Offset Noise LoCon
This is a LoCon model extracted from offset noise ckpt.
And can be used on anime model too.
What is offset noise?
Diffusion With Offset Noise (crosslabs.org)
What is LoCon
KohakuBlueleaf/LoCon: LoRA for convolution network (github.com)
How to use LoCon
Install this KohakuBlueleaf/a1111-sd-webui-locon: A extension for loading locon model in sd-webui (github.com)
Description
LoCon extract with linear 50% percentile and conv 10% percentile
FAQ
Comments (11)
Can this also just be used like a Lora or does it have to be used as a Locon to work?
you need to install locon ext and then use it like LoRA
Hello! I'm trying to extract an RPG model, for example, and I use this command:
python extract_locon.py --device "cuda" --safetensors --linear_percentile 50 --conv_percentile 10 "E:\SD\models\Stable-diffusion\v1-5-pruned.ckpt" "E:\SD\models\Stable-diffusion\rpg_V4.safetensors" "E:\SD\models\LORA\TESTED.safetensors"
and got a 2.5mb model, what did I do wrong?
you need percentile to be 0.0~1.0
I should use quantile at here, sorry!!!
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument mat2 in method wrapper_mm)
What should I do？
Same issue here.
Update: If you're using a1111-sd-webui-locon as an extension, then do not use sd-webui-additional-networks.
direct use <lora:LoconLoraOffsetNoise_locon0501:0.6> in prompt input box.
see: https://github.com/KohakuBlueleaf/a1111-sd-webui-locon/issues/12#issuecomment-1491531909
Same issue
same, nooo
days ago it happened to me with everything downloaded here but in huggingface it did not give an error
How you train this locon?
I don't train this
I just use extract.py in lycoris repo to extrat the noise offset model from sd1.5 model
So I can use this offset noise model to other models
Basically just like the "add differences" merge mode, but with locon