Check out my other Stuff
SD 1.5 SDXL FLUX CHECKPOINTS
unStable Illusion SDXXXL
Realism of unStable Illusion SDXL , improved with a lot of NSFW
V2
Recommended Settings:
Best Sampler:
ANY
CFG: 5
Use any SDXL resolution
A lot more NSFW
V1
Recommended Settings:
Best Sampler:
DPM++ 2M Karras 25 steps
CFG: 5
Use any SDXL resolution, i like 896x1152 the most
No negative Prompts
HiRes possible, 10 steps, low denoise 0.25-0.4, upscale 1.5-2
NSFW and SFW
Mostly for realism, but you can try some 3D or other Stuff.
Description
FAQ
Comments (16)
Do Illustrious and Ponyxl loras work? What about vae, do I need one or is it baked in?
can this do full nelson position
For those getting weird white/light blue blobs in their generation, load the default SDXL VAE because the one in the model is broken, most likely from merging models.
why would you use this without vae loaded?
@Peli86 If you use checkpoint loader node in ComfyUI it will load the broken baked VAE in the model.
@Part_LoRAs easy solution, use forge ui.
@Peli86 lmao. Tell that to the user called debri in the posts section. I'm telling people the solution because you were way too lazy to swap the VAE after rebasin the model. I wont respond anymore. The VAE that comes with the model is broken and nothing that you say will change that. lol
The model itself is great tho, thanks for sharing.
Hi, you say any resolution but for some reason I am often getting 'split' images where the output is 2 seperate images split horizonally (or vertically) do you know why this may be? I see no one else commenting on this so I wonder if its my reforge or something? I've had the same issue time and time again with bigASP which this model seems quite similar to in output (Is there a relation?)
Any ideas? thanks.
I've never heard of such a problem before, I'm sorry.
@Peli86 ok thanks, very strange, and you really use all sorts of resolutions? Do you use comfyui or reforge/forge or whatever?
@MrTitsworth Forge. Try to drag and drop one of my images into the PNG Info tab of Forge and use my settings. And make sure that your prompt isn't too long.
This could be explained by the inclusion of bigASPv2 in the mix. (I don't don't know it was used here specifically but it's a core component of most NSFW merges, I've seen this behavior in my own. Seemingly there were more than a few examples of multi-panel images in the bigASP training data to create a bias with some resolutions / aspect ratios). Adding multiframe, multi-frame, slit screen, splitscreen, multi-panel, to your neg propt should generally fix it.
@Peli86 Nah he's right. BigASP is notoriously bad for it, just like this model can be. It often happens when you ask for a typical body position that doesn't match the shape as in if you ask for someone lying down on their side and you have the picture in a portrait size then it will split it up etc, not sure how you've never had this issue but it is common.
In saying that, I have been enjoying this model when I don't have those issues, so thanks! good work! It is very similar in outputs to bigasp but has its own flavour which is nice.
@prodajie Yeah 100% despite that both are still great models, just annoying this issue occurs with some configurations.