SDXL: The тhe clever trick: makes AI images stop cutting off heads

12 Просмотры
Издатель
SDXL improves Stable Diffusion for high-resolution text-to-image generthesis by scaling the UNet, adding a second text encencoder, and conditioning on image size and crop offsets. This micro-conditioning stops training flaws like cropped objects from leaking into generated images, and a refinement model polishes fine detail.

Категория
AI Картинки
Выполните вход или зарегистрируйтесь для отправки комментария.
Комментариев нет.