Pony based with bit of animagine and loras
This model has VAE baked in.
Suggested settings:
I had CLIP skip 2 on every image
I had ENSD: 31337 all of them
All of them had highres.fix or img2img at higher resolution.
I don't use restore faces
I use afterdetailer for face details
Tiled diffusion for img2img upscaling and sometimes Noise Inversion for some more extra details.
4x-Ultrasharp upscaler
NOTE: if you find any prompt of the preview images look familiar it's because I've taken them from other model review images, Credits to the original authors. Thanks for the benchmark.
If you like my work then drop a 5 review and hit the heart icon. it's free and keeps me motivated
Commission open in Patreon
Get early access to my upcoming NSFW Lora in my Patreon .
Support my work by joining any one of them and get early access to all my upcoming loras and other perks such as fan requests and Discord role.
Join my Discord Server
Description
FAQ
Comments (10)
Clip skip on sdxl? What??
afaik, didn't we use clip skip cuz anyV3 had it's own clip training mangled or something? Is it really needed for an sdxl model or am I missing something?
It not quietly sdxl its pony and it very stupid)) I showed it below, it’s just falling apart into pixels, they not tested for that. So clip skip))
im just following what the original ponydiffusionV6 suggested. and my UI always have clip skip set to 2 so I never changed it for 1.5 for sdxl.
@agress74 for me it never fall apart . I haven't used fooocus so I can't say about it. And yah it's a pony based merge
@Goofy_Ai Thanks for the reply, weird, I don't remember reading clip skip for pony nor I've been using it on SDXL based models since it uses 2 clip models its kinda off to skip some of it's layers
@agress74 Well, this model uses clipskip so without it it doesn't works ig
@Goofy_Ai @agress74 nvm just reread, yup clip skip, imo it shouldn't have been trained that way, since clipskip was used for the NAI leak that had mangled it's training lol
Not sure what the drama is... doesn't the pony dev recommend 2
@l1408 Let's hope that in the 7th version the developer will take these points into account, and not go the way of Hachatur))
"I had ENSD: 31337 all of them"
what does it mean ? im trying sdxl models now, ive no idea as a person who is coming from sd 1.5
It's an setting you will find in automatic1111
Just type it in the settings and enter the value.