LoRA trained to recreate the style of Kittew.
Artist Socials: Linktr.ee
Activator token: 4RCH0N
Getting the best results:
first off; use the activator token (4RCH0N) and use the base PonyXL model. AutismMixes do not work well with it, and erase a good bit of the style, while giving you fry in return (bad deal). Second; this LoRA and the style in general has a lot of detail, so in order to properly bring it out, you may need to use highres fix. Base gens can be blurry, especially if the subject is far away or the face is small in the picture. I recommend trying ESRGAN_4x upscale by 1.2x, and mid denoising (0.45->0.6). Keep CFG around 7, any lower adds to the blurryness. If you want to squeeze more out of the lora, I can only recommend using Forge (Stable Diffusion WebUI Forge), and enabling FreeU and SelfAttentionGuidance with these settings:
Additional info about this particular model:
The model was trained on about 300~ images, though none of them were cleaned from signatures, so the little cat ears signature Kittew does is present on a lot of generations. If that bothers you, you can try using negatives like "artist name, signature". It is not very hard to inpaint out if you prefer that method.
It is completely stable at 1.0 weight, which i recommend, but it's still adequate at 0.8 if you have other loras that need "space" to work with.
Kittew's style is pretty consistent post 2023, so that is the main style the model replicates, even though a lot of earlier art was included in the dataset.
The model is also particularly good at NSFW and is stable with complex poses.
One of my personal favorite artists. Kittew really knows what's up 😏
Description
Initial version.
FAQ
Comments (21)
Oddly, I don't see this pop up along with my other Pony loras when I insert it into the folder.
Sometimes you have to restart the entire WebUI. Though, if you do still have issues with it later, send me a message and i'll see if there's something very wrong.
@RC0N i am facing the same issue; no matter how many restarts - it does not appear
Can you tell me what tool you're using? SDWebui?
Can you try this? https://pixeldrain.com/u/9Dy8M7sK
It's the exact same model, but i purged the metadata ( i added some custom meta, that's the only thing i could think of that might trip up some alternate LoRA loader ). It's an LoKr model, so if whatever software you're using doesn't support LoKr, that could be why.
Other than that, I'd have no clue.
@RC0N Forge
@RC0N The purged works!
I swapped the lora out for the purged one on the post. Thanks for reporting the issue!
@RC0N OP here, it worked for me too - to be clear, the version you linked in this thread is the same as the one now publicly available, yes?
@Sargovel yep. Exact same!
very accurate, good job!!
would you consider making a lora of @Madyaxx style from X? : )
Good job mate. :)
Not sure what I'm doing wrong, but anytime I try to use this lora, all I get is the standard cartoony/anime style. Changing weights, trying to use the FreeU and SelfAttentionGuidance extensions, and checking the triggerword all didn't help. I even copied some prompts off here word for word and nothing.
Soft and accurate. I love it!
Kittew really know's what's up for sure. Also one of my favorites. You have great taste. thank you for making this!
pls port to illus, the illus versions i got aren't nearly as good :(
👌
please, make for illustrous
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.