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
āØ 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 knowā¦ 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
A bunch of tweaks and upgrades to make things smoother and less likely to make you scream at your screen. Or, put simply: I changed some stuff.
FAQ
Comments (42)
uuuh new XL checkpoint... downloading ASAP. Thanks and Merry Xmas Cyberdelia!
Merry Christmas!
whoaaa the undisputed best model gets a christmas update. excited!
Hope you like it.
I just see this now!!! I'm super pumped and surprised!!!! Thank you
I love your models. Thanks for the effort!
Nice - thanks
Make it porn capable, and you will be the Emperor on every SD field. 1.5 and Pony are the best, make it for XL too.
Have you tried it? I don't generate NSFW sample images but it can run Crysis... sorry create NSFW images
@CyberdeliaĀ Yeah, its mild NSFW, nude only, nothing hardcore.
There are 2 great hardcore on XL, but they are not compatible with celebs and character loras. Nobody can make an ultimate version. Maybe you?
@culleyvirgilio346Ā maybe cyberrealisticPony_v7?
@culleyvirgilio346Ā which 2 models? lustify?
Is there any way to get the data set?
No sorry. I prefer to keep my dataset private. I hope you understand.
The only thing is that I would like to ask you some questions.
You should release your checkpoint in DMD2, itās really very effective. Great work!
Lol, dude! You seriously need to tidy up everything:)))
We have 2 hardcore checkpoint, which is good on XL. One creator is retired, other one is a crybaby, if i criticise something.
Make it man, show it!:))
So what is your request?
You are the only one, which i can ask for that.
As i mentioned, there is only 2 hardcore model on Xl, which are good. But they are merged with BigAsp, which is a catastrophe.
Bigasp trained with 6 million datas. Around 5.9 million is useless, low quality shit.
Your dataset is clearly superior, and high quality, only need traine for hardcore, and we get the best on XL.
People love loras, and celebs, unique girls, Bigasp erase all likeness from them. So its up to you.
Im very much happy with your Pony, but sometimes i want a more realistic Xl approach, and celeb girls of course.
@azlanshivank171Ā Ah ok - got it.
I do agree that Cyberdelia's dataset is far superior. I think a lot of people make the mistake of thinking that if you just throw everything into a dataset that it'll turn out good, but truth is, you need high quality data. That truth is been repeated multiple times by the best minds in AI, even in LLM training.
Dude great stuff like usual. I wanted ask you, where or how do you train models? I want to try and start training SD 3.5 Large and Medium, might become good one day.
Check SECourses. Great resource if you need to experiment. https://www.patreon.com/c/SECourses/home
@CyberdeliaĀ sure dude thanks. But can you recommend any specific trainer for fine tuning a model? And hardware requirements?
Hey. Thank you for cyberfix.
Yes, well... Itās not my creation, but Iāll gladly accept your thanks! ;)
@CyberdeliaĀ you saved anatomy :3
@EnigmataĀ If this gets me into the history books, I canāt complain.
@EnigmataĀ What is cyberfix?
@niwo439Ā Big anime finetune of IllustriousXL (NoobAI-XL) where SDXL 1.0 base was replaced by cyberrealismXL with checkpoint merge tools.
@EnigmataĀ Thank you! :-)
10/10 nice!!!!
Hello :) do you have any nice workflow for ComfyUI?
Not really. What I currently use is a complete mess, so I prefer using Forge as an alternative.
How would people say this compares to juggernaut? Is this model good at creating various ethnic women?
Please, train V-prediction (V-parametrization + zero term SNR) version, our anime saviour from realistic isekai realm š
Why do you need so hard Noobai, if its shit and need a lot of fix, merge,? It needed Illustrious, a fix from cyber etc. Ridicuolus.
@averyonolen361Ā Lol it is the best anime model that trained ever. But with swapped SDXL 1.0 on cyberrealism XL it is better in anatomy. But epsilon-prediction in recipe of v-prediction model makes picture brigther. It is the only weakness. No reason to use anything else except cyberfixes for NoobAI Epsilon and V-pred.
Not a lot of good SDXL around- but this is definitely one of them.
Seems to like more detailed prompts, but still haven't fully got a handle on it.
Amazing model, but it leaves a visibly lighter box around peoples faces when you use ADetailer.
No other model does it, including 3.1, and I've tried reinstalling Forge, ADetailer and redownloading the model?
i have the same, waiting for a new good update :) this one rocks
I hope you keep improving on this because I use this model more than anything. Seems to work the best for me on people's faces and bodies.
Details
Files
cyberrealisticXL_v4.safetensors
Mirrors
CyberRealisticXLPlay_V4.safetensors
cyberrealisticXL_v4.safetensors
cyberrealisticXL_v4.safetensors
cyberrealisticXL_v4.safetensors
cyberrealisticXL_v4.safetensors
cyberrealisticXL_v4.safetensors
cyberrealisticXL_v4.safetensors
cyberrealisticXL_v4.safetensors
cyberrealisticXL_v4.safetensors
cyberrealisticXL_v4.safetensors
cyberrealisticXL_v4.safetensors
SDXL cyberrealisticXL_v4.safetensors