Deep Anime
Patreon - Open Commissions - Newest Patreon Realease: Breat Helper 2.0 Alpha 2
After months of LoRA training, I decided to try my own model. After several hours, this came out and I am also very satisfied.
Under Downloads you find:
Pruned Version [FP16]
Full Version [BF16] +
TensorRT File (512x512 up to 1536x1536 - BatchSize 1-4)
Download under: Full Model fp16 ONNX-FILE
Onnx file (models\Unet-onnx)
TRT file (models\Unet-trt)
TensorRT only for Full Version not pruned
Changelog v2:
Train more Triggers
Bug Fixes
Enchant Details
Expand Dataset
Flexible +100%!
Model Info:
Focus: Anime
Clip-Skip: 1
Train Size: 768x768
CFG of Samples: 5
Full NSFW Support
Missing? Bugs? Feedback? Leave a commend please :)
Description
FAQ
Comments (3)
i dont understand why u use clip skip 1, everyone uses clip skip 2
I ask on reddit and 79% say Clipskip1
And a LoRA thats i create with NAI Clipsskip2 has lower dl/ratings as normal CS1
The question is why do you use 2? You can still use 2 with a model that does well at 1. Some models suggest you use 2 because they are broken at 1. Clips have something to do with how stable diffusion understands your prompt and translates into an image, with clip skip 2, you literally skip some prompt understanding and Ai will throw in random bullshit in your way so you will get more creative results that way.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.