
Join The Tinkerer on Whop. Membership gets you early releases, private tools and a bunch of extra stuff.
š Join on Whop
š¬ Join the community for support, free tools and early news on Discord
This is the model you reach for when you want super clean, photo-realistic results.
Itās sharp, powerful, and somehow just nails gorgeous images every single time.
āļø 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_200kāļø Recommended Settings DMD2
Sampler: LCM
Schedule Type: Exponential
Sampling Steps: 8/12 (12 for extra detail)
CFG Scale: 1-1.5
ā” These settings are not the same as the standard SDXL workflow ā be sure to adjust your pipeline!š” Whatās the difference between FP16 and FP32 checkpoints?
FP16 (half precision) uses 16-bit floating-point weights. Itās smaller in size, loads faster, and uses less VRAM - perfect for normal image generation. You wonāt notice any visual difference for inference, so itās ideal for everyday use.
FP32 (full precision) uses 32-bit weights, which means itās larger and heavier on resources. It doesnāt improve generation quality, but itās better if youāre training, merging, or fine-tuning models, since it keeps the exact weight data without rounding loss.
š§ Prompting
Example negative prompt:
cartoon, illustration, anime, painting, CGI, 3D render, low quality, watermark, logo, labelor use CyberRealistic Negative
⨠Optional Adetailer
Adetailer model: face_yolov9c.pt
If you only want the main face being refined set 'Mask only the top k largest' to 1.š Links
Backup location: HuggingFace
ā Fuel the Project
If this model gave you stellar results, consider buying me a coffee. Or two. Or, you know⦠a GPU.
ā ļø Heads-Up
This model can make mature or sensitive content. Youāre fully responsible for what you create, so use it responsibly (or if you canāt resist, at least be artistically irresponsible).
š¬ Want Better Prompts?
Use this custom ChatGPT to auto-generate polished prompts for CyberRealistic XL:
š [Try it now on ChatGPT]
Description
CyberRealistic XL V9.0 is focused on cleaner, more controlled output. Compared to Desire V2.0, itās less overtly horny, but still comfortably NSFW-capable.
All sample images were generated without DMD2.
This is not a successor to Desire V2.0 - itās a fundamentally different type of checkpoint, aimed at a different balance between realism, restraint, and flexibility.
FAQ
Comments (32)
Thank you, for both the LoRA and the settings to use it -
It is most appreciated!!!
author you are great the amount of detailing after generation of image is so good that my friends are not able to believe this is generated just woww
Oh wow V9.0 is the bomb - absolutely not regretting getting that one early! much faster and better results than 8.0 - just wow! (Like for my current thing I just switched from 8.0 to 9.0 and its running about 7-8 times faster and with better results!
P.S. Still seems unreal I kinda guess it's utilizing my AMD Radon RX 6800 XT a lot better than before (runs 20°-30°C hotter than before still within very comfy 75)
P.S.S. yea no more 6 finger hands as well....
@CatbeardaddyĀ Is this true? Please don't play with my feelings T_T
@Jackpot694581Ā It was certainly at that point - got a lot further since then overall, zluda running and such, then switched my whole system to Linux and now its even better.... Using more CyberRealisticPony V16 at the moment and prompts do behave different between that and XL so I'm not switching that much. Pony does make everything a bit too beautiful and cartoonish at times and sometimes needs more detailed description to get a good result (which then works worse on XL...)
What happened to your Patreon account? Everything is gone.
My Patreon page has been taken down by Patreonās Trust & Safety team. I'm now moving to a new subscription based provider.
I don't know how you can claim V8 is improved 'NSFW territory' its absolutely terrible for two person sex positions. In comparison V6 is perfect for NSFW sex positions. I hope V9 is better.
9.0 is simply fantastic! Best update yet.
9.0 is much better than 8.0 on 1-person portraits and accessories, but it looks worse on multi-person interactions, the model looks struggle with understanding the multi-person pose, actions, etc.
Please give me an example of a multiple people prompt.
@CyberdeliaĀ I have a list of test prompts especially for n-sfw. For example, the mmf double *enetration can be achieved 70% successfully in 8.0, but drop all the way to around 0 in 9.0.
SDXL, in general, has been always bad with interactions or poses.
@ibm20023Ā PM me please. Curious
@CyberdeliaĀ I modified my prompts by removing all slangs and modified them with describing the scene in standard English, in this way 9.0 performs much better than 8.0 and achieved several tests that 8.0 never had any responses before. Thanks for your hard work for this update!
@ElectricDreamsĀ try avoiding using slangs like c**girl positions or something else.. Use standard English to describe the scene and this 0.9 version can produce a quite stable result.
The only thing I really notice neg. is the eyes. Direct eye contact is great but seems to lose it the more the eyes are looking off in another direction. Still a good model, and I do enjoy your work
For 9.0 in ComfyUI, use clip skip 2 (i.e. -2). With 1 I was getting nothing but black.
u dont need to use clip skin in comfyui. if u r using them use -2
@NeuralKnitĀ -1 works on almost all SDXL models except the newer versions where people have merged with illustrious models a bunch and lost that.
@ollieicarus809Ā You really shouldnāt be giving this kind of incorrect information. Illustrious is a completely different type of model and itās not mixed with CyberRealistic XL. As I mentioned before, it simply doesnāt work with your ComfyUI workflow.
9 is good, but prompts and loras I used in 7 & 8 are turning to shit with 9. Weird colors, blurry, weird skin.
Really? Just normal SDXL Lora's? All my LoRa are working fine here
whats this new trend, that if you upgrade a model, every other will be on top again?
Yep, it's so fun....
From lots and lots of testing, the Cyber series is the only one that comes up above all others. I don't know how you did it, but thanks!
Great realism and lighting, but i have trouble with adherence or maybe it just cant do some stuff. For example it can't do pelvic curtains for some reason. your pony and illustrious do it fine. Desire cant do them either. maybe setup wrong? I follwed your parameters in description tho. thanks
The Danbooru tag "pelvic_curtain" is recognized in Illustrious/Pony models. However, SDXL likely doesnāt understand what "pelvic curtain" means, because it isnāt trained to interpret Danbooru-style tags. SDXL works differently and doesnāt rely on that tagging system.
@CyberdeliaĀ Ah I i didn't know that. So it takes natural language/writing? Like flux/zit or not exactly? Thanks
@TheCALĀ Yes, but not extensive like flux/zit. If you really want some niche stuff then use Illustrious :)
Details
Files
cyberrealisticXL_v90.safetensors
Mirrors
cyberrealisticXL_v90.safetensors
CyberRealisticXLPlay_V9.0_FP16.safetensors
cyberrealisticXL_v90.safetensors
cyberrealisticXL_v90.safetensors
CyberRealistic_XL.safetensors
cyberrealistic_xl_v9.safetensors
CyberRealisticXLPlay_V9.0_FP16.safetensors
0-cyberrealisticXL_v90.safetensors
cyberrealisticXL_v90.safetensors
cyberrealisticXL_v90.safetensors
cyberrealisticXL_v90.safetensors
cyberrealisticXL_v90_fp16.safetensors
XL_v90.safetensors
fb-c05.safetensors
cyberrealisticXL_v90.safetensors
CyberRealistic_XL.safetensors
CyberRealisticXL_v9.0.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.








