This is a merge of some of the best (in my opinion) models on Civitai, with some loras, and a touch of magic. 🪄😏
Some of the images were generated with 1 clip skip
Recommended settings:
Sampler: DPM++ 2M SDE or 3M SDE or 2M with Karras or Exponential
CFG: 5 - 8
Steps: 30+
Some of the checkpoints I merged:
AlbedoBase XL
⋅ ⊣ Realities Edge XL ⊢ ⋅
Realistic Freedom - SFW and NSFW
Explicit Freedom - NSFW Waifu
Colossus Project XL
WoW_XL
Some of the loras I merged:
LUT Diffusion XL
DetailedEyes_XL
NSFW POV All In One SDXL
Envy Cute XL 04
Perfect Eyes XL
🪄 The final touch of magic is that I used multiple "bad loras" with negative strength to push the model toward greater "goodness".
Description
Improvements:
Even better NSFW
Even better human anatomy
More stable, at least for portraits
Better image quality
No more plastic look which often happened with long prompts
After days of concentrated effort and several setbacks, I finally noticed substantial improvements. I made an entirely new merge, merged it with v1.5, added loras, and stabilized it by merging it once again with v1.5.
This version would be labeled as v3.2 if I uploaded all the versions in between 😎
FAQ
Comments (30)
Emergency download...
File is broken, fail to download multiple times on 2.8GB, you can resume it but SD don't want to recognize it after download.
the file is fine, something else is broken
Site might be having trouble, I can confirm this model works great!
Civit.ai is an extremely unstable website due to massive loads on their servers,
it has nothing to do with the model itself
File is fine today. 6.5Gb downloaded in just a few minutes.
New version coming soon, maybe... 😱
This is a really great model, I would love you forever if you added cunnilingus in your future versions, cause that's pretty much the only thing it really has trouble with. Though, from many I've heard, it's supposedly quite complicated for the AI unfortunately
this is easily one of the best SDXL models
I really love your model, only downside I have with it, is the fact it's not working with the tillable option from some reason, it just makes everything messed up and ugly once you turn it on. and it's a real shame, since it can produce some really awesome textures (without the tillable option on of course) hope you can fix it somehow? (hopefully by the time you release the next version ;) )
You can see two images I added in the Gallery for comparison...
I tried many SDXL models and I almost gave up because of how bad they were at skin details, they always resulted in generations that felt 3D and CGI with soft and unnatural details. This one is just different and gives very good, natural results, it's also good at other art styles, and good for overall use. I hope you can make one that's as good but isn't focused on NSFW, because even for some normal art it sometimes tends to make everything nude with boobies even for "alien" style lol, even when using some keywords in the negative prompt for nude, NSFW etc. 5 Stars! keep up the good work and looking forward for what you're doing next :)
Edit: Oh and BTW, I found that playing with clipskip on SDXL checkpoints have no effect on the generated images at all, I am using A1111, asking because you mentioned clipskip on the description.
The only problem I have is when I use my own trained LORA it doesn't work as good as on SDXL models. The trained face looks much younger and off for some reason. Any way to fix this?
I also want to agree that the faces are a bit too young looking, and i put a lot of (KIDS) in my negative prompt.
Can you merge HASSAN SDXL to your Model, please ?
The first truly capable XL NSFW model. Blows everything else away.
Which models did you merge in that includes watermarks? Sadly they weren't captioned with "watermark" either, so you can't easily remove them.
The POV All-in-One lora is full of watermarks but is otherwise head and shoulders more effective than any other XL NSFW lora I've seen. Worth having it in. "text, watermark, logo" in negative does help somewhat.
Feedback:
Difficult to make artistic demon porn with this, it's extremely oriented towards photographic styles. Still experimenting with LORA tho, and thank you for putting up a NSFW SDXL checkpoint
Weird and wonderful model, really sings when you build a prompt with it. Excellent detail.
took a while to figure out how to make interesting things with it. Very easy to make dumb bimbo pictures. Can make fun fucked up shit through im2img with a very low denoise at like .5
what settings and workflow are you using? :)
Thank you so much for this amazing checkpoint!
I have a question, there's a series of key words in prompts that lead to the output being weirdly distorted. In the example I added "sports bra" in the prompt. The checkpoint knows the concept apparently, but the output looks like in the image below.
It's quite a lot of concepts that trigger this. Do you have an idea of how this can be avoided?
Thank you a lot for your amazing work! <3
[EDIT]: Sometimes leaving out spaces works. Using "sportsbra" works. But there are also prompts without spaces that trigger the distortion.
best checkpoint ever, tbh
My images seem to consistently get text watermarks, despite putting "text" "Logo" etc., in the negative prompt. Same prompt in other models do not have the watermark. Any suggestions? Thanks
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.