I am admittedly kind of a noob at this, but if you enjoy my humble work, any donations are greatly appreciated: https://ko-fi.com/insult_to_ninjary
Your generosity helps me stay motivated to keep improving and to keep trying new things!
At a Glance:
Base model: Illustrated (but you can experiment with other kinds!)
Use a VAE?: Yes (baked or your choice)
Clip skip: 1 or 2 (experiment with both!)
LoRA weight: generally from 1 - 0.7 (higher has stronger style, lower has better color variety)
Hires. fix: use non-default upscaler; denoising strength 0.5 - 0.4
Trigger Words:
woodcut - greatly improves the authenticity of your results; muted colors and a more historical feel
no trigger - in general, you'll still get the effect, but it's weaker; more colorful and modern looking
You may need to weight the trigger word with numerical values or parentheses if the effect is not strong enough. I generally put the trigger word inside 3 sets of parentheses, like so: (((woodcut))).
My showcase images include examples both with and without the trigger so that you can see the difference.
Description:
This LoRA was trained on the famous Thirty-six Views of Mount Fuji series of landscape prints, plus the additional 10 prints that were later added to the series. Though not perfect, I think it does a commendable job at capturing this stunning art style!
The trigger word - "woodcut" - was not intentionally trained in as a trigger, but after seeing how well it worked with my previous LoRA (Woodcut/Etching/Engraving Print Art Style), I tried it with this LoRA as well and was amazed at how much of an effect it had! I definitely recommend using it if you want your images to be truer to the style. You don't always need it to trigger the effect, but it can be hit or miss without it, so using it ensures you'll get a strong stylistic effect right away!
Troubleshooting:
It tends to be a bit blue and/or green dominant, as you may have noticed from my showcase images. While this was also true of some of the original artworks as well, it's a bit too pronounced in the LoRA I think. After creating all my showcase images with a weight of 1, I discovered that lowering the LoRA weight can noticeably increase color variety. I generated a few test images at a weight of 0.7 and they definitely had more various colors without really losing too much of the style.
The LoRA also struggles with anatomy at times, most notably hands and fingers (surprise, surprise!). Just try your best with all the usual fixes - different prompting, try a different base model, use a VAE, try clip skip 2, lower the LoRA weight, use adetailer, etc.
I very occasionally got some weird blurring, which was definitely not in the training images. You can kind of see it in my first showcase image. It was pretty rare to occur though, and it's not really a major dealbreaker IMO.
Outro:
If you make anything awesome, please consider posting them to the gallery! That goes for all of my LoRAs! I really like seeing what kind of stuff you all are able to make!
If you have any questions, feel free to ask! I will try my best to help! Also, if you have any ideas for future LoRAs, please don't hesitate to share them with me!