Recommend:
vae-ft-mse-840000-ema
use highres fix to improve quality.
Source:
using MBW to merge Chillout and Dark Sushi
and also add a little NijiExpressV2
Description
FAQ
Comments (16)
v4.0 is incredible. thanks
They are great models, I like the V3 color style most.
Really nice checkpoint, handles styles and colours well!
Looks great :) Do you have plans for an SDXL 1.0 version?
For some unknow reason generating decent eyes with this checkpint seem impossible, i don't know what i'm doing wrong...
You need ADetailer extension
Great model, love it!
Quick question. What do you guys use for inpainting this model?
Making your own inpainting models is actually super easy! The main problem is that making an inpainting model for each and every checkpoint you use takes up a lot of hard drive space (although that's true no matter where the inpainting models come from).
this model is great.
the hands are ok without upscale but when I hires fix with latent upscale everything gets better and more detailed except the hands.
it produces extra fingers and mutations. does anyone know any method to get the effect of high denoising latent upscale without the hands getting ruined?
Try different upscalers, some of them have a really good knowledge of hands anatomy, and some of them just adding less unwanted details. For example: 4x_NMKD-Siax_200k - decent knowledge of hands - i often use it with inpaint to fix hands; 1x_JPEGDestroyerV2_96000G - it's designed to remove jpeg artifacts, and because of that it's less tend to add extra details.
@Kinkau thanks!
almost impossible to have dark skin with this model it is such a pain very sad
I HIGHLY recommend ensuring as low as possible Clip Skip scaling, and maybe VAE slider down to make sure that the prompts is used first!!!
By not using a low Clip Skip, 'dark skin' just gets swamped out by the randomness...
just discovered this!!!!! its great, how did i go this far without it , thank you!!!!!
Such a good model. Thanks for the hard work. Really versatile and such good quality outputs. Good stuff!
Any chance we can get the training data to make an Illustrious or SDXL checkpoint?
Details
Files
darkSushi25D25D_v40.safetensors
Mirrors
darkSushi25D25D_v40.safetensors
darkSushi25D25D_v40.safetensors
darkSushi25D25D_v40.safetensors
darkSushi25D25D_v40.safetensors
darkSushi25D25D_v40.safetensors
darkSushi25D25D_v40.safetensors
darkSushi25D25D_v40.safetensors
darkSushi25D25D_v40.safetensors
darkSushi25D25D_v40.safetensors
DSM4.safetensors
DarkSushi.safetensors
darkSushi25D25D_v40.safetensors
darkSushi25D25D_v40.safetensors
model.safetensors
darkSushi25D25D_v40.safetensors
寿司2.5D_v40.safetensors
darkSushi25D25D_v40.safetensors
darkSushi25D25D_v40.safetensors


















