Join The Tinkerer on Whop to download this model and many more — all for one monthly payment. Plus, get early releases, private tools and a lot more. 👉 Join on Whop
CyberRealistic Pony blends all the charm of Pony Diffusion with the striking realism of CyberRealistic. The vibe? You get everything from adorable to bold (sometimes both at once) with crazy-detailed textures, moody cinematic lighting, and a hint of AI flair.
🧠 How to Use It
Sampling method: DPM++ SDE Karras / DPM++ 2M Karras / Euler a
Sampling steps: 30+ Steps
Resolution: 896x1152 / 832x1216
CFG: 5
Clip Skip: 2✅ Recommended Prompts
score_9, score_8_up, score_7_up, (SUBJECT), Negative Prompt examples:
score_6, score_5, score_4, (worst quality:1.2), (low quality:1.2), (normal quality:1.2), lowres, bad anatomy, bad hands, signature, watermarks, ugly, imperfect eyes, skewed eyes, unnatural face, unnatural body, error, extra limb, missing limbsit's also possible to remove the normal pony tags:
(SUBJECT), Negative Prompt examples:
(worst quality:1.2), (low quality:1.2), (normal quality:1.2), lowres, bad anatomy, bad hands, signature, watermarks, ugly, imperfect eyes, skewed eyes, unnatural face, unnatural body, error, extra limb, missing limbs🛠 ADetailer Settings
Adetailer model: face_yolov9c.pt
If you only want the main face being refined set 'Mask only the top k largest' to 1.☕ Support
Enjoying the ride? [Buy me a coffee] – but only if the Pony delivered something you actually like.
⚠️ Disclaimer
This model might generate sensitive content. Whatever you make with it is on you. Don’t do anything weird and try to blame the horse.
🔗 Links
Backup location: huggingface
💡 Need Better Prompts?
This custom ChatGPT was made to whip up top-tier prompts just for this model:
🔗 [Try it now on ChatGPT]
Description
An alternative version featuring more moderate and balanced color tones.
FAQ
Comments (92)
Hi, what is the alternative?
Is this a continuation of the 5.0 line?
I realise it's a bit vague to use "Alternative" (renamed it to V12.7 Alt) it's basically V12.7 with more moderate colors, in case you find the original a bit too intense
Cyberdelia Thank you so much for the alternative version. I thought I was just hallucinating / prompting badly when I considered the colors "too strong" in the original. Much appreciated.
Cyberdelia appreciate it - there's a lot of newer models that have that saturated look with highlights overexposed, i imagine its an (unintended) pursuit of engagement and likes.. but natural is better, even if the kids flock to shinier/sweeter candy than what's generally tolerable. many people with serious workflows (i.e. for work purposes beyond porn) use editors in post, and that crisp colorful sheen kills what can be done with levels and details. just my two cents.
i'd love it if all creators released alts like this, cause with time constraints, relying solely on genAI clients can only get you so far. PSA: learn some basic image editing techniques like (Curves adjustments, clone/stamping, grid-warping, frequency separation...) with Photoshop, Affinity Photo, or even Blender/Krita etc, see how valuable neutral image aesthetics are.
12.7 alt has some really nice tone balance..
Really love your models, but seeing many platforms bow down to the abhorrent demands from the payment providers and the new strange EU regulations that are coming. I wonder which place would be the best place to follow your work in case shit hits the fan?
Good question and I don't have an answer. Civitai is my main website where I publish my models.
These are some sad & troubling times we find ourselves living in indeed ;(
ShadowCell Yep, getting drunk regularly helps take the edge off
cute
Version 12.7alt is another step in an amazing improvement in quality. Superb work. Thank you.
Does this model work great?
It’s not just amazing - it’s a mind-blowing, earth-shattering, history-defining masterpiece of cosmic brilliance that redefines the very concept of greatness itself!
Cyberdelia This might just be one of the best, most amazing responses that I've ever read on the internet in my life.
Thanks for offering this great alternative. Both version are great but I really love its neutral tones and it is a powerful in-painter.
was thinking of going back to 6.2 because of oversaturated colors and you dropped natural color version! How can you read my mind like that
Looks very realistic, great model
I was using V12.5 and V12.0 for a long while.
Recently started using V12.7 Alt - The quality is a huge jump from the prior, Was so good i made this comment!
Nice, appreciated!
This is my go-to checkpoint. I'm basically not using anything else anymore. It does everything I want it to do with astonishing realism, detail and fidelity. It's basically life-changing. If you're not using this and want photo-real images of girls then you're the biggest FOMO-Queen this side of Arcturus.
This model and every pony needs a photostyle lora.
To remove uncanny valley eyes and look and apply a 4k modern cellphone style, based on real common shoots, or just high quality photoshoots(not high fashion).
recommend one, we shalll try
Are you using a detailer on the eyes? I get perfect eyes this way with this checkpoint.
is this model designed to handle well 1024x1365 resolution?
TheNecr0mancer I've used the latest 3 models to create images in 960x1280, 1024x1344, 1024x1408, 1152x1536 and 1216x1600. If you use the chatGPT assistant linked in the description, it can help provide prompt tuning (reordering and small adjustments) to optimize when changing the resolution.
shinobitj89 I just recently 100% transfered to comfy over from 50/50 comfy, forge. and I just realized I can not set 1024x1365 resolution in efficiency node. Sad, as that is perfect 3:4 resolution with minimum 1024 small side
Any chance of making a Lightning version of the checkpoint?
Thanks for sharing your work with us, this is the best realistic model for Pony that I ever used :)
I noticed I can generate beautiful girls with this model w/o using Pony score tags at all.
Correct. I don't use these tags almost never anymore
What's the difference between 12.7 and 12.7 Alt ?
You can see the difference in the first 2 examples of 12.7 and 12.7A. The colours are "more" balanced.
Alt is better imo should be the default and the normal one should be the alt
Is it worth upgrading from 12.5 to 12.7?
What are the improvements?
Not really noticed a difference TBH. 12.7Alt and 12.5 feel almost the same. I avoid 12.7 though as the colors are just too bright.
Im having an issue here, all of my characters trained on pony, their appearance doesn't fully translate to this model or any other realistic model for that matter. I've also tried training directly on this model and others but it just ends up looking like my person lora but with 2008 mmo graphics when you turn off the shadows and lighting effects.
Anyone know how to fix this? With sd1.5 my lora always come out looking identical to the person.
Thanks
To follow up, the main issue is that the models are like "muting" the features of my character. They have a slightly longer facial structure with defined cheekbones and jawlline and a slightly longer pointed nose. Yet during generation they always get these features muted into rounded checks, squareish round head shape with button or potato nose.
jstark962 just put that version here: https://civitai.com/auctions
jstark962 Even though you find it "stupid" that I make it possible to put some checkpoints in the generator, I’ll try to add V10 next week. It’s just very expensive (+15,000 buzz these days).
Hey guys any response to my initial question? 💀
TheNecr0mancer I can't reproduce your problem. So older version doesn't have this problem?
Cyberdelia do you train character loras on your model, or base pony?
TheNecr0mancer mostly on base model - but also on my models. Trained 5 models last week on this version without any problems. Excellent results.
Cyberdelia can you give me some direction about settings for the trainer? i'd like to try it. (when watching the samples of my training it always looks correct, which makes this even weirder. even the shitty sample quality looks like that character)
Are you training locally (own PC) or remote service? If local, check out my detailed video on my very successful training process using this checkpoint. Link in all of my character LoRAs.
Generation3dX hey, thanks for the comment, great video. full of useful information for people who might be new to lora creation. I knew everything in the video already, but I watched the whole thing anyways. Great job. I will give training on this model another shot, i've always been a prodigy and adafactor user, but I will try a low learning rate Adamw run.
Same problem here, trained on cyberrealistic pony : character very strange with very few feature.
I noticed that with dmd2 all is good, very strange ...
TheNecr0mancer thanks! One additional tip I learned after making the video is to use Cosine with Restarts instead of a constant LR, and set your checkpoint saves to the lowest LR in the cosine curve. This gives you the highest fidelity results, as the model is nice and settled at these points.
TheNecr0mancer yeah, mate, same here, model seems to reproduce a consistent character based only on promt, trained a few loras, and when i apply them with certain character description, the model seems to overpower my loras and generate a consistent character, which is nothing alike from lora
doctor183 i've gotten some "decent" results now, but i have to hyperprompt the character for it not to come out a generalized mush. if i dont hyper prompt the character it just looks like a burnt to crisp mess. so now i need to get around this issue. (when i do hyperprompt i.e. every single detail of the character in training when actually trying to make an image in comfy, then it comes out great)
Little problem here, i tried to train a lora on your model and the result is great ... with dmd2. I tried without dmd2 and it appears that the result is like the lora just don't exist or/and is ugly (style 3D washed). I really don't understand how it is possible ...
Tried many version of my lora, without success...
Getting better each time I try it. I wish I didn't have to use prompts like masterpiece to get higher quality. As a newbie, I would hope this would be assumed that I always want it at the highest quality.
Cool
One of the best that exist.
[newbie pov] : yeah. i'd love to generate realistic people. but what does pony means? and how its fundamentally different from checkpoints like Realistic Vision V6.0 B1?
Pony is trained on lots of anime, so it will inherently be less realistic than a typical model. However, Pony is good at prompt adherence because it was trained on lots of specific booru tags (think of tags in images like 1girl, 2boys, smiling, blonde hair, blushing, etc.). So this is an attempt to make it as realistic looking as possible while still maintaining the booru tagging system. That's why score_9, score_8_up, etc. are used as tags because they were trained as booru tags
🔥🔥🔥Cool
Bro's cooking 🔥
I found that generating at 960x1440 gives some incredibly good results. Do not use score tags.
That's a non-standard sdxl resolution. I tried it and got weird results.
@AntiAntiDoubleNegatives Prompt adherence seems to suffer a little bit, but quality skyrockets in my experience with this resolution. Anything from 4-7CFG seems to work well depending on prompt.
I’m having issues with the outputs. When I set the CLIP Set Last Layer node to -1, it generates either a completely black image or strange brown patterns. Switching it to -2 or bypassing it results in lots of artifacts in the generated image. Any solutions?
Legend says if you set it to -13,6 you get to see big bang
i get the same issue. no idea how to solve it.
Why not remove the node? This is the problem?
https://www.reddit.com/r/comfyui/comments/19742ml/what_does_clip_set_last_layer1_mean/
Excellent model in terms of image quality. Thank you very much for your work.
Unfortunately, after version 8.5 (maybe later, but with 8.5 everything was definitely good) it started drawing anal sex very badly. Especially from behind. In 9 out of 10 cases it tries to draw the anus, and move the penis lower into the vagina, or somewhere in the middle. Neither increasing the positive promt nor adding negative ones helps.
It would be great if it were possible to fix this in future versions.
Could you PM me the prompt? On my Patreon I’ve got a version that’s based on an older CyberRealistic Pony with some new training maybe that version doesn’t have this issue.
I'd like to humbly campaign for version 11 to get bids for the auction if any Buzz Daddies are feeling generous.
Alright, sorted. But only ‘cause you dropped the Daddy card 😏
@Cyberdelia You're an absolute champion thanks so much for doing that!
i think this may be the best right now for generating AI porn pics, this is the most trending model for sure. it does work good, nice capabilities
It's the best
I noticed that the 12x models don't handle img2img upscaling well. With "Euler a" and 0.1 denoising, it ends up smoothing the skin too much, giving it a slight cell-shaded look. Any recommendations?
If you're using the A1111 webui, try the "Extra noise multiplier for img2img and hires fix"
i noticed that it works better at 720p than 1080. or is it my fault?
Great model! Would be amazing to see a controlnet model trained for this at some point
Love this model!
The best model I've used the last year!
Only one issue I can't seem to fix. All women I get are in their early 20's. I've added an age in the prompt, I've used an age slider. But I'm never able to get a women older than that.
Am I doing something wrong?
It's not a bug, it's a feature ;)
I’ll take a look to see if I can adjust things. Sliders should work though.
Have you tried booru tags? "mature female" and "older woman" tend to work decently at least on version 8, although old woman can become cartoonishly ugly depending on your prompt.
@oneeyed2 I did. It's seems to do just a little bit, not too much though
@Madman1980 @oneeyed2 I’ve uploaded a model on my Patreon that was made with a different method using another base and fresh training. I’ll be testing it today.
@oneeyed2 @Madman1980 Ok I tested the version of my Patreon and V12.7. Don't see a problem: https://civitai.com/posts/21604058
@Madman1980 if you remove the pony tags?
score_9, score_8_up, score_7_up,
update: nevermind... it's not the problem
The custom ChatGPT prompt is awesome!! Thank you so much.
This is the model youre looking for, this is the one that does the stuff the others don't do. Enjoy
Can anyone share some tips on how to create a consistent character with this? ideally with all the same features and while using this same model please! any help will be welcomed (still fairly new to this)
use a character Lora or something like this: https://civitai.com/collections/9670053
You can combine a character lora (low weight 0.3) and a TI. You get better consistent characters with CyberRealistic SDXL.
Great job with this custom GPT, is very helpful
I've been chasing the v8 magic with every subsequent release. It created perfect, sweet faces... am I the only one?
https://civitai.com/images/71441637
Yeah mate, same... But with v8.5. Quality of faces and bodies with my setup was on pair with FLUX. But every new release after that one continues to disappoint, no matter how hard I try to adjust :c
This model has long been one of my favorites, and if I have any ranking of the best models here, for it's use and fun, this one is still at the top. Thanks so much for the great work.
But I need some advice. I am unable to create a prompt for a picture of a woman standing on her feet with her hand placed between her legs and performing masturbation, but not naked, but over her clothes. Like a woman in an elevator, hand between her legs, masturbation and wearing jeans and a t-shirt. Maybe someone can give me some advice.
V12.7 Alt best realistic Pony I've seen so far. My go to if I wanna generate beautiful models and NSFW content. Played with different versions, currently happy with v12.7 Alt. If you wanna see a few (easy to replicate) favorite results of mine:
v12.7 Alt:
https://civitai.com/images/97977180
https://civitai.com/images/92571621
https://civitai.com/images/97559246
https://civitai.com/images/97649494
https://civitai.com/images/93444337
https://civitai.com/images/96814744
https://civitai.com/images/94213382
https://civitai.com/images/92878229
https://civitai.com/images/89051202
12.5:
https://civitai.com/images/86497682
v11:
https://civitai.com/images/80957008
https://civitai.com/images/79465612
v10:
https://civitai.com/images/76625441
https://civitai.com/images/79232766
https://civitai.com/images/77880015
https://civitai.com/images/77874970
https://civitai.com/images/77841594
https://civitai.com/images/77289459
v9.0 Alt1:
https://civitai.com/images/72069408
v8.5:
https://civitai.com/images/70421718
https://civitai.com/images/75882150
https://civitai.com/images/72473101
https://civitai.com/images/71932335
https://civitai.com/images/63538939
https://civitai.com/images/62309767
v8:
https://civitai.com/images/58107076
v7:
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.