PROzChoice

(51/50)

NOTE: BF16 Model is diffusion only. Read this guide on use.

FP8 Models are full checkpoints with CLIP & VAE attached.

FP32 PONYCLIP

Currently the FP32 version is exclusive to TensorArt

• It is a 11B parameter training distilled to allow a perfect blend of SDXL and PONY while using SDXL CLIP

• You can use SDXL named celebrity's or PONY characters

• SDXL LORA's preferred over PONY

• Should work with SDXL control nets as the model was quantized from FP32 CLIP-L and CLIP-G

• The diffusion (UNET) model will work with either SDXL or PONY CLIP-G and CLIP-L

Why 51/50 because 51/49 does not sound good, even if it is more accurate.