(also posted on NSFW, but not really that so posted here for more viewers).
Let me get the appreciation of the free service, hard work out of the way first. Thank you to the devs for providing this service etc.
It’s reasonable to offer constructive feedback.
I’m disappointed with the AI girl generator. I loved the generator up to end of 2024 / very early 2025.
I’ve seen other posts on here that I think are the same opinion as me.
Why was it changed exactly?
- Somehow the devs thought it needed improving? It didn’t
- The devs got spooked at what people were generating?
With the generator up to end of 2024 / very early 2025 I was able to generate beautiful females like this:
Now I can’t. Perhaps some might say: “good”. Well beauty is a personal thing. We’re all here to generate something we like aesthetically.
The generator now seems to be just a selection of photos. Doesn’t seem to be real art generation. Quite boring images sorry.
Often when “beautiful” is described you get the same girls with cheek bones, glowing skin, same face shape, messy thin wavey hair, striking eyes, pouty lips, really narrow stereotypical “beauty”. Quite boring.
I’m going to get criticised for criticising but it’s a shame.
I’ve seen responses before like “oh you shouldn’t criticise something that is free”, but why is there this forum then?
What was the AI model used up to end 2024 / early 2025? Is there anywhere else this could be found?
It seems a lot of people are downvoting this post without even bothering to answer the question. I believe the older AI image generation model used some variation of the Stable Diffusion model. The new model is based on FLUX.1 schnell (according to the dev themselves).
i like the “dev themselves” part. thats how i want to be referred to when im a dev with no known gender who switches between like 12 different personalities. The Dev themselves.
No 20 personalities. More lore… 🐙🍎…
𒐪🐙🍎𒐪
thank you. That’s what the model was and what it was changed to.
But I wonder why the dev changed the model.
I’ve updated the question subject title, to also ask: “why?”
‘I’ve seen responses before like “oh you shouldn’t criticise something that is free”, but why is there this forum then?’
i find this part of your comment supercute lol. like the purpose of this forum is specifically to criticize free things lol.
i agree everyone has their own sense of beauty! i love My Little Pony ponies and beautiful furries and some realistic ppl and for all of those there are other ppl who can’t stand them. my advice personally is to just focus a little on something else right now; generating landscapes and anything you could use which includes Text are both great right now. The dev mentioned elsewhere they know the styles for people are messed up right now and first to be fixed will be prompt token length so it stops being cut off, which may help some. And a little trickier so a little later they are tackling the problems with people right now and how they are all getting shifted to wierd anime and stuff.
Because of that, if you can wait it out, it should improve sooner rather than later. I never used the gen you mention, but I think waiting is the best bet. i dont think you’ll ever be able to learn the exact setup it had been using. you could also use No Style and make a prompt from scratch that emulates what you want. i think thats second best solution and others here can help u better than i can with that or i can link u to a post with a comment that explained how. tho it will be silly if u make prompts adapted for how it is now and then perchance fixes the system in a few days.
like the purpose of this forum is specifically to criticize free things lol.
not the only purpose of the forum, but a purpose.
Does making something free mean that criticism isn’t justified.
Remember, I thought the model was good up to late 2024/early 2025. So I have both praise (for how it was then) and criticism for how it is now.
Again, why was the AI image generation model changed around early 2025? (It was great before then and up to late 2024 - well for me anyway for the images I’ve shared in my original post above). I’ve added to the possible reasons:
- Somehow the devs thought it needed improving? - It didn’t!!!
- The devs got spooked at what people were generating?
- the model was too expensive
- the model wasn’t being supported any more
Because of that, if you can wait it out, it should improve sooner rather than later.
But it was fine back in 2024 up to the end of the year / early 2025!
If they’d kept it as is, then I wouldn’t need to wait because they already had it working well!
The update of August 2025 up till a few weeks was superb. It produced artistic looking images in a variety of serious academic styles that I have never seen elsewhere. Beautiful realistic beaux arts, belle epoque realism, art deco, vintage pulp art, art nouveau, modern realism, gouache, watercolor’s or even fine colored pencil and so on.
There was finally the possibility of creating images that weren’t only anime, manga or cartoons like in all of the run-of-the-mill AI platforms. Finally a truly artistic rendering platform.
And now, no matter what I try, it’s as if two or three steps backwards has been taken. Impossible to render anything serious anymore. I just don’t understand why these present changes were implemented.
That’s what’s not being answered: WHY was it changed?
Did they get spooked at what people were generating?
I think what should happen is that they continue to offer all of the previous versions as well as the new, that some people like.
I just think it’s awful. I know I’m not the only one who loves to generate AI girl art - there are several perchance pages on it.
But the results now are just awful: they don’t seem like the machine has actually had a hand in generating art, just a stock of photos. Grimy, distorted if you ask for certain preferances like curvyness, bland. Either just dull and grimy or high contrast stereotypical beauty: cheekbones, lips, freckles, curly fine hair, jewel like eyes. So narrow and dull. Frankenstein stuck together images. Before it seemed like the machine was actually rendering real art based on the prompt.
I don’t think there is a problem in calling out something that is free as having faults. Why is a free service sacred? Sure whoever makes it wants to know that people enjoy it, find it useful and want to take on feedback.
I just really enjoyed creating girl art I found it really meditative and absorbing, a form of delightful escape. Here are some more I used to be able to create. They’ve been up on Instagram for half a year and no complaints, some likes - an audience of nearly 4K followers:
I do say that first line to people who are bigots, but the real reason? Even I’am not entirely thrilled everytime. New stuff is like…, can be fun though. The majority of model changes have to do with a range of things. In summary it typically comes down too performance/funding/&efficiency – resource management.
Pre May 2025, i guess they use SD, this is the time where we can use (example:1.3) in prompt and use negative prompt, lot of character recognized whether it fictional or real.
Post May 2025, they say they use Flux, no more (example:1.3) or negative really being accounted to generation, but i still like it since it generate great oil painting or painting in general, still high prompt2image accuracy, prompt still recognize painter name style, like greg rutkowski or alexi zaitsev. me personally like this version much…
Post August 2025, thing gone ashtray, lot of generation artifacts or simply generating random stuff, gradually fixed and improved till now (that what they said), but imo never reach the peak pre august 2025.
Btw, i kinda like this recent one, it’s improved if you generate painting style image.
SD - stable diffusion is what someone else said they used to use.
Why, then, when I visit: https://perchance.org/stable-diffusion-ai and try that I just get the same results as the other ones in use today? Can’t get the images I use to be able to get up to early 2025.
now, here what i understand within my grasp about perchance, i’m not an ai expert or coder or programmer, just your yet another regular user.
those generator (all really) are simply just a “front end” to serve you to use real perchance generator, some are simple and some more advanced with fancy setting, all can have their own names. You can create one of this “front end” and name it “SD generator” or “SDXL generation” for yourself but in the end they use the same current model that being used by the server.
All of those advanced fancy setting are just adding extra prompt words to the generator without you knowing it, you can check it by generating an image with only one prompt words, after images generated, click the “i” on the top left corner of the images to see what prompt being added by using those fancy setting.
so no matter who generator you using now, they are still using same model, for this reason, you will never achieve the same image result.
again, i’m not an expert, just user, take this info with grain of salt.
They used stable diffusion and from what i remember it was shit .most characters were super deformed and don’t even think about doing anything apart from having the character standing around or sitting down .i always used the drawn anime art style because everything else looked uncanny
