Join The Tinkerer on Whop. Membership gets you early releases, private tools and a bunch of extra stuff.
👉 Join on Whop
💬 Join the community for support, free tools and early news on Discord
This isn’t just another version of CyberRealistic XL. Desire is a special edition, fine-tuned for artists who want to create stunning NSFW (and SFW) content with speed, realism, and flexibility.
So why go with Desire?
While the original CyberRealistic XL aims for pure realism across all types of work, Desire turns up the creativity. It’s designed for those moments when you want more freedom, more play, and a smoother workflow.
Desire captures the spirit of exploration. Whether you're pushing boundaries or just having fun with your art, it’s here to help you move fast and create without limits.
Recommended Settings
Sampling method: DPM++ 2M SDE Karras
VAE: is already Baked In
Sampling steps: 30+ Steps
Resolution: 832x1216 / 896x1152
CFG: 3-5
Clip Skip: 1
Upscale: 1.5 upscale (0.3 - 0.35 denoise)
Upscalers: 4x_NickelbackFS_72000_G / 4x_foolhardy_Remacri / 4x_NMKD-Siax_200kRecommended Settings DMD2
Sampler: LCM
Schedule Type: Exponential
Sampling Steps: 8/12 (12 for extra detail)
CFG Scale: 1-1.5
⚡ These settings are not the same as the standard SDXL workflow — be sure to adjust your pipeline!Description
CyberRealistic XL Desire DMD2 is a special edition, fine-tuned for artists who want to create stunning NSFW (and SFW) content with speed.
FAQ
Comments (48)
Aaaah this looks very notty!!!
i finally ran out of yellow buzz but I'm looking forward to trying this in 12 days! thanks for continuing to provide non-DMD2 version for us old-heads
Old bastards cant try dmd or what?:))))))
@akariraeshawn161Â no we can't :)
I can't get DMD2 models to work, ever. They always look super cooked, like someone set the HDR to 200000%.
@michaelthemenace You must use LCM sampler with Karras or Exponential scheduler.
Cfg:1-1.3, steps 8-12.
@lukisicarus357Â KSampler won't work?
@michaelthemenace Thats some comfy only feature, you need to select from additional samplers.
@kendrixriddick445Â thanks the LCM sampler worked. I event went and found a much better workflow for DMD2 that uses LCM - Exponential and the images look GREAT.
this model is SO fun! great work!
I have a problem in Fooocus - only ultrasharp noisy images with clip skipping = 2, more realistic at 6, black square at 1. Is this model not for Fooocus, but only for SDXL?
hai usato il preset "realistic" ? Se non lo hai mai usato prima scarica diverse cose tra cui checkpoint e LoRA
any reason you recommend clip skip 1 but most of your preview photos are clip skip 2?
Beautifully done. I am having trouble with it recognizing bulges / similar terms, but there are loras for there. The depth of these images is fantastic
We need blue era Cyberdelia:) The merge looks it has Biglove in it, just for the color, but its not so sharp.
I send this comment to @SubtleShader :)
But it's a mix from some XL models (XL - Catalyst etc.)
@Cyberdelia Ok:)
Here's an odd observation. The dmd model doesn't seem to know what pants are. I tried generating a woman wearing "slacks," "designer slacks," "pants," and "trousers," but every result had the same type loose warmups/pajama bottoms with a drawstring. Hopefully, I will have better results with jeans.
If you are using ComfyUI, try using dual clip loader and use own CLIPs, can improve such problems in majority of SDXL models. Also not sure if model is trained on natural language or tags. You can try to do natural language prompt, see if it sticks.
@Mescalamba Thanks, but I'm using forge.
@teppscan842Â Hm, maybe possible too, but no clue about Forge. Sorry.
Geez, i had same problem always, when i used pony. Now, it seems xl is way better, and i got only half clothes(ass out, loose), when i prompt jeans.
Give more steps(even 14) and add detailed prompt to clothes.
@kyreartyom823 @teppscan842Â Â use the non-dmd2 version if you want better prompt results
@Cyberdelia NO! im ok:)
Well I just tried the same prompt on the newest Gonzalomo (DMD), and it did pretty much the same thing. With a batch size of 4, I got 3 warmups and 1 actual pair of pants. So . . .
@teppscan842Â In case anyone else reading this wants to know: :)
In Forge you can just load external clip-L & clip-G the same way you load a different VAE. Just put the clip encoders inside the 'text_encoders' folder in the 'Models' directory. Refresh, and they'll be listed along with the VAEs in the same drop down menu. No other steps required.
@ironically_irascible That's helpful, thanks.
really fantastic model! I think it's better than Flux in most cases!
whenever there's a penis in a vagina the woman has a dumb open mouth expression. "closed mouth" prompts are ignored.
I think, the nsfw dataset has this default cheap porn expression. I always prompt for emotions, so i dont get it.
It's almost impossible to get something that isn't freakishly deformed. Any advice?
Use CyberRealistic XL
@Cyberdelia I did. Exact same issue.
@jesper123160Â then there is something wrong with prompt or settings. PM me the details
@jesper123160 or drop your prompt here. You can see in the images created by other users that they are ok.
It's a very beautiful, but difficult to manage model. It follows prompts to some extend, but it is bad at action and nsfw sex prompts. it often creates real mess. But creates stunning nude images. Without need of additional Lora's. As others mentioned - many open mouths scenes. I would like to see improved version soon.
Was looking for something that worked well with my older lustify lora (any modern lustifies did not). This one does surprisingly. Clean results. Cheers
I'm sorry, great images sometimes, but this model doesn't give a .... about your prompt...
Is there a way to use this in civitai?
Why there is no single model which can do uncircumcised penis??
That has to do with the type of training that was used. In most cases it’s U.S. training material. It’s already difficult enough to train penises, let alone get proper distinctions between them and not confuse AI models. For that, it’s useful to have a LoRA that can handle that. As far as I know, there are some options out there.
I've test this model, but for men, it's really difficult to create someone detailed with no deformation
Very finnicky, but once you get the prompt/negative prompt right it works wonders. Wish it wouldnt change the face so much, but at least I can do inpainting to avoid that
I have found the cyberrealistic XL checkpoints are the best for me (a beginner) Ive merged 2 of them and stay away from Lora's as much as I can (which I prefer) I look at everyone images and use there prompts to see what I come up with. Im learning everything is in the correct prompt. Id like to thank the creator well trained checkpoints and bringing the to us to learn from and play with. Thank you
This is a good model!
Any update in the future?
Yes, need to create some sample images
It's released!!
@Cyberdelia Thank You!