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
In view of a number of comments this Revised Version 2.0a featuring a customized VAE with reduced oversaturation.
FAQ
Comments (119)
Really really impressed with this model. Well done!
Version 2 is good, but not ideal for loras anime characters, minor inaccuracies are visible. The anime style is also supported, but only the general one; the more distinctive loras styles are not noticeable.
Thanks for sharing the comparison images. Much obliged. I'll look into it.
Great job on the alpha releases! I much prefer this more "natural" look :) thanks again.
2a is much better imo
Which one is that?
@seyik83688983 V2.0a
It is a pretty good model and would have blown me away a month ago. However, at the moment there are so many other ponly models released and I feel like compared to some of those, this one feels a bit off in regards to realistic looks. I have the impression that it looks slightly 3D rendered. Still, very good model that I'll try from time to time. Good job! Edit: I feel like this model makes particularly beautiful faces.
My aim wasn't to create a hyper-realistic model, despite its name, but rather to create a style that I personally like. It's a bit semi-realistic. I'm working also on a 1.5 semi-realistic which I really like. Hopefully this will be ready this month.
@Cyberdelia I see, that makes sense. I should also note that I feel like your model also makes particularly beautiful faces.
@bagofchips I'm currently developing V3, which has a slightly different feel compared to V1 and V2. However, the issue is that people often complain about me releasing too many checkpoints. Therefore, I might try releasing some samples first and see the reactions.
This checkpoint has ranked the highest for me in terms of realism. Which Pony variations surpass it for you?
@focalbluebell problem is that if you want more realism you lose the Pony characteristics. It's a constant process of fine-tuning and testing (which I love).The semi realism which it now has is what I personally like. @bagofchips also curious what are more realistic checkpoints with the original Pony characteristics.
what pony merged model you recommend for realistic looks ?
@amazingbeauty Cyberrealistic Pony v1, Real Dream (https://civitai.com/models/153568?modelVersionId=523031)
Cybers are my fav models for realistic, pony unlocks more; cute, cool or eros, you choose.
Thanks Cyberdelia :)
One of the best realistic models I've tried.
Just FYI:
If you go to Civitai's main page, click on Models, activate only those filters :
Model Type: Checkpoint
Checkpoints Type: All
Base Model: Pony
For some reason your model does not appear in the (rather short) list of results. Maybe it's worth investigating why. I "found" your model thanks to someone talking about it on some forum. But I would have never found it straight from Civitai.
This doesn't help the overall rating of Cyberrealistic Pony in the list of the best PonyXL based Checkpoints as you're not even there O.o. When your merge should definitely be in the top 2 with Real Dream (https://civitai.com/models/153568?modelVersionId=523031)
You must have something else checked as well or there's a bug (unless its been changed in the last 4 hours) because I get a ton of results for pony checkpoints including this model
Can you make this available for Civitai on site generation?
This is something I have no control over. This is decided by Civitai.
@Cyberdelia You do have to tick that box that says "allow on site generation" though for the model, I guess you did that already?
@diffusionfanatic1173 yes I did it when releasing the model.
I submitted a ticket
@Cyberdelia look forward to seeing this available, I think it has my favorite color balance of all realistic-ish Ponies, it looks MUCH less "washed out" than most of them
V3 is now available for site generation.
@Cyberdelia It's off again it seems
@diffusionfanatic1173 yes, only 2 models can be used for site generation.
And because Cyberrealistic and Cyberrealistic Classic are my babies, that's why I made the choice. I had to make that choice from Civitai.
Will try to fix this
Really like the look of V1 and it doesn't even look very oversaturated at all so far. I even used the Sharpspectrum VAE on top of it to slightly increase saturation and contrast and it looks great! I'll post some images once I get to it. Love your work!
problem of "oversaturated" VAE in multisteps processes, you get good image on first step, and hiresfix made it one time again and ruin everything.
@randomizer89 I'm actually doing a fair amount of passes though. But I'm using comfyui and keep the image in latent space throughout most of it. I could imagine that might help with the issue, cause I'm not VAE decoding-encoding as much.
Thank you for all your effort and time in making this model. I didn't use the model yet but I know it will be good ,cause everything you touch turns to gold..lol
Thank you :)
V3 is awesome, great work!!
My new favorite Pony Model 👍
Bit of an issue that I've encountered:
Generating consistent small breasts (lol) has been difficult. With other Pony models, I can usually get there by putting "small_breasts" into positive and "huge_breasts" into negative.
On this model, doing that will generate a lot of flat-chested women. And adding "flat_chest" to the negative then makes them very busty instead, lol.
Adding only "small_breasts" to positive generates a lot of busty women. And adding "huge_breasts" to negative at all seems to generate a lot of flat chests.
That's an odd thing and I also don't no where this occurred in the process of creating this model. I will do some test.
I've just tried it with V1.0a, V3.0 and regular pony side by side, with a simple prompt and exchanged only the small/medium/large_breasts and they all behaved consistently across the board. You could try weighting the tags or move them to a different place in the prompt. But there's probably something else in there that suggests a size, such as "cleavage" would increase the breast size to a certain level for it to make sense. Clothing items such as "corsets" can also heavily bias body shape, as well as many other things. Prompting often yields unexpected results.
Just yesterday I had a character always generate with flowers in her hair and lollipops in her hand. Turned out the lollipops came from the word "eyecandy" (huge recommendation btw. to get weird/surreal/cyclops images with about a chance of 1/100, while generally just enlargening eyes a little; in non pony models at least :D)
And the flowers in the hair were the result of using "daisy dukes" in her clothing and "bloom" in the lighting description.
@redpinkretro hmmmm perhaps, but I've tested this with the exact same prompt, across multiple (fixed) seeds, with several "Pony realism" tunes, and Cyber is the only one that trends so strongly to flat chests w/ "huge_breasts" anywhere in the negative prompt.
I'm aware that certain keywords can have an unexpected butterfly effect on the resulting image, but I don't believe that's what's happening here. My testing is far from exhaustive though, so I could be wrong.
@roe2 Oh, I didn't actually check the sensitivity of the negative on this. Maybe it was tagged in opposites, like "big_breasts"↔"small_breasts" and consequently "huge_breasts"↔"tiny/flat_breats", like a complete inversion of the vector. Interesting actually.
could you check if the new version fixed this 'problem'?
I'll double check, but "medium breasts" or "small breasts" in positive, and simply "large breasts" in the negative seems to do the trick. What does the underscore do to a token in Pony model? Have you tried just "large breasts" with no underscore? I noticed sometimes, the underscore between words will totally break an image for me - I just get blue or brown abstract pattern.
@Cyberdelia v3.1 seems to have fixed the issue, yes! Thanks <3
I didn't get a chance to try v3 yet, but just tried using v3.1 and getting weird artifacting throughout images. Thought it was a VAE issue but doesn't matter if using direct from the checkpoint or the normal SDXL\Pony VAEs. Tried multiple samplers and tried it set up exact like example images. Not having this issue with other animated or realistic Pony checkpoints. Nor did I have this problem with prior versions of this one. Maybe someone else can anyone else confirm I'm not just an idiot and doing something wrong? lol
I don't have any issue, maybe try another config, like Euler A Karras, cfg 5 on 25 steps
"Have you tried turning it Off and On again?" ;)
@C0rn_Fl4k3s Weird... I definitely already tried that... and yet it works fine now. Thanks.
Still getting that issue with DPM samplers. I only went with that because it was what seemed to be used in the examples. Oh well. Live and learn. Thanks again. =)
@theclassicalgod318 same problem after adetailer only : )
@Cyberdelia lol I did. I even tried flagging for fp16 since my old 1070 sometimes needs that for some things, but I guess my card and this checkpoint just don't want to cooperate when using DPM samplers. Not a big deal. At least we figured it out.
My favorite and most reliable model by far.
love it
Can anyone recommend a Disney character style Lora's that will work well with this? And thanks for this model!
Really nicely done. I love the lighting. 😎👍
Love it. The only thing I noticed is 3.1 version has a tendency to cut off legs in sitting or on all fours positions
Will this be available to use on civitai again? seems to have gone and it's a shame
Will try to fix this
@Cyberdelia legend. Cheers
@purelyforai fixed
@Cyberdelia Superstar!!!
good job, very versatile and creative.
Tried it cause FallenIncursio RTXSEKAIDEEE didnt gave me enough realism to achieve with some content, and i didnt got dissapointed. Some PonyXL loras act weirdly with this one but results are what expected mostly. For begginer like me its more than to ask for! Thank you for hard work with making this, its much appreciated. Working with this checkpoint was very satisfacting. My conclusion is that RTXSekaide is more reliable with loras but if someone want true realism that is breathtaking (literally) then here is the go to !
Well, it seems like version 3 is really better than version 2 in relation to loras. Although it seems to me that version 2 gives the characters a slightly more European appearance, and version 3, a little more Asian. However, I am not sure about the latter.
This is amazing. Congratulations!!!!
Since this is a merge, can you provide a recipe? Would like to know what you've used to create this, and also crediting original authors would be nice.
Hi coyote. I understand your question, and you should indeed credit the original creators when using their material. Technically, my checkpoint is a mix but not of other Pony checkpoint. Of course, I use Pony Diffusion V6 XL as I mentioned in the description, but for the realistic part, I did not use other models like your model "DAMN! [PonyXL Realistic Model]" (which, by the way, is a good model, I just tested it!). As I indicated, I used CyberRealistic XL with some modifications. So, once again, it is technically a merge. I do not use the term Checkpoint trained - I do use trained material in each model, but for example, I consider PonyXL a truly trained model while CyberRealistic Pony is not. Hopefully, you understand why I labeled it as a merge.
Once again, DAMN! is quite impressive, good work!
@Cyberdelia I didn't imply you used my model, just was wondering what did you use to create your model. Thanks for the insight, and for you kind words. Good work on your model as well!
@coyotte Yes, I know that, and that's not what I implied either :) Used only your model as an example.
👍 I think he's the best realistic model Pony, I love him.
Thanks for the Buzzz, but I think you need them more than I do and as I have a lot that I'm not using I can give you some if it helps you with development.
Keep it up, I love it! ❤️
Also have a lot :) and I like to give some buzz for awesome images.
@Cyberdelia ok, thanks a lot 😘
I keep getting this error, which basically crashes A1111 -
"A tensor with all NaNs was produced in VAE. Web UI will now convert VAE into 32-bit float and retry."
Other than that, I like the results when I can get it to work.
that's strage - I started using Forge a few weeks ago, and I must say, I should have made the switch sooner. It's much faster and avoids all the hassle I experienced with standard A1111.
Add --no-half-vae to command arguments in webui-user. I love Forge but the team will not really maintain it going forward, they will experiment with Gradio 4, etc. in that repo. They recommend switching back to A1111
@silverion aha good to know
@silverion Thanks. Are there any drawbacks to adding --no-half-vae? Any reason why I might not want to add it?
@zoesixxx I have not seen any material impact, however, people anecdotally reported that image preview is slower (you can control how often image preview is rendered during generation and I would recommend bringing the frequency down or disabling preview anyway). It might use a bit more VRAM.
The best base model! 5k+ images with SDXL and Pony.
Stick with this one :).
Tip: Use refiner and/or different checkpoint on hires. Use something like sdxxxl (suggestions?) that has both good realism and sexual concepts.
My formula is approx. this, but I'm still messing with it. 60 steps. Refiner to a normal SDXL model at 0.6 - 0.75. Hires to the same normal SDXL model, different negative prompt that works well for the model, 30 steps, 0.3-0.45 de-noising.
This will can give you amazing realism with the wild concepts available from pony. Unfortunately, it will not give you back the great backgrounds and scenery as the core fundamentals of the image is created by the point the normal SDXL model takes over.
Will try it - good advice!
Cyberdelia is #1 and deservedly so.
V4.0 is amazing!! Huge improvement over v3.1, outstanding work :)
Hi, congrats for the first place in base creators. A long time deserved honour. Outstanding creations and base models all around. If there's someone that can achieve photorealistic style on a pony model is Cyberdelia. It's just a matter of time.
Yes, I saw it, but I'm not really focused on that. There's a chance I'll drop down the list again tomorrow, which is totally fine :)
@Cyberdelia you'll never drop that place.... in our hearts!!!
@ElectricDreams this is so cute
⠀⠀Let's compare some PONY models 🦄
⠀⠀⠀⠀🌍 See here my detailed test 🌎
For example, I find CyberRealistic Pony to be clearly superior to Pony Realism. As far as realism goes, Pony is clearly in first place for me!
Leans too much towards very young girls. Even if you add mature, you get a weird mix of childlike and mature. In terms of realistic fantasy-like results, it does it well.
Consider giving this model a spin if you like the look of this one.
It is genuinely great, more tags would help this model get more visibility.
so pretty!!
What are the author's recommended sampler, clip skip, cfg and steps?
i havent been able to use any other model not sure ill ever need to
Failed to Generate, is this unavailable now?
It was remove to Civitai Generator. 🤷🏻♂️
Yes, don't know why. Maybe to NSFW?
@Cyberdelia Wait, Civitai pulled it without giving a reason? That's messed up.
@Cyberdelia It's the best checkpoint. I miss it already. Not sure how it's any different that any other 'real pony' type checkpoints as far as NSFW goes...
What's the reasoning behind pulling this from the onsite generator availability? Is a new version in the pipeline?
I find this very weird myself.
Not available? Is there a new model on way?
Will this be available onsite again? Its one of my favorite models.
Just found this model and it’s gone, sucks since it looks awesome.
I really don't know why they pulled it. Still no response from Civitai
@Cyberdelia Could you please consider posting them on HunggingFace or something ? I found your SDXL models there, but not the Pony versions. I read many positive comments about v.5 here for some reason. But sadly I missed it and worry I might never find it anywhere to give it a try and understand what's so different about it :/ Thanks.
@hansolocambo you should have checked hugging face
Come back soon. (˃̣̣̥ᯅ˂̣̣̥)
I really don't know why they pulled it. Still no response from Civitai
@Cyberdelia Not even over Discord? For a creator like you to be ignored in such a way seems very odd. =/
@Cyberdelia I really like this checkpoint it is a shame it not working currently in the generator. The sad thing is it shows in the generator but fails.
man.. i love your models! Great work and big thanks!
Incredible model especially version 5. I don't know what it is about the PONY models but they seem to be the best overall. Even so, I believe it can get better. Is it possible if you can make the model so that it is more receptive to the desi cafe PONY embeddings and Loras? Or perhaps version 6 can be made so that it can create characters from more races and skin types? Thanks.
Great to see it updated! but I don't see it in models. ):
Crossing fingers that 5.0 can be used in civit, devs!!!! Please allow this model again :'C
great model. in base res comparable to other realistic pdxl models, but the magic happens when upscaling to 4k. only model that doesnt get smeared or unsharp overall look and can generate great skin details or fur details at hi-res.
thanks a lot dude
Just asking but isn't it better to have Waifu2x and PhotoAX-Ersgan upscaler through it rather than doing it inside diffuser? I kept doing it for a while that way and it works magic asf and goddamingly always when i was upscaling via Auto1111 the image got destroyed in some ways.
@magmirant0011 For absolute maximum quality upscales, you'll want to use https://github.com/Coyote-A/ultimate-upscale-for-automatic1111 or something similar. The downside is that it takes a long time, since it's generating a bunch of images and stitching them together.
upscaler models dont add new details. and i tried a lot, but using comfyuis ultimate upscaler is worse than using controlnet (i.e. the new "one for all" controlnet ) and high res upscales. i tried up to 3072x3072 and tiled upscale is always worse. but of course this is very limited approach, because without a high vram grafka there is no other choice than using tiled upscaling...
Details
Files
cyberrealisticPony_v20a.safetensors
Mirrors
cyberrealisticPony_v20a.safetensors
cyberrealisticPony_v20a.safetensors
CyberRealisticPony_V2a.safetensors
cyberrealisticPony_v20a.safetensors
cyberrealisticPony_v20a.safetensors
cyberrealisticPony_v20a.safetensors
cyberrealisticPony_v20a.safetensors
CyberRealisticPony_V2a.safetensors
cyberrealisticPony_v20a.safetensors
CyberRealisticPony_V2a.safetensors
cyberrealisticPony_v20a.safetensors
CyberRealisticPony_V2a.safetensors