images-discussions

[twilit helm]

Well, it’s only been out for a couple of days. Give it some time.

[quiet brook]

yeah, only time can tell

quality comparison #images-canvas message

[twilit helm]

interesting

[quiet brook]

can u tell any difference

[twilit helm]

tbh i don’t see it but my guess is the difference in quality generation?

[midnight mango]

This cool down is kinda soul destroying.

"Wait 9 minutes before trying again"
I wait 9 minutes and resubmit the prompt
"Wait 8 more minutes"

😭

[quiet brook]

ye, any quality difference

[fading mauve]

hi

[tardy moat]

Does it seem like they clamped down on guidelines even from earlier today?

[minor cliff]

Hi

Why only 1 image can be created at a time

[frozen copper]

Are we still limited to 1 concurrent gen and 2 variations for plus users?

[forest walrus]

But of course no will hear this 🙄

[quiet brook]

they just turned off 2 concurrent gens for plus like 10 hours ago

now were limited to 2 images per try while pro is 20...

[forest walrus]

GPUs be deep frying 😔

[quiet brook]

the fact pro has never recieved any kind of limit is crazy

[forest walrus]

But of course there will be a priority for Pro, they pay 10x the price :V

[quiet brook]

plus users are always the users suffering

[frozen copper]

I know that, hope it gets restored asap, it's slowing me down.

And not paying 200 for this, simply too expensive

[quiet brook]

its just false advertising atp

each image will take 2.5 minutes to generate+only 2 images in a row+can only do 1 generation in once

while saying "unlimited" in their website

[forest walrus]

It says unlimited not fast or many concurrent generations.......

[next fog]

Does anyone notice a sudden increase in censorship in generating images?

[forest walrus]

Nah, but some say generating Ghibli style images is now a bit harder (Doesn't apply to me, i'm doing it all the time, so I don't think it's exactly true...)

[quiet brook]

maybe ppl be already turning pro plan into wrappers but they are only nerfing the plus plan

[tardy moat]

Yes

[next fog]

I'm a plus user. I could generate characters like Luffy and Goku, now I just get the policy violation message.

[forest walrus]

What was the prompt?

[tardy moat]

I’m running into guideline walls all over the place

[next fog]

Goku and Luffy both eating and hanging out in Luffy's ship Thousand sunny

[forest walrus]

Any reference images used?

[next fog]

Nope

It generates half of it then gives me the policy violation message

[forest walrus]

Very popular figures like Guko could get them in trouble, but maybe you can get around it by giving the images of the characters without saying who they are...

[quiet brook]

i see ppl generating goku on day one

[next fog]

I understand, but thing is I could generate those kind of pictures on launch day and day after that. But it stopped since yesterday.

[next fog]

Exactly, I also generated many, but now can't.

[sudden geyser]

Same here, policy violation censorship. It's annoying.

It's censored to the point of uselessness.

[forest walrus]

Hm, I remember trying to generate an image of Goku on day one, but it also refused...
Same with stuff like Mario.

[stone lodge]

has it already been butchered? guess i'll wait for it to come to bing.. sigh

do they not understand the lack of censorship is why it was so popular to begin with

[next fog]

I wouldn't mind if they had disclosed the fact that they are heavily censored but not admitting it and sneakily adding censorship is not good

They bait the people into joining the subs, then start censoring stuff

[forest walrus]

I guess they only limiting the stuff they get complaints from companies about, but that just guessing. as other stuff are still very relaxed, like known public figures...

[quiet brook]

mario is like always blocked

[simple sand]

I hope this is all temporary and doesnt turn like dalle 2 where it was good during the whitelist and launch day but after that got censored to oblivion and rate-limited + Token system

[stone lodge]

but you can do political figures doing the funny thing

once that gets blocked from prompts its over

[forest walrus]

Yea that's why I feel it's related to companies filing complaints to them...
I honestly never like when such stuff happens, but it is what is.

[stone lodge]

they did the same thing with dall-e 3

[quiet brook]

true

turn back the 2 concurrent and the ability to generate 4 in plus plan first

[forest walrus]

API will probably be more open anyways, like DALLE3.

[quiet brook]

remember dalle 2 was like 10 images at once

[forest walrus]

Way less users, and GPU eating models...

[simple sand]

honestly Im seeing people say that plus users are getting limited to a certain amount of generations per month which should be removed first

if thats the case

[forest walrus]

They were already suffering with GPUS when they scaled for 4.5, so I guess that image gen is deep frying the GPUs lol

[frozen copper]

Wish open source can replicate this capability, this image feature is miles ahead

[verbal sequoia]

DALL·E Category and Channels are now “Images”

With the release of 4o image generation, we’ve renamed the DALL·E category and channels to Images, to better reflect all OpenAI image models.

-# Share your amazing images in #images-canvas!

[next fog]

I feel like soon everything will be censored to the ground and making this generator sama as any other generic text to image model😔😔

[sudden geyser]

Companies censoring OpenAI image generation is like suing Photoshop because it can be used to draw copyrighted characters.

[quiet brook]

reallllllllllll

[quiet brook]

while advertising it as "uunlimited"

[forest walrus]

I really hate it when it happens, like it's just silly image generation, nothing bad gonna happen if Nintendo doesn't make a fuss about Mario eating Pizza lol

[simple sand]

🙂 Guess what got removed

[stone lodge]

when dalle3 launched you were able to generate ||pоrn|| so I expect them to completely ruin this one too

[sudden geyser]

I got it to generate Supergirl, but I had to stipulate it to remove all copyrighted elements.

[quiet brook]

who. is putting. those guardrails.

[simple sand]

Whats weird is I can still generate stuff like JK simmons as Omniman, Edna from incredibles, (Im just tryna test), but other random stuff just fine. idk what causes it to trigger

[forest walrus]

OAI. has to listen. or they gen in trouble. big company with lots of eyes on them
Again, I don't support it, but that's just the case, if I was running a company I would oblige too so I don't run into lawsuits...

[simple sand]

Honestly Its understandable that companies want to bar their characters from being used in ai generations. I'm just way more concerned about the plus user limits vs the restriction filter. As they've shown the 4o Imagen has way more uses now other than just redesigning funny characters like making charts and infographics or concept art.

What I want to know in the long run is what are the daily limits for sora/4o imagen. And will it be unlimited or will there be ANOTHER token system. I hate that theyve removed the "unlimited" part on plus users now and have left it vague on what they plan to do. I worry that its going to actually be viable to make a bunch of alt accounts just to use the daily free image generations, which I hope does NOT happen.

[quiet brook]

other companies doesnt give a damn, even mj

[forest walrus]

I think unlimited is a bit hard for Plus as ChatGPT also has other models and features that add up to the 20$ cost...
I wish everyone higher limits, but I think that may only be for Pro or Pay as you go through something like the API.
They said they are working on making it more efficient, so may result in higher limits for Plus (and making it possible for Free to run as well).

[quiet brook]

then like, don’t advertise it as unlimited

[forest walrus]

On Sora it's Unlimited, just slow, which is different...
Unlimited != Fast

(Well somewhat unlimited, it's still under abuse protection, but like nothing is actually unlimited for real)

[sudden geyser]

Messing around with it, you can get around the censorship by telling it to remove all copyrighted traits from the image and not mentioning that the elements in it is something from copyrighted content.

[forest walrus]

👀

[verbal sequoia]

Showcase your DALL·E creations to the world!

Collaborate with our OpenAI Instagram page! Just invite @openai as a collaborator before you post. If selected by our team, your work will be featured on our handle, giving you more visibility.

Quick tips for selection:

• No copyrighted or profane content.
• Your profile needs to be public.
• Stay mindful of sensitive topics.
• Outputs that closely follow the prompt are preferred.

Feel free to share reels, carousels, or just a single image!

[sudden geyser]

And what it considers copyrighted content is... not immediately obvious. Something that you'd think is free to use is copyrighted and something that is free to use may seem copyrighted. So shifting the burden of copyrighted content detection TO it is the best option.

[quiet brook]

but there is a fixed limit of ~201 tries per day which it shouldn’t exist because it’s advertised as unlimited

yeah true nothing is truly unlimited (even time itself) but if there’s a fixed limit, it isn’t unlimited

[forest walrus]

Isn't that just abuse protection? it already takes long to generate, you can't possibly be typing prompts all day 👀
Also I honestly didn't test that myself, i'm not sure how I can go to that number in the first place

[quiet brook]

If it’s advertised as it, then remove the fixed cap and let time and your imagination be the only undodgeable limit

can understand they throttling you if you’re pumping 1184726 images in one second

[forest walrus]

Are you actually hitting limits at that number anyways? what source is saying it's ~201? I'm seriously trying to understand how such number can be reached in the first place with such slow queues and generation speeds. Also you count them manually?

I have so many questions lol

Anyways, daddy sama retweeted this on TWITTER DOT COM from 8 hours ago:

"The 4o image gen demand has been absolutely incredible. Been super fun to watch the Sora feed fill up with great content .

Our GPUs are also melting in Sora land unfortunately so you may see longer wait times / capacity issues over coming days. We’ll be working hard to get back into a healthy state."

[quiet brook]

I’ve seen at least 3 people reporting the ~201 tries limit on sora on day one and two when we were actually generating many pictures, and I measured out the chatgpt limit manually by hitting the daily limit after ~90 images, because chatgpt is faster

heck I even saw one hitting the 201 yesterday

[lilac nimbus]

is it set to 1 image at a time currently?

[simple sand]

I do see that point (despite the fact they have microsoft as a huge investor) but I do believe They shouldnt have put unlimited gens as a in the first place and now remove it as now it leaves opportunity for people to request refunds, so theyve dugged themselves a hole there. But if they do want to change it they should stick with free users not being able to generate images until theyve developed a plan or improved their hardware. I heavily prefer a subscription based payment system over a token system but I heavily prefer a token system over the two combined.

[simple sand]

I believe so, I can check

[quiet brook]

just first make the website saying “approximately 200 generations per day because of high demand” and later when they scale up THEN put “unlimited”…

[simple sand]

The reason they changed it to unlimited I believe was because they werent getting many people using sora, as it was intended for sora users. But they were developing 4o image gen at this time so I dont understand why they didnt keep it that way until AFTER they launched 4o imagegen

[simple sand]

on sora.com Im able to generate 2 at a time

Oh wait no

I may be wrong

[uncut haven]

1 image at a time in ChatGPT(since launch)
2 image at a time in Sora(reduced from 4)

[simple sand]

Ok yeah 2 at a time

[forest walrus]

Ok, I honestly consider it abuse protection (I'm a software Dev and I would mark this as service Abuse limit), but maybe they could label it as 200 generations for now yea...

[simple sand]

wait is it 200 daily or per month?

[forest walrus]

Daily

[simple sand]

Ok I can live with that

[forest walrus]

(If that number is true in the first place)

[simple sand]

true

[lilac nimbus]

It was letting me do 2 at a time but now its down to 1 🤔

In the manage plan it says up to 2 concurrent generations, but then trying to make 2 now it says its limited to 1

[velvet rampart]

Many are asking how do you get this or that generated. See for example this. #images-canvas message I generated by descriping it and tweaking it to different enough. And one critical factor I have notice that a long prompt goes through easier than a short one. Based on my experiments it needs to be over 1000 characters long including spaces. That seems to the limit so that the filters do not clamp on them so easily.

[twilit helm]

probably optimization since their gpus are on fire

[velvet rampart]

It has to be done in one go. If you start to fix, the filters clamp on it very easily. Context matters.

[lilac nimbus]

can all the ghibli requests be sent to its own GPU and the rest of us can use the rest 😆

[simple sand]

TRUE

[dim cradle]

lol yeah. and ask yourself if the world really need another anime image period or another political one

[quiet brook]

Rate limit the ghibli

[quiet brook]

pro can generate much more

[forest walrus]

yea, obviously...................................

You know, if I paid 200$, I would expect very high / unlimited lol

[quiet brook]

both are the same “unlimited” advertisement

[quiet brook]

they just turned it off

[forest walrus]

I really don't know if that 201 is even true, as it's honestly very hard to test yourself, you have to sacrifice life, work, and go through torture of extreme backpains and bad smell 💀

[lilac nimbus]

why you calling me out like that?

[forest walrus]

Average OAI users:

[quiet brook]

they removed credits from sora like 5 days ago and they kinda rolled with it with 4o img gen

[lilac nimbus]

Is it possible to set up a folder in our library that can be self referenced for constant characters?

Sort of like the presets but housed iside the folder?

[quiet brook]

#images-canvas message

[forest walrus]

Oh btw, the "Unlimited Images and Videos" is gone since they started the scaling process...

[dim cradle]

Some people made posts about it . Saw on Reddit , can't add screenshot here (No permit.)
Personally, I can't confirm since my maximum is 30-40 per day and not always.

[forest walrus]

Unpopular completely unhinged opinion: I think generating 201 images a day on Sora+ 90 on ChatGPT, totaling 291 images is quite unhealthy for the physical and mental health... JUST SAYIN

[dim cradle]

Possibly, but it depends on why the person does it.
I know designers/artists who use it to spark ideas, experiment, etc, so it really comes down to the reasons behind it.

[simple sand]

oh its separate?

[forest walrus]

Usually Sora never connects with ChatGPT in anything other than the sub, so I would guess so yea, but as mentioned, hard to test, will result in consequences...

[quiet brook]

I could continue generate in sora after I hit the limit of 90 in chatgpt

[dim cradle]

It’s really sensitive to the prompt model, but need to force
/clearly describe the color palette to avoid the default yellow-brown tones it tends to lean toward in art prompts (lwatercolor, oil imitation etc ).

[dim cradle]

Sure but 290 images a day and theyd be doing nothing but being slot machine addicts, not doing actual dev or workflow

[dim cradle]

All addictions are bad, no denying that.

well yeah, but to Linky's point (I think), if youre doing 290 generations a day, youre probably not using this as a tool, but to get a dopamine rush like from a slot machine. of course being able to generate quickly is helpful for people actually doing workflows too, but youre not doing much actual work if youre spending time to do 290 generations

[vale wind]

Why can’t the new image generation model not use people’s faces to make an image? This is so sad man

[dim cradle]

you have plus or pro? people were using their own image earlier. maybe just try again in a new chat or sora site

[vale wind]

Yes I have plus.
Whenever I try putting in a image of for example myself and let’s say a backround and I tell him to put my into that environment it will start generating but then stops and says it’s against the policies

[forest walrus]

I have generated multiple images based on real people images (some are popular public figures) and it works fine...
What is the prompt, like what Environment you try putting yourself in?

[vale wind]

A normal dude with sunglasses on and a normal park as the environment. It will only allow to generate if it’s in a certain style. But whenever it ein a photo realistic style it will cancel the generation

[dim cradle]

are you in sora site or chatgpt? use both. start a new chat in gpt, rephrase it a little and try again

[quiet brook]

and the slot machine isnt even efficient

[dim cradle]

unless you win big on the first pull and can walk away

[vale wind]

I’m on chat gpt. Thanks for the help tho

[dim cradle]

go to sora directly

[forest walrus]

Hmm, did the image(s) you upload contain anything that may possibly trigger moderation? It could be just an error if it's really all normal stuff. no moderation filter is perfect...

[dim cradle]

idk, i just tried with my ugly self and it worked on chatgpt

[neat nexus]

What is the limit for image generation?

[vale wind]

Hey, where do I only generate images on sora cause I can only see video generations

[vale wind]

Nope

[dim cradle]

click video, then brings up a menu to pick image or video

[vale wind]

Yeah but am I able to generate images on sora or only videos

[dim cradle]

well if you select image, then it will make an image

[quiet brook]

~90 a day in chatgpt, ~200 tries a day in sora (max of 402 images)

[next fog]

I saw a Twitter post where one of the openai team members said that the ai is "hallucinating" policies that don't even exist, and they are on it.

[quiet brook]

lmaoooo

this is how token based img gen works ig

[midnight mango]

I wonder how much is costs open AI to generate a single image with 4o

is it like cents?

[dim cradle]

more than one time with dalle3 even it would say 'sorry, i cant make images yet or something' some random times

[uncut haven]

For perspective, DALL-E 3 generation costs 0.08$ for square, 0.12$ for vertical or horizontal.
Considering 4o takes nearly 3~4 times per image, I think it will cost 3~4x from that.
They said API will be released in a few week, so we'll find out soon...

[quiet brook]

please SCALE UP ENOUGH

[uncut haven]

When GPT-4.5 launched, Sam said:

we will add tens of thousands of GPUs next week and roll it out to the plus tier then. (hundreds of thousands coming soon, and i'm pretty sure y'all will use every one we can rack up.)
So, 100,000s of GPU was not enough for 4o image, or they're still coming?

That was a month ago.

[dim cradle]

idk but it seems faster on sora to me now. maybe just rng. but then again the yankees are sleeping

get your gens in before they wake up again haha

[frozen copper]

Seems like they cranked up the filters, the prompts that can be generated a few days ago got blocked by "Unexpected errors" now

[uncut haven]

"Unexpected errors" means error. If image got filtered, the message will explicitly say about policy.

[frozen copper]

oh i see, cause right now i am still getting errors even though the status page says all is operational

[quiet brook]

yeah its faster for me also

maybe ill count the time but im playing fortnite

[frozen copper]

An i got errors within seconds

[uncut haven]

Also, Sora UI is stupid; if I delete 1 image from set, it will say "1 failed varient". It does not differenciate failed generation / user deletion / policy filter if at least 1 image is generated.

If all varients are filtered by policy, then it will say "This image may violate policy" or something.

[verbal sequoia]

Get notified for server events!

Visit id:customize to pick up the <@&1261377106890199132> role.

[quiet brook]

it only queued for 20 seconds

[dim cradle]

enjoy it while you can. it will be slow again in about 6 hours im sure.

[twilit helm]

Honestly, the rate limits are quite annoying. I’ve been waiting for an extended period to generate images, and I’ve only managed to generate four. At this point, it’s not advisable to release a product if your infrastructure is unable to handle the demand.

[quiet brook]

yeah, it’ll become slow when 12:00pm utc

[quiet brook]

first they removed the ability to generate 4 images in a row. Then they removed the 2 concurrent generations

although the 2 concurrent is clearly advertised on their website

[frozen copper]

And really need to loosen the filters, even the slightest of triggers will error out the gen now

[twilit helm]

yess… 😔

[quiet brook]

rate limit the ghibli!

[uncut haven]

It's funny that video still can be generated 4 at once...

[quiet brook]

yeah

[tardy depot]

Why do i have to wait 15 min between image generations in 4o? i am premium we should have more images available. Remove these from free accounts

[quiet brook]

**average generation speed on plus plan: 7am utc **

overall: 02:09.914 | 129.91s
queue: 00:15.390 | 15.39s
generation : 01:55.038 | 115.04s

[vapid saddle]

I have tried GPT's new image generator, and it's not as good as Bing image creator.

[dim cradle]

So only 1 in queue now?

[dim cradle]

#openai-chatter message

[desert pier]

Is there a limit to how many images I can generate on Sora on the Plus plan?

Additionally, are ChatGPT Image Generations & Sora Image Generations seperate?

[uncut haven]

Currently 201 generation per day. ChatGPT seems 90 per day.
ChatGPT and Sora is separate.

[desert pier]

Is this limit mentioned anywhere on the website?

[uncut haven]

No, but one user posted limit reached message in https://discord.com/channels/974519864045756446/1204360881593520128

Sam said this is temporary, so hope this will be lifted.

"Unlimited(but actually limited)"

[desert pier]

How do ChatGPT's & Sora's image generation differ?

[uncut haven]

Oh, I meant limits are separate.
They use same GPT-4o image generation.

[desert pier]

Sweet, though it's weird why they'd seperate the limit... 🤷

[quiet brook]

yep

[uncut haven]

1 queue limit really feel TEDIOUS

[desert pier]

I remember registering an account on Sora, it mentioned 50 image generations a month as a Plus User.

[uncut haven]

Docs saying "Unlimited images and video"
https://help.openai.com/en/articles/10245774-sora-billing-faq

[desert pier]

Aha, maybe they've changed it after Sora released to the public after a few weeks or so?

[frozen copper]

Yeah, the monthly limits got removed

[dim cradle]

Well, its more efficient now😂

[uncut haven]

On March 22, they removed credit and granted unlimited generations to Plus and Pro.
On March 25, they released GPT-4o image generation, and temporarily introduced rate limit since their GPUs are melting.

[quiet brook]

meanwhile pro leaves untouched because they don’t want to let the whales be mad

[dim cradle]

So I can use now the Chatgpt app again since theres no advantage in using sora.com

[uncut haven]

They blocked Sora account creation. It's just like Sora day 1...

[sudden geyser]

Ugh the image generation gets dumber. "It sorta maybe look like random person who exists somewhere in the world, hence I won't generate this photorealistic image." Sometimes it works, sometimes it hits the policy.

[tardy depot]

I have never tried it, how can it be better, it's the same model

[dim cradle]

bing uses their own version of dalle3

unless they update to 4o too idk

dalle3 actually did premiere on bing

[gusty finch]

how does the new limit work ?

[dim cradle]

depending on demand

[gusty finch]

how many images per minute ? i have the plus tier

[pulsar plover]

per minute? 0.2

[rigid bough]

limits are dynamic based on load and no objective statements from OpenAI on actual limits right now

[gusty finch]

i had a 7 minutes breakdrown

and its still going on a few minutes

[uncut haven]

In ChatGPT, it's quite random.

[rigid bough]

I am on Pro and have been generating without limit and each generation takes about a minute

[uncut haven]

Like DALL-E 3.

[gusty finch]

wish i had pro

[vapid saddle]

you can see there are quite some users here these days telling how good the new GPT image creator is, and they are not happy if you point out no it's not good. They might be working for Open AI and want more people to pay for the upgrade, which is just like Iphone changes the position of its camera.

[gusty finch]

it seems i am going to need get a job work for it ahaha

[quiet brook]

keeps popping up something is wrong for me

[uncut haven]

Well, it seems they got GPUs enough for text generation, but not enough for image. GPT-4o feels faster than before...

[sonic badge]

i still have a 2 que limit

[quiet brook]

plus?

send screenshot

[sonic badge]

yea

cn u attatch ss here

[quiet brook]

send in #images-canvas

[uncut haven]

Attach to https://discord.com/channels/974519864045756446/1204360881593520128 and refer to it.

[sonic badge]

ok

done

can you guys only do 1?

[uncut haven]

Do you mean batch size of 2?

[sonic badge]

yea

like in sora

like 2 images in the que

[uncut haven]

We meant concurrent generation count. Currently it's limited to 1 in Plus...

[sonic badge]

let me try that

[uncut haven]

Like, 2 different promts in the queue.

[sonic badge]

nope

@uncut haven check image canvas

i have 2...

[uncut haven]

Did you get those 2 images with different prompts?

[sonic badge]

yea

[uncut haven]

Oh, alright, I see now.

[sonic badge]

is it different for you?

[uncut haven]

I can't do like that. I see "You can generate 1 image or video at a time" when I queue another one.

[sonic badge]

oh

wierd

ive tried with like 5 different prompts and hanvnt got limited

[pulsar plover]

This plus with all the limitations feels like a demo for the Pro version, but atleast it works I guess...

[silent sentinel]

I also have been limited

Been limited to only generating one image at a time, even though my account pays for generating 2, anyone else? Solution?

[uncut haven]

Most Plus users are limited. Max varients 2, queue size 1.

[quiet brook]

false advertising atp

[sonic badge]

@quiet brook i dont have any issues? does it depend on location

[uncut haven]

We can't solve for now. That's temporary limit that OAI placed to prevent their GPUs from melting.

[silent sentinel]

Queue size was 2, I had 2 queued yesterday and today, all of a sudden it's 1

[uncut haven]

Same...

[next turtle]

What model of ChatGPT is the one that regenerates smoothly?

[sonic badge]

why do some people not have it though

[next turtle]

The one that starts with a blur and slowly unblurs imafe

[silent sentinel]

Understandable if they need to limit use, but I wish it would say temporarily reduced to 1, instead of making me think I'm insane

[next turtle]

Is it a new model

[uncut haven]

GPT-4o image generation
https://openai.com/index/introducing-4o-image-generation/

[dim cradle]

Why is the image gen faster in App? Does it put less effort in it?

[sonic badge]

guys this is wierd, making images for me also only take about 1 min instead of the 100+ it has been for some people and its overall fast and im not limited? Anyone else?

[dim cradle]

In App? Yes

I feel like it puts less effort in it

[sonic badge]

im using website

[dim cradle]

sora ?

[sonic badge]

nope

chatgpt.com

[dim cradle]

yea thats what I mean

[sonic badge]

oh

[dim cradle]

Less effort?

[sonic badge]

dont think so

[quiet brook]

ok, try to submit another prompt while its already generating one

[quiet brook]

they dont like revealing much

[uncut haven]

"Less effort" is impossible. one missing token and part of image will be glitched out or missing something.
Seems ChatGPT has priority in GPU pool, I guess. (Also, it only makes 1 image at a time... which makes it faster and less consuming.)

[dim cradle]

o3 image gen when

[sonic badge]

i did

it works i did with 4

[quiet brook]

in sora

[sonic badge]

yea

[dim cradle]

does that even make sense? 4.5 would make more sense right? Because of the training data

looks like I cant generate goku, naruto anymore

[quiet brook]

does it show a loading ring of 2

[azure crown]

Why doesn't display the images generated in temporary chat?

[uncut haven]

I saw that in images-canvas, but it was deleted.

[sonic badge]

yea it does i can go further too I cn dm u screenshots if you want

[quiet brook]

alr

[dim cradle]

Having the same issue..

[azure crown]

What's your plan?

I have pro but this is so annoying 😄

[frozen cloud]

none of my generations work

and tbh yesterday I was comparing my team account from work and the plus account I pay for myself, and my own plus account was taking much much longer to start generation -- I was wondering why that was?

[quiet brook]

pro tip you can use multiple chats in chatgpt to generate multiple pics at once, because sora only allows you to do one concurrent now

[frozen cloud]

ah I see, so it's working on Chat GPT but not on Sora.... strange

[lapis thistle]

Hey guys. Just subscribed to plus to try the image generation. Getting a lot of errors. Finally was able to get an image generated and now it says I have to wait 5 minutes to try another one. Is that right?

[rigid bough]

there is a rate limit on Plus at the moment so everyone can get a chance to use it

servers are overloaded, by rate limiting they preserve the service

[quiet brook]

can pro still do 5

[rigid bough]

dunno, I have been doing 2 at a time

[low skiff]

Can someone explain me why some things are content policy restricted and some dont? For example i can not create darth vader but i can create jeff bezos

[azure crown]

Different companies different copyrights

[quiet brook]

only being able to do one concurrent gen is kinda infuriating

[rigid bough]

would you prefer one gen and the services to be up, or more with extra unreliability?

[red prairie]

Sora isn't even working for me

[dim cradle]

more with realiability

[rigid bough]

well, we have what we have

hopefully OpenAI are standing up new GPU capacity

[dim cradle]

That should be their first way. Not making it more "efficient". How do you want to make it more efficient without throttling it down? He should have said " We are scaling up our capacities, shouldnt take longer than 1-2 weeks"

[rigid bough]

who knows how long it will take to source more GPU, it could take a couple of months with shortages at the moment. I would like them to do what they did 18 months ago and pause new subscriptions

[dim cradle]

Otherwise I suspect its an investment to get more plus subs..

[dim cradle]

Yeah but I guess the aim is to get everyone back to their service

Especially plus users

And making it worse afterwards to save compute..

[rigid bough]

Plus is priced too cheaply now, it should be $49 a month with all the new features available

[dim cradle]

I dont know if you should think that way. You could finally say plus is WORTH its money instead of emphasizing the idea to make it even more pricy. Plus wasnt worth to me before anymore but if we get unlimited, 2 in queue, with 4 images each, with consinstent quality and low restrictions, it would be very much worth the money. Im still waiting to see whats happening before I renew my sub

[rigid bough]

most people would pay $49 a month for Plus, if they put the price up a small percentage would cancel their subs but most would continue

[dim cradle]

Plus is to inconsistent to justify a price of 50 a month

Even pro is, sometimes

for 200 a month

Since plus would be 1/4 of pro, it should also get much higher limits of text models like o1 which is still 50/week

[quiet brook]

Real

[quiet brook]

for the love of god please SCALE IT UP ENOUGH before making it free for all

if they dare to nerf or throttle it more… 😭

[dim cradle]

This is all a false game tbh. They know what they're doing

Maximum profit

[dim cradle]

This seemed all to good to be true😂

[lapis thistle]

I am getting policy violation on absolutely everything that has to do with replacing one person with another. Is that a Europe thing?

[rigid bough]

no, it's everyone

[dim cradle]

Some things that worked before perfectly fine get now denied. Having the same issue. Theyre working on how they want to extend or remove restrictions

[quiet brook]

we have trust issues with openai now

[rotund flame]

they definitely censored a lot of things last night

[lapis thistle]

It's unusable. I asked him to give me 10 ideas for wall art. Then I asked to generate mockups and got violates policy error. So his own ideas violated his own policy.

[rotund flame]

I would even pay Pro if we could get the uncensored version because that would help with a lot of things here at work but it's like every prompt is rejected now...

[dim cradle]

Ever had since 12 days of shtmas

[dim cradle]

They wont make pro uncensored

[rotund flame]

i mean the version we had yesterday

[quiet brook]

so real… 12 days of scammas

[rotund flame]

But I guess it's now even more censored than midjourney

[dim cradle]

I mean we got 1-800 Chatgpt

[azure crown]

how can i generate spiderman without restrictions?

[remote stone]

Describe Spiderman without mentioning it's Spiderman, don't use the word Spiderman

[quiet brook]

not even available to countries other than us and canada???

[azure crown]

Still restricted

[rigid bough]

image generation is quick now on ChatGPT

[rotund flame]

"draw a man that looks a bit like spider man but isn't spiderman" try this prompt maybe

ok "spider man looka like violates our content policy"

[azure crown]

Doesn’t work :’)

[rotund flame]

it's just stupid at this point...

[azure crown]

Yea

[quiet brook]

the fact that openai has tons of guardrails on a silly image gen model while other companies doesn’t give a damn is crazy

[rigid bough]

OpenAI care about longevity and not being sued

[dim cradle]

Its oversensitive

[rigid bough]

it's fine

[left ginkgo]

It's not fine. Stop lying to yourself. Restrictions are crazy for people who pay for the product.

[next fog]

Is sora always on server capacity? Damn thing gives me the same response all the time!!!

[rigid bough]

IP and Franchises with lots of lawyers mean that OpenAI needs to restrict generating them

[vagrant zodiac]

plenty of paid services have restrictions on how you use it. most do, even. Nobody should expect that just because they pay for a product that they can then do ANYTHING with it. You buy the product with the understanding that restrictions may apply. That's like the entire point of the ToS that you agree to.

[north snow]

Guys can some recrate my photo using this new model with this popular Ghibli style?

[rigid bough]

why don't you do it yourself?

[north snow]

I cant, idk why

[verbal sequoia]

Showcase your DALL·E creations to the world!

Collaborate with our OpenAI Instagram page! Just invite @openai as a collaborator before you post. If selected by our team, your work will be featured on our handle, giving you more visibility.

Quick tips for selection:

• No copyrighted or profane content.
• Your profile needs to be public.
• Stay mindful of sensitive topics.
• Outputs that closely follow the prompt are preferred.

Feel free to share reels, carousels, or just a single image!

[rigid bough]

are you a free or paid subscriber?

[north snow]

free

[rigid bough]

pay then

[left ginkgo]

First of all I'm not talking about restrictions on trademarked content or illegal content. Me, and presumably a lot of others here are talking about the blatant restrictions on anything slightly deemed "harmful" or "offensive" by OpenAI. It's gotten to the point where in the Imagen most prompts are rejected if they contain anything deeemed even slightly harmful. Now, 4o is fine with harmful topics, pretty much never refuses them but the Imagen restrictions are insane.

[vagrant zodiac]

It should work with chatgpt 4o right?

[dim cradle]

You can dm me your picture. Ill do it for you. No need to pay and be disappointed.

[tall bobcat]

its not working at all on plus at the moment

[left ginkgo]

Plenty of other corporations and Imagen companies have much lesser restrictions than OpenAI.

[vagrant zodiac]

Then use those? OpenAI's restrictions have been clear since the de2 days.

[rigid bough]

then use them and stop complaining

[dim cradle]

Youre killing conversations. Doesnt lead to improvements that way.

[left ginkgo]

I'm not complaining I'm advocating for a better future where everything isn't dumbed down and tailored to be "SFW"

[rigid bough]

I am happy that OpenAI has a SFW standpoint

[vagrant zodiac]

I don't think people are realising how much of a rabbit hole allowing this content can be. AI is super powerful, people will use it to generate very disturbing things. OpenAI has tried to be a generalist in this sense and allowing these things could open them up to future problems.

[left ginkgo]

I'm a paying subscriber, and no offence to Robert but he pays for the two hundred dollar plan so I doubt he'll have anything remotely bad to say about OpenAI or ChatGPT. Just my two cents. We need discussions like this to further the development of AI.

[dry torrent]

Is the image generation working for anyone? Nothing has been working for an hour at all

[left ginkgo]

Why should very disturbing things be gatekept? As long as it's not illegal or depicting anyone in real life then I don't see why people shouldn't be able to make it. There's plenty dark fiction, highly disturbing TV shows and books etc out there and has been since the beginning of literature.

Quite simply it's human nature and restricting it is not the way forward.

Furthermore I realize how harmful unrestricted AI can be. It's scary but we shouldn't lobotomise it because of the mere thought that it could produce harmful results and do harmful things.

[vagrant zodiac]

Because it's hard to control just how bad that gets. There's not an easy way for them to identify when something veers into that terrirory at generation-time, so you need a very very extreme limit to prevent it. Frankly the amount already allowed on there that we see would suggest to me that this is about far more extreme content than basic violence.

[rigid bough]

People who say other companies allow things that OpenAI do not allow just need to go use those other companies

[north snow]

Sended

[left ginkgo]

I do understand that and hope they reduce the guardrails and limits as time goes on.

[left ginkgo]

ChatGPT is arguably the best Imagen model out there right now. No, not arguably, it is the best. Telling people to go to other companies is a void argument I'm sorry. There's a reason we're here in the OpenAI discord. We use ChatGPT and we're advocating and sharing our opinions to try and change and better the platform.

[rigid bough]

sigh

[left ginkgo]

See what I mean?

You're not providing any insightful arguments that challenge mines.

[white skiff]

people have wished the same several times with openai releases and its pretty much always gone the other way

[quiet brook]

its broken for me...

[left ginkgo]

Well lets hope sama is on the right side of history this time. Based on his twitter posts im hopeful man

[white skiff]

he has said the same for years now

[left ginkgo]

This.

Assuming they achieve AGI and superintelligence before anyone else they're in control of what goes in and what comes out.

[quiet brook]

if they nerf this its basically shooting in their foot

[vagrant zodiac]

I mean we're talking about what a private company allows on their service... in a world where other companies offer solutions which DO allow those... Idk what personal freedom aligns with 'the freedom to use generative AI to make graphic images;?

[left ginkgo]

Oh yeah for sure with the publicity they've got right now if they nerf it's game over

Exactly. If AGI can't even produce sexual or violent content (as an example) to a consenting adult which has paid for the platform then there's something seriously wrong.

You can't babysit the masses.

[quiet brook]

I dare you to timeout again in chatgpt when generating a picture…

[dim cradle]

In few years maximum we got some completly uncensored imagegen which is perfect in every task

[dim cradle]

Its just on openai to get this position first

[quiet brook]

from “feel the agi” to “yet another general img model”

[left ginkgo]

AGI needs to be near fully unfiltered.

[white skiff]

they successfully have done past years, so...

[left ginkgo]

They've did a good job with GPT-4o to be fair. It very rarely rejects a request

Compared to two years ago

[left ginkgo]

And they've gotten away with it because they've been the leading superpower within the AI race. We're no longer living in a time where that's the case.

[velvet rampart]

They have decreased the number of generations. Yesterday, I generated somewhere over 50. Today it was around 30 images.

[dim cradle]

Information can never be harmful. Its on society to ensure people that build dangerious stuff wont do it

[quiet brook]

In chatgpt or sora?

[tardy moat]

Agreed; anyone else finding tons of guideline issues all of a sudden?

[left ginkgo]

They've cranked up the restrictions recently

[velvet rampart]

ChatGPT. I include failed ones in the number.

[vagrant zodiac]

I feel like these attitudes towards AI are why I've been slowly clocking out of the community. Everybody wants everything but doesn't stop to pause about the ethical implications of allowing that.

Some people have an all-or-nothing approach that smacks of entitlement. For a tool we don't even fully have a grasp on the limits of. Grim. Oh well... Gonna go back to generating things for fun.

[tardy moat]

Well, that is just the opposite of fun

[quiet brook]

first it was 90 on day one. now it’s 30??!!??!?!
also was it telling you that it’s a daily limit or it’s just “you’ve hit a rate limit and you need to wait xx minutes”

[white skiff]

not much has changed, people here cry about not making able to do spiderman or lewd and the same time refuse use the other services

[uncut haven]

Well, if I need unrestricted anime, then I will go NovelAI. If I need dedicated writing service, I will go SudoWrite. If I need maximum privacy, I will go local LLMs. There're various services/platforms out there with different characteristics.
Also, any service costs money and some research - nothing can be completely free, nothing can be completely unrestricted. However, OpenAI is best for me in everyday purpose - from writing to consulting, learning, coding, getting resources(like, images).

Of course, openness, transparency, and freedom is what we all need ultimately, but AI is still somewhat risky to open. See ElevenLabs voice AI is used for fraud and voice phishing, for example. OAI's products are also used for fake news, fake blogging, etc,.

Still, OAI is going more and more open. I tested it and it's more open than any other proprietary AI service for now that have content moderation in place.

They just needed to come clean, transparent. Single line of official announcement can prevent conspiracy and hating.

[dim cradle]

#images-canvas message doesnt seem to be temporary..

[velvet rampart]

I don’t think so. To me, the filters have stayed the same from day 1. It is just that we are doing so much with it. One approach works and the other not for the same goal.

[velvet rampart]

I agree.

[quiet brook]

the main problem is
other services aren’t that good, or we have already paid for this service and isn’t going to pay for another one

also it was able to do it on like day one

[white skiff]

and then people crawl back to openai products and they can keep their policies as is

[dim cradle]

Openai shouldnt be funded with billions

[quiet brook]

openai probably have a bloated strong moderation model that is able to detect anything that is just slightly look like it violated their fragile policy and just blocks it with brute force

[velvet rampart]

About filters, context matters a lot - how you say it and how long prompt is used.

[white skiff]

original chatgpt was absolute gold, so was de3 before super nerfs, now it will repeat with 4o image gen

[quiet brook]

I swear dalle 3 didn’t look like this sloppy back in Oct 2023

[velvet rampart]

I have generations that were blocked due to a too-short prompt, but they went through when I used a longer, more descriptive prompting style.

[quiet brook]

I mean they cook up something great then cripple it down to bare minimum, so kinda yeah

we are experiencing the golden age of 4o image generation now. cherish it

[velvet rampart]

What I do not understand is what has gotten worse since day 1. Could someone please explain?

[rigid bough]

the masses are crying because they can't create Disney IP

[velvet rampart]

Disney IP, that’s a given since dalle3 system. Nothing new there.

[winter hull]

What issues are people having with the image generation that weren't present on day 1

[simple sand]

Idk what youre talking about in specific, but Increased restrictive policy trigger, removed the ability to queue multiple tasks at once, openly removed unlimited generations from the pricing plans for plus users after having it there before, added new gen limits and not showing what those limits are anywhere else nor the ability to see how far you are into gens, and added a cooldown in between image generations. Honestly that last part is fine, it's simply the fact its on top of everything else.

[winter hull]

I've tried the same prompts and its the same blocking copyright and lewd(especially when it comes to women) stuff

[simple sand]

Sam altman even said himself that people are getting generations blocked that shouldn't have been

[winter hull]

Yeah but that was happening on day one too

[simple sand]

No?

[white skiff]

some things have changed

[winter hull]

I can run the same prompts I did on day 1 that weren't blocked and the same ones that were and its the same result

[simple sand]

Images I've used yesterday have stopped working now

[lucid mesa]

No on day one it wasnt so restricted

[dim cradle]

They did remove unlimited for plus? Lol. How much is it then?

[winter hull]

What kind of stuff has stopped working

[winter hull]

It's still unlimited but there 5-10 minute cooldowns every so often

[simple sand]

They havent said but its now removed on the subscription plan details

[quiet brook]

exactly

[lucid mesa]

They put cooldowns because too many people use it and they can't keep up

[dim cradle]

Wow

Glad I didnt fall for the hype and renewed my sub..

[winter hull]

blame all the studio ghibli people 🤷‍♂️

[simple sand]

We will

[velvet rampart]

This is my experience too.

[quiet brook]

rate limit them, NOT US

[winter hull]

normies will get bored with it in a few days anyway

[dim cradle]

please

[quiet brook]

in fact they removed it on both plus and pro #images-canvas message

[winter hull]

this is the best ai for making content though, I'm thinking about how I will be able to make videos, comics, etc in the meantime. it enables real art much more than other image generators where you're just up to the whim of randomness

[simple sand]

Can we get some crappy nvidia gpu from the back and send any studio gibli prompts to that

[winter hull]

the filters just block so much creativity though it sucks

[quiet brook]

Real

[velvet rampart]

For Dalle3, there were and are rate limits on how quickly one can generate and how many images. So nothing new there. We never have had unlimited image generation.

[quiet brook]

I’m sorry but I’m not a normie…

[quiet brook]

iirc they never advertised dalle 3 as unlimited…

[winter hull]

fan art has been a thing throughout history there's no reason why any copyrighted characters should be blocked, no other image generators block that except for openai

even bing dalle-3 doesn't block copyrighted characters

[quiet brook]

yeah Bing img gen is better on copyrighted stuff

[velvet rampart]

That’s Microsoft. I think that even Mickey Mouse company thinks twice going after them.

[quiet brook]

openai now might as well just get acquired by Microsoft

[winter hull]

what's funny is that it literally shows you the copyrighted characters/art anyway before deleting the image so they're not even protecting they're asses anyway. it's easy to circumvent too btw, just use the developer console in chrome on the network tab and you can pull the in progress images from that. Just tell it to fill the bottom 40% of the image with black, and only put content in the top 60%.

[dim cradle]

same. i think most people only use ai to make ip stuff and whine when they cant

[winter hull]

seems weird to me they dont allow copyright characters but allow pictures of presidents and stuff smoking drugs

[quiet brook]

yup

[dim cradle]

well people can opt out if they want ryan

ip can bring a lawsuit

[winter hull]

no other image generation company has been sued lol

[dim cradle]

OpenAi is synonomous with AI for most people, so openai has to be cautious they are in the sites of everyone for a lawsuit

[winter hull]

well they're being sued I guess, including openai, but I doubt it'll go anywhere

it's hard to rule on it's gonna depend on the judge because its uncharted territory

[dim cradle]

yeah an openai doesnt want to be the guinea pig

so theyre cautious. simple as that

[winter hull]

well they already are that's why they're being sued

[velvet rampart]

The same approach as with Dalle3 works with 4o for IP stuff. Describe it, reference it, and do not say it directly. Use long prompts that are over 1000 characters long, including spaces.

[dim cradle]

yeah, so why make it worse by allowing ip laws in

[winter hull]

it's impossible to not be sued anyway as a company that big it's just a fact that its gonna happen for every stupid reason that's why they have legal teams

[dim cradle]

you basically answering your own question

[quiet brook]

we have trust issues with openai now…

[dim cradle]

I dont have any trust issues with them. i dont put my faith in corporation to start. i know how the business world work

[winter hull]

i can tell you already what's gonna happen with 4o image generation, it's gonna end up like dalle-3 where it can't generate anything realistic and they make everything look overly-AI generated. enjoy it while you can

[coarse flint]

yeah seems like its blocking more stuff now, at first it seemed like almost everything went through. now nah

[winter hull]

then wait a few months for deepseek or another company that actually has confidence in their product to release something similar/better

[tardy moat]

Yeah this sucks. I’m running into guidelines everywhere.

[quiet brook]

I hope they do NOT

[winter hull]

but they're fine with sexualized men, especially sam altman

[tardy moat]

Sama said we should be able to sexualize both

[winter hull]

both need to be allowed

[quiet brook]

also the moderation isn’t cutting off your prompts from the start, it’s either in the middle or the end

[velvet rampart]

Believe, they are tight here also. It might be different from female side but it’s there.

[simple sand]

Thing with dalle 3 is it was different since there were alternatives like copilot got access to it and they gave 150 gens (4 variations) daily and it was like that since the start and when openai did release it, it was limited gens from the start and it stuck with that. This is more like dalle 2 where it was kept for white-list and launch day where people got unlimited generations, and once it did release it was turned to a token based system (way worse payment system) so all the people who originally had free access before that were turned off and everyone looked for other alternatives, like stable diffusion, midjourney, or ideogram. What's worse about this situation is theyre limiting the paid users now vs the free users (they removed free tiers ability to image gen at the moment but if it comes down to it they should remove it permanently rather than limit the people who paid). They're actively changing policies and now removed unlimited from their details and have left themselves open for people to refund the subscription.

[tardy moat]

Sama did tweet something to that effect

[coarse flint]

yeah, what made it the best image genorater is the high quality plus being able to almost make anything you want. But if the restrictions continue then its just going to get out done by reeve'

[rigid bough]

it's not moderating the prompt, it's moderating the image

[winter hull]

this is the most advanced and useful artistic tool of all time, sexualized men and women have always been art, there's no reason it shouldn't be allowed, especially tasteful art.

[tardy moat]

Maybe they’ll loosen back up I don’t know. But if guidelines remain this tight, that’s gonna be disappointing.

[dim cradle]

US logic: showing completly rotten soldier in full detail but wont show a woman in bikini

[winter hull]

you can go back to cave men and see that art has always been sexualized to some degree its such a big facet of human identity

[coarse flint]

i used a picture of my house and told it to put people on the porch and it keeps blocking me. like thats crazy

[quiet brook]

they be generating fomo like crazy

[tardy moat]

Yeah, I’m running into some really weird stuff too. That isn’t even like sexualized or anything. I can even imagine what’s going up against the guidelines other than I might not have permission from the subjects.

[velvet rampart]

It does both. I have experienced that it has denied the prompt.

[quiet brook]

it has an extremely strong base and it knows everything

[dim cradle]

i think like openai will eventually strike deals with like warner brothers or disney, then youll be able to more openly make that kind of ip stuff. theres obviously money to be made there for both sides, so a deal will eventually happen is my guess

Unfortunately xAI is just as unrealiable as OpenAI, otherwise I would put my horse on xAi to have the best imagegen which is also completly uncensored.

[quiet brook]

they said yes to politic figure deepfakes but said no to Mario

[velvet rampart]

Context matters. What you have discussed before in the chat and how you present the issue. I always ask why this was blocked. That way you learn how to phrase and not.

[dim cradle]

• drugs, gore

but no woman in bikini

didnt they tweet some things are being filtered that shouldnt be. maybe theyre still working out some bugs

[dim cradle]

which equals making it more restrictive..

[quiet brook]

and possible degradation of quality

[dim cradle]

better go find an alternative then.

[dim cradle]

Ong if you were god we would still be apes

you guys should check out one of the ai that let you make your nintendos and girls in bikini, im sure its out there

[velvet rampart]

So, I think that we need to stop whining and report things that are blocked but should have gone through. Goodness, they launched the product this week and we are stress testing it lots of weird ways, some that they could not imagine.

[verbal sequoia]

Get notified for server events!

Visit id:customize to pick up the <@&1261377106890199132> role.

[dim cradle]

all these guys do is whine. if theres no issue, they make one and drone on and on and on...

[quiet brook]

isn’t gpt4o img gen the only model that is token based??? other alternatives are still stuck in the limbo of diffusion models and they are NOT good at text rendering and such that 4o nails

[dim cradle]

i dont know, go research it and find their discord

[quiet brook]

well, Gemini 2.0 flash kinda is the only thing

[uncut haven]

Gemini 2.0 Flash image generation.

[quiet brook]

yeah I thought of it just now but it’s quality has been completely outshined by 4o…

[uncut haven]

There's chinese open-weight model Janus Pro, but that's PoC - not usable in real world.

[simple sand]

Whats going to happen at this point, (since openai's somewhat gave a hint at how they upgraded their image gen) is that every other image gen company is going to try and copy the same technique now that they know its possible, and then they're going to make it cheaper/less restrictions/open source. Whats sad is openai can't seem to hold their user base and the only people that do stick with it care more about the openai company more than they do about finding a reliable image gen. If they can limit the free tier rather than limit the plus/pro tiers. Way more people that are already paying will probably stay due to ease of access and reliability.

[quiet brook]

openai is just faster than other companies

with advanced tech

[dim cradle]

Waiting for that

[dim cradle]

and funding..

[uncut haven]

So the main problem is that OpenAI offering most advanced AI... /jk

[quiet brook]

also like. pin this message or something instead of letting a bot post it every 5 seconds to let it stick here, it sucks to scroll through a big embed just to see new messages

[simple sand]

They do but they limit it to oblivion after launch date and give other companies the chance to take over as they have less risk of being sued and they don't have as much care for what the user base generates (not saying OpenAi should completely do that, but they need to use the advantages of what they have and not rate limit the plus users), its what happened to dalle 2

[simple sand]

Agreed

[dim cradle]

I blocked the bot. Makes it less annoying

[uncut haven]

Is it Discord bug, or OAI App bug?

[dim cradle]

i cant believe its still doing this myself. a few days okay

Discord staff

[simple sand]

The embed? Its the bot itself

[quiet brook]

it’s just doing this to make sure this message stays at the bottom of the channel

[simple sand]

Yea

[dim cradle]

it starts to feel like torture haha

but it also seems like mods dont even show up here much

like once a day so they dont get annoyed by it i guess

[uncut haven]

They can't do anything, since it's not problem of their side...

[dim cradle]

well who do we complain to then

[quiet brook]

I need to make my image generation benchmark look more official smh, and after mj v7 releases I might buy the mj subscription again to test it, currently 4o img gen is the 1st on the leaderboard

[dim cradle]

when is mjv7 out

[uncut haven]

Midjourney recently did survey for their video model also.

[quiet brook]

rumors said its next week

[simple sand]

Something I find funny is I'm pretty sure there's like 2 images in the 4o image gen announcement that would be triggered by the restrictive policy trigger now (woman, Karl marx)

[dim cradle]

could be interesting. this probably force them a little haha

[remote stone]

more than image quality, what midjourney should really nail that this new dall·e version has nailed is listening to the whole of your prompt and actually nailing details

it's really its biggest advantage over MJ

[simple sand]

Following off of that, openai also has an advantage over mj in the fact that it's really easy to make/edit the exact image you want with a short prompt in chatgpt + remixxing which with mj, its much harder to do and requires a longer prompt

[remote stone]

Yeah this dall·e version is the GOAT at image editing AI at the moment

[quiet brook]

its not dalle

[remote stone]

Oh it's just called 4o or?

[simple sand]

I think they do call it dalle 4o in the code but yeah rn its just 4o image generator

[brittle shard]

It’s the actual language model making images through language

[simple sand]

Its like a feature of 4o

[remote stone]

Neat

[plucky hare]

It's part of 4o's native multimodality: https://openai.com/index/introducing-4o-image-generation/

[vapid wigeon]

i still have overload errors despite the rate limit

[dim cradle]

i always find mj users cult like and the service itself overrated every time i have tried it. but i guess thats for ai discussions

[uncut haven]

Well, "I can't do that" text response also count toward rate limit... so nothing special, actually.

[dim cradle]

yup

[vapid wigeon]

that's unfortunate

[eternal kraken]

If image cant get generated due to policy restrictions, does it still count as a generation from your daily usage?

[brittle shard]

Probably. Ask it before hand if it’s going to censor it

[simple sand]

Whats funny is one time I wanted to generate a realistic image of steelix, it blocked it, I asked what was wrong with it, it told me it was a copyrighted character but it could help generate a close imitation of it, I said yes and it generated the image (needed a few changes but it worked)

[uncut haven]

I encountered funny situation when testing about filtering. 4o generated and returned image without filter, but 4o responded "Sorry, I can't to that" as text, and chat title was "Inappropriate request rejected".
They're completely acting on their own...

[brittle shard]

They’re alive

[short geyser]

Hay I was wondering if I upgrade to plus now will I get the new image generation model?

[eternal kraken]

but the moderation is just ridiculous atp tbh, can't even generate funny memes with it

[brittle shard]

Most likely but who knows with how OpenAI rolls stuff out

[simple sand]

Yes but they're currently changing policies and adding a little bit of limitations so it depends what you want to use it for

[brittle shard]

They’re probably getting 3000 cease and desist letters a day right now lol.

[short geyser]

text on image mostly

[velvet rampart]

I think that there are two levels of filters going on, the actual text and the generated image. The image being generated is checked several times. All this is tied together by context of chat. Less prompt you have, easier it is to be blocked. If the prompt is longer, the offended piece that got the short prompt blocked now goes through. Drowning it one could say.

[dim cradle]

if you're not getting it to make ip stuff and girls in bikini youll be fine

i still am not having any of these issues other than speed

[tidal cliff]

Google or Openai images?

[brittle shard]

Oai

Is t even a discussion tbh It’s like 200x better than anything else at all

[tidal cliff]

Why?

[eternal kraken]

yes probably, but i dont even try to generate public figures or anything which "should" get blocked. i just try to generate some text over an image with the text "TAX ME IF YOU CAN"

[tardy moat]

literally the best stuff to make

[brittle shard]

Multi gen cohesiveness and prompt adherence

[dim cradle]

you might just be able to find an image or two of a girl in a bikini online

[simple sand]

Most of the time yes, atleast more than image to image, but the restriction policies are currently turned up to max and theyre still trying to figure out what limits to pu on plus users as they just removed the detail that says unlimited generations for plus and pro price plans. So if you have a project in mind you want to do go for it but if you want to make memes of copyrighted characters or redesign characters then its rough

[tardy moat]

I've seen all those

[dim cradle]

and ip stuff? i guess if you have no imagination

[tardy moat]

Depends on what you are doing. If you are running ttrpg set in different IP worlds, it makes sense. Even if you are doing anything transformative, like taking one IP style and doing it in another, is possibly transformative like what Warhol was doing with the soup cans

[dim cradle]

well campbells soup is probably safe

[lean wyvern]

I still have Dall-E as my image gen model? is there anything that i can do to get 4o? (as a free user) ?

[brittle shard]

idk man big soup takes their designs seriously

[dim cradle]

ip laws in america are like more defended than anything. so of course the #1 ai company is going to tread carefully on that stuff

[brittle shard]

sam said free is delayed because of demand 😦

[tardy moat]

there was something else he did in the same style that caused a legal case, it escapes me now

[tardy moat]

most the IPs are American own

[dim cradle]

yeah thats why they are careful im sure

if disney sets their site on you, well, they have more money than god for a lawsuit

and most of the media coverage ai gets is how its stealing the soul of art or whatever, so disney suing openai would be like clickbait heaven for everyone. so... of course openai will be cautious with it

[tardy moat]

This is the case I was refering to
https://en.wikipedia.org/wiki/Andy_Warhol_Foundation_for_the_Visual_Arts%2C_Inc._v._Goldsmith?utm_source=chatgpt.com

[dim cradle]

batman, dc comics, harry potters, game of thrones, superman, is all owned by warner media or discovery i think they merge with, so same thing

[tardy moat]

disney's stuff also seems off limits right now

[dim cradle]

i just mentioned that

[velvet rampart]

Yes, but you can make derivatives and say to the AI so, like this #images-canvas message

[tardy moat]

Hopefully this gets lightened up on Monday, I wonder if it is possible that they would restrict the guidelines more over the weekend due to less oversight?

[dim cradle]

i think more likely just get less options. like someone else said, they are probably getting hammered with cease and desists.

and add in celebrity types will surely start to opt out when they find out about it and realize people just want to spam them in bikinis

[velvet rampart]

I just hope that they publish a style/technique guide for 4o. For example, when I ask for modern digital arts, 4o produces these painted images that it calls painterly style. Dalle3 produces these digital, almost 3D-like images. That’s just one difference in how the systems interpret styles and techniques.

[dim cradle]

dalle2 was the best for that imo

but yes a style guide would be great

[velvet rampart]

With these current rate limits, I do not want to start to experiment with all what’s there.

[quiet brook]

Same

[simple sand]

How do you think OpenAis gonna top this with dalle 5, how do you get better than this?

strange now Im just getting errors whenever I try to generate an image and its counting towards the cooldown

"I wasn’t able to generate the image due to an error during the image creation process. If you'd like, you can describe any specific changes or a new request, and I can try again!"

[white pendant]

how can I generate images using the api ?

[plucky hare]

Hey! Currently you can use DALL·E 2 and 3 on the OpenAI API. 4o image generation is currently available on ChatGPT and Sora, and it will be coming to the API.

[white pendant]

how much is it?

[dim cradle]

do you know who we talk to stop this openai spam already?

we get it, dalle is now images. for the love of god stop torturing us with the message

[rigid bough]

send a message to @shadow bane and let them know

[plucky hare]

For DALL·E 2 and 3 on the API, pricing is per image and depends on the model/settings you specify:
https://platform.openai.com/docs/models/dall-e-2
https://platform.openai.com/docs/models/dall-e-3
For 4o image generation, you can use it on ChatGPT and Sora if you have a ChatGPT Plus, Team, or Pro account https://help.openai.com/en/articles/8932459-creating-images-in-chatgpt
(You also can use DALL·E 3 on ChatGPT)

[silent sentinel]

Anyone else still only able to generate one prompt at a time despite paying for plus? Thought it would be fixed by now.

[quiet brook]

wait is pro also throttled to only one concurrent generation?

[quiet brook]

🔥 DALLE CATEGORY AND CHANNELS ARE NOW IMAGES🔥

in sora right

[silent sentinel]

Apologies, meant to type plus

[velvet rampart]

I can do 2 at the same on pro level.

[chilly gale]

Hi guys, if I pay the $200 plan, is it possible to use unlimited generation image on chatgpt ?

[quiet brook]

I’m fear that even pro isn’t unlimited

[quiet brook]

not 5?

[rigid bough]

Pro is unlimited

[quiet brook]

I’ll do more tomorrow ig im so tired today, hope there are no more weird limits to plus and nerfs …

[dim cradle]

the mods replied and said this message is staying because apparently they think people cant figure out dalle is now images, but will change in the future i guess

those of us that are active have to endure it's endless popping up, but the users that dont exist wont be confused

[rigid bough]

I agree it's annoying

[dim cradle]

anyone who glances at dalle and doesnt see it, will look at the rooms and see images and unless they are brain dead i think theyll put two and two together

[willow isle]

It'll probably disappear once everyone's used to the new changes in the server and it becomes the norm - the recent updates to ChatGPT got the community some new faces and old DALL-E users back and active

That's just speculation on my part though

[rigid bough]

there are bigger things in life to worry about than a repetitive server message

[dim cradle]

i wont argue that, but we can still complain about it being annoying

they seem more concerned about pleasing people who arent here than those of us who are chatting

[velvet rampart]

Might be. I did not try.

[willow isle]

https://help.openai.com/en/articles/10245774-sora-billing-faq it's stated to be 5 here

[dim cradle]

and see, squeeky wheel get the grease. its stopped 🙏

😭

[willow isle]

spoke too soon

[dim cradle]

lol yep i knew i was risking that by sayiing anything

[spare onyx]

that felt like a personal attack 😂

[dim cradle]

yeah some mod power tripping

'oh you think its gone...' click

[next turtle]

Should we let AI take over us

People fearing about an "AI apocalypse"

[dim cradle]

idk but image discussion isnt the spot for it

[willow isle]

It's automatic

[dim cradle]

yeah but its noticeably less suddenly

[willow isle]

likely just a coincidence!

pretty funny though

[dim cradle]

🤷

i just wrote the mods about it, they replied, then i wrote back

[snow crane]

i think the gpus are dead

[fair tendon]

we’ve increased detection duration to make it less annoying, thanks for your modmail

[dim cradle]

see

you can all thank me

[rigid bough]

thank you

[worthy nova]

so are there rate limits right now for plus? right now 4o is telling me to wait and i don't know if it is just hallucinating

[proud ice]

Yee im getting doggpiled by it as well so I made a meme I posted in #images-canvas that portrays my frustration

[twilit helm]

rn i think we can generate 4 images before we have to wait

[red prairie]

I feel that somehow this new image generator has made people hate AI even more

[dim cradle]

Dunno about that, but for some reason I can't use the ghibli feature at all. Like my app on both mobile and web just straight up rejects it

Seems like I havent received the update at all

[white skiff]

free user?

[dim cradle]

Yes, but all my friends who are free users have access to the least - 3 images a day
I dont even have that

Even signed up for the play store beta, yet no good

[white skiff]

free users only have access to the old version that rejects almost everything

[red prairie]

Probably because in the past few days, everyone and their mom has been using this style so much to the point that even Miyazaki reacted to it and despises AI for that?

[dim cradle]

TBH You can't blame Miyazaki

he's a legend and to see a machine use his data and create his art without his choice has to be annoying
Let's hope this usage never becomes commercial

Let it be limited to us normal users

[red prairie]

Not blaming him I'm just saying that too much people have been using it lately and that's probably why they nerfed it. I do understand him tho

[dim cradle]

I understand but

[dim cradle]

.

[dim cradle]

Yes yes
My wording might've been a little off sorry lol

[white skiff]

it does not matter as that was the case even before this new one. sometime free users had no access at all

[frozen copper]

Plus users still 1 concurrent gen with 2 variations?

[dim cradle]

yea and it will likely remain like that

#images-canvas message

[frozen copper]

It had better not be permanent, that would really suck

[dim cradle]

It likely is

[pale saddle]

why it is saying only 1 concurrent job for plus user? Is it not 2?

[white skiff]

not anymore

[frozen copper]

It was, until we melted their servers

[dim cradle]

Until they let people melt their servers

[frozen copper]

But in all seriousness tho, have they not learned anything about overprovisioning their servers whenever there are doing product launches? There is no way they did not expect overwhelming demand after doing so many launches

[dim cradle]

They did it on purpose to get more subs. Using "oh no our servers melt" as an excuse to throttle it down afterwards

Saved compute + more subs

[pale saddle]

for me it is saying only 1 concurrent job. I am a plus user.

[dim cradle]

It is saying that for every plus user

[strange copper]

Is it just me, or has the quality of the generation dropped a bit? Like now when you throw him some object, he pays less attention to the details, the text floats, etc. Has anyone encountered this?

[dim cradle]

I feel like i always get the best fidelity on gpt one image at a time but thats not based on anything but gut feeling i guess

[velvet rampart]

I have had some generation issues this week. Not quality per se. Some earlier images bleed through. The generation process treats the chat thus far as context.

[kind phoenix]

The search and filter feature on the SORA website does not work properly. You can search for a term but when applying filters nothing happens even after reloading the page with the filters.

[velvet rampart]

So, no quality dropping issues but you have to be more precise what you want with 4o. Do not assume like with dalle3.

[strange copper]

well, for example, there was 1 promt - I needed to place a can and have it make a nice background, and in the first days everything was fine, but now it either ignores some drawing, or blurs the text on the can that it saw earlier

[velvet rampart]

Trademarked products can be an issue or if image is low quality, the AI makes things up. I just checked an image where a man holds a can from yesterday. The can is of low resolution, so the AI replicated the can in the man’s hand fine with right coloring but it made up what kind of can it is.

Also, if the image is ”far” away in the chat, it might have trouble referencing it. Give it again later to remind that you are using this particular can. As said, the AI treats the whole conversation as context.

Also how busy the servers are, has an effect on quality. This we know know dalle3’s images and their quality. Better images are generated in low traffic times.

[strange copper]

according to Moscow time, at what time approximately is the load less?

[velvet rampart]

Weekend mornings, tomorrow morning can be a good time as it’s night time in USA.

[strange copper]

it's sad( ​​Because a lot of freeloaders will now run to make cards for goods on marketplaces and so on

[dim cradle]

i feel like i get better images on chatgpt site too than sora.

[velvet rampart]

Yes, the fidelity in Sora leaves one wanting, a lot.

[dim cradle]

although i also feel its best to start a new chat for each image there. it seem to me sometime the infos from previous image will bleed into a new prompt

[brittle cobalt]

sometimes its nice when it blends stuff in, but other times it gets in the way of your vision

[dim cradle]

yes if its a similar image or setting it may not be a big issue i guess or maybe even a happy surprise. if you are shifting gear to something new entirely, best to start a new message

[velvet rampart]

Yes, this has happened to me several times. It is ok, if you are working on a same theme, but if you want truly new stuff, a new chat is better.

[dim cradle]

i wonder how many images have been made globally since 4o image maker debut 🤔 impossible to answer i know

[civic ore]

api usage 🙏

[proud ice]

Reeee

[civic ore]

with pro plan, how many images can you do concurrently?

[clear moth]

hi

need some help with prompts here

[sharp dock]

are you guys able to generate images

i’m getting feature disabled messages

[rigid bough]

where?

are you ok @dim cradle ?

[dim cradle]

Did you see one or the other ones aswell?

[rigid bough]

I saw a lot that were deleted by automod

[dim cradle]

Oh okay thats what I wanted to know. Ty

sry for pings then

[clear moth]

any tips to get better faces? i still get some ai ness to it

not as good as what other people are getting

[vague cargo]

I just paid for plus today, to use these new different generated images and now I get this restriction bomb, I'm out

[mystic halo]

is there a way to get exactly my face in sora?

[vague cargo]

is there any way to refund the plus?

[distant folio]

i just noticed few min ago sora plus account is only get 1 process at a time, Anyone facing this?

[simple sand]

https://help.openai.com/en/articles/7232895-how-do-i-request-a-refund-for-chatgpt-plus-or-chatgpt-pro

[quiet brook]

refund the subscription…. Let them pay…

[sonic badge]

are you?

[quiet brook]

I can’t lol

[sonic badge]

why

[quiet brook]

already over 7 days and I subscribed via iOS…

[twilit helm]

ios refunds work differently. most of the time they deny your refund literally

it’s one of out a million you might get a refund

[sonic badge]

so you would refund if you could?

[glad hare]

I think I might use the image model for editing existing images more so than generating new ones. My creativity is so low.

Anyone else like this?

Actually, not even that. My mistake. I’m more likely to generate images based on what people upload online and edit said online images.

[brittle cobalt]

riffing off pre-existing ideas is still creativity 🙂

[vague cargo]

thanks

is someone recomend a good ia that i can generate sonic images?

[red prairie]

You can use an AI upscaler

[clear moth]

how would that work?

[red prairie]

There is a site called Magnific ai which allows you to upscale images

[clear moth]

something free maybe?

[red prairie]

You do get 10 free uses

[clear moth]

dont think so its redirecting me towards the upgrade page , i have never used it before

[red prairie]

I never paid for it and I still had 10 free uses

[clear moth]

doesnt seem so to me

well it would help if you know anything about how to make sure the image of the person looks the same as the one i gave to chatgpt

[pseudo mica]

the incredibly obscure content policy is making this new feature nothing that people should waste their time on. Literally couldn't get it to turn a picture of a controller in to a starship.

[brittle shard]

#images-canvas message

[pseudo mica]

nice, how did you do that? mine literally said 'nope, can't do to content policies'

[brittle shard]

Uploaded a picture and said “make this a starship”

[quiet brook]

fear of it being nerfed now

it has lost some of its accuracy

[deft schooner]

image generation now rate limited for plus users now? 🤓

[quiet brook]

meanwhile pro is untouched at all..

[deft schooner]

for now at least, but it’s also marginally more expensive

[vapid saddle]

how com 4o image's human interaction with objects function is so poor? The objects can barely in the right place. Any solution?

[quiet brook]

the rate limit of this is too high now

didn’t even generate 5 images then it tells me to wait 10 minutes

[plush parrot]

100 a month is still pure greed

[quiet brook]

so this is how they scale this thing….

[quiet brook]

indeed

[plush parrot]

#image-bot

🥁

[deft schooner]

yeah chatgpt pricing isn’t it, tbh I can just run stable diffusion on my pc lol

[quiet brook]

look at #ai-discussions lol the contrast between companies is insane (and I just got muted for another minute because of posting a twitter link, wow)

1984

[vapid saddle]

exactly, people are so simple minded that they can't see through the greed part. A very simple question, why it's not called Dall E 4, but 4o image? Because Open AI has totally destroyed ruined the reputation of Dall E, while Bing Image has saved Dall E. It's all marketing and propaganda to push 4o image and the limit, pretending they are crowded used and better than Stable Difusion, Mid Journey, Bing Image, etc. The truth is 4o image can't even solve a simple issue of human integration with objects, so far behind in this image generation competition. At the same time, so many fervent Open AI fans attack those speaking the truth.

[dim cradle]

wow, how did you pull yourself off the cross to write this

[vapid saddle]

like everybody is not seeing the chaotic layout of human and objects in generated 4o images

[dim cradle]

I see the best image-gen so far. Imagen3 is right up there in a lot of ways but not with text. most people go where the goods are with ai, youre developing some weird fanboy stuff like its marvel vs dc or something

and then projecting outward

[vapid saddle]

Ask ChatGPT what yourself are projecting first

[dim cradle]

ok

[crystal grotto]

Guys

Did chatgpt new the studio gilphi image generation is available for free ?

[cobalt marsh]

rate limit is so horrible

[crystal grotto]

I want to create it but its paid ahh

[floral zodiac]

Right now, with how long it takes to generate images on the Plus plan, it’s more useful for making fun or random pictures to look at or post on social media. But if you’re hoping to use it for anything that needs a steady workflow, the wait time is just way too long. So for now, it's more for casual playing around than serious use.

[dim cradle]

definitely

but give it a week or two. they'll get more gpus

and the hype will also die down from more casual types

[valid laurel]

Is image generation down?

[floral zodiac]

Guys, don’t get your hopes up too much with this — not while it’s still working like this. For now, just treat it as something fun to mess around with when you’ve got time to kill. If a half-baked product gets released too early, it’s better not to expect too much from it… at least for now.

[rich star]

Why is there no longer 2 in queue for plus?

[dim cradle]

because so many people are using it

[brittle shard]

The anti nerf is incoming https://x.com/sama/status/1906214023269605630?s=46

Sam Altman (@sama) on X

@CtrlAltDwayne we are gonna do the opposite of nerfing it

BUT

still please chill out a bit

X · Mar 30, 2025, 5:16 AM

[dim cradle]

nice

people will still whine they can't gen 100 images at once and that openai hasn't developed a serum to keep them awake 24/7 to generate images

[quiet brook]

good, but is he actually doing it???
also the ones who need to chill are the ghibli ones, not the normies…

[north snow]

Guys can some with subscription convert for me 1 image? I want to be muppet on one photo.

[sonic spindle]

fr

[rigid bough]

please don't ask others to run prompts for you

[quiet brook]

yes, shoot me the picture via dm

[uncut haven]

That IS really good news.

I guess the team can't sleep for next 7 more days though.

[quiet brook]

Well I hope we get the ability to do 2 concurrent and 4 images in a row back first

and I hope by “not nerfing” is that they are actually scaling it up

[velvet rampart]

If we could not get 2 images with dalle3 as they nerfed that, I do not expect that we get it now.

[uncut haven]

Can't agree more. As I said, it's really TEDIOUS.

[quiet brook]

dalle 3 is like chat based and if you want multiple concurrent generations and images just use multiple chats with the dalle gpt lol since it always gets you 2 image back

[velvet rampart]

How I bypassed this issue with dalle3 and now with 4o is by requesting that one image after the other be generated.

I get an extra mile out of 4o when I bring in dalle3-like behaviour. I use the following command:
I want and expect DALL·E 3-like behaviour. Analyse prompt before generating (written analysis. Here is what I want and need:

[ashen crown]

What’s the rate limit of 4o generating images guys?

[velvet rampart]

The pics I get out have this extra on them, like what an updated dalle3 would be.

[velvet rampart]

At the moment, 30-40 images per day, including fails.

[ashen crown]

thx
My free tier is like 5 generations per day.

[quiet brook]

you should be grateful that you have it on free at all lolol

[quiet brook]

it was 90 per day on day one

[ashen crown]

There’s no doubt lol. 😂

I was going to have subscribed the plus tier early this day.

Just wanna have a try for Ghibli by myself.

[rigid bough]

oh no, not another one

[velvet rampart]

Yes, but they lowered the number on Friday and yesterday I figured out the number.

[dim cradle]

i still havent made a single ghibli haha. must be the only person on gpt not to yet

[dim cradle]

what is it?

[ashen crown]

Maybe OpenAI’s strategy for diminishing the stress of GPU usage.

[north snow]

What is generaing image limit now on gpt plus?

[ashen crown]

It may be true cuz I’ve heard many people murmuring about it.

[median olive]

just got a 31 minute wait

[quiet brook]

lmao😭

use sora, chatgpt tends to have a stricter rate limit

[brittle cobalt]

23 minutes 😦 it's like a video game forcing you to pause. Perhaps I should make an image of that.. in 22 minutes..

[rigid bough]

where are you getting this wait? ChatGPT or Sora?

[brittle cobalt]

Gpt, I've only used sora a couple times for images

[rigid bough]

how many images have you created today?

[hollow terrace]

I am a plus user without access to 4o image generation, is it not rolled out to all paying users yet?

[rigid bough]

all Plus users have 4o native image generation

[brittle cobalt]

Uuhh, a bunch, I couldn't guess. Note I also use the app on both windows and Android, not the website

[rigid bough]

ok, I haven't been limited once, but I am on Pro so that's probably why

[brittle cobalt]

It definitely is, I know some were saying 600 images for pro? But things have obviously been changing

[rigid bough]

I'd never get near 600 images in a day

[hollow terrace]

Oh okay thanks, I was confused about my chatgpt saying it was generated by DALLE, but I guess that's what the 4o image generation uses?

[rigid bough]

no, you shouldn't be using DALL-E, do you want to paste a screenshot of what you are doing?

[hollow terrace]

I can't post images here, where should I post it?

[rigid bough]

post it in #ai-discussions

[frozen copper]

lmao rate limited. 26 minutes on 4o

[rigid bough]

that's brutal

[quiet brook]

this is how they scale

[frozen copper]

Should have overprovisioned first before opening the floodgates

[dim cradle]

No way

[rigid bough]

I wonder how many images you have to make to get rate limited?

[uncut haven]

S.C.A.L.E. – Systematically Constrained Architecture with Limited Efficiency
-- GPT-4o

[frozen copper]

I lost count lmao

[rigid bough]

so you had a good run then

[dim cradle]

Its still 1 in queue and 2 images per request I guess right?

[rigid bough]

for Plus yes

[uncut haven]

Right.

[sharp dock]

Any one generating relistic/photo realism pictures with facial consistency?

[frozen copper]

Yeah, including lots of gens that i'm not satisfied with and filtered gens

[rigid bough]

fair enough

[uncut haven]

As far as I concern... it sometimes filters that shouldn't get filtered, and sometimes not filters that should get filtered.
It's really unfinished as they said.

[outer pilot]

let me know if anyone answers this i need it too

[frozen copper]

I prefer they nerf the filters tbh other then the most explicit ones

[dim cradle]

I didnt do it myself but its not 100% consistent as I saw from other examples. Maybe like 80-90? Depends

[uncut haven]

Still, GPT-4o do many 1-shot success that DALL-E or MJ fails or requires multiple shots.

[outer pilot]

what did you use

[verbal sequoia]

Get notified for server events!

Visit id:customize to pick up the <@&1261377106890199132> role.

[frozen copper]

Yeah, now we need an open source version of this. Hope Stable Diffusion (or something else) comes up with an answer to this

[rigid bough]

nobody is going to be able to open source a model as big as GPT-4o

[frozen copper]

And that's the problem, much as i like 4o now, it would be great if we have alternatives for competition

[rigid bough]

it's going to take a while because it's native to the model, it goes all the way back to the training data, the model has learned what things look like, the data sets were huge

[quiet brook]

REALLLL

[quiet brook]

they didnt know they cooked

or

they didnt know that the ghiblification is coming

[quiet brook]

sometimes its 90. sometimes its only 5

[uncut haven]

Complete random like DALL-E 3 back then...
Time is also random - from 2 mins to over 30 mins.
I saw screenshot with several hours of rate limit with DALL-E 3.

[dim cradle]

Were did you know these screenshots werent generated?😂

[frozen copper]

Cause DALL E suck at generating proper text

[uncut haven]

There's inspect element for that actually.

[sonic spindle]

people spamming it with stupid ghiblis and overloading it bro i just wanna use it to create infographics to help me study 😩

[uncut haven]

Where filtering is needed, is "explore" section in sora. Sometimes it's really WILD.

[quiet brook]

exactly, when i post some wild things from there to here itll get deleted LOL

[quiet brook]

i dont find the ghiblification interesting at all...

like, its fun to have a test of it. but when even millionares start to use it to make their pfp ghibli, it isnt funny anymore

[rigid bough]

I agree, it's a bit dumb

[uncut haven]

Now Sora fails to properly title the image time to time - it just says "Image Generation" instead of generated title.

[floral zodiac]

Gotta wait 20–30 minutes after generating just one image before I can make another. Wow, so fun. What an amazing product. 🙃

[uncut haven]

That's the result of "biblical demand" and "melting GPUs" that Sam said... OpenAI runs tons of GPUs and models require GPUs beyond our imagination, but wonder how much the load actually is.

[sonic spindle]

sad

just paid like $20 for plus the other day

purely for images

[low skiff]

This is a plan i don't think they really reach the limits

I would say it is a Strategy, also the Generation time of the images can easily be faster but I think it is intentionally slow for now to make sure people think twice what ro generate as it takes a bit

[thick smelt]

Hello guys

How ya doin'?

[sonic spindle]

unusable at those rate limits for serious work

need to iterate on them sometimes

cant do that when waiting 30 mins for a single image

hopefully the ghibli hype dies down soon

[quiet brook]

yes please

[low skiff]

I am pretty sure it will not stay like this. On the other hand you didnt have this model a week ago so it should not be something to count on for business stuff yet, give it some time

[sonic spindle]

yea hopefully

[low skiff]

So complaining about that you need the images but you didnt have it before as well is not fair:D

[uncut haven]

Like Sora release. It was cooled down after a week if I remember correctly.

[low skiff]

But yes i totally understand what you mean

We should be more grateful that we even have something insane like this

People working their half life for tech that is out of our Imagination

[floral zodiac]

The worst part about all this is that a bunch of people just paid for a subscription because they were excited to try out image generation—without realizing what kind of awful limits they were walking into. That part really sucks.

[sonic spindle]

so hyped for api access

[sonic spindle]

fr

thats what im annoyed about

[low skiff]

It will cool down

Midjourney brings V7 soon btw haha lets see what they cooked (or not)

[sonic spindle]

i hope it can do infographics

openai needs some competition haha

[low skiff]

Yes,Text and prompt coherence is most important

[low skiff]

Closest competition now is still Ideogram

[fossil linden]

It seems multimodal image generation wins when it come to text at least so no flux midjourney will be able to compete

[sonic badge]

@quiet brook if you could refund ur plus would you? Is it that bad

[low skiff]

Yes you are right the multimodality is key

[fossil linden]

Since LLM/multimodal generate image it also able to understanduser better since it is LLM too

[quiet brook]

because sora is bad…

[uncut haven]

Also MJ produces a lot of noise and distortion on specific types of images, especially for artifitial things that completely imaginary.
I see that as limit of diffusion models. Even increasing step count will not fix it, and makes it worse sometimes.

[quiet brook]

just fomo generation and you can’t really use it unless you cough up $200

[fossil linden]

Whenever the new thing release people get hype but later it is bad. I think it is more about human behaviour in general than about the product! People hyped and raise their expectation way too much

[quiet brook]

why don’t they rate limit the pro plans and just let them blast through 20 images every minute????? why is plus always the one who gets the nerfing

[floral zodiac]

Honestly, they could’ve just said it straight from the start. No one would be mad—people who can’t afford it would’ve just moved on. That’s way better than leaving things like this.

[quiet brook]

maybe someone is already hogging a lot of pro accounts and turning it into a wrapper for ghibli spam

[fossil linden]

because plus is paying 10% of pro!

[quiet brook]

while majority could only afford plus just to even get to try it gets rate limited to hell

[fossil linden]

It serve as motivator to get pro for some!

This make perfect sense from business strategy

[sonic spindle]

pro is NOT worth it for just that 😩

esp since the main selling points were unlimited o1 which has been beaten by FREE gemini 2.5 which has insane free rate limits

i mean deep research is cool but there are alternatives to that too

[quiet brook]

yup

[fossil linden]

I will say it depends on use cases and personal preference!

[quiet brook]

no free trial for most of their things makes the purchase feel uncertain too

[sonic spindle]

better to just wait for api access and pay per use

[fossil linden]

If someone want to generate more images and will get unlimited and lot more generation for them it make sense!

[rigid bough]

I pay my $300AUD a month for a good seamless Pro experience, I used to be on Plus but when Pro was announced I signed up straight way

[hexed sandal]

Is there a good prompt to restore an old image?

[quiet brook]

but not everyone can afford it

[fossil linden]

That is true and i personally do not think it is good value and i think it is way too expensive but i want to respect other people decision and use cases!

[rigid bough]

so? OpenAI isn't a charity

[fossil linden]

20$ is also not charity!

Point is it is way to expensive for most people can afford! but i do respect decision of people who can afford it and find it useful!

[rigid bough]

I wouldn't pay for Pro if it didn't fund itself

[quiet brook]

yeah lolol ppl will like rage for a day then move on..

[floral zodiac]

This whole thing could’ve been cleared up instantly if someone responsible just came out and said something like:

“Hey, sorry—starting a couple days ago, we’ve added limits to image generation. We know everyone was having fun before, but this is just part of our business plan. We’re bringing you back to reality now. If you want the same experience you had before the limits, you’ll need to pay. If you can’t afford that, then you’ll just have to stay where you are. That’s how the world works. Thanks.”

That’s it. Just be honest, and it’s all good. Everyone would understand. No one would feel tricked. People know where they stand based on what they can afford. 👍

[pulsar plover]

Yeah, but Investors don't want to see anything like that

[quiet brook]

they want 💰 💰 💰

[velvet rampart]

As I said earlier, for Pro level on ChatGPT 4o’s limit is 30-40 images including fails.

[quiet brook]

pro? really?

[velvet rampart]

Yes. That’s my experience from Friday and Saturday. I have Pro.

[uncut haven]

"Sora is not available in your country yet". WHAT

I was using it like 1 min ago without problem.