Osorubeshi -> anime-like models trained on SDXL
If you like my work you can tip me on - patreon.com/Natakaro or buymeacoffee.com/natakaro :)
Feel welcome to join my (still in progress) discord server - https://discord.gg/bPRpQQRc4Y
Nsfw_Hyper_v0.1 - NSFW oriented model with more "web hentai thick line" style
Kakkoii_v1.1- Fixed version, more 3D like, more colors, small differences (with some promts there are quite big but it is rare)
Rain_v1 - new semi-realistic, "thick line" style
Pastel_v2 - More trained "Pastel" style model
Great at nsfw, nudity, manage well anatomy, really not the worst in hands. If you have problem with fuzzy eyes just add more steps or use ADetailer.
Old models:
Z - alternative for the Y, more vivid colors, slightly better eyes, more styles etc
T - sometimes better than Z, sometimes not, better eyes and more vibrant colors
D - new, vibrant, little bit weird
Semi - new experimental model, more 3d, semi-real
Kakkoii_v1- Best model so far, good for sfw, nsfw, female characters especially, diffrent name because diffrent specific style ;)
Pastel_v1 - More anime like, cool style inspired by old "pastel" sd models, really good at nsfw stuff
Kakkoii_Turbo - Test model for turbo version of Kakkoi style
v1 - Just better in everything (especialy in nsfw), but almost lost male characters in model training (will bring then back probably in next model)
Description
FAQ
Comments (18)
Always renders with almost no saturations and white dots everywhere? Not sure why its rendering this way.
Please send generation parameters, I will try to check.
Same problem here. The Previews look fine in 1111, but the final image is very muted and has those white dots. I tried several 3M Samplers with 40 steps & CFG 8
@kurap1ka This means you are using the wrong VAE or no VAE at all....
@olivetty ah yeah to used for them to be baked. Thanks!
@olivetty I see, was using mage.space
@olivetty what's a good VAE to use with this model?
@gik0geck0 @kurap1ka https://huggingface.co/stabilityai/sdxl-vae/blob/main/sdxl_vae.safetensors Stabilityai has a VAE that works fine.
@highsodium7229 no idea how to use this, I put this in VAE and load it and get errors all over the place... isnt a vae supposed to be .pt or .ckpt?
@natakaro I'm having the same problem with the dots: https://ibb.co/PZfSMnj
I use Fooocus, the parameters were Euler Karras, CFG 7, 20 steps, seed 5429448612200702317
I have installed sdxl_vae.safetensors, so I'm sure is not a problem with VAE
Your checkpoints are beautiful! I hope this helps to solve the problem
@Madxis Thanks bro, I did also find this out, posted a new comment with these instructions for new users. Keep sharing knowledge!
Amazing
Please also make SD 1.5 model.
something to be aware is xl doesn't get the same image gen as 1.5, the biggest difference is how many tokens (so distinctions) 1.5 can get and have as details/options compared to XL, so 1.5 is ideal for single style/theme while XL (despite many being focused) beats it in multistyle/theme or if going to only 1 of those, quantity of details. seen amazing 1.5 creators that had awful xl and vice versa, training/tokens are quite different. so if you like a xl model, accept it is a xl and vice versa as they'll work different (personally there are some 1.5 models I'd love to have xl that sadly aren't)
@upseerac 1.5 is suitable for simpler styles and sdxl is really just spreading its wings. The potential, however, can be seen from the hands, etc. where on sd1.5 I often needed to use controlnet and at least a couple of attempts, on sdxl at most a second attempt at inpainting
@upseerac sdxl needs a powerful system, at least you should know comfyui completely, but sd1.5 models are not like that, of course, I can name some sd1.5 models that are almost as good as sdxl models. However, not all of us have a powerful system. Also, thank you for your attention.
@natakaro I agree with you.
If we look at the number of users and also the number of Loras, the sd1.5 models are still popular because they are easier to work with.
@saman_a464 agree there, I prefer xl for most gens but i do keep 1.5 models that seem to trump in quality over xl models sometimes, for certain aesthetics, even on large size images (as i tend to go 768 or 1024 for base, if not using an upscaler), I'm all pro options, and love having both 1.5, 2.1 and XL models, but I just disagree with requesting alternative versions of checkpoints since I feel each will have differing goals
@upseerac Yes, each version has its own beauty and results that others do not have. I wish sdwebui had a speed like comfyui for sdxl models. I hope it will be better in the future
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.