence or storage capacity, but there are 2 things checkpoint makers should add and improve urgently.
>1- Add more secondary objects to their datasets! Background scenes like everyday objects, animals, landscapes, forests etc.
>2- Add more slice of life scenes to their datasets!
I find it hilarious that any model has the intelligence and capability to generate any NSFW scene or position but when I put in tags like "character cooking, holding a frying pan" that's when the real problems start. Character grabs the pan but with a 10 finger hand, or SDXL adds an extra arm to the existing 2 just to hold the pan, or if the character is a girl her arm turns into a man's arm. Pretty funny honestly.
Anyway I think these checkpoints are way too overbalanced toward '1girl, goon' model stuff but idk if that's checkpoint makers being lazy or SDXL having some max learning capacity.
>1- Add more secondary objects to their datasets! Background scenes like everyday objects, animals, landscapes, forests etc.
>2- Add more slice of life scenes to their datasets!
I find it hilarious that any model has the intelligence and capability to generate any NSFW scene or position but when I put in tags like "character cooking, holding a frying pan" that's when the real problems start. Character grabs the pan but with a 10 finger hand, or SDXL adds an extra arm to the existing 2 just to hold the pan, or if the character is a girl her arm turns into a man's arm. Pretty funny honestly.
Anyway I think these checkpoints are way too overbalanced toward '1girl, goon' model stuff but idk if that's checkpoint makers being lazy or SDXL having some max learning capacity.