"I like big knocks and I cannot lie..."

Maybe, the CLIP Skip of v23 Ultra version should be 2.

ill try give ULTRA a shot soon

非常好用

so good

Which version will you recommend for Training LoRAs? so we can avoid Training LoRAs for Each New Versions every time

Beautiful.

I still prefer 2.2, I'm trying to refrain from using another lora for furry gens. But when i tried using the other versions, it like the furry is trained out, like it pushes more for a human look/skin when I'm trying to make furry/non furry img.

How do you get good images result. All of the images I tried creating are all absolutely terrible?

GGG

Great generator!

great

This is not just another PONY checkpoint; it's a checkpoint in AI Generation history! Wow! Amazing, thank you for making this!

May I ask if the eyes of this model are not very good, and it needs to be combined with eye lora to achieve better effect? ​​I am using the same workflow, and when I change the large model, the eyes of other models are normal, but the eyes of this model do not look right, and sometimes they are even closed. It is very likely that there is something wrong with my usage method. I hope to get some guidance. Thank you!

IMHO V22 is always better...or the creators are borning now about the reality with ai :D

Make a sideversion for amateur realism. You have a ton loras for that. Skip furry whining and erase that semireal look, which is from v1 and didnt change a bit.

Great! Really makes awesome figures

bad, does not know artist names or styles, really bad at artistic images

Ultra is an inferior version of standart Realism, lol

‌Feedback on Model Versions:‌
I tested the ‌2.3 Ultra‌ version and noticed it struggles significantly with ‌male character generation‌—especially compared to ‌v2.2‌, which performs better but still frequently defaults to female outputs even when prompted for males. This issue is particularly pronounced in ‌NSFW content‌, where female figures dominate unexpectedly.

‌Question:‌
Is there a specific version optimized for ‌male/male homosexual content generation‌? I’d love recommendations for builds that handle such prompts more reliably.

Love using this one

Anyone know how to standardize settings across different software frontends/setups so images come out the exact same whilst using the exact same prompt data, config, etc?

Oestensibily, I import my own prompt data, etc, from my own images into different frontends; like SwarmUI, Forge, etc, but the images just aren't generating the same.....

Heard a bunch of explanations for this, but never a means/guide to actually produce the exact same results.

flux cant do nsfw,hidream too slow ,

Why was V2.3 removed from generations on Civitai?

it WILL NOT do portraits, it always is like from the knees up.

Not to sound totes dramatic or anything lol (I'm in a somewhat sentimental, sad mood though), but I just wanted to say how much I've come to really like this community (Pony Realism page in particular, because of course it's my fav model by far ^^).

It's looking like somewhere is getting the ole' ban hammer treatment real soon though ;( Sure can maybe use workarounds, I guess we'll see.....thought I'd just say something before that happens just incase ;(

Anyway, stay safe ZyloO & the rest of the AI community ;) <3 Best of wishes from this crazy broad ;)

PS - politicians really are a bunch of absolute useless cucks lol.

I am totally lost trying to figure out what ControlNets you can use with this model (newbie). Suggestions?

What is the difference between ponyRealism_V22 and ponyRealism_V22MainVAE, and why can I not find the latter for download anywhere, yet I see people using it?

after a dozen different models back and forth, i have to say thats the best of all v2.2. ill get near perfect hand, feet, eyes and face. but the best is it does what i want and not something random. and hey i generate pics on my shitty 5yrs old ryzen 4700u iGPU🤣absolutely unsupported cant work, should crash, but i did a 11hrs batch rush yesterday, its alive...

thanks for all your efforts!

tomorrow i go to do a comparision run with v2.3 then i decide which will win but for now its v2.2

I love using this model!

This model used to be good. What happened?

2.1 is back ! I love it !

Good model!

What is the best upscaler for this?

Highly recommended!

v2.3 is actually awesome!!!

Part of the perception/expectation problem people have with a lot of checkpoints is the version numbering.

Something like RealDream goes from v16 to v17; but there's usually barely any difference.... Hell, I've downloaded something like v12 through to 16, & there's still barely that much of a difference (& that's not even necessarily a criticism either).

Meanwhile, Pony Realism goes from v2.2 to v2.3, & a lot of people are probably thinking "Oh, a minor new update".....lol.

v2.3 is in actual fact quite a huge upgrade/change!

I didn't like v2.3 as much as v2.1/2 at first, but it is great! It's just a big enough of a departure that you can't just copy & paste all your old prompting over from v2.1 & expect great results. Just stick in there & learn to accept it's a different beast; & if you play to it's strengths & changes you will get great results.

TL;DR - Don't throw in the towel too quick without giving it a fair chance; because it is capable of achieving superior results if you learn to master it & tame the beast ;)

2.3 is really good. Thank you.

Can someone help how to use this on Draw things iOS app? I’m going to use it to generate NSFW with Juggernaut XL v9. Any help to improve my setup also would be great! Thank you so much in advance < 3

I'm using comfyui. does anyone know why the faces look so mushed or morphed? the bodies look great, but all my faces look deformed with ponyrealism. cyberrealistic looks fine, but it's terrible for sex scenes. ponyrealism is better, but has face issues for me.

Why are later versions getting less and less realistic? 2.3 almost looks like cgi, while older versions look photorealistic. I'm honestly a bit sad that 2.2 got taken down...

It's not just the realism either, as 2.2 is very versatile, being able to create just about anything, human or otherwise.

Meanwhile, 2.3 struggles to create even humans that don't look off, like oily cgi dolls or something... And just forget about anything fantasy coming out alright.

I’m gonna be honest, 2.1 is still the one I get the best results, and I’m always excited to go back to it when it becomes available in the generator.

So how many voted for 2.2 can't see results so am curious, also would like to know when 2.2 would be back up?

How do I know which loras work with this merge and how should I filter for them?
Is it SDXL 1.0 or 1.5 and can I use both?
Do I only need to filter for pony or better for pony 1.5 or so?

I don't understand why the generated results turn out different for me while i'm trying to make the 100% same results as your images. I used the exact prompts and settings for the images as yours, but they turn out different and worse than the one you showed. I'm clueless what i'm doing wrong :/

Anyone else find DMP2 work great for onsite generation?

Is there a list of camera angles that can be used for this model?

Please bring 2.3 back. I love Pony Realism but had better results with 2.3 (the non Ultra version).

Really Good models i find that version 2.3 Ultra is really good even if some of the most explicit NSFW need rework with loras and face detailler if theface is too much at a distance but definetly WOULD reccommend

How do put this model into SDAI on my android I can't get to my stable fusion storage on my android to put it in that way either

It's sad that fewer and fewer LoRA are being created for Pony. Plus, many of the old ones have been deleted, literally forcing people to switch to Illustrious. For example, for the recently released anime Clevatess only 21LoRA and all of them are only for Illustrious.

Hi! Using Pony Realism v2.3 + IP-Adapter Face ID Plus V2 in ComfyUI.

Issue: txt2img works fine with reference image, but img2img

creates character duplication (same face on both characters = twins).

Question: How do you prevent face duplication in img2img when

using IP-Adapter with multiple characters in one scene?

Current setup:

- IP-Adapter: weight 0.6, weight_faceidv2 0.9, end_at 0.75

- Negative: duplicate, bilateral symmetry, cloned face, similar features

- img2img denoise: 0.65

Any workflows that prevent this?

So recently dug up some old images made with this (2.2) and have to say it's still far superior to Illustrious for realism. Don't get me wrong, Illustrious is much better but I still use this in hi res or img2img

In my experience over a year: Version 2.2 ( I will be always thankful for that!) is alsways the best because of the balance between lora using an the checkpoint delivery. Version 2.3 ist ok with some fixes but v2.3 ultra is not really useful (experimental is really the good label). ;)

How inpaint version supposue to work? Examples are empty workflow. When i try usual inpaint workflow, this model ingornic prompts

Hi, how can write good position like "front" "back" etc... ? Thanks

2.3: what a fun model! it takes steps without breaking down (200-300+ steps). if you are prompt heavy its great to get everything in at really high quality and detail without oddities. so worth playing with. ty!

v2.1 Alt is still one of my favorite models. I wish I could get the same aesthetic from Rillsum.

2.3 ULTRA is still unmatched. What a masterpiece. Thank you very much for creating it.

GGUF version ?

Can someone help me please? I used the workspace from the image for v2.3 ultra above and apart from the filename for the checkpoint being different to the named download everything is the same. During ksampler the preview of the image looks fine but after going through vae decode it's coming out pixelated and multi coloured. I can only assume the version uploaded to civitai was not the one used to generate the images or I'm missing something.

I fine-tuned my model using the above base model, and it successfully generates images that match my intended prompts. However, I am facing an issue during image generation: the output does not accurately match the face of the person used in the fine-tuning images.

Is there any solution to improve facial similarity and ensure the generated images better match the fine-tuned face?

Unfortunately, it's still bad at generating groups of people (two, three).

Hello friends! It's a great model so far but it seems I am having trouble with feet! I thought at this point feet would be rendered normally but the toes are either clumped together or just lacking in detail. Can someone please help me out?

It's been a great model to use in stability matrix. One of my go tos when things don't wanna work right

Literally my favorite pony checkpoint ! Very good work 👌

Is there a list of recommended control net + ipadapters to pair with the checkpoint? I want to use it for some inpainting and detailing - where can I find the best/most compatible components to plug in?

I use v2.2 and v2.3_ultra, but not create correct image with bra, panties, hair color. My prompt it's correct?

score_9, score_8_up, score_7_up, female, latina, milf, red_hair, medium_hair, green_eyes, blue_panties, red_bra, looking_at_viewerdepth of field, highly detailed, high contrast, film grain, Rim Lighting

Hello! where can i find the workflow for this? I'm new to this and don't know

good at prompt adherence,but extremely bad at face generations. If you want to make a 3/4 shot, or your character should be full height, the face will be distorted at 99% of time. Use with Adetailer, or loras for better face generation. If comparing to stable yogi, I would recommend to use the second, as it has better results with faces.

Anyone have a dataset workflow to pair with this checkpoint, like one reference image to multiple image dataset workflow

I created and published a workflow to use the inpainting model. Check this out ! https://civitai.com/models/2483064?modelVersionId=2791685

just wanted to say, i am broke af, but if i had money, i would donate a bunch. your models rule since ages! TY!