Alright, I need to get something off my chest because I’ve been testing, experimenting, and re-trying the same thing over and over again, and no matter what I do, it just doesn’t hit like it used to.

Let me be clear from the start: I used the exact same method to describe what I wanted. Same wording, same structure, same prompts — nothing changed in how I approached it. But the results I’m getting now are completely different. And not in a good way.

Let’s Talk About These Two Batches of Images The two image batches I posted above were all meant to show the same character.

Same outfit: black tank top, tight jeans

Same setting: outdoors, in motion

Same identity: blonde woman, serious or determined expression

But that’s not what I got.

What I got were completely inconsistent renderings:

Faces that don’t match at all from one image to the next

Body types and proportions jumping all over the place

Lighting and tone shifting randomly

Styles flipping between semi-realistic and full-on plastic

It doesn’t feel like one woman doing different things. It feels like ten different women who vaguely meet the same description.

And if you’re trying to build a character-based scene, this is a dealbreaker. It’s not useful. It destroys continuity, and it makes actual storytelling almost impossible.

Now Look at This One Image That Worked That one image I keep pointing out — the one I actually like — that’s how it’s supposed to be.

Same hair

Same style

Same expression tone

And most importantly, she feels like a real character, not just a generated output.

That image came from the older model. The model that actually understood what I was trying to do.

It didn’t try to give me endless “fresh” variations. It gave me a consistent look. It gave me her.

And This Isn’t Just About One Character I’ve got more characters like her. Full sets. Different faces, different personalities, different stories — but all generated using the same approach I used for that one character.

The old model handled them all just fine. It let me build a cast of recognizable characters that stayed visually consistent across scenes.

I wasn’t making random image prompts. I was building a story panel by panel, and I needed those characters to hold their identities from frame to frame.

The older model made that possible. The current one does not.

The New Model Doesn’t Respect Identity This new model gives me something like: “Here’s a blonde woman. That’s close enough, right?”

No. That’s not close enough.

Because I’m not asking for random beauty renders. I’m asking for a character.

The one I designed. The one I wrote for. The one I want to appear in multiple images in a consistent, believable way.

And the new system doesn’t treat characters as something to preserve — it treats every prompt like a brand new concept. That completely breaks the flow for anyone trying to build a visual narrative.

Want an Example Everyone Understands? Look at Ben 10. Look at Ben 10 — the original series. Sharp art, clear lines, a consistent design that made the characters feel real and memorable.

Then look at the new Ben 10 reboot. Same names, same basic characters — but the style is all over the place. Watered down, simplified, cartoonish to the point of being unrecognizable.

That’s exactly what this new image model feels like. It’s not that it’s completely broken — it’s that it lost the essence that made the old one powerful. It forgot that consistency matters just as much as creativity.

What I Actually Want I want the ability to:

Lock in a character’s look

Control how she acts and moves

Adjust her facial expression

Place her in a sequence of scenes

Keep everything in the same tone and style

I don’t want to re-roll twenty images and pray one of them looks close enough. I want to build scenes, not hope for lucky pulls. This isn’t about complexity — it’s about continuity.

Are There Slight Differences? Of Course. Nothing’s perfect — I get that. Even the old model had quirks.

But at least with it, I could generate 200 images in a session, scroll through them, and actually find what I wanted.

I'd go: “…ummm nope… nah… mmmmm maybe… huh… oh that one… and… oh, I like that one… that one… ummm no…” And that process worked — because they were all still in the same style, tone, and form. There was cohesion.

Now? Everything’s scattered.

Side by side.

Two batches of failed generations

5 images that got it right ( the singles)

Same prompt. Same intent. Same user

This isn’t a user problem — it’s a generation model problem.

And if the new system can’t deliver character consistency anymore, then someone needs to step up and either bring back the old capability or build something better.

Because this isn’t a nitpick. This is the difference between random images and real storytelling.

Some of us came here to tell stories — not settle for scrambled visual noise.

for the love of god give us back the old model

And just to be clear — it’s not just this one character. I have other characters I created using the old model, and if I tried to describe them the exact same way now, I’d get the same broken results like in those inconsistent batches.

The old model helped me a lot with my storytelling. It gave me characters I could build with — characters I could rely on to stay consistent.

So please, with all due respect: Bring the old one back.

I don’t like this new model. It’s lost what made the old one so effective.

Bring back the version that worked — the one that understood character design, visual continuity, and story-driven consistency.

Bring back the ((())). ps the one at the bottom is supposed to be at the top with the others

Idk what happened but it said that verification failed, and when it does that, anything didn't come out. I already turned off the ad protection thing, and it still didn't work out. I didn't use VPN in case u wandering...

Look at how users of perchance had been decreasing and then ai art update and whoosh, dovnward trend goes upward just like that.

Remember the time when the tldraw plugin was released, the app literally creates anonymous canvases for your generator when you import the plugin? Well, tldraw is apparently making those canvases read-only by July 1st, just got this notice when opening one of them.

This popup will appear every time you load any tldraw canvas frame from the tldraw plugin. Thought of something like putting a "warning" in the plugin page or something else.

Haven't been posting on this community for a very long while now (been focusing on a lot of new things) so I apologize if I made some mistakes.

Hi, I'm looking for a way for the AI to understand a hex color in the format #fff000, for example, if I get it from a color html element. If you directly ask the flux plugin to show a #fff000 shirt he will use a random color instead, so ideally I need to translate it into words, if that's possible at all? Or perhaps, there's another way.

Just looked at the stats & I see that the users have dropped from 111 to 69 this week...i've been away for a week and was hoping things may have improved on here in that time - they haven't - still painfully slow,still a lot worse than the previous version...not good.

I'm practicing more and more with the new FLUX model and something bothers me: I find that the faces of "random" girls are still very stereotypical like in fashion magazines, especially when it comes to lips which are always oversized. It looks like the girls have had botox injections for ten years.

I tried with prompts containing terms like "random girl, thin lips",, and in the negative "oversized lips, botox, surgery, implants, silicon, fake", but nothing works, I always get a clone of "emily ratajkowski".

Has anyone ever managed to make a generation with these thin lips like Kirsten Dunst's?

By the way, does anyone know why every generation, even without style, always adds "natural beauty" in the prompt? I think it distorts the action.

At first, only the fingers were distorted when the model was released. Recently, finger generation has improved somewhat, but now I'm noticing that hand and limb distortions are increasing. The limbs used to be fine before—only the fingers had issues—but now the limbs are getting worse over time.

Hello everyone, I just joined and I got to say that these AI image generators are awesome and addictive. I love this website. I only have one problem. I am trying to create multiple characters looking at me but I can't seem to do this. I tried commands like: They look at me. Or: Their eyes are focused at me. I also tried replacing "me" with the center of the image. Or the center of the scene. It does not work. I can keep generating images and hoping they will have their eyes focused on me but the more characters the more chance one will look the other way. Can anyone save me and tell me how I can make them look at me?

Hello. Please, I need the help of a dev. I posted an image to public gallery by mistake. I couldn't delete it. I have use this generator: https://perchance.org/ai-text-to-image-generator and the image in question is https://aigc.uploads.dev/image/445c01874e1cffb30195647a4c82b05ed3293bb821685606d8138edcf91a5c87.jpeg Thanks a lot!

Here's my container or iframe for multiple plugin or link

...
  <div id="chatIframe4" style="width: 100%; height: 100%; overflow: auto; zoom: 0.55; border-radius: 0.25rem;" hidden>[image(settings.galleryOptions)]</div>
...

Only the [image(settings.galleryOptions)] would be stuck in initial load in color scheme; the [commentsPlugin(settings.defaultCommentOptions)] ; [settings.introMessage] ; perchance uploader or private gallery would sync with the gen theme toggle.

Before this I use long src url methods and it could have theme updated too!!

What's the trick here?

There have been many banned users—whether from chat or gallery—who claim they never committed the violations in question. Maybe they’re not telling the truth, but sometimes it really seems like they share the same IP ID or are using the same VPN. I don't know if that's actually possible, but I tend to believe them.

This makes the need for individually deleting or hiding gallery images even more urgent. We shouldn't ban everyone affected just because of someone else’s actions.

I need a new AI text model asap. With this model, the AI very easily forgets things often. For example; on one message, we go sit on the couch together to talk, and 2 messages later, the AI completely forgets that we are on the couch and goes: "Let's go sit on the couch, okay?". Like, we're already on the couch, there's no need to go and sit on another one. It might just be because I have a huge description for my character, but I need that long description, and I can't change it. Please tell me the AI text model will be updated soon, it genuinely sucks ass as of right now.

I am not sure how I feel about image renderings containing the actual signatures of the artists they must be trained upon. I did not specify the rather popular fantasy illustrator "Greg Rothkowski" here. I've noticed the new model is good at rendering words, and I actually think it's doing an overall better job at rendering good results, so congrats to dev on this triumph!

However, this is really pushing the argument for AI a bit into a bad light. If curious, I used the "Fantasy Portrait" style with the following prompt:

the Krovians are a sturdy, ((four-armed)) species with rough, scaly skin. They are known for their resilience and adaptability, having evolved on a planet with extreme temperature variations. Their cybernetic enhancements are often a blend of necessity and aesthetic, a testament to their technological prowess. The driver, with his grizzled appearance and cybernetic arm, seems to be a seasoned individual, likely with a wealth of stories from his time navigating the undercity. The Krovian driver, a grizzled Krovian with a cybernetic arm, looks at you with curiosity before nodding in understanding as you glide over to the open side door...

No negative prompt used.

So are celebrity images gone for good? Just checking.

Hello everyone. I feel kind of dumb asking this, but I'm a little lost. I'm not here to weigh in on the whole t2i old vs new model stuff, I just wanted to know...how everyone learned about the new model in the first place? I don't see a specific post announcing it here or anything. Did I miss something, was it announced elsewhere, or was it just that the site changed and people got it from that?

I'm asking because I'm not so worried about text to image stuff myself, but in a past post the dev said they would get to text gen after images, and I would like to stay up to date on that.

https://perchance.org/ai-character-chat

generator link; https://perchance.org/gbjbu5kj4u

no other clues atm beside that it happens maybe 1% to 2% of generations and if i recreate it exactly including same seed the recreations are fine and nonblack

I understand the new model is new. I also understand the intent of progressive AI is to learn from the prompts, and that detailed prompts offer less variables. That being said, the batches generated under "Pro Photo" filter provide little to no variation at all. They all look the same no matter how the prompt is altered.

Just started happening within the past 12 hours or so. Anyone else having this problem?

I know that @[email protected] was working on gallery filters.. maybe this is related and hopefully temporary?

EDIT: This seemed to be fixed for a while but it is happening again. This time, the "Network failure" error message takes a lot longer before being displayed.

I miss the old model, but we can't do anything about it anymore. I think dev made up his mind on having Flux.1 be permanent for all of Perchance AI image generators.

The people who adore the new model have won over the Developer of Perchance. It's pointless of asking for the old model or requesting SDXL as a secondary image model. I've mostly given up/ forever paused my image generator like some others have with their own generators.

It's not going to change.