This is a merge of bigASP and LUSTIFY, which, in my opinion, are the best non-Pony SDXL checkpoints for NSFW images. This merge produces results similar to bigASP but with improved image quality and aesthetics.
Links to checkpoints merged:
Description
FAQ
Comments (29)
Interesting... Still suggest clip skip 2 and pony score tags?
BigASP did not need/want clip skip of two. That鈥檚 a Pony thing. (Score tags yes.)
Nice job! These SDXL checkpoint are getting so good!
Thanks for actually giving credits. Really appreciate that
I already loved Lustify, and this merge is even better! 馃敟
When trying to create a merge with sd_xl_base_1.0_inpainting_0.1.safetensors + sd_xl_base_1.0.safetensors I get this error: Error merging checkpoints: argument 'metadata': 'dict' object cannot be converted to 'PyString'. Merging with bigaspV1_v10.safetensors works.
Merging to ckpt, or disabling the 'Save Metadata' allows the merge. But, I still get errors when trying to use the model. I have not found a solution yet.
will you make V2 based on the new lustify that came out?
i hope he does
Based on my testing, this model is definitely more than a mix of Lustify and BigASP. Can we please have whatever the underlying model is/was, since it has got to be amazing?
When it's not putting out junk (50%), It's THE BEST so far for the stuff I'm doing (babes in lingerie). Just one thing - teeth - it consistently and only puts out terrible teeth on my beautiful girls! Needs focus for next iteration pls. ;)
yep.
Inpainting img2img with "perfect teeth" helps a lot
@jacob42聽Yep, I'm getting round it with Adetailer now ;)
One of the best models for NSFW!! Thanks!!
I notice that every 5th or 6th image includes a strawberry! Presumably an 'Easter egg' thrown into the training data to mess with people's heads. If that's true, it's just a wee bit childish, and a shame.
Must be your prompting, never seen one to notice it.
How would that even work?
not being able to inpaint, it produces grainy output,far from realistic,im using comfyui ,any solution to this ?
I'm also use Comfy. The model is very realistic. The autor of the model has given us the workflow, I use it with some minor modification, Then I upscale my best images with Clarity Upscaler, I love to use the model in img2img. Denoise should be rather high (0.8, e. x.)
@makeshka1008779聽could you please share the workflow for inpainting
Hey guys, try out the model with my NSFW Lora
Works very nice!
i do like the outputs but man, i just can't get it to follow a prompt. probably the bigASP genetics doing that to me, i've never been a danbooru tag prompter
What are the recommended settings to use? 馃檹
can you redo this with v3 of lustify? I tried to myself, but it ends up giving messed up results when inpainting, so I probably did something wrong
New versions don't have to be better. I like it as it is now. Using it to train LoRA's on. Inpainting is always tricky with XL if you go above 0.6 denoise.
@Lucifie聽ok
You can use the BigASP pre-prompt "photo \(medium\), score_7_up" and negative "score_1, score_2, score_3" to radically alter the output for what would otherwise be the same prompt.