Goddess Project

This model uses BF16 base UNET with minor changes to allow for full feminine beauty.

• Works with LORA's Fully

• High detail using BF16 UNET Blocks

• Fast using NF4 TE/CLIP and TXT Blocks

This model is a mixed precision model that can fully fit in a 24GB card. Don't have a 24GB card this model still runs extremely fast without sacrificing detail.

How you ask, by compressing only the Text adherence mechanism in the model we are able to sacrifice a small amount of prompt adherence for 1.5-2.0x speed or more.

If you would like to see the base DEV checkpoint in mixed precision be sure to comment below or support the model by posting your pics to the goddess project model,

SFW or NSFW welcome