🎊NEWS🎊
I made a v6.0 ZIT.
zit - Maybe there aren't enough learning steps
anima - I think I set the training parameters wrong! (*ノω・*)テヘ
boogu - booooooooooogu! turbo lora is HF repo
🎏TODO🎏
Create a new dataset for V6.1
Adjusting the bias in the v6.0 dataset and the training parameters.
Creating the ZIT Version.
All glory to Boogu!
🐷About sample image🐷
standardized the sample images for V5.4.
Same prompt.
Same resolution. (3:4, 1.5M pixels)
No upscaling.
Different seeds.
Using distilled LORA.
🎴Version History🎴
V6.0 New concept
|2D|ー🐈ーーーーーーーーー|2.5D|ーーーーーーーーーー|3D|
V5.4 jitome return!!
|2D|ーーーーーーー🐈ーーー|2.5D|ーーーーーーーーーー|3D|
V5.0 boogu test
|2D|ーー🐈ーーーーーーーー|2.5D|ーーーーーーーーーー|3D|
v4.1 ill int8 only! 🍦I'll make bf16 later too (if I remember...).
|2D|ーーー🐈ーーーーーーー|2.5D|ーーーーーーーーーー|3D|
v4.0 jitome love
|2D|ーーー🐈ーーーーーーー|2.5D|ーーーーーーーーーー|3D|
V3.0 change base model
|2D|ーーーーー🐈ーーーーー|2.5D|ーーーーーーーーーー|3D|
V2.0
|2D|ーー🐈ーーーーーーーー|2.5D|ーーーーーーーーーー|3D|
thanks maxfeifei8’s one obsession
V1.0
|2D|ーーーー🐈ーーーーーー|2.5D|ーーーーーーーーーー|3D|
P.S.
Since I'm training on a potato PC, the lack of training data is really noticeable!
Training boogu… 10 hours per 1,000 steps with an RTX 3060 and 12 GB of VRAM…
あとがき
じゃがいもPCで学習してるので学習データ不足が目立ちますね!許してちょんまげ!
booguの学習…RTX3060 VRAM12Gで1000STEPあたり10時間…
Description
jitome
background enhanced
FAQ
Comments (2)
In all of your showcase images, you have grey borders at the right side or bottom of your images. This is caused by using "wrong" resolutions. You should stick to width and height values that are dividable by 32. Illustrious works great with 1 megapixel, so the following resolutions work without creating those edge artefacts:
1024x1024
1056x992
1088x960
1120x928
1184x896
1248x832 (3:2)
1312x800
1376x768 (close to 16:9)
1440x736
etc.
Thanks! I'll check it out!





