Sign In

FP32 BasePony (CLIP-G, CLIP-L)

68
1.1k
33
Type
Checkpoint Trained
Stats
679
Reviews
Published
Dec 19, 2024
Base Model
Pony
Training
Steps: 5,555,555
Epochs: 5,555
Hash
AutoV2
1820A14B32
SDXL Training Contest Participant
Felldude's Avatar
Felldude

FP32 Pony v6 One Million Block Restore (13GB Full Checkpoint)

For Ultra Realistic FP32 PONY go here

  • THIS IS A FULL CHECKPOINT DO NOT LOAD ADDITIONAL CLIP OR VAE (Separate CLIP's are for merging into other models)

  • Over 1Million Zero values have been restored (When compared to up casting)

  • CLIP-L & CLIP-G have been restored also while maintaining the PONY tokens

  • This is the base model of PONY with no additional changes to the UNET or TE - Beyond the Million+ Zero Block Restore (1034415 zeroed values restored on UNET)

  • This model is highly sensitive to Scheduler and Sampler settings

It has pointed out that this model does not like certain style LORA's - With the higher precision score and source is followed greatly on this model.

  1. If a LORA is having trouble working do not use score.

  2. Use the correct Source for your LORA, source_anime, source_cartoon, source_3D etc

Can be run in any UI (Forge, Auto1111, Comfy UI) the model will be downcast by default. These settings improve complexity but are not required. (They will slow down IT's in most cases)

  • Comfy UI --fp32-text-enc OR --force-fp32

  • Forge/Auto1111 --clip-in-fp32 OR --all-in-fp32