Hello ! This LoRA is my first one, I have finally started to build my own LoRAs. I'm just a beginner so the LoRA isn't that strong, weight around 0.8-1.0 didn't introduce bad noise on the composition with my models.
The style is rather close to winged_cloud arts but the influence of my models is still very strong. On the aftermath, the result is pretty good.
Two versions :
1) Test_2 : 6000 steps 8 epochs and basic settings (max res 768,768 and clip skip 2)
2) Test_3 : 7500 steps 10 epochs same settings (+ withdraw words like shiny, shiny skin, so that the shiny skin should be more normal to the LoRA as I understand how it works)
Both are good, I think Test_3 is closer to the artist style since it is more trained.
75 images were used without hetero sexual content, just girls and 10 steps per epochs for each images.
If you have tips for making style LoRA, I'll be curious, let me know in the comment of the model.
Have fun !
PS : no control net in the examples, 100% the impact of the LoRA and the checkpoints.
Description
v1 was a mess, forgot to withdraw steps limitation, v2 is the first good outputs I get in the first try (or second if the first fail count).
FAQ
Comments (6)
Seems a lot more stable than my old one. I like it!
I think so too, I compared your LoRA with mine with weight 1.0 and yours was giving weird stuff but I think is still a bit better with the shiny skin
@NaClO_74 It was one of my first LoRAs, so I didn't really know what I was doing yet. I kind of still don't lol, but I'm definitely getting better! That one really came out extra weird though with anatomy for some reason. I'm hoping to give it another try later on if I have more time, but less important now bcz I can use yours!
@Insult_to_Ninjary Yes, not so intuitive to use koyaSS, have fun though.
Would you consider making a Pony version?
I am currently no more doing AI stuff, but maybe will be back one day...