rasusurasu Style LORA V0.1
rasusurasu_LORA_Style_V0.1
If you like what I do, please consider making a small donation, thank you 🙂
Trigger Word: rasusurasu
Second Lora model, hope you'll appreciate it :)
This Lora is based on the twitter of rasusurasu, trained aroud 223 images
https://twitter.com/rasusurasu
Recommended Settings:
-> Checkpoint : Anything-v3.0
-> Trigger Word: rasusurasu
-> Scale: around 4
-> Recommended sampler: Euler a
-> Steps: 28
-> Clip skip: 2
----- High Res:
Denoising strength: 0, Clip skip: 2, Hires upscale: 1.15, Hires steps: 34, Hires upscaler: R-ESRGAN 4x+ Anime6B
-----
Method of Training :
Due to my inability to run a stable diffusion locally I trained this lora with this colab repo
Fast-Lora-Trainer -> https://github.com/Linaqruf/kohya-trainer
Got any questions ? Contact me on Twitter https://twitter.com/Tyzze2
Description
Lora based on @rasusurasu twitter
FAQ
Comments (5)
Goodness. 360mb for a style LORA? Seems a bit high, yeah?
Yeah, its because of the network dim that I set to 320 as there are 200 images
Maybe ill do a lighter version ...
You can fix your problems by including regularization images (of things that the artist doesnt tend to draw much) , and properly tagging your images.
Thanks ! I will try that :)
I've never went through the tagging process, any recommendation or guide ?
The creation of this LoRA seems a bit sketchy but I absolutely love it !
It just works <3
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.