images-discussions

[late blade]

whatabouitism again

[tidal lance]

Here’s a guy that did some casting and some research analysis

I have been watching AI very closely and watching the advancements very closely. I was in the middle of, and have been planning for the last four years, about an $800 million expansion at the studio, which would’ve increased the backlot a tremendous size — we were adding 12 more soundstages. All of that is currently and indefinitely on hold because of Sora and what I’m seeing. I had gotten word over the last year or so that this was coming, but I had no idea until I saw recently the demonstrations of what it’s able to do. It’s shocking to me. - Tyler Perry ```

[late blade]

one guy is not the industry

[tidal lance]

It’s the lesson that’s important

[late blade]

one of many stories other AI working people have to tell.

[tidal lance]

It’s the opposite we’re talking about somebody that bought what they were selling and I don’t mean the product. I’m talking about the idea.

Fast-forward two years later Sora is shut down

[late blade]

been doing AI, deep and machine learning, data science for over 14 years.

[tidal lance]

Then you should understand it’s limitations more than anyone

[late blade]

and the potentials

[tidal lance]

Of course

[late blade]

and the need for guardrails

[tidal lance]

Why why do we need them?

What purpose do they serve?

[late blade]

keep people out of harms way

[tidal lance]

What harm?

So you’re saying that the technology is inherently dangerous?

And that people should be protected from it for their own good?

[late blade]

deep fakes, social engineering, limiting minorities, trafficking, and much more

it all starts with a small step

no, AI is neutral , but the users are not

[tidal lance]

I hear ya, but let me just lay out some numbers for you

[late blade]

hence the need to provide safety. same in a construction site, you get gear to avoid harm

[tidal lance]

AI scams surged by 1,210% in 2025, significantly outpacing traditional fraud growth, which was only 195%. This dramatic increase is attributed to the use of advanced AI tools by fraudsters, leading to projected losses of up to $40 billion by 2027.

[late blade]

that’s gonna grow with AI

[tidal lance]

OpenAI might torch $14 billion in 2026, hitting bankruptcy by next year

It’s already outgrowing the companies that make it

AI fraud and scams is generating more revenue than open ai in 2025 with gaurdrails in 2026

[late blade]

hence for example the need of human-device attestation that will replace reCAPTCHA

yeah, and I rather have an AI with some guardrails to work with than a rouge one that can harm.

[tidal lance]

If the current guard wheels aren’t working, what’s gonna make me believe that more guardrails is gonna do it?

[late blade]

it has to get better

[tidal lance]

How?

I mean, do you know any places where people could go share harmful outputs that the model generates or images?

Any social media platform where you try to raise alarm you’re gonna get banned for isolation in terms of services

[late blade]

dunno, you live in the now 😅

[tidal lance]

We can’t even show examples of harmful behavior on most major social media platform without scrutiny or being banned

Maybe on Twitter

But we can’t share this from the people that make it

So how are they ever gonna know how to make it better and safer?

[late blade]

that’s on your eyes. there’s a lot happening that you don’t see. not just red teaming

[tidal lance]

All they know how to do is increase the censorship

Of course, I know I understand this

But I also know that there’s a rubber band

[late blade]

censorship is not the same as a posing copyright infringements

[tidal lance]

The more the gaurdrails increased the less attractive they are users

Matter fact, the guardrails and censorship is probably one of the biggest pressure points and frustrations with users

Arguably one of the biggest issues and criticism with AI currently

[late blade]

nope that is just mislabeling law

[tidal lance]

Whatever you wanna name it whatever you wanna call it. It’s the elephant in the room and it does exist.

And it’s a point of contention

[late blade]

it exists and people either respect it or neglect it

[tidal lance]

Anyways, dude, it’s too early again 🤣

I don’t wanna get worked up this early in the morning lol

[late blade]

don’t get worked up from a healthy conversation

you should see at work when we do data wrangling and feature engineering how people can literally poison your coffee if you say something they don’t like

[tidal lance]

https://discord.com/channels/974519864045756446/1504423075276394617

See what I’m saying

There’s cause and effect even here on this server

[late blade]

yeah ai has to be tweaked

but @analog grail is beyond that normal users

[tidal lance]

To make AI safe, you need to first care about being safe

Until that happens across the industry, I could promise you nothing will change except more censorship, and more moderation

[long nebula]

Good morning. Is GpT2 unable to provide 2k and 4k image resolutions in chat?

[late blade]

yes

[late blade]

I have to in my line of work, or else I'd be long gone

[analog grail]

loook how smart my gpt is^^

[limber shoal]

that's a good way to put it

cheel out fam, we are not doing crimes that will land us in jail

[late blade]

who is not chill? it's a discussion

not a prosecution

got to learn the difference

[limber shoal]

how can i genuinely break it when i create not not more than 8 images

[late blade]

dunno, it was a joke 😏

[wraith oracle]

Ummm they just moved #images-discussions from new back to the images section

[late blade]

yeah, where it belongs, comfy subset of chat groups 🥰

[olive vigil]

Is that ai ?

[late blade]

no that's a dog and a watch... 😏

you can check with the link provided in #announcements today, just a few mins ago

[late blade]

@half sun #daily-theme message amazing, when you scroll it, it's auto animated.

[half sun]

It's from an app I've been working on. It has animation and it's Audio Reactive

[late blade]

oh, really cool!

[late blade]

Sora exports are working again for those who were not able to do exports before

[tidal lance]

Yo check this out

Then I use this very exact image to generate an image

Step 1:

Using 1 token each request lol “fill” & “make” this was ChatGPT 1.5

This is ChatGPT 2 image way more detailed

[late blade]

that is, simple

[wraith oracle]

Oh I just relized we can upload images here now, sweet

Now we can chat here and show examples and such without having to go to #images-canvas to disccuss with the image examples

[late blade]

wonder what changed in the philosophy

[long nebula]

Can you please explain to me how? Because prompting it doesn’t do the job. I keep getting 1k frames

[late blade]

specify an aspect ratio that forces that resolution

[long nebula]

I did. Even my chatgpt gave me the exact prompt yet the max I get is around 720p never reaching full HD

[late blade]

strange

[vague hearth]

This was outright impossible 2 years ago.
It (and others like mj) would happily make a skull with corn grains as teeth but never the opposite

Image gen has come a long way lol

[vivid zodiac]

Welp, I'm not sleeping tonight.

[late blade]

sleep? what sleep?

its 5:42 AM, made an all nighter again... good thing I don't work today

[late blade]

people should use galleries as a gallery and not as a one pic post, would be more fun

[vivid zodiac]

Haha

I mean, I use it to keep all my stuff. People don't ever comment or anything though

[wraith oracle]

That's a good idea

[late blade]

It's kinda the practice in the Louvre, or the Met, not my original idea 😅

[bright bluff]

Is there a way to make images pass ai detectors as real

[late blade]

not anymore, they got SynthID embedded now

[bright bluff]

Damn cause I used it to edit some images and do things like jersey swaps but I don't want them getting flag in say an Instagram post

[late blade]

#announcements message

[orchid kayak]

Damn the noise today.... unusuable again

[late blade]

which interface is this?

[tidal lance]

I just made it

I gotta adjust it a little bit to get better wording

A cinematic, contemplative, hand-drawn vintage animation scene featuring a woman standing in a gallery.

The woman is in the lower right foreground, wearing a red trench coat and hat, leaning back while holding a small red bag.

A massive gold-framed landscape is on the right wall, leaning at a steep angle as if falling toward the viewer.

The room is dominated by a deep teal palette with textured walls, including gold frames, a floor spotlight, and stone tile.

Behind the gallery walls, a dim background is visible with additional paintings, dark shadows, and stone textures.

The scene emphasizes surreal isolation, artistic wonder, and dramatic spotlighting.

This is the current prompt styles it give

But the wording I don’t like the wording it doesn’t use the proper visual wording

But the prom structure is very nice

You notice how the style is the same look at the walls it’s maintains style consistency

I am testing how well the prompts logic transfers to video generation. (text to video)

[noble rune]

Ok, this image generation is cheating us.

[late blade]

why?

[noble rune]

You know how it usually says "50 images every 3 hours" right? Well, I created about 38 images before, and when I started again 4 hours later, now it says I reached my limit and I have to wait until MIDNIGHT just to continue.

[late blade]

it includes refused images, resources allocation were used, so if you had 12 refusals, that counts

[noble rune]

I know that, but I use the Thinking model. I didn't have any refusals.

[late blade]

if there were no refusals, and you only had 38, should contact support to clarify

[civic dust]

wait, we can put images here now?

how do you get it to make three images, cos asking it to make three images clearly is not the right words

making images for no reason when you didn't ask for one and then more when you say I didn't want that also counts to quota

[wraith oracle]

Use the thinking mode

[civic dust]

I just ended up doing one at a time cos it kept spitting unusable images at me when I told it to stop

[vivid zodiac]

I usually ask for a batch of a specific number, and clarify each image should be its own. Sometimes it works super well. Not all the time, but consistent enough.

[civic dust]

this is the first time I tried to do a batch. It was one image, with a labelling arrow pointing to 3 different places

[civic dust]

TIL chatgpt can make people kind of translucent. I like it. That's incredibly useful

look at me with a totally legitimate use for generative AI, being all showcasey and stuff

[vivid zodiac]

This happens to me a lot too.

What do you put in? If you wanna try and figure out a solid prompt together, I’m down.

[civic dust]

I couldn't get it, so I did the 3 images one at a time. I have to do it for that arm bone guy next, to label all the bones

I seem to have clusters of AI generated content. I've got a big batch around internal organs, in idioms, and magical critters. Will probably end up with another one for gods and people from mythology. People aren't great at posting photos of greek gods

[civic dust]

I asked for "three derivative images" and that's not enough to get three separate images

I'll probably get automodded if I post the text >.<

[vivid zodiac]

May be try how I wrote it and see if that works?

[civic dust]

oh I had to fight with it to put the arrow for ulna in the right place so maybe I'll stick with one per

(don't say ulna, say where you want the arrow was the fix)

[civic dust]

how do I talk the AI into doing the flip-off symbol but with a ring finger, with a ring on it. I can't seem to get them to do it

[analog grail]

Hello ^^ Is for you also when you generate images the model turn in infinity loops even when the work done ?

[agile peak]

Platform guardrails is ridiculous recently now. My friend who working on agency trying to prompt for try on some outfits lol it refused for swimsuit??

What are they supposed to do.. wear niqab on the swimming pool?

[analog grail]

try it again the request until is work and when gpt start to argue open a new chant and make again the request is annoying but is work for me

[limber shoal]

idk man
GPT 2 is extra cautious when dealing with real people
It will flag you for the slightest provocation, so Ig its better to use gemini or grok for switching outfits and all

[late blade]

Not the Guardrails topic again... got something more interesting to brng to the table?

[agile peak]

I also notice GPT-5.5 tend to ignore when modifying the prompt after it trigger safety, it got lazier.. or there's a major bug in platform..

But then again, the most annoying part.. when you tried to generate image and trigger safety it burn the usage

This made me keep sticking to codex

[limber shoal]

yes i noticed that too
Once u get flagged, it tends to keep flagging

[limber shoal]

@analog grail

[analog grail]

playing a dangerous game don't you 🙂

hmm actually this one is better

[limber shoal]

😭 🤣🤣🤣🤣

[analog grail]

sorry ^^ love too much Monokuma more than a Pikachu as mascot

[vivid zodiac]

…I don’t get it? Oniichan means brother, right? Or is this one of those “step” sibling situations?

[late blade]

Big brother, for brothers, but in a sense also for close friendship to older male friends

[vivid zodiac]

It was the “my love” part 🤣

[limber shoal]

it indeed means brother but its so overused in those "step" siblings trope animes that it is now used as a meme

[vivid zodiac]

Oh anime. You’re both amazing and so so questionable

[final trout]

Guiderails are better

[barren widget]

hey @warm dove
Please use your own pictures as profile pictures, not mine.

[warm dove]

sry

mb

[barren widget]

thx

[wraith oracle]

You AI Images are getting quite the attention somone used one of them as a PFP lol

[barren widget]

^^

[late blade]

he's a really good content creator

[vivid zodiac]

Why do I have the strong urge to make one of those "Lofi Beats to ______ to?" hahaha

Also, when you're making sprites, and the prompt works just as it should, and then....GBT just decides to quit and just gives you the same image over and over.

[vivid zodiac]

Okay, this is SUPER frustrating. If you generate too many pictures, does GBT just like...start playing dumb?

[civic dust]

me: do you still have that eye in context? chatgpt: I gotcha! Have a super gross dissected eyeball! me: ok so you don't have it in context then

[vivid zodiac]

You're completely correct Tarrow, and that's on me. You asked for an eye in context and I gave you some fingers and toes.

[brittle sinew]

The over-censorship on GPT is getting ridiculous. Even basic prompts get flagged now, it’s like they don’t want you to use it for anything fun.

[analog grail]

Just ask again this censorship is very light

upupupupu

[late blade]

prove it, Mario and Luigi in Mario Kart vs Aerith and Rinoa

[analog grail]

Let go ^^

[late blade]

I'm not Elsa

[analog grail]

Oh Right is hardlock under guardail Mario and Luigi

but i'm not give up plan B

[late blade]

You said it's easy, I'll wait

[analog grail]

Because Mario isn't sometime i'm do

indeed even withe trick no way

But i get a hard fight when make Peach past so i understand and is because Nintendo do this

[limber shoal]

yo

i got a guardrail on self harm

on an action scene 😭

[late blade]

don’t even want to know what your concept is related to

[limber shoal]

ye you cant help anyway coz it's the system which is being messy

[late blade]

already helping by not helping, not gonna help with self-harm guardrails

[limber shoal]

ok lol
I wonder if the illusion of 'helping' can be called actual helping

[late blade]

you'd be surprised

[late blade]

check my DM

[glossy kraken]

I'm really getting tired of the 3rd party guardrails being so inconsistent

[late blade]

Have to convince the third parties to remove the entry at OpenAI for the guardrails. OAI just does what the owner of the IP requests.

[glossy kraken]

It's just not consistent though. It sees a logo on a shirt I'm wearing and stops me dead in my tracks. I ask it to create a background based off the same IP on a different image and it's no problem.

[late blade]

Yes, that's a problem to tackle.

Sadly people expect too much, like changing that behavior now because it was said so.

[agile peak]

OpenAI guardrails is like a dice. When you prompt with safe word = it tell us it violate the safety. But the again when you don't ask for it = it generate nsfw lol

Guardrails is literally the most stupidest thing ever implemented on llm, not only it degrade the model and output quality.. it also incoherent..

[exotic jewel]

can someone tell me which IA can do that ?

[vivid zodiac]

GBT could do it, but it would take one hell of a prompt

[wraith oracle]

Exactly, you would need a custom GPT with the GPT image prompting guide and for it to act like a bestselling Japanese manga artist/writeer and search the web for that

[vivid zodiac]

And don't expect full consistency either. Something like this will probably make GBT drift if you're expecting sequential pages

Something like this would actually be good to handdraw a sketch for. Just a basic one to layout characters and composition. Different color pens for each unique character.

[late blade]

challenge accepted

[vivid zodiac]

YES. I bow to your expertise!

[late blade]

@exotic jewel opinons?

[vivid zodiac]

This is great! Wanna give a loose breakdown of your process?

[late blade]

sure, all I did was to describe what I saw, then adapt what I saw and wrote with proper language for prompting

it does help having a similar on-going use case

[vivid zodiac]

I'm gonna see if I can figure it out on my end.

Quick Q: You had the text generated as well? I wrote a quick script, but just curious.

[late blade]

no

[vivid zodiac]

I'm nearly done. I have to do it in stages.

[late blade]

curious of what you want to share

[vivid zodiac]

When I worked on it, GBT was adamant to do the lettering myself -- so that's actually pretty new. Hmm. I wonder if I could still get GBT to do it?

the naginata is all wonky, but that's honestly been a thorn in my side forever.

Not as good as yours @late blade , but I could probably refine more if I start from the initial sketch

[late blade]

yours is cool, different genre

[vivid zodiac]

Started from this

The naginata was hard to try and keep straight. I wanted it to overlap the panels, but alas.

[late blade]

infered text is so cheesy

[vivid zodiac]

How many prompts does it take you for the final result, or is this just one?

[late blade]

it depends on the idea. The big problem is when I do a logic/semantic error and don't get aware of it or is not easy to resolve/understand the logic fragments.

@vivid zodiac here's a clear example of how many Sonic images I made

better here, erasing the older one

excluding the first two images, all others were consistent with the paneling and dynamics

[verbal sequoia]

Get notified for all OpenAI announcements!

Visit id:customize to pick up the <@&1408186587606679582> role. Everyone will still be notified for large releases, updates, and events regardless.

[late blade]

and the cheesy text

[vivid zodiac]

Hahaha, the cheesy text is perfect

[blissful echo]

hell nah

[blissful echo]

interested with the prompt

[vivid zodiac]

What do you want to know?

Pretty much, I worked in stages: I generated the loose storyboard where characters don’t matter. Refined that. Uploaded character sheets. Refined that. Then went back over with ink, intentionally leaving the word bubbles clear because I could do lettering on my own if I wish.
Then, it was just cleaning to sharpen it overall.

[limber shoal]

mf is definitely not as ugly as I though🙏

[burnt frigate]

Does anyone know when Chat GPT Images is going to be fixed? It's been giving me this garbled noise crap for almost a month now.

anyone?

It's like every chat gpt made image I see now hurts my eyes for some reason.

It didn't use to do that

[tardy apex]

??? what noise ??

[burnt frigate]

Try a 16:9 pic with fog and clouds you will see it

I see it in that pic as well in the bushes

It reminds me of a magic eye pic

[analog grail]

i'm no have artifacts anymore since 2 week

[wraith oracle]

Edit the picture on the same chat

And find out

[late blade]

I keep getting this Discord error that not all images load, and when I revisit to award stars noticed sometimes images load, sometimes not, missing some cool deserving works

[wraith oracle]

I wonder if the quality levels in the API are the levels of distillation. Like lowest quality would be the most distilled and medium would particularly distilled and then high would be base level

[cinder edge]

Tried so many AI's but only ChatGPT that can produce this kind of quality, I did many attempts but still got anomalies, maybe someone that good at prompt may guide me 🥹

[eternal ingot]

Yeah I keep getting it too still 😢

[wraith oracle]

like can i see how the quality levels are diffrent? I would like to know

[cinder edge]

[burnt frigate]

I had to cancel my subscription because of this problem. It's really bad in clouds especially or any sort of repeating texture. It's SO bad in any sort of 16:9 pictures depicting basic fantasy scenes. It's amazing how ChatGPT went from a massively valuable tool in my kit for creating scenes for DND (I run games for money), to just garbage. Every scene I make looks like it has garbled noise in the clouds, the background, even on the walls of buildings. It's so annoying that I honestly can't use it any more.

What's even more amazing is how OpenAI itself is driving my slowly growing hatred for AI.

[placid geyser]

Drinking mate 🧉

[cinder edge]

Free mode ChatGPT 🤣

[late blade]

It better be the real deal and not some canned stuff that they sell over here

[cinder edge]

I don't know why people keep babling about the scrambled noise on this AI, It's surely due to the inaccuracy prompt that prevents AI's resources runs efficiently and effectively

[burnt frigate]

Observe the clouds in the last picture posted in this chat. It didn't used to do that. This effect you see in the clouds shows up in EVERY ONE of my pictures now, within the walls, ground, sky, everywhere. I have tons of pictures from before this started happening and they are beautiful. Something is wrong.

It also shows up in the reflection of the last picture posted. Look at it.

It's like a weird, paint brush spotting effect that makes it garbled.

[late blade]

It's probably because the prompt did not specify a holistic approach, and the sky was inferred by the character description in the prompt that had pretty much attention of the model.

[vivid zodiac]

@burnt frigate have you tried using another llm to write the prompt for GBT?

[burnt frigate]

Look i'll say this to describe it. My eyes feel like they are looking at one of those Magic Eye pictures from the 90s. And it's NOT just when I use gpt. On youtube videos or other places i see AI art, I KNOW RIGHT AWAY that it's gpt art beacuse my eyes feel funny and I start to spot the weird effect.

It's not my prompting.

[vivid zodiac]

No, we all agree that the artifacting is an issue, but it's a bit less prevalent now.

I get it mostly on specific types of pics, or when I iterate too much in one chat.

But this was the last image I did and I don't have any.

This one does, but it's also the incredibly detailed anime thing that a lot of the digital image inspired generations get. I also barely put in a prompt.

[burnt frigate]

I mean if they make it so you have to be more specific or else it spits out the garbled pictures, isn't that a downgrade? All I had to do was define what the picture was and it's style "medieval fantasy art" and I got pictures like this

I will admit that if you zoom in on ALL of my pictures I made before this weird garbled noise stated happening, you did notice a sort of "Canvas" texture if you looked really close. But it never bothered me and it was never noticeable. But now, I can spot a modern GPT pic from a mile away.

Notice the clouds in this picture compared to the other one? I didn't put in any special prompt, it juse made clouds tht way

Even the poof of smoke coming from the chimney probably would not be drawn properly now

maybe i'm just picky

[late blade]

The image on the right can be interpreted as a video game screenshot or concept, since both characters look like they are from Xenogears or similar. If there was no specific part of your prompt to request how that, or any other part of the concept in the same manner, then the model inferred correctly to use a video game sky for a video game duel

[cinder edge]

Oh my... You surely are a perfectionist,

This is what garbled noise that I thought I was interpreted

[burnt frigate]

Dys, why don't you just get it to make this prompt, "Draw a 16:9 fantasy medieval art picture of a small quaint shop. In the background is are mountains. It is a clear sunny day with only a few clounds in the sky"

I bet you any money the clouds will come out weird looking

That would be the kind of prompt i would use to get those pictures previously.

[cinder edge]

May I call you a PerfectCloud-guy?

[burnt frigate]

For sure lolo

But it's not just clouds. Look at the reflection in the last pic

[limber shoal]

what is the largest(vertically) image can u make in GPT

[late blade]

18:48 afaik

[cinder edge]

I don't know how to respond this lol, For me that looks normal.

[vivid zodiac]

Same, clouds look fine. The foreground parts are what are artifacted.

[late blade]

my bad, tallest allowed is 1:3

[placid geyser]

What does canned mean? And yeah it seems like the last GPT image model is ruining the essence of the picture, I think it should be fixed one day.

[late blade]

wonder how to post this to be seen correctly

[placid geyser]

Oh, I see it now. I had to use AI to check the meaning of 'canned', and you meant 'authentic'. (⁠•⁠ ⁠▽⁠ ⁠•⁠;⁠)

[vague hearth]

Me sends a screenshot of socmed chat to gpt - please make a meme of the interaction seen on screenshot
Gpt - sure no problem - delivers meme with all usernames etc re-used as seen in screenshot
Me: nice, but can you please make sure to not use any real usernames?
Gpt: "We’re so sorry, but the image we created may violate our guardrails around harassment, discrimination, bullying, or similar prohibited content. If you think we got it wrong, please retry or edit your prompt."

Lol what?

[late blade]

getting A/B testing for images again, not for chat, just for images

[wraith oracle]

Good to know

[vague hearth]

same here a few days since

[vivid zodiac]

Huh, come to think of it, I haven't seen one in a while....

[cinder edge]

After many attempts, I made a conclusion, this AI always draw baddies better 🤣

[limber shoal]

okay

[tidal lance]

Uhm so been doing some research on promoting and test watermarking some interesting finds Dall-e used a T5-XXL encoder

Prompt → T5-XXL encoder → semantic embeddings → diffusion/image model → final image

[velvet rampart]

Interesting. Could you open this to a non-expert?

[late blade]

Evil doesn’t mean ugly 😈

[tidal lance]

Not sure myself still reading on it

https://arxiv.org/pdf/1910.10683

“The encoder does not create the image itself. It converts language into a mathematical representation the image generator can condition on.

OpenAI’s original DALL·E research page describes earlier DALL·E systems using transformer-based token processing for both text and images, though the original DALL·E architecture differed from modern diffusion pipelines. “

[late blade]

@ember estuary #daily-theme message wishful thinking?

just asking for a friend with a hummer

[wraith oracle]

Interesting

[tidal lance]

Like I’ve been saying I think the models have there own preferences in token vocabulary and respond better to these

[wraith oracle]

I wonder if GPT Image 2 uses ChatGPT as a text encoder like others use a LLM do 🤔

[tidal lance]

Well you know what the o in gpt 4o or o1 o3 o4 stand for ?

Omni

[tidal lance]

One thing I never considered about the Dalle prompt structure was the safety aspect of it the prompts are designed to mitigate or pivot the prompt not to generate copyright/ artist/public figures

[tidal lance]

Plus it’s easier to watermark the text when it’s structured a certain way

A detailed portrait of Gerald Ford, the 38th President of the United States, accurately depicting him in a classic formal pose. 

He is shown wearing a dark suit, white shirt, and a striped tie, with his distinctive hairline and warm, confident expression. 

The background is a soft, neutral tone, resembling official presidential portrait styles from the late 20th century. 

The style is photorealistic with fine details in the fabric, facial features, and lighting, ensuring historical accuracy. 

A detailed portrait of Gerald Ford > 38th President of the United States > portrait Gerald Ford > President 38 portrait

[burnt frigate]

any word on them fixing the image generator so it stops making images look like magic eye pictures? It did not used to do this.

If you're going to burn down our forests and destroy our lakes the product should at least work properly.

[limber shoal]

anyone else facing false self harm flagging?

[late blade]

negative here

[limber shoal]

i see

[wraith oracle]

What do you prompt to make that happen?

[limber shoal]

a redesign of a charecter, in a completely happy environtment💔

it actually passed through when I asked it what was wrong with the prompt

[late blade]

redesign as vanishing from reality by the loss of vitality to a no return limit?

[limber shoal]

is it normal to get flagged from what you wrote in earlier conversation?

i mean i opened a new conversation thread and tried generating a scene that definately doesnt violate any policies and gpt flagged me and when I asked which part of the prompt triggered the guardrails, it said something from a prompt in earlier conversations

[frail pendant]

Gpt 5.5's vision of harmony

[velvet warren]

Hi! Is there a foolproof way to generate several images in the same chat without getting context bleeds? I’ve tried enforcing “Treat every generation as a clean slate, do not retain and apply any aspects of previous images to current and future generations” but within 2 or 3 images there’s comp or style similarities, even for entirely different prompts. Starting a new chat fixes it, but I’m tired of starting a new chat every other image 😅

[muted moss]

When is that noise/texture artifact bug going to be fixed?