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
Fully revamped version made compatible for the Civitai image generator.
FAQ
Comments (63)
nice
How come the latest version aren't available for generation on here? Such a good model
It has been a decision by Civitai. I respect it and hopefully today's version will be added back into the generator.
@Cyberdelia Thanks for the reply. It's a shame. I really hope so too
I'd love a hyper/lightning version with low steps.
You can get some good results with 10 steps when the subject is close in / zoomed.
I don't use civitai's generation and use your models on other platforms so is it worth it to download 6.1 ? any major difference ? currently I have your 6.0, 5.0 and 4.0 models and I much prefer unrestricted or neutered versions so what's your opinion ?
You have to decide yourself. I personally like 6.1
What other platforms are you using? Never used online generators before
@Cyberdelia I used be on mage but now I mostly use tensorart and automatic1111 webui and thinking about creating my own image generation platform.
@Cyberdelia so I'll have to get 6.1 now too cuz I'm not gonna stay behind on your models which are by far my fav.
@Shah_G I now realize that not everyone has a fast internet connection. While I can download a checkpoint in a few minutes, it naturally takes much longer with a slower connection, and then you have to make choices.
based on the comparisons 6.1->6, it looks like cyber got skin texture, tone and lighting dialed in more naturally (notably freckles are less 'IG filter' and eyes are brighter in those mid shots). i think it's a pretty obvious side-to-side improvement.. i wasn't planning on grabbing this but i think it's worth it
@Cyberdelia Thankfully I do have a decent connection so it doesn't take much time for me to download. I just control my urges and only downloads models that are worth it. Your models are the first ones that I actually downloaded and then duchaiten models
@Narz I just downloaded it and will be experimenting it with all other versions as well. I just have only problems with sometimes weird eyes and mostly weird hands which I can't figure out a decent way to prevent those. Might as well try cyber negative with pony and it might help other than that don't have an idea which loras can work with pony models fairly well atleast for handling bad hands and faces (not considering adetailer)
@Narz one thing I like about cyber pony models is for me in cases cyber pony keeps a perfect balance and doesn't take the results to an extreme as compared to other pony models.
@Shah_G I don't use cyberrealistic negative a lot. Just some simple negatives.
what is your ideal sampler and vae (if not baked in) for v6, v5, and v4 models. I don't see any specific settings mentioned other than basic positive and negative prompts.
Custom VAE is baked in, but you can change that. Just use your favourite VAE if you like. Sampler: DPM++ SDE
@Cyberdelia thankyou so much. I was mostly using DPM++ 2m and 2m karras but i'll try sde ... just need to compare my prompts through different versions of your model. Maybe a different setting might suit my needs
@Shah_G DPM++ 2m karras is also fine
Crossing fingers devs allow online generation of 6.1🙏🤞
Btw if they allow it, what sampler would you recommend for site gen? I've using Euler A with 2.0
That should work fine. I use DPM++ SDE but you can experiment with others
@Cyberdelia gotcha, will experiment but I must say I saw 6.1 on another site and tested it (with DPM++ SDE), really nice version ^-^
@Tinchisx oh what site? Just curious where they use my models :)
@Cyberdelia In Yodayo site.. Btw in that one I could use DPM++ SDE and the 5 girls I normally do here with 2.0a version, they get different face and with Euler A it's a little different too.. Now I want them to allow it here to test with different samplers!
@Tinchisx will check it out.
It's back in the generator! Happy happy joy joy!
@Cyberdelia I'm happy they allowed 6.1! Well deserved! I've been testing here on site and "DPM2 a" is nice, just wish civit have DPM++ SDE D:
V6.1 is a big improvement, it looks so good! Genius work!
I had made a stupid mistake in the previous versions. The result was somewhat semi-realistic, which I actually found quite okay. Removing the stupid problem + some fine-tuning made it a lot better.
@Cyberdelia That's great! I feel like overall detail is impressive here, especially skin looks really nice and realistic, I'd dare to say it even beats some non-pony sdxl models
The image quality with version 6.1 improves a lot, really appreciate your effort.
I had made a stupid mistake in the previous versions. The result was somewhat semi-realistic, which I actually found quite okay. Removing the stupid problem + some fine-tuning made it a lot better.,
@Cyberdelia would you mind sharing what that mistake was?
I found 6 very good on structure and detail and could jump between real and comic with ease. Compared to realism model teh skin texture was more smoothed and a bit fake BUT not as burned as realism so it was either side of the line. Adding a loRA solved, but I will try 6.1 see what changed maybe this is the ultimate.
I was back and forth between everclear, realism and now cyber.
Just make a decision already! ;)
@Cyberdelia LOL the problem is there is rarely a clear winner. I was leaning to archiving the other 2 and using this, I suspect 6.1 might tell me for sure,
@creeduk I know :) every checkpoint has their pros and cons.
@Cyberdelia - after a lot of side by side testing in comfy and doing the slide compare it seems 6.1 has better skin details and hair is typically more realistic. 6.0 has the blemishes but very faint, 6.1 brings them out, but I found eyes and teeth seems consistently better in 6, 6.1 seems to have more artifacts, does that make sense with the training? Any setting tweak that would account for the difference to pull the best out? I am running at dpm++ SDE 30 steps cfg 6 same seed.
Nice. I'm very glad to see CRP return for onsite use. <3
Thank you so much my friend great to see model back up and better than ever ...cheers
Thank you. Appreciate it!
6.1: Bring back Kristina Pimenova to the dataset!
Cyberdelia is the best!!!!
This is exactly what i've tried to say for so many months but nobody listened to me!
very good
Does this model know any celebs in its dataset?
Thank you for the 100 tip, this one the best models in the pony line up. I am able to get a lot of great detail and flexibility with Loras.
I will be using this to learn more
Hello! I made a comparison of Pony models. 1 test prompt so far. More ongoing.
https://civitai.com/articles/6491
Comments / suggestions are welcome.
the best model I have found.
Can you make this checkpoint in to a Turbo or lightning checkpoint? Thanks for considering.
have you tried using the pytorch lcm XL weights lora?
I feel like this model struggles with bright/sunny/clear sky background, deals great with darkness and less colourful ones.
Yes, but the world is also dark and not so bright nowadays ;)
Thanks, I didn't really notice. I'll look into it.
With version 6 I can get nice faces, but as soon as an open mouth with a tongue sticking out comes into play, the checkpoint no longer knows how to display it. Both ADdetailer and 'face restoration' only produce an immature image! With version 6.1 I can create a maximum of three images, after that there is only 'colorful fog'. I have been using your SD 1.5 checkpoints since the first days, but you can forget Pony.
And one more thing..., with DPM++SDE you make a pony checkpoint incredibly slow, which might not be a problem with 'nicer' results, but the results are limited
Can you please make an Inpainting version of this - your inpainting models for 1.5 were the best!
Best model and creates amazing results without burning too much buzz. 10/10, recommended.
Really Appreciate the model and design revisions in 6.1! Fantastic work!
Where can I find the CyberRealisticPony_V61a used by the example pics?
CyberRealisticPony_V61 and CyberRealisticPony_V61a are the same.
Has anyone managed to successfully prompt for skin tones/ethnicities? Everything else than "asian" is completely ignored. I can't get anything with triggers like african, latina, dark skin etc. Only get white or asian girls. Funny enough hair colours and lengths work well.
I think it has to do with how many ethnicities are in the base model. In my experience most model lack it but some are predominant to use their own basic face and even with lora it is hard to remove it. My best experience is a skin slider that interestingly makes the AI kind of use some dormant images from ethnicities with such skin. Maybe using a well known person from a certain ethnicity will work too, as base for mixing.
I usually find getting more specific can help, i.e., colombian or kenyan instead of latina/african. Also different combinations and weights of "dark skin" "very dark skin" "dark-skinned female/male" "tan" can sometimes work in unintuitive ways. Can also try a dark-skinned "innate" character like karin (blue_archive) or nitocris (fate) and then tweak their features.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.