Now having really done alot of ai codings recently...

When i tried to make a program where the ai created 3D models, the code assist ai could create the program for the normal ai to use flawlessly BUT the normal Perchance ai that was supposed to feed data in to it is not smart enough to format it's responses perfectly enough for it to work. So are these two different Ai? the code assist ai and the normal ai text plugin ai? Also my guess is this is one of the good ideas that doesn't work because ai text is still in need of update.

Another big noticing is that, on phone, it doesn't make sense to stay on the page waiting for ai code assist to generate since it can be 10+ minutes. So I often end up with maybe 3 projects on different tabs either typing for the ai or waiting for it to think. So, tho I've downloaded a firefox addon to reduce risk, tabs still occasionally refresh, which is terrible. The probably 10 minute wait for the ai thinking is gone, the input i had asked the ai is gone (maybe remember last input if page reload?), and only if the last change the ai had made had been something worth saving do i not need to then go in to backups and do all that fileopening removefirstline copy then paste to get the current project back online. And often it feels weird to instantly save something the ai just did without really checking it out first. So anyway, there is indeed some suffering from some annoyance factors in a full workflow for using the ai code assist.

main thing number 3: I remember Perchance saying a Multiplayer Plugin is coming soon. Even before ai text update if I remember correctly. Now having built a bunch of stuff, I've slowed down on the games; and I realize it's because of no ability to add multiplayer functionality. So I agree with Perchance a Multiplayer Plugin now feels about as high priority as updating ai text to me; it basically means the difference between enthusiastic game makers like myself making games or not yet. I also see huge potential in that, when multiplayer ai assistcreated games begin plooping out, it will be like a new era of awesomeness.

Anyway, thus thoughts. enjoy!

Particularly the Cinema and professional photo ones.

I played with the model a bit and I'm really impressed. And I googled a bit. It seems you guys keep changing the model every now and then, so older posts that asked this question are probably outdated by now.

What model is being used right now? Is it custom. Are there loras applied, etc? Also, if it's custom, is it possible to share, or even sell?

I've been using AI for a while (locally and online) and I can tell you that this is possibly the most model I've ever used.

When backup is made, it should detect if the Perchance side or HTML side is entirely empty and, if so, not create that noncode intro line.

i say this getting in to teh ai code assist and having blank perchance sides frequently cuz teh ai apparently avoids using it... and saving often breaking so i have to go back to previous backups. Would be nicer being able to just select all instead of deal with needless first line that is irrelevant in this situation. thus it should detect if it's irrelavent.

loving playing with teh ai code generation nao; is superfun.

the 'new' button disappeared tho

The gallery plugin already does the reverse, you can ban prompts, but doesnt seem to have a way to research by prompts. So doing something like 'i want images generated that are about baseball' is almost impossible, or you do like https://perchance.org/prompt-hunter and query all images and after the prompt shows results, hide all of the results that dont apply.

Hi everyone I have very little knowledge of coding but I can learn! I want to have an offline AI that is very smart and can actually hold a conversation not just recycle lines I have used earlier, I want it to actually remember the conversations we have, please tell me how to do that thank you

----If I have a list of text options in my script that includes:

myList here, there be apples second, a plum third, a pear.

---and I have a place in the t2i-framework that specifies user input options

settings userInputs foo label = foo type = select remember = true options = {import:myList} //i know this isn’t correct, but gives a sense of what I’m trying to do

Is there a way to have settings.userInputs.foo.options. reference the complete set of options in myList? One item of note is that the myList options are full sentences, and thus contain punctuation such as commas. So joinItems statements would not work with a comma delimiter.

thank you!

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

When i ask the ai to describe something it really seems like it just wants to get to the point and move on.

Like i want it to describe a tennis match and it will go 'the player in white won 2 sets and then won the match. Next prompt'.

I exagerate but it feels like whatever the time frame you give an event, would it be a process that takes minutes or weeks it will wrap it up in a paragraph. And that is if you havent said anything else in the prompt.

What i do currently is basically breaking it down like for a tennis match point by point or set by set. But its getting annoying fast.

Is there a good way to say to the ai 'take your time, describe it in vivid detail over 10 paragraphs'

Has anyone experimented with secret annotations in prompts?

In older SD models, we could include things like \[some message::0\] in prompts — text that would be ignored by the model but still included in the raw input. This allowed us to embed metadata or trigger filters without affecting the image generation.

For example, we used \[ContentLabel::0\] to signal age restrictions. Our filter system treats this as a PG-13 banned prompt keyword, so users could self-label their content by adding it.

However, it seems that in newer models, this trick no longer works. Possibly because prompt weighting syntax like :: is no longer interpreted the same way — or ignored entirely.

Did anyone find a working alternative? It's important for us since we rely on this system to label and filter user submissions.

Hello,

I'm looking for information about the commands we can insert in our prompt with text to image generator

by asking, searching, testing,... I already got some useful ones.

(saveTitle::: TITLE)

(saveDescription::: DESCRIPTION)

(guidanceScale:::{5-9})

(seed:::123456789)

(resolution:::512x768) Portrait

(resolution:::768x512) Landscape

(resolution:::512x512) Square

(resolution:::768x768) Big

and some "useless" ones (these only impact how the generation will look on your screen, but not the image itself once saved)

(size:::700) for Portrait

(width:::700) for Square/Big

(height:::700) for Landscape

(style:::border:4px solid black; margin-top:7px;margin-right:7px)

But now I'm trying to guess if it exists one of these for the number of images generated, because I sometimes want to try my prompt on a larger sample than offered

I tried (numImages:::24) (images:::24) and that kind of thing, but it don't work ^^'

Did someone know if a command exists for the number of images ? and, is there other of these lines that I can add to my "library" ? ^^

Thanks

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.

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...

Hi. I'm using a line like this [fileName="{0-9}{a-z}{A-Z}".selectMany(12).join("")] to create a set of numbers and letters at the beginning of the prompt. In the previous model I used to set the priority to 0, so the AI ignored it, but I noticed that the new model ignores :: and these numbers started to appear on objects inside the images. Could I assign a filename in any other way?

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.

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

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.

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.

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.

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!

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?

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.