This style is inspired by the work of manga artist IZAYOI SEISHIN.
Description
In the previous version, multiple view images were often generated. Additionally, there was overfitting in the background. Therefore, in this version, we edited and optimized the dataset.
Trained on 200 images
3 repeat,20 epochs, 12,000 steps
This version has 16 epochs, a step size of 9,600, and a loss of 0.112.
U-Net only
U-Net Learning Rate: 0.0001
Network Rank Dim: 16
Network Alpha: 8
FAQ
Comments (7)
So, what is the trigger word？
Coolio.
Also, up for art style suggestions?
Any advice on how to avoid getting white borders/frames on almost all images?
Already added the following to negative prompts with no real change:
(border:2), (frame:2), (white border:2), (white frame:2),
Write about the background in the prompt.
@franklin_sports322 All the prompts I tried have detailed background descriptions, but I keep getting white borders or frames. Only about 1 in 10 images comes out with a proper full background.
@UJLammy There's probably something wrong with the prompt.
@franklin_sports322 You mean prompts, I tried several different ones, all with same results.
Even tried the LoRA at a lower weight, sadly with mostly the same results.