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
V13.0 takes things in a slightly different direction. It dials back some of the moodiness and intensity from earlier versions, aiming for a smoother, more versatile feel when you’re working with it.
It’s not a straight-up upgrade. Think of it more as a tweak that brings a different vibe. If you liked the bold, dramatic style of the older builds, you might still stick with those. But if you’re after something a little softer and more balanced, this version might hit the spot.
There’s also some focused training under the hood to fix specific issues from past releases. That makes it a bit more stable for certain workflows.
Overall, V13.0 isn’t here to replace the earlier versions. It’s just another creative option to suit whatever mood or style you’re going for.
FAQ
Comments (70)
Regarding the fp32 vs the fp16 version: was is the to-be-expected different outcome in the generated images?
(amateur here, unhealthy half-knowledge at best)
Is it better understanding of the prompt? Details to the scene? Are there some simple terms to describe it’s advantages/disadvantages?
Obviously it’s double the size. So that’s a disadvantage on the effort end.
It’s minimal and almost negligible. It can be useful if you want to tweak CLIP or do other exciting things. But my advice is to just use FP16, even though I personally use FP32. Nowadays, I always try to offer both versions.
@Cyberdelia 🙂Thank you!
EDIT: thx @Cyberdelia for the advice. Adding "ultra photo-realistic, best quality, ultra-detailed, realistic" to the prompt helps V13 to get realism more closely to 12.7 Alt.
Original Comment: V13 quick comparison and first impressions (examples see down below or on my profile). I would say... Less realistic, sharp and detailed but more vivid and vibrant. Tends to go more towards 2.5D CG aesthetics instead of realism. Of course could also be just prompt / LoRAs for my examples. We'll see what people get out of V13 tailoring the prompts and LoRA more to this version. For me personally... only first impression: I think I will stay with V12.7 Alt for now and occasionally test V13 again. So not worse or better, just matter of taste.
V13: https://civitai.com/images/98139024
V12.7 Alt: https://civitai.com/images/92571621
...
V13: https://civitai.com/images/98140427
V12.7 Alt: https://civitai.com/images/98098845
...
V13: https://civitai.com/images/98139412
V12.7 Alt: https://civitai.com/images/97559246
...
V13: https://civitai.com/images/98145841
V11: https://civitai.com/images/80957008
...
V13: https://civitai.com/images/98140177
V10: https://civitai.com/images/76625441
...
Damn, appreciate your work, this is super helpful. Will be a pass from me due to the noticeable step back in realism.
@welcometoea @Guthwulf It’s indeed partly about prompting.
Starting with “ultra photo-realistic, best quality, ultra-detailed, realistic” helps, for example.
The effect you see in 12.7a appears to make it more realistic, but it’s not truly more realistic.
By removing that effect, you now have more freedom to create that same look using a LoRA instead.
Think of it like buying a car:
Car with a built-in navigation system → convenient, but you can never turn it off or remove it, even if you’d rather use your phone or another system.
Car without a built-in navigation system → you can later choose exactly which system you want to use, without being stuck with something that’s always in the way.
This way, the second option gives you more freedom and flexibility.
@Cyberdelia thx for the explanation. I usually try to avoid using to many LoRAs and I never made one myself. So I would rely on hope that somebody made such an LoRA to give me that sharper, more detailed and realistic look that 12.7a had built in. Any suggestions?
EDIT: "ultra photo-realistic, best quality, ultra-detailed, realistic" indeed helped btw.
@Guthwulf I don’t use LoRAs either, so I can’t really give any advice. But specifying it in your prompt should work. If you look at the first example image, you’ll see what I mean. It’s a bit more work, but it gives you a lot more freedom.
@Guthwulf But as I already mentioned in the description, it’s possible that some people might prefer a slightly different style. That’s perfectly fine, of course.
Put the magic prompts into negative. There is no magic prompt? Its not true. People back then tried everything with the first sd models and gathered a lot of great prompts.
I would add to positive: photo \(medium\), real life, hyperrealistic too:)
Really like the output of V13. Silky smooth and crisp outputs,. This would be a great checkpoint for someone who likes to pile on the LORAs when they work
Exact, that's exactly what I intended. I saw an incredible number of images with a ton of LoRAs (sometimes way too many), and a clean checkpoint is perfect for that.
this ckpt is very compatible with a lot of loras from other model based series, i like that.
Yo this Version 13 is really good. like its almost FLux Level of detail.
wow... 13 looks realy good, nice color, nice realism, nice contrast... very nice Picture :) Thanks for it
From the beginning. Version 12.5 was already amazing, especially with realistic images. I almost completely abandoned Flux for this version! 12.7 only confirmed that I didn't need anything more for realism. But 13 is literally heaven. Not only is the realism excellent, but this version follows my prompts better than anything else! Please keep creating more versions of this wonderful tool.
Have you ever tried CyberRealistic SDXL?
If this already excites you (and rightfully so), that one will absolutely blow your mind.
@Cyberdelia I will try it.
@Redurtni_Lana and? :) I’m really curious about your experience. Personally, I use my SDXL model 90% of the time, and to me, it feels superior yet somehow, the Pony model is way more popular, and I’m not quite sure why.
@Cyberdelia You mean this: https://civitai.com/models/15003/cyberrealistic or that: https://civitai.com/models/312530/cyberrealistic-xl ?
@Redurtni_Lana This one https://civitai.com/models/312530/cyberrealistic-xl
@Cyberdelia sorry, have you tried generating futa images?
@Sebi73 Nope, but I have generated a lot of questions I regret answering
Thank you for your brilliant work. This model and series of CR models are nothing short of spectacular. I raise my hat (and other things) to your awesomeness xxx
I have better results with 12.7 Alt than with 13. Good for landscapes. For humans, CyberReality has many problems with hands and feet, other realistic models are better for humans.
It strongly depends on the composition in what I have tested so far. But I have no formula, its a bit hit or miss.
@mad_rooky You're right. The model is good, just more unpredictable.
@_8_I_8_ unpredictable is the word, it really does struggle with with some concepts unless you're willing to use skimmed cfg nodes and conditioning concat but then you will need to refine to fix problems. Most realistic models have moved so far away from pony base they just don't have the ability for a lot of nsfw concepts and that is just too bad. The best we can do it use a none realistic model then through controlnet into a realistic model such as cyberrealistic, but then that can mean a lot of swapping which slows every down.
@rogerstone442 I've found Pony Realism (PONY), JedpointReal (PONY), Real Dream (PONY), Fucktastic Real Checkpoint (PONY) and some PONY and ILLUSTRIOUS models by Stable Yogi to be surprisingly excellent. And not only for NSFW. The main limit is always the pose of the character, it is quite difficult to obtain a dynamic stance. But the quality of the rendering with these models is pretty impressive in only 20-30 steps. But I agree, still we have to swap between several models.
Hey, I'm having a problem where the faces look terrible. They're blurry, shifted, or even missing. Does anyone have any advice on what I'm doing wrong?
Since you are not saying what you are doing, I can only tell you are doing it wrong, but how is your secret^^
@mad_rooky It happens with every generation I've tried so far, regardless of whether I enter a lot of details in the prompt or none at all. I'm using the Automatic1111 Stable Diffusion web UI with the basic settings. I've tried various sampling methods and schedule types. I'm new to this. I just have no idea how to proceed.
@wlkoelsch486 show your image prompt details, with scale, steps, etc
@wlkoelsch486 If I am not missing something, A1111 is hopelessly outdated, you should switch to Forge UI, its a fork of A1111 with some extra features but most importantly its still maintained. Perhaps the problem starts there. Because normally getting basic images with a model like this that come out half way okay should be super easy. If you look at the images around here and the prompts they are using, and you can't replicate that at all, its pretty certain that the prompt is not the issue.
Had the same problem, and simply adding "ugly, deformed" in negative prompt made it way better
Have you tried using the recommended face detailer face_yolov9c.pt?
Cool model
This checkpoint is pretty good at understanding general languages prompt! good job
The model is really interesting and allows me to create images with the level of realism I prefer. Unfortunately, I just can’t figure out how to generate multi-subject images — it seems to ignore any syntax (or more likely, I’m just not doing it right). Can anyone help me?
Have you tried using BREAK between subjects?
@marcin3005 only between two subjects? I tried using it to separate the parts of the prompt, but it doesn't seem to work.
@SpankingFake
It usually works for me :(
Read this for an explanation of how it works:
https://www.reddit.com/r/StableDiffusion/comments/197vffn/how_does_break_work/
does not work on civitai generator
Question: why do you recommend adding "score" to the prompts? Is it a sneaky way to convince the model that it should give good quality results?
About the “score” tags - Pony XL was trained on booru-style data, which includes metadata like score_7_up, score_8_up, score_9, etc. They’re not some magic quality switch, but they can steer the model toward images that were more popular (which often means cleaner lighting, stronger composition, or familiar styles).
That said, you really don’t need them. If you check my latest samples, none of them use score tags, and the results hold up just fine. If you're curious, you can always test one like score_9 to see what it does.
very good
Is inpainting possible? How to add an additional character to image? What settings should I use in ForgeUI?
I'm personally getting much better results from this than 12.7 Alt, but I use it exclusively for big butt girls to animate. My X profile has examples of V12.7, I'll be posting some V13 very soon to compare.
Great work, OP.
I seem to be having more trouble than usual with CP generating four-toed girls. Both alt-12 and 13. Any tips?
Jeez, use a different term or acronym.
@Protonium True got scared there for a sec at least CR or smt
@moosie88 No, just go for it - everyone should cherish their fetish. These days, there’s already so little we’re allowed to enjoy, so treat it with care :)
And maybe, just maybe, try not to call CyberRealistic Pony “CP.”
It makes it sound like you’re into something that requires an FBI disclaimer instead of four-legged ladies.
@Cyberdelia Oh crap, that acronym went right past me. I am sorry.
I'm not trying to cherish a fetish. I'm trying to generate normal feet and toes, and having difficulty. It seems to be more of a problem than in earlier versions, so I wondered if anyone else had similar issues, or better had a solution.
Getting black images. Using sample workflow. Any ideas whats wrong ? Thx
Are you using Forge/ComfyUI?
I had the same issue with ComfyUI. Generated black images only. Removing the "CLIP Set Last Layer" node solved the issue for me.
@Damaskar Yet it helped. Thanks
Trying to use this with futanari concepts, getting frustrated when I see excellent work from other authors and a total garbage from mine :(
Guess the problem is in my crooked arms and not in the checkpoint. I am so upset ;(
Gg
Nice
Vers nice
Has the default realism been toned down for 13? If that even makes sense lol
If feels like I need to front-load my prompts with realism tags, otherwise the subjects look almost plastic
12.7Alt seems to be realistic out of the box.
I know Cyber said it's not an upgrade, so just sanity checking myself ❤️
Noticed a huge downgrade in realism from 12.7alt to 13. Very disappointing. Bring back 12.7alt!!!
I don't want to oversell this checkpoint by any strecht, but Cyberrealistic Pony V.14. has changed my whole life. Best buzz ever spent. This is truly a work of art from a master mind. Absolute must. Just buy it now.
A bold statement, giving it a shot.
This is the 10. model, what changed your life, but i still waiting for EventHorizon 2.0, which continues the v1 aestetic.
Once you buy it, can you generate through the site? I can’t see any create button for it even though I paid
@username9960361 On-site generation is separate and not all models allow it. It has to have the paintbrush next to the version. For v14 you'll need to use an external generator like Forge or ComfyUI.
@username9960361 @WarmKnowledge6820692 I will put this model in the generator this Friday. I think it should work.
@Cyberdelia Thank you!
Details
Files
cyberrealisticPony_v130.safetensors
Mirrors
CyberRealisticPony_V13.0_FP16.safetensors
cyberrealisticPony_v130.safetensors
CyberRealisticPony_V13.0_FP16.safetensors
cyberrealisticPony_v130.safetensors
cyberrealisticPony_v130.safetensors
cyberrealisticPony_v130.safetensors
cyberrealisticPony_v130.safetensors
cyberrealisticPony_v130_1.safetensors
cyberrealisticPony_v130.safetensors
cyberrealisticPony_v130.safetensors
CyberRealisticPony_V13.0_FP16.safetensors
CyberRealisticPony_V13.0_FP16.safetensors
cyberrealisticPony_v130.safetensors
cyberrealisticPony-v13.safetensors
cyberrealisticPony_v130.safetensors