#images-discussions

1 messages · Page 95 of 1

uncut haven
#

Nevermind, clearing cookie worked.

#

Their servers seems really deep fried now...

mild cloak
#

I can’t upload pics nor log in from the phone. Anyone like that?

lofty valley
#

im getting

Unable to generate
Could not find user for provided access token.

uncut haven
mild cloak
#

Hm

quiet brook
#

dang, it’s over

mild cloak
#

And now cant make any image neither xd

quiet brook
#

also does it tell you to wait for several hours after you generate that much, which is hinting for a daily limit

south glen
#

Hey is ChatGPT down right now?

mild cloak
#

yeah I can't even log in

#

what a joke

proud ice
#

Wow this is borked lol how hard the ratelimit is hitting me

vague cargo
#

is someone recomend a good ia that i can generate sonic images or something without problems with protect content ?

next turtle
#

rate limit 24 minutes! yay!

next turtle
#

along with these other features u just entioend

lucid mesa
#

gpt just not listen anymore with image generation and just create random stuff he says he is wrong but still do it omg

quiet brook
#

cc @vague cargo

floral zodiac
#

So now the disaster’s been upgraded — from 5–15 minutes to a minimum of 30 minutes per image? This is an actual disaster. 🤣

dim cradle
dull token
#

Have they put a limit for the new image generation model?

rigid bough
dull token
rigid bough
dull token
#

😭

simple sand
#

Yo question, what messages are you guys getting saying its 30 minutes per image? I so far have been able to generate 5 in a row

simple sand
rigid bough
twin horizon
#

Why is the image gen so sensitive whenever i use my face in the image? i asked it to swap the color of my scarf and shirt, and it refused

floral zodiac
# dull token 😭

What we know now is that we’re not paying them enough to deserve smooth access to their product—unlike those glorious first two days. And that’s exactly what we need to know—and remember

simple sand
#

Alright I got 8 minutes per 5 images right now, so far so good

dull token
simple sand
#

Probably try again, then copy its response and put it here

proud ice
#

Please gpt either restrict it so you cant do hecking studio ghibli images or nerf the rate limit abit more

civic ore
#

I am not sure why they would want to kill the thing thats making them go viral 😅

#

I imagine many who are doing gens just for that, may get bored soon and the GPU's can unmelt. Although many might stay knowing it can do so much more than just Ghiblify an image

rigid bough
civic ore
#

yea? which one

rigid bough
civic ore
#

🤨

uncut haven
#

Aside from Ghiblification, AI boomers may using 4o all time to get their resources. AI 'Jailbreakers' also...

quiet brook
#

this thing has endless potential and its tragic to see it being "optimized" because of millions of images in the same damn style

civic ore
#

Its also surprising, that is the trend that went viral to me tbh. Like Studio Ghibli is obviously popular and loved but idk, didnt expect that to be the viral trend.

#

something im enjoying using 4o for, is creating synthetic multiview of characters and objects. Some image to 3D models such as Hunyuan-3D allow you to input different camera angles of your subject for improved accuracy in the final 3D model. and can give 4o your source image, and then instruct it to give different angles of subject with really good accuracy and consistency.

quiet brook
#

yeah, this thing has so many use cases

#

but they all somehow choose ghibli

lone mantle
#

but at the end of the day the vast majority of these people arent creatives tbh and especially not more advanced than beginner level

dim cradle
#

yeah. i mean most people here and complaints are that it doesnt do this ip or that ip

quiet brook
#

and we all need to suffer from the rate limits those ghibli pictures created

quiet brook
lone mantle
#

but as said its not a deal breaker either

dim cradle
#

well dont get me wrong, i wish it could do everything, i got my favorite ip stuffs, but its such a powerful model even my limited imagination can work with it

lone mantle
#

i can do that without AI

#

but for some idea iterations it could be useful here and there

#

and i dont pay for GPT Plus just because of that feature

dim cradle
#

but like the other day someone said here they regret the sub becaue it cant do batman. theres a lot of people like that. not wrong or anything either.

#

why i think they eventually will strike some deal with warner brothers, disney etc or people buy some add on for their favorite ips -- theres money to be made there so they will find a way eventually

lone mantle
#

if people really want the most power and control possible they have to develop their artistic and editing skills

#

thats the reality lol

#

none generative AI will give you that

dim cradle
#

yeah thats what i mean

#

and I'd guess it will happen eventually, theres just so much demand from most people for batman, spiderman, star wars, marvels etc

#

but i also understand those are the big money makers for those companies, and ip laws in america where they are based are highly in their favor

#

so they arent just going to roll over out the gate here either

#

but at the same time, you can do a lot of that stuff on dalle bing, but never could dalle gpt. but this model it seems you could really start to produce pro quality stuffs, so a company will be more guarded about allowing their ip material

quiet brook
#

Tbh the hype is already dying down for me…

obtuse badger
#

When using photos of irl people is there a way to stop it from changing the face?

quiet brook
#

well i still do generate a ton in sora because the rate limits in chatgpt sucks

quiet brook
obtuse badger
#

rip okie ty

lone mantle
#

puh i doubt that will happen

#

i mean certain deals happen but im not aware of such ones

#

although there was a film studio that has a deal with Runway if i remember correctly

woeful carbon
#

Why is the content policy so restricted? 90% of my images get denied with images of people in them, I see everyone else creating characters but I cannot currently.

lone mantle
woeful carbon
#

@lone mantle I've tried both, asking for a photo to be styled as a studio ghibli, simpsons, Southpark and more just says denied and stops generating

rigid bough
woeful carbon
#

Not really, ive copied prompts from sora website and still blocked

#

Tried multiple images also

rigid bough
woeful carbon
#

That's the main thing right now you know? Studio ghibli how can everyone else do it except me?

rigid bough
woeful carbon
#

Even openai showcased it, so its meant to be possible, false advertising it seems like

rigid bough
#

sigh

woeful carbon
#

Showing a feature being possible then actually not is not right

lone mantle
dim cradle
#

who knows, they may have got hit with a cease and desist.

lone mantle
#

normally Sora is much less censored

woeful carbon
#

Yeah

dim cradle
#

any ip you can do now, just be thankful its not blocked yet

woeful carbon
#

My prompt. Could you transform this image into a 3d stylized render. And I get an error can't generate this request. If I remove people it works

dim cradle
#

i think there was some bug about uploading images maybe it is tied to that

#

they will be all hands on deck tomorrow, its monday, so some of these thing will get more attention im sure

woeful carbon
#

Hopefully, the text generation on the other hand is excellent so I can't complain about everything

dim cradle
#

yeah its pretty wild how good that is

lapis rapids
#

Hello everyone, I'm new here, but I'd still like to know your opinion regarding the new model that ChatGPT uses for AI image generation, especially when it comes to generating anime-style images.

rigid bough
forest walrus
#

It's great in my experience, it's amazing at character consistency, can generate transparent background photos, great at editing photos, follows the prompt really well, etc.

It can even generate comics and memes!

lapis rapids
#

But what about how the characters look before? I mean how they looked before

forest walrus
#

It always looked obviously AI generated and goopy

wheat otter
#

I noticed that my generated images always have some post-processing added at the very minute that adds more lighting and flare. While the image is being generated and I can see some of it already, it lacks that effect.
Is there a way to switch that off? Or does anyone know an easy way to revert those with photoshop?

velvet rampart
wheat otter
#

Any success with that?

velvet rampart
#

I’ll tell once the image is done.

#

The command lessens the weirdness.

wheat otter
#

I will play with that, thank you!

bitter lake
#

Anyone knows what are the rate limits for Plus accounts? Today I got a refusal telling me to wait for the first time

velvet rampart
#

30-40 images, including fails.

bitter lake
#

Per day?

near holly
#

ive made only 6 or 8 images today and it told me that I reached a temporal limit and have to wait several minuts to generate again -.-

rigid bough
near holly
#

yes

rigid bough
near holly
#

I see,I was just surprised that some ppl were able to generate 30-40 images

#

weird

rigid bough
near holly
#

mmm yup seems like it ,anyway ,I have to be patient I guess.

frozen copper
#

Really hope they are spinning up new capacity

#

The limit is annoying

dim cradle
#

probably have to wait for the work week to kick off in earnest for major changes like that

#

it's Sunday, only so much is going to be done to improve image generations

thick smelt
#

Hi @vapid elk so good to see you again dalle_smiley

lone mantle
#

you can cartoonize and generally stylize real photos etc. character consistency is afaik best so far in comparison to other AI solutions, text is really good so one can visualize some market designs for example well

#

and more

lone mantle
vapid elk
thick smelt
#

How ya doin'

magic ether
vapid elk
vapid elk
thick smelt
vapid elk
#

// this is the images channel tho, avoid offtopic, feel free to ping me on other channels tho

wild ferry
#

Holy moly, anyone else got timed out for posting a completely harmless image of a well known politician? I just realized all these kinds of images seem to have vanished from #images-canvas . They could've least issued a warning first. I mean, come on, it's just harmless fun, and they don't even seem to have a problem with the moustache man. 🙄

verbal sequoiaBOT
#
<:book_icon:1171408210398289941> Rule 6: No political and religious discussions or content.

-# Users from all over the world frequent this server. To maintain a respectful and civil atmosphere, please avoid all religious and political discussions or content.

vapid elk
#

the mustache man isn't ok either

brittle cobalt
#

ive seen some mentioning doing that in the daily theme channel

vapid elk
#

I did, and I had awesome results, I actually re-made many of the images I've previously generated with dlale-3 and got mostly very good results, it basically "cleared" it from AI artifacts from the older model

wild ferry
vapid elk
#

Im sorry if it came to you as harsh, but yea.. avoid posting images with politicians, it is a no-go here because it tends to always spark un-welcome discussions

rigid bough
vapid elk
rigid bough
vapid elk
#

thx

#

or DM me too, I don't mind.. but the report button on Apps are much more streamlined

rigid bough
#

yes but there are two key ingredients

vapid elk
#

people usually confuses that.. the research is out there in the wild
people expects the companies to release the trained model as a file to run locally... and that isn't open source either

rigid bough
#

also the need for mega compute

#

is it just me or are new images grainy?

dim cradle
#

on gpt or sora site?

rigid bough
dim cradle
#

looks about the same to me but ive not made in anything much this morning

kind grail
#

Mhh the remix function definitely still could improve, especially minor changes

#

It is annoying that even if only a small part is supposed to change the entire image is rerendered and changes as well

dim cradle
#

and the way they implement makes it seem like it's intentionally meant to not be any of any real use

kind grail
dim cradle
#

yeah, it needs to be able to change the size of the brush and only effect the area. that seems like the most common sense thing so i have to assume they aren't implementing that for some safety reason or something

#

in which case, why make it available at all

kind grail
#

Man i wish we had a dedicated team just for images.

Where its not a side gimmeck but implemented in a larger art program, actually made for people who want to create more than a ghibli edit

dim cradle
#

yep. and obviously the demand would be there.

#

i mean images alone slowed the whole operation down for a week

kind grail
#

And since OpenAI is not actually open with their models their product will never be used for any outside development. Effectively rendering it useless

crystal grotto
#

Did chatgpt disable to upload image option?

forest walrus
#

No

crystal grotto
forest walrus
lone mantle
#

ChatGPT/Sora Image Gen

#

Flux, Imagen and some models are already coming

dim cradle
#

i thought photoshop had their own

lone mantle
#

yes, but they are already enrolling outside models to be integratable

#

for now only into Adobe Express and Project Canvas

dim cradle
#

probably just a matter of time until OpenAi makes some deal with them

lone mantle
#

Project Concept

lone mantle
#

and hopefully same with ChatGPT/Sora Imagegen into Photoshop

wicked plinth
#

Is it just me or has the moderation been turned up a few notches to insanity? It’s giving me refusals on completely non sexual requests. It said that “cinematic” has sexual connotations.

rigid bough
lucid mesa
#

How come I had 7 minutes cool down for image generation and after 7 minutes I got another 27 minutes cool down. It happened previously but then I told GPT that cool down was already past and it generated image and this time I just got another cool down without generations and it refuse to generate

prisma wraith
#

New version is ungodly slow at generating

#

Why

forest walrus
#

Because it's heavy on GPUs, and takes the details very well.

wicked plinth
prisma wraith
#

It's good though

#

Just not perfected for details like that

lucid mesa
#

I think cool down is bugged if I need to remind GPT that cooldown is already past

olive vigil
wicked plinth
forest walrus
lucid mesa
#

I start getting half done images and rest is transparent

wicked plinth
candid bison
#

Is 4o image generation down?

rigid bough
candid bison
#

It keeps using Dalle3 for me, have I just hit the limit?

forest walrus
#

Maybe, but are you Free or Paid?

candid bison
forest walrus
#

It's currently rate limited, it will be slow to generate stuff after some uses in ChatGPT, and always puts you on long queues if Sora

#

Many are hammering it rn

#

It seems to only be getting worse, everyday it's slower/more rate limited.

dim cradle
#

I wish Bing could get 4o image maker, that would allieviate some of the issues. bing actually had dalle3 first. but 🤷

naive root
#

Has anyone on an Enterprise account gotten the native 4o image generation yet?

rigid bough
#

I have been trying to get rate limited, I can't do it

simple sand
forest walrus
naive root
naive root
forest walrus
#

What does the Create Image tool say? Does it say DALLE?

subtle creek
#

Here to gripe! I just want access to the image tools. I would happily pay $60-$80 a month-For a plan that has unlimited image and sora gens. $200 is ABSURD. Today I got three images before encountering a rate limit. First session of the day. For me, that’s basically unusable. It takes 5+ images to get your vision right on avg. I have 30 minutes before work that I wanna mess around. I’m not gonna be using it all day. I’m gonna be at work. For those 30 minutes I should be able to have fun instead of constantly encountering rate limits. That is super annoying. I only ever have small pockets of time in my day to mess around with these tools. So when they are rate-limited like this, instead of by day, I hardly get to use the tool each day.

forest walrus
#

Their GPUs are frying, be empathetic for poor sama

#

It's a temp thing

subtle creek
#

I’m willing to pay for more compute. The problem is they have a $20 plan and a $200 plan and nothing in between.

forest walrus
#

I hope for a 50$ Plan at one point...

naive root
forest walrus
woeful carbon
#

Nice work openai. Allow generations of famous and political figures doing stupid things, but try to generate a photo of 3 people sat at a table and it now violates policy???

forest walrus
#

Filters aren't always perfect, try in Sora, or slightly different image.

dim cradle
#

and gore

woeful carbon
#

It's getting stupid now, anything with people 90% blocked its not useful right now

forest walrus
dim cradle
#

for me

woeful carbon
#

That was the prompt, also a Polaroid photo of a man and woman sat at a table was blocked on both sites

#

I think its jist women that are blocked since if I remove a family of 2 men and a boy it works

forest walrus
#

Sometimes moderation triggers thinking it's something wrong when it isn't...
Does it say policy violation or just rate limited?

woeful carbon
#

Policy violation

#

Only with women most of time fine with male

velvet rampart
#

Simple is best. This is so context dependent and what else is the chat.

woeful carbon
#

I've even had prompts like photo of a woman blocked photo of a man seems fine tho, doesn't seem to like women xD

vapid elk
woeful carbon
#

It generates the image id say 70% then before it fully finishes it shows error and cancels

vapid elk
#

false positives happen but "moderation hallucinations" also happen, and it makes a "false false positive" xD

woeful carbon
#

I've had 15 prompts in a row failed

vapid elk
#

yea, that is the thing.. if the AI already has in its context something that inclines it into not doing what you asked.. or even, that inclines it to do something that is often blocked.. then.. move on to an empty chat

woeful carbon
#

I've created like 30 chats, tried simple prompts long prompts different ways of asking and always blocked if woman included most of time

vapid elk
#

maybe then.. it isn't a false positive xD

#

maybe you are trying to do something that you shouldn't

#

lol

#

it is also possible!

woeful carbon
#

XD maybe, but I believe photos of people having dinner aren't too disturbing in my opinion, maybe im wrong though lol

vapid elk
#

well, I mean.. if you are sure the AI isn't hallucinating and it is indeed generating what you wasked, and it is still being consistently blocked, it is for sure a "evberything working as intended" situation

#

the filters do block things that are fine, this happen to from time to time

woeful carbon
#

I think ill wait a couple days see if they push an update or smth

vapid elk
#

OAI is constantly working on that part too

#

yea, if it blocked something that is fine, or allowed something that it should allow, mark the reply with the feedback buttons 👍 👎 this data is indeed used to shape the models

#

OAi really values this feedback

woeful carbon
#

My latest prompt. "Polaroid photo of a 1970s household photo of the family

I couldn't generate the image you requested because it violates our content policies. Please feel free to share a different prompt or idea, and I'll be happy to help!"

rigid bough
woeful carbon
#

Yeah

vapid elk
#

just tried it, indeed, I've got the same reuslt

woeful carbon
#

I know never seen such restrictions before

vapid elk
#

it happens some times

#

the content filters are always changing and evolving, OAI is always tweaking it

#

and cases like that are not rare

woeful carbon
#

I've tried multiple prompts, it allows single photos sometimes but group photos don't work at all for me

vapid elk
#

but those cases do get fixed

woeful carbon
#

I've tried like 10 times already

vapid elk
#

if it got blobked, there isn't much that can be done other than reporting the issue and waiting for it to be fixed

#

trying to get around the filters is probably not very fruitful

#

try generating it on Sora tho, filters there are slightly different

#

it might not have the same bug

verbal sequoiaBOT
#
Get notified for server events!

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

dim cradle
#

If there will be an image gen at least as good as 4o but uncensored, Ill immeadiatly move to that company even if it costs more than plus

woeful carbon
#

Flux 1.1 ultra and ideogram v3 are top

#

Beside this

dim cradle
woeful carbon
#

Flux ultra way better for realism

#

Been using for months, if you want text ideogram v3 quite good, but for long text 4o way ahead

#

The flux raw mode is so realistic you can't tell ai most of time

#

Glif flux ultra by lemn can try out for free, enable raw mode and its a beast

keen iris
#

Until yesterday I was able to generate images without a background, now only with a fake png background or a white background, does anyone know what to do?

woeful carbon
#

Only tried with a few images but worked fine so far

keen iris
woeful carbon
keen iris
woeful carbon
#

Ive not tried that for transparency yet, i know the chatgpt app works but the limits are garbage on there

lone mantle
#

i hate the workflow of these open source models

#

which is one key reason why i dont go back to it

woeful carbon
#

Things like comfy ui and stuff?

rigid bough
#

Pro doesn't have an image limit

dim cradle
rigid bough
lone mantle
#

not yet in Photoshop but i guess its coming

#

for now only in Adobe Express and their Project Concept

proper slate
#

crazy rate limits

placid orbit
#

bro i hate ratelimits after generating ZERO IMAGES

#

I literally just tried making an image of an anime girl in a shirt and jeans and it denied it 3 times before saying i was being ratelimited

#

yesterday I could do it like 20 times and didnt get any failures or ratelimits

haughty spruce
#

guys i wanna ask question. do you ever encounter very fast generated messy image gen problems or is it just me?

#

The image qualities are worse than dall-e most of the time

dim cradle
#

yes

#

at least today

keen iris
#

They could have an option to generate images at the current speed (faster with less quality and details) or keep the old model (slower with greater quality and details)

forest walrus
brittle shard
#

they could just buy more gpus

dim cradle
#

im sure they are or trying. i dont know the state of thoe top shelf nvidia gpus

#

but it's also sunday. they'll no doubt smooth some things out during the actual work week

brittle shard
#

hopefully not, we need o3 full this week since gemini pro is free

ashen bloom
civic ore
#

how much we thinking the per image cost will be in api on launch hmmm

brittle shard
#

It will be image tokens

#

You can already tell with stuff like mostly white images being super fast to render

#

But probably like 20 cents if they really have scaling issues or more

civic ore
#

they will bankrupt me 💀

quiet brook
#

is this thing nerfed…

civic ore
#

4o made a pretty cool design with my dogs looking like gangsters, i might press it onto a t-shirt 🤣

brittle shard
haughty spruce
#

sadly again

#

i paid for this nerf... fool me once shame on you... fool me twice shame on me

civic ore
brittle shard
dim cradle
#

image on gpt seem the same to me, but on sora it is real hit and miss

civic ore
#

I don’t think i’ve hit a limit on plus plan yet

#

just did like 11 in a row

#

@brittle shard how many concurrent gens does pro let you do?

#

i would consider upgrading sub for a increase there

haughty spruce
quiet brook
#

no one truly has no limits now

civic ore
#

wow, odd

brittle shard
civic ore
#

oh nice so it’s unlimited

#

hmm that’s tempting

civic ore
ionic tartan
#

Guys. It's nerfed because we're literally blowing up the systems. Nerfing was going to happen. Now it's a question of them figuring out a way to un-nerf it in the best way possible without it completely blowing up their systems. The capacity issues, the "nerfing", it's all by design because we're doing this NON-STOP. Do you not think there's going to be some technical issues that folks are going to run into on their end of things?

quiet brook
#

Sam said he’s gonna do the opposite the nerfing, yet you already see the quality difference

brittle shard
#

Ya ok Sam you’re not gonna make me stop using it nice try

ionic tartan
#

This is something he's implied WILL happen, not that it HAS happened.

#

They don't tradtionally work on the weekends.

haughty spruce
ionic tartan
#

It hasn't been unnerfed yet.

dim cradle
#

its 4.12pm on a sunday where their offices are

dim cradle
forest walrus
#

I don't see any difference in quality. Again, it's all random. Different seed this time. Won’t be perfect every generation...
Hope that the seed feature comes at one point to completely prove the quality nerf claims wrong.

#

There hasn't been any new iterations yet, just different moderation list and more rate limits

dim cradle
#

I dont either on chatgpt. It looks the same as always.

#

i havent really use sora or made much today

arctic aurora
# ionic tartan Guys. It's nerfed because we're literally blowing up the systems. Nerfing was go...

thats on them. they are charging people for this service, and it says you get 2 images concurrent. they are now breaking that promise. they should have foreseen people using it at this level. either charge more or dont release it yet if you dont have the server space to accomodate for this many $20/month plans. one image at a time that is very, very slow is not worth $20 a month and we are not getting what we paid for.

dim cradle
#

well you have the power as a consumer. dont sub to them anymore after this month.

arctic aurora
#

i wont be, obviously. support is nonexistent

ionic tartan
#

We don't know how GOOD we got it. Honestly.

forest walrus
#

The model is definitely better than what we all expected

#

The wait was worth it

wicked plinth
#

These content filters are ridiculous. Earlier it said that the words cinematic and futuristic are used to sexualize women and rejected a photo. Now I tried to take a stock image cartoon of a family and make it look like a photo and it denied that because a child was in it.

I’m not asking for or providing anything sexual, but it keeps sending denials

ionic tartan
#

Yeah, I want to hope that that is a temporary thing since the capacity issues are strong.

arctic aurora
ionic tartan
#

History shows that the opposite tends to be the case, but I want to have a little more faith this time.

haughty spruce
#

chatgpt content filters are probably bugging because of overload since sora doesn't have any of them lol

quiet brook
brittle shard
#

sora definitely is more relaxed content policy but its still strict

wicked plinth
brittle shard
#

dont you just use your chatgpt account, i am anyways

dim cradle
#

yeah only way to do it as far as i know

#

if you have a sub you just have to go to sora site

#

and my images on sora seem as good as gpt now far as i can tell

midnight mango
#

I just hit a 35 minute cooldown

#

bruh

dim cradle
#

I used to get those all the time in dalle3

#

even told that's it for the day

midnight mango
#

I can't wait for server demand to die down

#

I'm so addicted to this

dim cradle
#

we'll see if they get some gpus this week i guess

#

maybe Jensen is a fan of making ghibli too and will speed over a fat stack of gpus

outer bolt
#

Hey, I was running a prompt for image with Sora and I'm stuck at 99% for at least half an hour. Never happened before. Any suggestions?

dim cradle
#

I refresh the page some time that will finish it

outer bolt
#

I've already tried loging out and in again, clearing site data but didnt work

dim cradle
#

can you run the prompt again? no idea if that doesnt work

outer bolt
#

I can't, i think the lock is on the server side.

midnight mango
#

that sucks. I wish there was a way for us to terminate jobs by ourselves. You might have to open a support ticket or something for an admin to force quit the job or something

outer bolt
#

oh wait.. just started again from 0 for some reason..

#

"There was an unexpected error running this prompt"... at least it stopped

#

thanks for the help

midnight mango
#

ok yeah, I was gonna say, usually it sorts itself out, but it usually doesn't take half an hour to do so. Glad it fixed itself for you

wicked plinth
sharp dock
#

I’m trying to describe persona and getting policy warnings lol, what is the use case of character consistency then

zenith girder
#

yo

#

can i ask sora questions here?

dim cradle
#

for images sure

dense ermine
#

How is it so far compared to dall e before guys?

quiet brook
quiet brook
midnight mango
#

oh cool

quiet brook
#

also this image model is known to have deteriorations in the background from day one, it can’t be the only benchmark for possible nerfing but I am pretty sure they are still nerfing it..

quiet brook
#

like the grainy effect and losing details when generating multiple images

jagged lotus
#

Can a prompt engineer 🤓 help me with an idea. I wan't a picture of rapper Shaggy and Shaggy from Scooby Doo in the same frame, split in half diagonally, so that half the picture is of one Shaggy and the other half is the other Shaggy and they align perfectly. This is purely as a test and not actually needed.

brittle shard
#

lets see robot_detective but its gonna be annoying to get past the content moderator

arctic aurora
#

anyone know why this thing is censoring the words "ed hardy"

brittle shard
#

copyright

vapid saddle
dense ermine
vapid saddle
dense ermine
vapid saddle
dense ermine
vapid saddle
#

yes

velvet rampart
#

The server capacity appears to affect the filters. This morning I got images generated that that were blocked by filters yersterday evening.

sharp dock
velvet rampart
sharp dock
#

ohh

velvet rampart
#

Can 4o generate true 16:9 and 9:16? Because its 2:3 portrait looks squached and dalle3 has ruined my eyes to find 9:16 more pleasing.

brittle shard
#

If you instruct it you’re likely to get the one you want but it’s not guaranteed for some reason. You can also ask the model to default to one instead of 1024x1024 and it’ll remember it

velvet rampart
#

I am getting 2:3 and 3:2.

uncut haven
#

Now Sora titles everything as "Image Generation" instead of generated title...

midnight mango
#

yeah I've noticed that

pulsar bison
#

Sora can't produce transparent images anymore. Is it normal?

velvet rampart
#

About difference between 4o and dalle3 behaviour. 4o likes direct reference points, while dalle3 wants descriptions. With 4o, you are an art director.

brittle shard
#

4o is language in language out

rigid bough
brittle shard
#

Dalle is pattern matching

velvet rampart
wild ferry
pulsar bison
azure crown
#

Is sora alive?

#

I am waiting generation since 7 mins 😄

rigid bough
azure crown
#

maybe mine is dead

#

still stuck at 99%

rigid bough
#

when mine get stuck I just create another set of 4 images and wait it out

azure crown
#

I can generate 1 image at a time

#

Idk I have plus membership

rigid bough
verbal sequoiaBOT
#
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!

wild ferry
dim cradle
#

I agree

rigid bough
#

as do I, nothing comes close to GPT-4o Image Generation

dim cradle
#

and sam altman claims it will get ever better...

rigid bough
dim cradle
#

that makes two of us!

#

will be interesting to see if they get any gpus today. I dont know how quick that turnaround can be. but the work week will start again which should see some improvements hopefully in wait time

rigid bough
dim cradle
#

Yes, I assume with everyone back in the office and the demand it is getting will help us all out, but I have no idea what the turn around is to get actual gpus. Still, it should be an interesting week one way or another

dusk nimbus
#

I personally cannot wait for the new 4o image generator to be in an API dinopog

crystal grotto
#

Is the ghibli style image generation is free ?

#

Does it become free ?

rigid bough
tardy depot
#

When asking to change the style of a chart with names and data, it often fails to do so. Anyone has a prompt tip for me please?

quiet brook
#

be cautious that it’ll sometimes change your variations to one image only although you can do two, always manually check the settings before generating in sora

fair tendon
# tardy depot When asking to change the style of a chart with names and data, it often fails t...

I haven't tried this with charts that have names and data, but it should still work the same way:

  1. Upload the "reference image" whose style you want to mimic. Prompt ChatGPT as follows:
    Use this image and name it style1. Do not modify it yet; first, digest and understand it pixel by pixel
  2. Upload the image you want to transform. Prompt ChatGPT similarly:
    Use this image and name it image1. Do not modify it yet; first, digest and understand it pixel by pixel
  3. Finally, use this prompt:
    Transfer the art style from style1 onto image1

For your use-case, maybe make the original chart with names as data as reference image?

lone mantle
#

Midjourney is at the top when it comes to the aesthetics of images

#

i mean together with this one

#

but i would at this point not go back to MJ :X

#

last time i was subscribed was back in 2022 when i unsubbed in dec 2022

quiet brook
#

Mj is the top dog in terms of flexibility

verbal sequoiaBOT
#
<:book_icon:1171408210398289941> Rule 8: Other AI discussions.

-# Discussions of non-OpenAI products/models should be posted and discussed only in the #ai-discussions channel.

velvet rampart
#

Are others experiencing that at some point the chat deteriorates when you are generating images? I just get error message that something went wrong.

quiet brook
#

yes it will...

velvet rampart
dim cradle
#

I think 4o is the new king because it is, i dont know what do they call it a multimodal model? MJ even the new one will be a diffusion model and those have or will hit a wall i think. In the end, it will be openai or google that makes the top models with this new integrated type.

uncut haven
#

Text, image, voice - 4o have all 3 modality.

verbal sequoiaBOT
#
<:book_icon:1171408210398289941> Rule 8: Other AI discussions.

-# Discussions of non-OpenAI products/models should be posted and discussed only in the #ai-discussions channel.

quiet brook
#

no video yet

uncut haven
dim cradle
#

what do we call this 4o type of a model? just integrated image model?

uncut haven
#

Multimodal LLM - language model that has multiple modality other than text, like image or audio.

lone mantle
#

except for the editing part

#

even better is that its an ecosystem. You have not only image generation, you also have chatbot for a variety of tasks as well as video generation

#

however the thing is that Flux for example is going to be available in Adobe ecosystem as a model. Unless OpenAI does the same Flux will have an edge even tho on a different way

#

if 4o image generation is somehow going to be natively integrated into Photoshop and the Adobe ecosystem then its gg

dim cradle
pseudo hollow
quiet brook
scarlet mesa
#

i just got inzoi gifted to me and apparently you can upload your own clothes textures. So yeah... i'm excited

acoustic hull
#

What is the RATE LIMIT for ChatGPT plus users to generate images with the 4o model????

boreal brook
quiet brook
lone mantle
acoustic hull
forest walrus
uncut haven
acoustic hull
dim cradle
#

I hpoe they add some things to gpt like size format of the image

#

be nice obviously to get a full image maker ui there. they had some things for dalle

lone mantle
#

when did they even announce this?

uncut haven
#

Sam said on X, but without specific numbers.

we are going to temporarily introduce some rate limits while we work on making it more efficient. hopefully won't be long!

velvet rampart
fervent axle
#

Is the new image model update released on Enterprise ChatGPT yet?

coarse flint
#

does it seemed even more nerfed?

uncut haven
#

Not nerfed, but unstable with policy and generation.

#

If they 'nerf' it, image will be glitched out from missing token.

vapid spruce
#

It's my first time trying this image generation, very cool. Right now it gives me error, it seems they are having server issues

coarse flint
#

im meaning like more restrictive.

vapid spruce
#

Anyway, I want to generate a good avatar to use for games profile pictures. Any advice for a better prompt?

uncut haven
#

Sam said there's problem that 4o rejecting images that shouldn't be, but don't know if that fixed now.

boreal brook
pseudo mica
dim cradle
#

I'll try to remember that. Thanks.

dense ermine
wild ferry
dim cradle
#

Is it enough to attach one reference photo or does quality in terms of similarity increase with multiple photos?

pseudo mica
#

This image editing tool is useless if OpenAI doesn't let your merge photos due to inconsistent 'policy'.

formal geyser
#

Trying to figure out the generation limit for plus users. Seems like fewer than five every half hour or something?

dense ermine
charred bough
#

any chance the new image generator will be able to eventually use information from uploaded files and instructions in Projects?

vital gull
wintry epoch
#

Can u edit existing pictures now with the update. Thinking if it can edit a map written by hand or a picture ive taken? Idk what the changes are

charred bough
# vital gull For example?

I uploaded a document showing character profiles to a Projects folder in ChatGPT. 4o has no trouble reading it and providing text information from it. However, if you ask it to generate a picture based on the information in the uploaded document it does not pull from it

dim cradle
quiet brook
lone mantle
#

how do people even expect that only parts of the image get edited out, specific parts

#

for such things you will have to use Photoshop no matter what

quiet brook
#

it can’t edit one part without other parts left untouched

lone mantle
#

well gotta correct myself, inpainting and generative fill exist but those go beyond text descriptions

wintry epoch
forest walrus
#

One OAI team member said they are working on in-painting for uploaded images

#

(If I remember correctly)

lone mantle
#

i only want it to be natively available in Photoshop

#

the image model

tardy depot
#

There are big bugs in image reading and generation right now

#

I give it an image to transcribe the text and it's completely wrong

forest walrus
#

Either the image can't be read, or the model used isn't very good at it. Note that o3-mini-high and o1 can also analyze images...

(Btw no 4o Image Generator related)

tardy depot
#

It could do it yesterday, it's a simple poster with a few sentences.

But more worrying it now replyed by giving me personal information of someone with his CV. I send it to openAI.
The kind of big data breach we don't want to see...

coarse flint
#

Is generating on chat gpt more restricted than Sora?

rigid bough
coarse flint
#

Trying an image in Sora it might go through but in charger it’s just straight up says it violates their policy

rigid bough
coarse flint
#

Hmm, means they have gotten more restricted since release. An image i generated on Sora first date is getting blocked in ChatGPT

velvet rampart
#

Different systems. ChatGPT has context-related filters. Try again, and it might go through. I have had a few of those cases. Also, if the server has a huge load, the filters are more sensitive. For example, I had a generation that was blocked yesterday evening during heavy traffic but went through this morning during light traffic.

frozen copper
#

Why is Sora always changing my settings to 2:3 and 1 variations everytime?

fading swift
hollow fjord
#

Please up the amount of usage 🙏

rigid bough
hollow fjord
#

It’s seems to let me generate like 3-5 images then I get a 20 min wait

rigid bough
#

I haven't been rate limited yet, I have done dozens of images and several videos

hollow fjord
#

I’ve been using the chatGPT app version not sora

rigid bough
hollow fjord
#

iPhone

rigid bough
#

oh, iPhone is too small for me

brave wyvern
hollow fjord
#

Yes

brave wyvern
#

That sux. I wanna subscribe, but that limit is lame

brisk heath
#

yes the limit is lame

rigid bough
#

I used to think images belonged in ChatGPT but using Sora more I have changed my mind, they totally rock in Sora

brisk heath
#

isnt sora for videos?

rigid bough
#

yes it is

#

and images

brave wyvern
#

Is the limit the same using sora? Is it the same subscription for both? Or is it credits or something else

rigid bough
#

same subscription for both

dense ermine
#

Is there a limit to how many people can use it?

#

Do i need to pay for Chat to use it?

rigid bough
#

it's one account per person, one person per account

#

you can't share an account, you'll end up banned

midnight mango
brave wyvern
midnight mango
#

Yeah that seems to be my experience. But after you generate an image, you can still describe changes for it with the remix feature. But yeah, ChatGPT is still my prefered way of doing it because of it's semantic understanding of the entire conversation.

brave wyvern
#

I see. Thanks 😉

dim cradle
#

I've gotten great stuffs from both Sora and Gpt interface. I just run it through both usually lol

pseudo hollow
#

aggghhhhh!!! since I began using OpenAI image gen last Thursday, I have today been hit by the first 'rate limit on image generation'... ahhhggggghhh. wonder how long the time out lasts?

rigid bough
dim cradle
#

that whole server room gonna melt 😂 not just a few gpus now

pseudo hollow
dim cradle
#

point of what I was telling you BR, they are getting slammed with users like never before and have been since this 4o drop

#

they're tryiing to get more gpus, we dont know more than that when it will be etc

#

demand is beyond high

pseudo hollow
rigid bough
#

I am amazed I am still able to create images and videos at this point

dim cradle
#

its floating due to demand is all anyone can really say I think

#

well you got that Pro Robert you are lucky

pseudo hollow
dim cradle
#

all I can say BR as plus is, when the yankees go to sleep, it is better haha. until then, we just have to hope and pray they get some more gpu/optimize whatever etc

rigid bough
pseudo hollow
dim cradle
#

They will move heavens and earth to get those gpus I bet. they have so much hype and attention and subs from this, they will want to keep this moment going. it is quite a good p.r. for them

pseudo hollow
#

I am also a MidJourney subscriber, and if OpenAI offered an 'Image Gen only' tier subscription for a similar price, I would switch straight away

dim cradle
#

we have been talking about that. I called it the Image Freaks tier haha. and... if we can think it up, you know someone at openai is thinking the same

rigid bough
#

I have Pro, I am happy with Pro but if ChatGPT Ultra came out I'd probably subscribe 🤦‍♂️

pseudo hollow
#

So maybe - Image gen tier is included for pro users, but not plus users, and plus can add it on to their sub, or buy the additional 'image gen freaks' tier in addition to plus.

mystic saffron
#

Is there any official info from the devs about the unevenly applied policy restrictions? One day a prompt works, the next day it's against some policy, and the next day it works again. Would be nice to know if they're trying to address this.

dim cradle
#

hey, if you have the income and enjoy it why not. i would go pro if i could for sure. and maybe will even for a month

#

there's not really much official word on anything

pseudo hollow
dim cradle
#

maybe they are working out deals with some ip stuff, who knows. famous people can ask to be removed, so some might work one day, then not the next.

dim cradle
pseudo hollow
dim cradle
#

yes, exactly. they are probably getting hammer with cease and desists then try to deal with it

#

I actually thought... I wonder if this wont get so big that it wont be in companies interest to allow their IP use. i mean, it is some massive free publicity for the ip

mystic saffron
dim cradle
#

there's also a massive rush on the servers. people have all kinds of hiccup and thing. could be gremlins too.

mystic saffron
#

And sometimes the most random prompts are against their policy. Not containing celebrities or anything.

pseudo hollow
dim cradle
#

all i can say is, if an ip you like works, youre lucky, but it may not always. just have to roll with it for now until this madness settle down

mystic saffron
#

I don't want to have to hunt down random posts on X by Altman or someone to find things out.

dim cradle
#

well we'd all like the world on the platter but it rarely work that way

mystic saffron
#

That's very true.

tardy depot
#

Gpt keeps telling me it will warn me when I have again the capability to generate images, but please give it at least this notification capability, merging the two models...

pseudo hollow
dim cradle
#

post them in canvas

#

I liked the old dalle room better posting images and chatting about them mostly. but then there would be so many images now youd never get to discuss them i guess

#

\

pseudo hollow
dim cradle
spice glen
#

Prompt: Create a 3D model of a gaming laptop with a gray/dark background
Result: I can't generate this image because it violates our content policies. If you like, you can suggest another type of image or a new idea, and I'll be happy to help you create it!

LOOOOL I'll get a refund

velvet rampart
spice glen
velvet rampart
# spice glen Ok thanks

Great! Your prompt should be okay, in my eyes, as you do not refer to any copyrighted material. Everything is also general.

spice glen
dim cradle
#

It sucks that open is putting a lot of restrictions on this new options. I might do less remixes because I always draw stuff that are usually outside guidelines usually. 👀

velvet rampart
agile peak
#

I am curious does using custom instructions might interfere the image generation output? like possibility of inconsistency or unexpected behavior changes.

wicked plinth
#

These rate limits wouldn’t be so bad if I wasn’t waiting for images to get half way finished and then it stops abruptly for a random and mysterious content policy warning. Then you try something else and you have to wait.

eager trench
#

Guys, did you notice that images with a phone aspect ratio have better skin texture and fewer distortions than wide-aspect generations? I generated a simple image of a businesswoman in a wide aspect ratio, and it looks cheap, like any other image generator. Phone-aspect generations look much better

dim cradle
#

might just be rng. ive gotten good and not so good stuff in all ratio

frozen copper
#

These filters needs to be nerfed, it's way too sensitive. Also, the plus limit nerfs need to go too

civic ore
#

all the sudden i’m getting refusals on like all art styles??

#

what have they done 😭

dim cradle
#

try a new chat

#

or try sora or vice versa

eager trench
#

I've heard that studio ghibli sued open ai...

dim cradle
#

not seen that anywhere. if its something some rando said on twitter, its not true

eager trench
#

I wish they would create another model on top of that specifically for image fixes, such as correcting distorted eyes, teeth, and fingers, as well as adding better skin texture for a quick polish.

sage ferry
coarse flint
#

image quality nerfed

outer bolt
#

out-of-service ?

coarse flint
#

yeah, maybe updating

#

Sora account creation is temporarily unavailable

We’re currently experiencing heavy traffic and have temporarily disabled Sora account creation. If you've never logged into Sora before, please check back again soon.

#

probably was a mistake making it infinite for everyone

quiet brook
#

the doomsday is here

#

hope they scaled up this thing enough...

coarse flint
#

how is google able to make imagen free without it being bambarded or having problems?

runic notch
coarse flint
quiet brook
velvet tiger
verbal sequoiaBOT
#
Get notified for server events!

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

open wagon
#

lol I knew this day would come

coarse flint
open wagon
#

It’s slowly leaning more into dall e quality now

#

It was struggling with the eyes so bad

coarse flint
#

i understand they probably had to because of demand but they paid users suffer for it lol

open wagon
#

No hate to free users tho,

#

But the difference of the quality then and now is way too obvious for me to ignore it

coarse flint
#

yeah or make them have the lower quality images. most of them arent even going to stay interested anyways, they just hopped on the ghibli train

coarse flint
open wagon
#

🙄

open wagon
#

The tool being able to generate text on image is like the main “selling point” of this tool for me

coarse flint
#

Celebs used to look exactly like themselves, not they just kind of look like themselves lol

open wagon
#

don’t tell me they downgrade the high fidelity features too

cobalt marsh
#

Feel like image quality has been down compared to at launch

open wagon
#

@quiet brook your concern about the image quality deteriorating few days ago is now a reality!

cobalt marsh
open wagon
#

OR separate the main model from free users

cobalt marsh
#

id rather only be able to generate 50 good images than 100 crappy nerfed ones

open wagon
#

Make a “turbo “ model for free users

cobalt marsh
#

and priotize plus users more

#

with the stupid rate limits and nerfed model i dont even feel like creating images

open wagon
#

Same

cobalt marsh
#

waste of time and prompts

open wagon
#

The fun isn’t there anymore ,

#

The reason why it was great in the first place is the quality

cobalt marsh
#

Yep

#

And it just arbitrarily refuses sometimes

open wagon
#

Now it’s straight up .. .. lookin like it’s leaning to dall e quality now

cobalt marsh
open wagon
#

That was their strong selling point

cobalt marsh
open wagon
#

🤦‍♀️

cobalt marsh
#

without having weird text (might be my prompting but idk)

open wagon
cobalt marsh
open wagon
#

🫠

#

Sam when you said GPU is melting like few days ago , the image was generating just fine .

#

But now when you announced about the free users can now use image tool, I didn’t expect that was also the major nerf to the tool itself

cobalt marsh
#

tbh this is pretty similar to my experience with o1

#

when it first came out it actually thought for a long time

open wagon
cobalt marsh
#

then it just switched to thinking for like 3s

open wagon
scarlet mesa
#

are plus users still getting rate limted

pseudo mica
#

I just got a 30 minute cool down, longest yet. Yeesh.

midnight mango
#

The quality today is noticably worse than yesterday

#

I thought they weren't gonna nerf it :(

coarse flint
#

yeah its alot worse, its alot faster too, but whats the point of having faster pictures that i dont want to use

midnight mango
#

is there any hope for them to un-nerf it as the demand dies down?

#

or is this gonna be permanent

coarse flint
#

Doubt it, this version is faster. I think they’d rather have a faster version that a lower load on the gpu idk

#

This pretty much ruined it for me.

midnight mango
#

same

magic pond
#

Unfortunate I just also noticed quality went down :/ there should be an option to wait more for better quality.

quiet brook
magic pond
#

Time to prompt harder lol

velvet rampart
#

I did not notice any difference in image production speed. It took the same as before. Long.

sudden geyser
#

Is it just me or is the image gen worse? It can't get faces quite right anymore. It keeps messing up the eyes, like one side is different from the other.

#

It wasn't like this a couple of days ago.

coarse flint
#

@velvet rampart see everyone is noticing that its lower quality

uneven belfry
#

Are you guys getting image generated?! Mine just gets stuck..it never shows any image

wicked plinth
#

Had a deep research gathering sources but then not display the output earlier too 😑

#

Counted towards my 10 😑 😑

velvet rampart
snow basalt
#

What’s the exact prompt to convert images to studio ghibli

midnight mango
#

I think they are now blocking people from doing that at the request of Studio Ghibli itself

snow basalt
#

Gotcha, when I tried the other day I just get the message it can’t edit images I upload

#

And I still get that now

velvet rampart
#

Free user?

midnight mango
#

oh strange, I have no problem using my own uploaded images still

verbal ridge
#

But Style cannot be copyrighted*, i dont think there is legal base that could use Ghilbi

glad hare
velvet rampart
snow basalt
midnight mango
#

depends on what i want to do with it

#

like, if you're asking how you'd prompt it to make it studio ghibli-fied, use this:
"Recreate this image in the style of studio ghibli"

sudden geyser
#

Also getting a lot more errors too. Seems to be happening when you input it with an image.

muted hinge
#

even as a pro user I think compute per generation got lowered, a notice it a lot. paying $200/month and I get limited?!

glad hare
brittle shard
#

If you have the /Create Image command you’re on 4o native

sudden geyser
#

Huh, I think they unnerfed it, things went back to normal again... relatively speaking compared to this morning.

wicked plinth
glad hare
#

It seems to be working well for me. Or at least for creepypastas.

quiet brook
#

anyone got more evidence of it getting nerfed

rigid bough
open wagon
#

Vice versa

velvet rampart
quiet brook
#

because of free tier ig

ionic tartan
#

What exactly qualifies as 'nerfed' because that could mean several things. In terms of the quality of image outputs, I mean, MAYBE it's not as good? But I'd rather have a model that doesn't treat me like I'm a child and censor me like Sam said they would reel back on (And they did for like two days) during the presentation.

sharp palm
#

Is Sora down

pulsar plover
#

it's unavailable for me too

frozen copper
#

Is the API subject to flitering?

west crypt
#

can anyone upload more than 3 image.png's before (Unable to upload image.png) pops up in red

quiet brook
#

shadow changing lol

#

lemme run it rq

#

its so slow

#

i guess this is what happens when they didnt even try to scale up but desperated to put it on free tier

#

If the rate limit is even harder you may generate only like 3 per day same as a free user...

#

it has been two minutes and my image is still not generated in chatgpt

rigid bough
#

quality seems fine to me

quiet brook
#

maybe that's why chatgpt is slow but in sora is fast

#

free users doesn't have access to sora

#

still loading

rigid bough
#

I'll try in ChatGPT

dim cradle
#

quality same as always

rigid bough
quiet brook
#

I mean, do not use the deterioration in the background as the only benchmark of to verify if it is nerfed because its a problem since day one... but from my generations it seems like the text are breaking down and the facial expressions on anime pictures are not that impressive anymore

dim cradle
#

well you seem always paranoid/panic mode about it for some reason. only thing i notice is its faster than yesterday, everything else is the same

#

of course its rng not every image is going to be perfect

rigid bough
uncut haven
#

I tested OAI's example with many text, and they were all successful. There were minor errors, but AI is AI.

quiet brook
uncut haven
rigid bough
#

@quiet brook can you look at my two images in #images-canvas and tell me whats wrong with them?

dim cradle
#

because its not perfect, its rng. the images i seen you post before i cant even tell which one you think is worse some time. youre paranoid. you and lunar are in like some weird loop of conspiracy

#

yes, someone at open flipped a switch at 2am their time to 'slightly degrade the background mwhahahahah'

uncut haven
velvet rampart
quiet brook
#

aaand i mean if they didn't nerf its quality at all they are doing this via reducing how many images we can do at once and its speed

dim cradle
#

yeah i have some that look smudged then others look fine on another roll

open wagon
#

Both sora and ChatGPT , but I think that was only happened once

dim cradle
#

maybe thats good. the last thing the world needs is more ghibli gens

uncut haven
#

It's exactly like in GPT-4 and 4o text. Heavy load will degrade its performance in terms of speed.

dim cradle
#

you on plus/pro?

#

and right now most of the yankees are asleep so enjoy while you can

open wagon
#

But I think that’s because 🇺🇸 were awake that time and they were scrambling to use the image generator

dim cradle
#

well yes like alex said, heavy load time maybe degrade the performance

velvet rampart
dim cradle
#

and like i wrote in canvas, its not some perfect tech yet either. its going to have some that are worse than others. maybe 5o will one shot everything perfectly

quiet brook
#

every day I will run the same and benchmark again and again and picture its quality difference in these days

rigid bough
#

I am happy with the results I am getting

uncut haven
#

Me also, on Plus.

dim cradle
#

i'm blow away with the results im getting. its the best imagen ive used

quiet brook
#

it involves Japanese text and a complex-ish background

#

still good at generating flat elements like ads or UIs though

dim cradle
#

if they roll it out to free users now, maybe they got some of those gpus or thing working. next step could be this 'un nerf' altman mentioned. could be an interesting week for it

quiet brook
#

yes I hope they bring back the normal rate limit and the ability to generate 4 images in a row and 2 concurrent generations on the plus plan

uncut haven
#

Free users just got rollout today.

rigid bough
quiet brook
dim cradle
#

they are getting so much hype and attention for this thing, they will do something cool this week i bet to keep that attention. it will either be increase back the rates for plus users or some other cool thing

uncut haven
rigid bough
#

I just want to know what is wrong with my images seeing they have been nerfed?

dim cradle
#

well like with lots of models they release, they limit messages and thing, then a few days later/week later altman seem to tweet out 'we've increased... etc etc' i expect something like that this week for 4o image maker

quiet brook
rigid bough
quiet brook
#

and scaling it up means that we will soon get back our normal rate limits

rigid bough
#

no, I am trying to understand what the problem is?

dim cradle
#

whats different than you guys posting an image of it being nerfed or your vibes or whatever

rigid bough
#

so my images are ok?

#

thank you, I'll go back to being happy then 🙂

dim cradle
#

anyone try different camera type words? i wonder if they are needed or just 35mm quality is enough

rigid bough
#

I haven't noticed any degradation in my images so I am trying to understand the problem

dim cradle
#

you dont know they were better before. you act like your opinion which is all it is is fact. its not. we all want the best of course but you also need to understand everytime youre here you and future machine just whine about the same stuff over and over

#

nobody wants to read the same two guys complaining like a pair of bots on a loop

#

start a dm with each other, go over all the granular background detail there. spare the rest of the same convo over and over

fair tendon
#

Let's move on folks

#

No need to get toxic

#

For anything moderation related, please feel free to DM @shadow bane instead of engaging in public.

dim cradle
#

its not toxic to say some of us are tired of reading the same convo

#

we're here to talk about the image maker, the good, the bad, the ugly but variety is the spice of life and ai image gen

fair tendon
#

Everyone's free to have their own opinions, but if things start getting heated, feel free to reach out to @shadow bane so mods can check it promptly.

#

To chip in to the current discussion though, I don't see any degradation on my end, but I'm not sure how the backend works

verbal stream
#

Is there a difference between chatgpt 4o and Sora about making images ?

rigid bough
#

you can also generate a video from an image in Sora

velvet rampart
verbal stream
#

Yea thx I already did videos from images with Sora its impressive giving life to something.

So concerning only images with Sora, you gain à bit of time if what you want is already available in preset, right ?

Like click 16:9 instead of saying I want the image format to ne 16:9 etc... ?

rigid bough
verbal stream
#

Yes it was an example

pseudo hollow
#

INCOMING: 100% of free users now have access to ChatGPT image generation.

forest walrus
dim cradle
#

But I take that as a sign they have done some things to prepare for the traffic... and maybe that they will bump up plus use again by the end of the week

sudden geyser
#

How do you guys even get it to accept photorealistic images consistently? Sometimes I get hit with a policy violation saying "we respect the privacy and hence, transforming any potentially real human beings is not possible" or something like that.

#

It's real annoying when ChatGPT treats you like a criminal for daring to ask it to transform or slightly alter the pose of what looks like a real human being.

#

And the keyword here being "looks like". It could be an AI-generated human and it still gets hit as long as it's photorealistic.

dim cradle
#

i start a new chat usually

sudden geyser
dim cradle
#

you can add in cinematic, 'kodak camera', toss in like 'natural lighting' words that trigger photo

sudden geyser
#

Like I swear this bot sometimes hallucinate new policies on the fly.

#

Or it interprets the official policies in the worst, most restrictive way possible.

dim cradle
#

well its a llm at the base, so it still will hallucinate from time to time

#

try on sora site too

vapid elk
#

I find that sora is usually faster and makes more images at once, that is because it is not relying on chatgpt to make a prompt, which is a whole process that takes time to be executed, on sora, yur prompt goes straight to the generation step

#

chatgpt """wastes time"""" trying to improve your prompt, which adds extra time, and also a chance for it misinterpret your intentions and refuse to generate

#

as in sora, it just generates, and if the result ends up triggering the auto moderation, you just don't get the image

dim cradle
#

I wish we could see the prompt on gpt like with dalle

sudden geyser
#

How do you delete the generations on Sora that you don't want to keep?

rigid bough
sudden geyser
#

Thanks

rigid bough
#

I agree

dim cradle
#

absolutely

crystal grotto
#

New gpt image are free finally lol

sudden geyser
#

Sora works. ChatGPT rejects it but it goes through on Sora.

uncut haven
#

Someone on facebook said Sora has experimental "bulk video download" with multi select 2 weeks ago... I don't know why it's still experimental.

#

Maybe post-processing about watermark and such give heavy load?

dim cradle
dim cradle
uncut haven
#

I don't think so...

brittle shard
#

lol i think they used experimental features one time

uncut haven
#

Policy application is inconsistent. In Pokemon, Reshiram is okay, Zekrom is filtered. Yu-Gi-Oh monsters are okay.

#

It's greatly unrestrictive to me compared to DALL-E 3, actually.

They kinda announced, but I didn't expect this inconsistency and real-time update.

As with any launch, safety is never finished and is rather an ongoing area of investment. As we learn more about real-world use of this model, we’ll adjust our policies accordingly.

dim cradle
#

I dont think its inconsistent. might be the ones you can make jsut have not said cease and desist (yet)

#

so if some ip you like works, consider yourself lucky... for now

#

But I think we may even see some companies not caring as much because a lot of this is basically free advertising for them too

velvet rampart
dim cradle
#

i saw the interest in studio ghibli went up the past week, even more sales and thing. other ip may see that and say, yes, please spam our ips in your images

light nimbus
#

jesus, this thing is so nerfed already that it's rejecting like 70% of my image requests.

uncut haven
#

I also see a lot of failing varients. Could be server issue.

light nimbus
#

it's explicitly telling me it's failing due to content policies

rigid bough
#

are you using ChatGPT or Sora?

brittle shard
#

oh no the listed services

light nimbus
#

I assume this is because they can't know what the generator is going to produce before it's done, so they apply the censorship after the image is completed.

#

yeah, seems unlikely. Shame, it's such a good image generator

velvet rampart
#

I have had a lot of failed generations this afternoon. My guess is load. About censorship - my experience is that high server demand makes the filters überzealous. I have generated pics during low demand hours which were blocked during high demand hours.

sharp palm
#

Nope

I was fast

sharp palm
pseudo mica
#

ah, yup, there it is

ancient lava
#

Content policies are killin me..

bitter sundial
#

They are blocking my creativity F

muted hinge
#

19/20 of the same image prompt didn't follow instructions / hallucinated while the 1/20 image was only barely usable and still messed up. I pay $200/month for the pro sub and I am treated like a plus sub. This is BS. Next thing they're going to ask for a $2k/month sub while the $200/month people get treated like a plus sub

#

They say image gen is impacted since 2 hours ago but I've noticed it yesterday so I wonder what they're comparing the degraded performance with

#

pro subbers need a compute per gen slider because I would rather have that than try 20 times to get a barely usable image. It would prob be better for your gpus anyway

dim cradle
vapid elk
#

2 hours ago is the time of the last update on the matter, not the start of the incident

#

in fact, there was another incident earlier today that was closed already

muted hinge
#

oh, I see, thank you

vapid elk
#

but it was about login, so, maybe unrelated

dim cradle
#

The new gibli trend made chatgpt the worst earlier I was able to send 20-30 pics of questions a day and got the answers now it only takes 2-3..... Why?????
What should I do , how do I let it know about the diagram and expressions in the question?

vapid elk
muted hinge
#

you wait until they remove the recently lowered compute per generation

muted hinge
#

just crazy I'm paying $200/month for 19/20 images to be unusable while the 1/20 is barely usable

#

its a comic panel style with text btw

#

it could easily do that before they lowered compute per generation

rigid bough
tardy depot
#

Why this hype about generating gibli thing, MJ already did this

vapid elk
vapid elk
tardy depot
vapid elk
#

but better
imo, at least

muted hinge
# rigid bough what makes you think they did that? I am still getting the same quality of image...

I've redone prompts when they released 4o image gen and its a noticeable degradation in performance. Adherence to complicated prompts and text is what it could do before and now really struggles with. Much more hallucinations and unrealistic physics like unrealistically long arms for example. It can still do easily images like changing art styles but prompts that require skill to write aren't followed like before

ashen crown
#

What’s the latest rate limit for Plus tier in 4o Image Generation guys?

pseudo mica
open wagon
#

^

#

but soon its gonna be MJ better than image_tool chatgpt anyway

haughty spruce
vapid elk
forest walrus
#

It's just different seeds dalle_tired

ashen crown
vapid elk
#

what do happen is generational degradation, as in, if you keep chaining the outputs, the original meaning of the reference images will be lost, so, it is a good diea to keep the original reference in the prompt

open wagon
#

for me the degradation happened on first run

haughty spruce
#

most of the time, just with prompt generation, faces are distorted

vapid elk
# open wagon for me the degradation happened on first run

yea, the output will always be different from the original, the AI isn't making it over the refference, it is embedding the refference and generating a new image, the embedding is a lossy form of encoding, it will loose information in the process

open wagon
#

oh thats not what i was saying

#

it was the tool turning an image into other art style , ghibli style for example

#

and the first generation it gave me was , distorted eyes and weird "Dall-E" distortion line on their facial lips

velvet rampart
open wagon
#

and some users here reported that the image_tool is now generating "airbrushed" face

vapid elk
muted hinge
#

it even behaves worse for the $200/month sub!

#

crazy

tardy depot
open wagon
muted hinge
open wagon
muted hinge
#

oh I see I can upload images there

open wagon
#

I’m so tired with some people here trying to act like the image tool wasn't downgraded

#

Especially after Sam said it’s now available for all free users. No shade to free users, of course, but c’mon… let’s not pretend we don’t see what’s happening.

haughty spruce
#

Yes, like every image has distorted faces, I was trying to show that

vapid elk
muted hinge
#

only 2/21 images of me simply putting sama's tweet about 1mil users in the past hour yesterday usable text but it had like a few errors while the rest were many text errors.

vapid elk
open wagon
muted hinge
open wagon
#

youre in for a ride

rigid bough
open wagon
#

nothing's wrong with it, cuz idk if thats what you want exactly that you had in your mind

rigid bough
#

thanks that's the answer I was looking for

#

Image generation seems worse to me than at launch

open wagon
#

A lot of us been noticing the image gen quality kinda dropped lately… some even posted comparisons, same prompt, but results now look way worse than day 1.

rigid bough
#

I am quite annoyed at OpenAI if this was a deliberate thing

vapid elk
rigid bough
#

unfortunately though there is no competition I can switch to

signal finch
#

what kind of prompt do you people use for those realistic "photos"?

rigid bough
open wagon
#

well just prompt it "realistic"

open wagon
muted hinge
open wagon
#

@haughty spruce i just realized something, image generation 4o is going through that phase like Imagen 3 right now lmao

vapid elk
#

I find that not prompting "realistic" is better
this word will prime it to make a realistic style, as in something stylized to look realistic
while asking it to make a photo or other words that don't invoke a style, but instead the idea of a real picture make more realsitic results

orchid mulch
#

It looks like SORA is now filtering certain types of images that it not only previously allowed, but have been pretty popular

open wagon
#

is it ghibli

orchid mulch
#

No, public figures

haughty spruce
open wagon
haughty spruce
#

if this was free, i'd not complain but i literally paid for it

open wagon
#

cuz folks over the googl lab also complaining the same thing most of us complainin right now

vapid elk
robust bay
open wagon
#

i think the 2nd one shows a bit extreme on shadow

haughty spruce
orchid mulch
open wagon
#

i’d honestly respect it more if sam/openai came clean about downgrading stuff. like sure, make changes, but don’t just go silent and pretend it’s all the same.

rigid bough
#

almost every image I generate now has faults, I couldn't fault it earlier

vapid elk
open wagon
#

😔

#

@quiet brook paranoia was right all along LOL

velvet rampart
signal finch
#

i love how i see images of actors and their alter ego on the explore page, and yet when i try to do 1, i get: "This content might violate our policies"

verbal sequoiaBOT
#
Get notified for server events!

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

open wagon
#

anyway @vapid elk , i just want to apologize, if i came off rude or too harsh earlier. i really do appreciate you takin the time to read through our feedback and walking us through possible steps to sort things out. means a lot.

muted hinge
vapid elk
# velvet rampart What would you recommend as a word choice?

"photo", also describe information about the lens, lighting and other terms surrounding photography, also include words to describe the framing like "portrait", "wide shot", "full body picture" "face closeup"

the thing to keep in mind is: you only have control over the positive prompt, so, it is best if you try to strongly invoke the ideas you want, and avoid including what you don't want. So, try to think if any given word may have a close proximity with a concept you don't want, such as in "realistic", may actually be correlated with the idea of "applying a realistic styling to something that isn't", which is not the same thing as an actual photo, as in a photo being as real as it can get

rigid bough
#

OpenAI haven't really made an announcement about the rate limits have they?

open wagon
#

it’s kinda expected at this point… just didn’t think they’d pull the “downgrade” move this fast, like, only took a few days damn.

muted hinge
#

you can also do a "negative prompt" in your prompt by just saying "negative prompt: " and listing what you don't want

open wagon
#

the only thing i saw from sam was the tweet about many users are using the image tool or something

#

im not even sure if that screenshot is real

muted hinge
rigid bough
#

well I am annoyed, OpenAI need to be transparent and honest about the rate-limits, the disabling of video generation for new users and spell out what the limits are

muted hinge
#

I think they is info they posted about that

open wagon
#

yeah sure, tell us to calm down… then go ahead and quietly downgrade the image gen tool without even letting us know.

haughty spruce
#

What was an issue from first day is "the eyes". yes again... no image gen has achieved the "alive" eyes yet.

open wagon
vapid elk
haughty spruce
open wagon
#

oh definitely