This LoRA will help create images in SakimiChan's style.
For better use, You need to install Addition Network Extension
Update 19/9/2023 v3.0
Update dataset
Increase Steps training
Increase Epoch
Update 25/8/2023 v2.1
Reduce the epochs from 40 to 20 to make it more suitable with many models XL
Update 25/8/2023 v2.0
Improve to SDXL
Installation
Open "Extensions" tab.Open "Install from URL" tab in the tab.Enterhttps://github.com/kohya-ss/sd-webui-additional-networks.gitto "URL for extension's git repository".Press "Install" button.Restart Web UI.
How to use this Lora
DowloadPlace this Lora in stable-diffusion-webui\extensions\sd-webui-additional-networks\models\lora\After Installed Additional Networks you will see this section below Seed text BoxClickrefresh modelsto load LoraTick EnableChoose LoRA in Network module 1, Choose Lora Name in Model 1Adjust the Weight at Weight 1
More Information
Using with my Model for best result: https://civarchive.com/models/73945/nova-ai-art-v1
Have a look at my other Lora: https://civarchive.com/user/chuong1224/models
My Others page: https://linktr.ee/NovaAIArt
My TensorArt page https://tensor.art/u/626657680337064597
Description
FAQ
Comments (12)
Looks very-very good! How many images did this Lora train with? ;)
i train with 180 images. Maybe more at V2 😄
@chuong1224 Hehe, I thought it would be more :D Looking forward for new version ;)
@Romazeo Thanks you 😄
@chuong1224 I tried with custom char, looks oretty close to sakimichan style. But do you tested it with addition of specific character Lora? Because I noticed, that some char's loras goes very bad with some models like this (nova)+ style lora. It starts to break character design and background :(
@Romazeo could you give me the link of charater lora which you use to combine with sakimichan lora? i will try to debug and if can, i will fix in v2
@chuong1224 For now, I have problems with characters of this author https://civitai.com/user/RiuKi_MK1/models
I can create well with, say, YesMix model+sakimi or zumi style. I think it's because of more "drawing" properties in YesMix. But when it goes with slightly 3d+photoreal models (speaking of Nova or Puffy)...Everything goes bad ^_^
@chuong1224 And in particular, for example, this lora https://civitai.com/models/80192?modelVersionId=85014
May be it is all because this one is very tuned to TV series anime style?
@Romazeo I have to do some test and maybe the problem here is the wweight. Because 2 Lora was made with different Network Dimension, so when we combine them, at first we need to find out which weight is the best for them. For ex: Motoko Kusanagi Lora should use 0.2-->0,4 and my Lora sakimi just need 0.2-->0.6 when using combine with that. I upload the testing result here: https://mega.nz/file/0bVyCARb#z4WptMyl4JeF09jFX2MstS4k6ZLJUcMyPXCICbEfpvA
@Romazeo get yourself lora block weight extension, set character lora with MIDD and artstyle lora with OUTALL
@chuong1224 This is exactly what I was doing this night :D But not with Motoko, with another char. But nevertheless. I found out, that I should defenitely sacrifice artist visual style lowering his lora's weight...Otherwise, character in char's lora will have bad visual or very far from canon outfit ((
@fizzballs But if I down artstyle lora, then I get very low artstyle I needed :D