@Cyberdelia yes, you know what I mean, cannot provide example as this should not be allowed here lol

@leepeter1231 I think he doesn't know about animals, maybe about young girls with penises, yes, that would be more likely....and the style is not realistic, its kind of CGI (computer-generated imagery) with girls with penis and no animals....lol....all and that the model is not all bad, it have a nice textures.

@tupu I mean this model cannot generate girl $%^$^% with a dog

@Cyberdelia https://civitai.com/images/924327 something like this, but I wonder if this can be done in realistic style models

@leepeter1231 Yes, Ponyrealism or the illus version of the creator. Its semireal at best.

Further you push realism, less adherence remains. It is the law for Pony, Illustrious and any anime based model.

And there are dedicated furry models, search for them.

If you want more realism, you need postprocessing, or search for realistic helper loras.

@leepeter1231 I know what u mean

V4

OMG!! You broke gaussidorov score with this one!! First 10000/10 I've seen!!!

В чём заключается алгоритм?))

@AhuyKek  В том что модель ищет лицо например разрешением 120 х 120 в своей базе( она не будет искать лицо лучшего формата а будет искать подходящее по размеру) а у меня ищет лицо 1024х1024 и уменьшает его до нужного формата и вставляет на место

@AhuyKek  и так с остальными частями фото

മുംബൈ യെല്ലോ ഗേറ്റ് പൊലീസ് സ്റ്റേഷനിലെ കോൺസ്റ്റബിളായിരുന്നു രാജേന്ദ്ര ധോണ്ടു ഭോസ്‌ലെ. ആദ്യ പോസ്റ്റിങ്. നഗര വീഥിയിലെ തിരക്കുകൾക്കൊന്നും മാറ്റമില്ല.

@Cyberdelia Its 2.0 fault, bad resolution? V9 pony was same, but not before and v10 is normal too.

@timthomas131 I don't get this problem with most newer Pony versions, but see it constantly with almost all Illustrious-based models. You can kinda prompt around the long/elongated torso issue with negatives, but Illustrious still has a bad problem with overly defined torsos (appears very muscular/boney and generally, unnatural body definition).

@AFD_0 Just the newer versions of Illustrious or the old ones too? I dont use them much, so if just the new version have this fault, its resolution problem. 2.0 advertise itself as from 512x512x to1536x1536. I think none of them works. But ususal 1024p has problem too, it seems.

@timthomas131 I've noticed it from the start of realistic Illustrious models and just tonight noticed it in a Pony merge that obviously had a large portion of Illu mixed in. Early realistic Pony models had the long/stretched torso issue as well iirc, and it's mostly improved with Pony over time (until Illu started getting mixed), but I've never noticed the overly defined torso physique issue (oddly pronounced ribs/muscles) with Pony, just Illu. So far, I've only come across maybe 3 or so realistic Illu models that don't have this problem too badly (at least not as much as others). Perfection Realistic v3.0 and v3.1 by @6tZ is fairly good in this regard, Untitled v1.0b by @klaabu does this, but in a more natural manner, and same with Reality Master v2.0 AniMax by @hotfantasy_ai

Other than a few 2K-specific models from @Felldude , I've been doing everything in 1152x960 from SDXL to Pony to Illustrious, so I honestly have no idea if resolution is playing any role.

Most of my negative prompts use "(tall), (skinny), (long body), (long limbs), ((long torso)), stretched body, muscular", so that's probably having some effect on the stretched torso issue, but not sure why it's not as effective for Illu as it is for Pony.

@AFD_0 Bad resolution example, but here its none of this, just torso.

https://civitai.com/images/75153280

@robertreeve359 Yes, that's a very extreme example! Having gone through every single Pony (and Illu) test from the @Lizardon1025 test samples, most recent Pony models usually aren't like that, unless they're very recent Pony-Illu hybrids (though I'm honestly not sure if that's the case with ZyloO's PR v2.3, but the earlier versions seem fine). Some very early realistic Pony models had stretched torsos as well (without the exaggerated bone/muscular definition), but until recently it hasn't been bad imo. I'm sure with time, the realistic Illustrious creators will make the same improvements as they did with Pony.

@robertreeve359 @AFD_0 chiming in - the elongated torso is usually caused because a) bad resolution chosen by the user, b) lack of training images with said resolution, c) lack of concept prompt, leaving room for hallucination.

I've tested ny illustrious model up ro 2036px, it works great, but - it can and does make long torsos if you're not locking concept in with prompts.

I find often that people have got very used to the way Pony prompting works - just 175 tokens of vomit, and when working on Illustrious, they often struggle cause they lack the proper prompt architecture.

@klaabu Good info, thanks! Any idea what's causing the overly-pronounced muscular/bone structure on Illustrious torsos? Even on results with little or no torso stretching/elongation, many Illu models still exhibit oddly exaggerated ribs and/or distended stomach that's not present in any SDXL or Pony models. Some (like yours), do this in a more natural, realistic manner, but with most others it just looks weird and somewhat alien.

@AFD_0 It’s due to a combination of dataset bias, aesthetic compression, and latent overfitting. Use targeted negatives or body-focused sliders to rebalance this effect. I'm actually releasing one over the next few days that helps with natural toned look :)

@klaabu Awesome, looking forward to it, thanks again!

@klaabu "I find often that people have got very used to the way Pony prompting works - just 175 tokens of vomit, and when working on Illustrious, they often struggle cause they lack the proper prompt architecture"

Illustrious clips are nothing special. They are not t5 or LLM. You can add 500 token gpt noise, the base is booru tags. Nothing to do with long torsos.

@jasongriffis394 that’s fair, but missing the point slightly. My comment wasn’t about the CLIP’s intelligence—it was about prompt tuning. Pony gets away with bloated prompts because it leans heavily on pattern saturation. Illustrious needs tighter architecture because it doesn’t interpolate the same way. It’s not about LLMs or intelligence—it’s about how token relevance impacts latent injection. Prompt sloppiness on Illustrious breaks the image. Simple as that.

It's not that hard to solve this problem, I made a model that specifically solved this problem, but it was decided that the character was too young to be killed off.

Use the lora in the link to make N long torsos/short torsos (with the same seed and cue word) with various camera angles and character poses, train a lora to use the long torsos as reg images, and uniformly add a tag such as “long Torso” to the original cue word.

It's only difficult to prepare the image, because it has to consider the character generalization problem, and it's difficult for current real style models (it can't recognize people's race like other sdxl models).

@Cyberdelia Thnx!

have you tried not being gay ?

Thank you! this is much better now.

I do fine with 4x_NMKD-Superscale-SP_178000_G on A1111, but ill try out that nickelback aswell

i can recommend these 2 i tested as best ones when sharpness is important https://openmodeldb.info/models/4x-realsr-df2k and https://openmodeldb.info/models/4x-UniversalUpscalerV2-Sharp

@xperia256 check out your result with, since this is a more realistic model https://openmodeldb.info/models/4x-PurePhoto-RealPLSKR

If you didnt already know you can use loras for that :D

@Cyberdelia Great! Thanks!

yes

The AI is trying to tell you something...

@Cyberdelia Oh, i didn't knew about the auctions until now. Thanks for answering me! I hope it becomes available soon! I like this checkpoint a lot!

@Cyberdelia That's got me curious, how come there are new generations being made and published on site considering it's not available at the moment?

@Cyberdelia How does one do that? I'm still learning the ropes.

@Cyberdelia Thanks man

Also, how did we go from sd1.5 models with 15 fingers to illustrious models with exactly 3 fingers? Shit is crazy its 2025