General:
As simple as that, a lora in style of the kawakit artist. I tried picking images that were close to each other in style, and I only chose the ones that had the most "finished" looking style - so the sketches, partially colored or rendered images were skipped altogether
Tags and information
Use tags:
@k4wakit_style
+ optionally
asian, makeup, eyeshadow, censored, mosaic censoring
Support:
If you like what you find here, please consider making a donation https://ko-fi.com/hat0ri, thanks!
Description
FAQ
Comments (9)
Did you upscale/sharpen the images you used for training? I'm getting quite a few over sharpened looking output images.
That's a good observation. It's from an unsharp mask. I'm using .py scripts to automate parts of the workflow, and I meant to remove it from one of the scripts, because otherwise both scripts do it making it essentially run twice. That's definitely gonna be one of the fixes for v2
@hat0ri Yeah a v2 without the sharpening would be awesome.
Other than that the mouth can be a bit weird.
@HaloSkull What exactly about the mouth would you say could be the issue? Is it when the mouth is closed or open?
@hat0ri Open, The teeth on the top were melding together with the bottom teeth when I tagged for torogao. The teeth are a bit hit or miss. You are training with 1.0 base right?
Eyes also are a bit iffy when viewed with a tag like from side
@HaloSkull I couldn't quite recreate the things you mentioned. You can probably drop the weight a bit and maybe that would help. Either way, we'll see what I can get with the next version. And yeah, it's AnimaBase 1.0. I'm not sure if it actually works better with training then preview 3. I've also seen someone made "Any Anima" for training, so maybe I'll try that as well
@hat0ri 1.0 should be the only version people make loras on unless a newer base comes out.