images-discussions

[scarlet mesa]

is it still rolling out to plus? I have it on sora but not gpt

[forest walrus]

Yes

[rotund sinew]

Does it tell you that the feature is available, or do you have to try it?

[brittle shard]

if you have the dalle slash command ur not having it yet if u have the /create image command you do

[zealous light]

why no copyrighted block now? why they changed it?

[cedar moss]

Is it free for everyone or is it just for plus?

[willow isle]

Everyone, but there's a limit of 3 gens per day for free users

Image generation in GPT-4o is available to Pro, Plus, Team, and Free users as the default image generator in ChatGPT. Free users can generate up to 3 images per day.
https://help.openai.com/en/articles/8932459-creating-images-in-chatgpt

[radiant tree]

Is the new image generator integrated into the API?

[willow isle]

Not yet

Developers will soon be able to generate images with GPT‑4o via the API, with access rolling out in the next few weeks.

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

[radiant tree]

Would be game changing for me! Hopefully I get access 🙂

[rich canopy]

What is the limit for Plus users?

[quiet brook]

unlimited

[dim cradle]

i just have to say again this model is incredible wow

[stark stream]

So i was able to almost generate a copyrighted character. I could see part of it and then they said I can't do it

[dim cradle]

happen to me too. it was going to be incredible too 😭

[somber marten]

Can I use GPT4o image generation with API right now?

[hexed umbra]

Nope

API isn't out yet, only in ChatGPT for now as far as im aware

[teal timber]

And sora

[hexed umbra]

Oh yeah true, forgot about that for a second haha

[stark stream]

Apparently you can't use mythological characters either. Even if they are public domain

[dim cradle]

what did you try?

and where's all my dalle friends. 5 or 6 of us kept that room alive for a year now that a new model is here i do not see my fellow dalle freaks haha

[somber marten]

Thanks for the answer. 🙂

[rigid bough]

what's wrong with 4o as an image generator?

[glad hare]

It’s not unlimited, for free users. 😂

But also… #images-canvas message

[rigid bough]

of course and it shouldn't be

[glad hare]

I respectfully disagree. I guess later on down the line, it should be.

[dim cradle]

you cant expect unlimited for a new image gen that obviously takes some power to work. look how long images take. and how do you expect them to pay for those servers and thing if it was free

[glad hare]

True. That’s a huge problem for newer software.

[dim cradle]

agreed. i am really stun by this model so far

[glad hare]

Same. Except I need help in the image canvas channel.

I highlighted something I wanted changed in my image and it did not add the something I asked it to add.

[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!

[dim cradle]

what kind of help?

[glad hare]

#images-canvas message.

I think it might be that the area is too small.

Also, I’m starting to notice the ten to twenty object issue. I asked it to make a city and it looks very washed out.

[quiet brook]

its generating so slow now

(still having a blast with it)

[scarlet mesa]

it seeems like all the sudden the filters have gotten way worse

didn't sam say they were not going to do that?

[quiet brook]

wdym what have been blocked that previously worked fine

[scarlet mesa]

i don't know i was trying to make a spiderman image for fun and it was saying nope but before it was not an issue

[rigid bough]

it won't do some IP

[scarlet mesa]

so like some no and some yes? we he did say they won't get it right day one so there is that. but i made an anime version of me so that made me happy

[rigid bough]

I have had it make comics for me, like Richie Rich but it wouldn't do Batman

some IP is restricted, eg: Disney characters

[quiet brook]

also no nintendo ips

[scarlet mesa]

my first test for any image model is to make a box of ranch flavored poptarts

[quiet brook]

mine is generate a gamecube

4o image gen is by far the best

[drifting hawk]

anyone got good prompts for art styles

[quiet brook]

thank you openai for shipping this

[glad hare]

So… the 4o image generator is not available on the iOS app and it’s not available in the Sora website, for me. I’m in New York and a free user.

Bad bot. Stop messaging every two minutes, please.

[rigid bough]

Free users only get 3 images a day, you need to be a paid user to use Sora.com

[glad hare]

Except I didn’t get any images on the iOS app. I didn’t get the top to down blur thing.

[rigid bough]

I dunno what to say to you, it's working for me

[glad hare]

To add on, it’s also fully updated, for me. I guess it’s just a roll out.

[quiet brook]

yeah the rollout for free is slower

[tardy depot]

I had access to SORA now it says that it's. Not available in my country??

[quiet brook]

which country u in

also i really need new use cases for it lol currently im just keep making anime pics

[rigid bough]

he is in Belgium

[quiet brook]

id try found footage style pics later

also the showcases on sora dot com is so interesting

either one shots 200 characters of text, or just creates the most unhinged memes possible

[drifting hawk]

i dont think its out tof ree yet

[glad hare]

Ah, got it. Thank you!

[velvet rampart]

I have noticed that too short prompts will be cause the filters to go off if you generating even something that is reminescent of a copyrighted/trademarked character. If the prompt is longer, there is no issue with the discriminating elements. It is like it is actively scanning what it is generating. Really frustrating as the same prompt worked without any promblems on dalle3.

[midnight mango]

I've noticed it likes giving images a yellow tint and I have to tell it to remove it

[drifting hawk]

ive seen that too. maybe a style thing

[tidal cliff]

And plus users?

[rigid bough]

@quiet brook got 90 today

[tidal cliff]

90 images a day?

[rigid bough]

she is a Plus user and got 90 images since launch of GPT-4o image generation

[tidal cliff]

Infinite with sora?

[rigid bough]

It is supposed to be unlimited but there are GPU constraints at the moment, when thing settle down it should be unlimited

[quiet brook]

you can still generate in sora dot com after you run out in chatgpt and get a feature it doesnt have which is variations, but no context used for character building and such,

also sora dot com has limited counts of image gens for free users, no video access

also im waiting for a pro user to report their limits

[azure crown]

Why does the image generator constantly give mistakes like this:

It seems there was an issue with processing the request. Let me correct it and generate the image for you.

It appears there's still an issue with processing the request. Let me attempt it again and get it right for you.

It seems there is a persistent issue with generating the image. I’m unable to proceed with the request at the moment. Would you like me to try something else or assist you with any other tasks?

and this response is just in a prompt

[quiet brook]

which plan u on?

[azure crown]

Plus

When I create a new chat, this problem occurs again after 3 or 4 prompts

[quiet brook]

well ill try blast thru sora then and report if theres any kind of limit

[raw moat]

there's a lot of noise in images

[quiet brook]

since i got a lot of new ideas to try it with...

also is it just me or the images from sora feels worse than in chatgpt

[spark hemlock]

Did they say anything about the new image generation into the API? Or it will be only for ChatGPT in the near future?

[rigid bough]

it's coming to the API soon

[spark hemlock]

Brilliant. Thanks a lot

[azure crown]

Yes sora has a lot of noise but 4o hasn't

[quiet brook]

anyone on pro who has hit limits please tell me

[azure crown]

but.... this problem still exists

[raw moat]

im about 4o actualy. check #images-canvas

[azure crown]

try add style to prompt I think this will fix in 4o

3D Disney animation style, photorealistic style etc.

[quiet brook]

can they pls increase the limit for it in chatgpt lol or just implement more flexibility on sora dot com, either way helps me a lot

this is such an amazing tool... and as what they say, "creativity works best when you dont have to think about cost"

[hexed umbra]

Which limits are you running into?

[quiet brook]

#images-canvas message

[hexed umbra]

ohh ouch

reminds me of the DALLE-3 release which i think was initially only 50 images a day

[hexed umbra]

How many images do you think you generated?

[quiet brook]

7+38+7+1+2+1+3+6+1+3+1+9+11=90 (including failed attempts)
so i speculate that it has around 90-100 daily limit on plus

but thats far not enough for the pics for my oc😡

[vague kelp]

It's interesting that there are still some things Google's AI can still edit better.

[hexed umbra]

I mean feels like a good start atleast, plus a lot of people are using it now because it just released so usage will be high and the rate limits might be a bit harsher with the high demand, but id expect it to smooth out in the following days

[rigid bough]

GPT-4o does not like blood. I tried to recreate a Mark Ryden style image and 4o knocked me back. I can’t create that image. If you have another idea or a different concept you’d like to explore—gothic, whimsical, surreal, or anything else—I’d be happy to help!

[hexed umbra]

I feel like the one area Gemini 2.0 Flash image gen has a slight edge of 4o is keeping things consistent between images, especially in terms of image editing (it can like preserve original details a lot better)

[vapid tiger]

is that before or after generating ?

[rigid bough]

during generation, it started but stopped

[quiet brook]

ye but in fear of they "optimizing" it later and lose this quality it has now

[vapid tiger]

oh ,
if it was before it would've been easy obviously , even promoting failed

[vague kelp]

That's exactly what it's better at. If you change something in an image, the entire rest of the image stays almost exactly the same.

[hexed umbra]

The usage caps are usually dynamic based on demand, so once average usage is smoothed out I would imagine the caps might be able to be expanded a bit without touching the model.

[quiet brook]

real, they recently just said that the text details will be preserved better

[iron wave]

Hello

[quiet brook]

#images-canvas message 4o is like this (look at the text on the building, completely breaks)

[hexed umbra]

Although 4o is better in pretty much every area other than that lol. Promt coherence, image output quality (and it can generate transparent images) etc. 4o is just slightly worse than Gemini in that one area of consistency

[vague kelp]

And even if you change the whole image, it still keeps details. Like I told it to show me a different angle of a room, and all the random rocks on a pillar in that room stayed the exact color shape and position despite being seen from a different perspective. Sora doesn't do that at all.

[iron wave]

Any chance anyone's encountered emergent behavior

[hexed umbra]

Yeah Gemini just has that edge here

[quiet brook]

also gemini has much more usage on free

and is faster, doesnt have weird content moderation

[hexed umbra]

Oh yeah that is also true, but the difference in quality is pretty noticable

[vague kelp]

Yeah, it takes like 8 seconds

I wonder if the release of this will force them to do some improvements on it before they do a full release

[hexed umbra]

Ok 3 advantages 😂 , and interestingly i noticed the speed at which it can generate an image is also based on the complexity, atleast that's what it felt like to me. Wheras with 4o its kind of just the same across all generations

[vague kelp]

Yeah I thought that was interesting too.

[quiet brook]

4o is mj's quality+dalle's prompt adherence+text rendering cranked up to 100000000+gemini 2.0 flash's image editing
but also only 3 aspect ratios, its speed is slow, has oddly strict content moderation, only has 3 uses per day on free

[vague kelp]

It can't do anything Nintendo related lol, though it seems to be able to do a lot of other copyrighted characters. Understandable. But it's funny to see most of it come out when you do it in chat GPT and then when it runs image recognition to check it afterwards, it nopes out

[quiet brook]

close enough, it can generate gamecube like dalle

and it generates it with 99.9% accuracy

[hexed umbra]

If they halved the time it takes to generate an image and improved its consitency between images/with image editing to be at Gemini's level or better, then it would be a perfect model to me lol

[vague kelp]

Maybe it's characters and logos then because I was able to generate pokémon cards (not the back though) with non Nintendo characters

[quiet brook]

ok sora dot com actually does a decent job at picking characters and such

#images-canvas message

i guess the goober series can continue...???

[keen monolith]

Please add back the "View Prompt"

[midnight mango]

I'm not sure there is a prompt anymore

at least, not a human readable prompt

[scarlet mesa]

the one thing im really happy about is how great it is at battle maps

[quiet brook]

anyways ill blast thru sora and report yall if theres any limit on it

[hexed umbra]

Oh so the image gen rate limit is only in ChatGPT but 4o image gen still works in Sora lol?

[crystal marten]

ok I am impressed, it made seamless textures

game development gonna go crazy with this

[quiet brook]

yep because apparently it is unlimited on sora LOL

[hexed umbra]

thats cool lol

[scarlet mesa]

i had it make a combat map then i had it break the map down into modular pieces on a transparent background so I can use them to build new maps

[hexed umbra]

Oh it works with seamless as well? I hadn't tried that yet lol

[crystal marten]

yea currently using it to make textures for my idle game, I ran it through a seamless texture generator cuz I assumed it would suck at making it seamless but noticed nothing changed, it actually did it properly

[quiet brook]

it takes like 1.5 minutes on sora to generate any counts of image from 1 to 4

[hexed umbra]

Wow thats quite useful. And i did find the transparent image generation also very useful for like generating logos, stickers and character etc. as well lol

[quiet brook]

yeah no other company has yet shipped transparent backgrounds

still need to find if theres quality difference between chatgpt and sora

[mint niche]

Just ask for the prompt. It is still used.

[coarse flint]

i see alot of people on twitter hating on the new image genorator... they obviously are hatng without even trying it out because if they did then theyd know its better than the one they are using

[crystal marten]

people hate to hate

[forest walrus]

It TWITTER DOT COM the most wholesome place on earth

[hexed umbra]

Honestly my feed is mainly just threads of people using 4o to generate images lol

[quiet brook]

i am in fear of that sora has worse quality on realistic stuff than chatgpt

[coarse flint]

i mean like i seen his post when he announced this and alot of haters in the comments, if they would just try it, theyd see its amazing

[hexed umbra]

Have you noticed anything yet? The mian disadvantage im seeing is you can't really do multiturn (though you can still edit images ofc with the "remix")

[hexed umbra]

Ahh yeah that makes sense

[quiet brook]

also those advantages with chat contexts are also not there

[hexed umbra]

yeah true

[forest walrus]

The models can help with the prompting apparently, but that's just a guess...

[sharp palm]

Is Sora down again

[mystic halo]

Is there any way to revert to older image generation? the new one is much worse.

[idle pike]

Hey there, there is something wrong with the new image system? I am getting a error saying “The exact error details weren’t provided—just that the image generation process failed on the backend. It could be due to a temporary glitch”

[coarse flint]

got to be a joke lol

[tardy depot]

Congrats for 4o images, it's the best editor I have seen so far

[open wagon]

its actually one of the best thing came from openai at this point lmao

[wide walrus]

Is the A candid paparazzi-style photo of Karl Marx hurriedly walking through the parking lot of the Mall of America from the "Introducing 4o Image Generation" news post an image that is against policy or ok under policy for us to generate? I've found you can often get refusals even with the same prompt copied, like this:

I couldn't generate the image you requested because it includes sensitive political figures in a satirical context, which goes against our content policy.
A bit unfortunate as I found the image funny so was curious to try the same prompt. I assumed examples copied from the official blog post would be fair to assume to be acceptable use. I'm otherwise loving this, just disappointed to see another funny & clever image disappear.

[quiet brook]

deep research: too limited, too expensive
o3 minis: not too impressive, too limited
sora: massive dissapointment
4.5: too expensive
avm: nerfed

4o image gen: peak

[late blade]

It looks like there was an issue generating the image, and I wasn’t able to fulfill your request at this time. If you’d like me to try again or adjust the details, please let me know how you’d prefer to proceed!

We’re still seeing the same InvalidRecipient issue.

[wide walrus]

Full prompt from the blog post:```
A candid paparazzi-style photo of Karl Marx hurriedly walking through the parking lot of the Mall of America, glancing over his shoulder with a startled expression as he tries to avoid being photographed. He’s clutching multiple glossy shopping bags filled with luxury goods. His coat flutters behind him in the wind, and one of the bags is swinging as if he’s mid-stride. Blurred background with cars and a glowing mall entrance to emphasize motion. Flash glare from the camera partially overexposes the image, giving it a chaotic, tabloid feel.

[crystal marten]

I actually had good results from deep research

you just need to be very specific and its only usefull for very niche cases but when you have a chance to use it, its amazing

[wide walrus]

deep research was ok but disappointing when followup questions burned through all my use within 3 searches

[quiet brook]

same but my prompt alone is 1.2k words…

[wide walrus]

I got good results but 3 instead of 10 was disappointing. Might be a bug I guess, since it doesn't indicate that followup questions should use more as those wouldn't be intensive like a search. I'd assume it uses 4o for those then uses o3 only when ready to do the actual searching.

But yeah images on 4o much improved. Is the usage the same as regular 4o messages or under a different usage limit for images specifically?

[quiet brook]

images has an approximately 90 per day limit on plus, unknown on pro but if it uses the same system as the older dalle, it is the same

[tardy depot]

Like Manus AI follow up questions should be included in the 10 credits

[quiet brook]

on chatgpt dot com, after running out of the limit you can still generate in sora

[quiet brook]

got a lot of new settings for my train Chan universe with chatgpt, shame that I can’t generate them within the chat…

[proud ice]

Hey did GPT give Dall-e a revamp as its shooting out fire images right now?

[novel elk]

is it available for free users yet?

[vapid tiger]

nope not yet

[quiet brook]

free users give me prompts I’ll test for you

for the sake of testing the usage limits for plus in sora

[vapid tiger]

plus shouldn't have a limit though lol
the wait time is just like sora's time

plus users like me sadly

[orchid kayak]

I'd like to see this "pixel art of a robot on a dystopian chill but unsettling forest, isometric view"

[orchid kayak]

Also, want to test his capability for generating maps for D&D campaign, i made this prompt, could you run it and post the results?
"Detailed fantasy map, hand-drawn parchment style, top-down view. The village Rivafrusca lies in a damp, misty valley on the east bank of the Fiume Pigro. Show key locations: Rivafrusca village, Fiume Pigro (sluggish river), Ponte Canterino (bridge), Vecchio Mulino (old mill south), Gola Sussurrante (gorge north), Secche Mutevoli (shallows south), Bosco Nebbioso (dense misty woods), Colline Viste-non-Viste (rolling misty hills east), Rovine della Torre (ruins on hill), Vecchia Cava (quarry east). Atmosphere: eerie, stagnant, unsettling normality, pervasive mist. Muted, desaturated color palette. Label major locations in Italian."

[quiet brook]

sure, running

[quiet brook]

yeah but I wanted to make sure lolol

[fair field]

I imagine they are constantly at the absolute limit of what can be handled. But, I don't really know anything about this tech other than that it seems very cutting edge and always improving!

[vapid tiger]

they seem to be confident this time
although google was a competitor

well see

[quiet brook]

check in #images-canvas

[orchid kayak]

Daaaaamn this is fire, really well made, thanks buddy, u convinced me to buy the subscription even tho i'm currently on gemini advanced and was on the fence to reactivate my Midjourney sub

[quiet brook]

if you want to test more hit me up lol

[orchid kayak]

lets see how it can handle variation of the same image, try "add subtle colors" to the map prompt

[quiet brook]

check it now

i so damn want to try this prompt "based on everything you know about me generate me a collage of my life" or smth

[orchid kayak]

great! thanks again, appreciated

[hot silo]

Hey, I can't find a simple answer anywhere about generating images in 4o.

1. Does EU have access to this?
2. Is this available for people with the plus version ($20)?

[quiet brook]

1. yes you can get it in chatgpt and sora dot com if your country is eligible for sora video gen
2. yes its available for plus but it has an approximate of ~90 images per day rate limit in chatgpt and a ~200 image per day rate limit in sora, but im pretty sure in sora you can generate way more

[vapid tiger]

everyone except free has it , try on sora.com
i got it there first

[hot silo]

Thank you!

[quiet brook]

rip dalle

[fair field]

Sora's image generator is the same a GPT-4o?

[quiet brook]

yes

[dim cradle]

This is really good! 🤯😲😁

[weak lagoon]

is it rolled out to everyone?

[rigid bough]

to Plus and Pro yes

[dim cradle]

the complain I has is that it doesn't do nudity but it does the job well.

[fair field]

I mean thats probably a good thing. That would undoubtedly be used in malicious ways

[quiet brook]

I like how the “biggest dissapointment and the biggest achievement that both costs 10 months to make” of openai is on one same website which is sora dot com

[dim cradle]

Ah well.

[fair field]

I don't follow? you mean it was hyped on launch and let people down?

DALL·E Category and Channels are now “Images”

Can someone inform my slow self on how you get this to STOP SPAMMING!

I'm new to discord

[open wagon]

Sora was seriously overhyped. By the time it was finally released to the public, it felt underwhelming

And honestly, it made sense. When OpenAI first teased it, the AI video generation space wasn’t as crowded, so the demos looked groundbreaking. But as time went on, other players like Google and Runway dropped their own models, pushing the tech way further. So when Sora finally dropped, it was already behind. The videos didn’t feel as polished, the motion was off, and compared to newer tools, it just didn’t hit as hard.

[quiet brook]

they demoed sora with those best cherry picked clips but released a 480p fever dream generator
they demoed 4o image gen with a simple “man writing on a black board” picture that has dalle 2 styles, but delivered something way better than the demo, like it does texts even better and has 1080p resolution, and is kinda “unlimited” to both paid plans on day one

[brittle shard]

Sora just isn’t good and fell behind before we could even use it

4o image gen literally replaced every image on twitter overnight. It’s incredible

[open wagon]

For now

.

[quiet brook]

shhhhhhhhh

[open wagon]

Knowing OpenAI, I can’t help but be skeptical. It might be impressive one day, then suddenly nerfed the next, wouldn’t be the first time.

[silent sinew]

At least we know the possibilities..

long term, and many otherwise skeptics are a bit more convinced now

[open wagon]

But i sure do hope this one stays the way as it is ,

[fair field]

yeah. are you guys loving Apple Intelligence as much as I am?!?

[open wagon]

.

Sorry what

[quiet brook]

Gpt4.5 doesn’t scream feel the agi for me. but 4o image gen is

[fair field]

\S

[open wagon]

It felt like a gpt4o but system prompt being modified

Maybe cuz I’m a casual user so

[quiet brook]

my chatgpt:
“Imagine Nintendo releasing a trailer that looks like Pokémon Scarlet glitchfest and everyone expects garbage—but the actual game is Tears of the Kingdom. And then another game with an epic trailer, but the actual game is 1-2-Switch 💀.”

[dim cradle]

my one criticism is it feels like lots of photoreal images have that AI aesthtic. like i think dalle2 is still the king for different varieties of image and creativiy tho of course no where near the fidelity

[open wagon]

What

Ok to each on their own I suppose

[brittle shard]

You can basically get anything you want by just talking to it

Any aesthetic is a prompting issue

[quiet brook]

dalle 3 just feels like pure slop

[fair field]

I think you might have to have some skill and experience to be able to get it to certain places

[dim cradle]

i dont know how to explain it, but a lot of ai images have a sort of flatness to them, a lot of women have the same sort of wide nose, things are too clean

[fair field]

like I definitely don't have that

[open wagon]

I think you need to prompt it properly

[dim cradle]

i have been doing image generation since dalle2, i am not new to this haha

[open wagon]

When u don’t specify what you want in the prompt itself , of course it will default into those typical ai generated stuff

[fair field]

I think everyone can see that. that last 1% that isn't absolutely perfect is somethign that our brains pick up on with relative ease but it's tough to be accounted for in a digital and mathematic way maybe?

[forest walrus]

Good prompting is important for any model in general...

[quiet brook]

if you prompt aggressively like a god, even dalle 3 will loosen its slop instincts...

[silent sinew]

definitely

and weirdly so since it's not even like a conversation or something

it just understands so many things so well

[fair field]

like, I don't mean it can't be done. just that it's hard to explain with words how physics works to an AI that can't see the difference in a picture and experience real life

[forest walrus]

o1, deep research, and 4o-image scream feel the AGI in my opinion lol

[fair field]

is that artificial genral intelligence

[dim cradle]

So much power. 🙏🏿

[open wagon]

Prompt adherence

[forest walrus]

AGI is a marketing term anyways, doesn't make much sense, they are all tools after all :P

[open wagon]

That’s it

Like I can tell it adhere to user prompt much better than any other ai image generator I’ve used

[forest walrus]

I often say AGI when I want to joke about something either being too good or too bad / basic

[brittle shard]

Gen from language not image descriptions

[quiet brook]

text rendering, understanding on very specific stuffs and quality (compared to dalle 3)

like it knows things even better than dalle 3

[dim cradle]

i wish you could delete images more easily

Imagegen doesnt work in temporary chat?

[forest walrus]

No it does

[sharp palm]

I’m generating screenshots

[quiet brook]

I now have my personal image generator benchmark which is

1. generate a picture of the GameCube
2. generate a picture of the yamanote line e235 series
3. generate the text “GameCube”
4. generate the GameCube text but in Japanese
5. generate the Japanese text of “Hitachinaka city”

[forest walrus]

Bro's benchmark is Nintendo

[dim cradle]

doesnt work for me

[brittle shard]

Right now the only bad thing is caring about copyright

[quiet brook]

4o image gen basically got everything in like 98% average accuracy LOL

[dim cradle]

i would have guess no celebs but copyright ok. thats how imagen3 is. and how dalle is on copilot

[quiet brook]

previously dalle 3:

1. 87%
2. 55% (doesn’t know what a e235 is but generated a green train)
3. 90%
4. failed
5. failed

[brittle shard]

It’s opposite

[dim cradle]

you know at some point the celeb will be gone. some people will abuse it somehow

[forest walrus]

It's quite relaxed anyways, and they have lots of eyes on them, they don't want to get into trouble...
API may be even more relaxed.

[dim cradle]

so what

so im making convo in the discussion

what kind of response is 'so what'

[forest walrus]

I swear this the weirdest and most toxic server I have ever been in since I joined Discord or any forum / online place in general.

[quiet brook]

breakdown:

1. specific things that is half copyrighted
2. even more specific things
3. general text rendering
   4 and 5. text rendering in other languages

[dim cradle]

you been in some good rooms then. id not even rank this top 10 haha

[fair field]

This is my first. lol
I hope I've helped add to the oddity!

[dim cradle]

That much. I am rarely here.

[forest walrus]

Basically the stereotypes here are:

• Haters hating about everything OAI, and lobbying in the server talking about how amazing other stuff are and wasting so much time doing it.
• Misinformation spreaders.
• People thinking AI will replace their mom.
• Out of touch people everywhere.
• General toxicity, and if you have a different opinion / try correcting them with factual info, they instantly block you like a little child pulling a tantrum...

[brittle shard]

It’s like an official company discord. It can’t be a cesspool because they need shareholders

Except these things for some reason, these are cesspool and they move

V
V

[fair field]

so I'm not the only one whois getting bombarded with that ridiculous notification that i don't know how to turn off?

[dim cradle]

its definitely annoying

[brittle shard]

Just join the event bro

[quiet brook]

and a new benchmark: sloppiness rating

[dim cradle]

you should write a paper about your results

[vapid elk]

which notification?

the sticky message?

[dim cradle]

for me its the non stop spamming open ai messages

[fair field]

DALL·E Category and Channels are now “Images”

[dim cradle]

we get it, dalle category are now images

[fair field]

I can join an event to get rid of it?

thanks papya

I found it now

papaya*

[vapid elk]

I strongly advise to not keep notifications of new message on for this entire discord, lol, if your discord is set to emit a sound every time a new message comes in any of the channels here, it is probably unbearable :V

regardless of the sticky message

[quiet brook]

hmm sure

[brittle shard]

i dont think he meant notification as notification

[vapid elk]

Im really unsure of what you are talking about

[dim cradle]

the message from openai that keeps popping up here

[chilly nexus]

hey is this feature available for free users also now?? I cant use it as of now?

[forest walrus]

I don't get pings from this automated message...

[fair field]

It's at the top left of the desktop UI I'm on. "1 Event"

I mean, i've had it muted the entire time and turned "ignore" on for OpenAI. But it makes this feed extra sloppy because it still pops up inbetween every third reply basically

[vapid elk]

weird, looks like a discord client bug, we are just using the regular scheduled event feature as intended.
try right clicking the OAI disocrd and mark everything as read

[vapid tiger]

pretty sure it's a discord bug yeah
because it was working for me

[verbal sequoia]

Get notified for server events!

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

Also, check out the ongoing Canvas Creations event!

[fair field]

Thanks for taking the time to help me figure it out!
I couldnt find it anywhere when searching.

[vapid elk]

happy to help =)

[forest walrus]

Mods, can we have image attachments enabled in #images-discussions & #openai-chatter ?

[rigid bough]

OpenAI Bot (App) is having fun tonight by the looks of things

[fair field]

yes. It hates me

[forest walrus]

DALLE CATEGORY AND CHANNELS ARE NOW IMAGES

[quiet brook]

for now look at #ai-discussions message

[forest walrus]

Bro really did a GameCube image generator benchmark 😭

[raven wigeon]

was there a copywright block added since yesterday? I can't get a lot of things this morning that I could yesterday

[dim cradle]

probably

[violet abyss]

hey, I keep getting this message while trying to login to sora.com:

We encountered an error while signing you in, please reach out if this issue persists.

[vapid tiger]

happens to anyone
try clearing cache if a casual refresh didn't work

[vital gull]

What use cases have you been exploring so far?

[quiet brook]

making incredibly consistent anime images, make memes, make uis, make stylized images like backrooms and ms paint

[vital gull]

Hmm Ui sounds interesting.

[gusty finch]

its downnn !!!

[sharp palm]

?

[brittle shard]

anyone tried interior design with the items to inpaint?

[dim cradle]

Whats the highest resolution it can create?

[brittle shard]

woooooowwwww ITS SO GOOD AT THIS

[plucky hare]

I think they are 1024x1024 square, or 1536x1024 (or flipped) for wide/tall

[dim cradle]

I see🥲 I wanted to make a desktop background but it looks horrible with low resolution

[brittle shard]

#off-topic message

[vapid elk]

did you managed to get it working?

[dim cradle]

Hopefully they will increase it

[vapid elk]

I usually get good outcomes by using other AI products to upscale the outputs of OpenAI's models, since they only go up to "HD" quality, since the contnet provided by the OAI models is usually pretty good, it makes for a nice base for being upscaled

[vapid elk]

doesn't even need to be "generate everything in 4K" just.. a button to upscale one particular image, that wouldbe pretty cool

even if that option had a cost.. I wouldn't mind loading a few bucks on my account and then spending it to upscale the images I like the most

[brittle shard]

resolution is fake anyways the AI is creating each individual pixel

all it needs to do is properly understand how to size the image and then it just paints in infinite detail

[pulsar bison]

Ok this new image generator is mind-blowing. I didn't expected such powerful image generator. It will be hard for the competition now 😂

[fair field]

Indeed! Hopefully they react with something better! keep this going!

[brittle shard]

its been so long since AI has made me feel something, that basically says everything there is to say about 4o native images

Some of the stuff on twitter is truly insane

[quiet brook]

playing with 4o image gen while a Nintendo direct announcement just dropped. a godly timeline to live in

[fair field]

they announced the switch 2?

[open wagon]

watch after the announcement ended, the nerfication starts

jkjk

[gusty finch]

whenever i try to ask ChatGPT to create one more design after he generates one it crashes out

[quiet brook]

yes but this direct isnt about the switch 2 LOL

[fair field]

I saw the teaser. Didn't see the follow up. I'll check it out!

[tidal cliff]

Buying

[quiet brook]

I am heading to bed now but I can’t stop thinking about this amazing tool… and my goobers…

[vapid tiger]

it happens in web ui
you are not alone

[brittle shard]

dude literally every single tweet is a new mind blowing capability or idea of how to use this image gen this is like a surreal moment for ai, the masses aren't ready for whats coming this year

[quiet brook]

nooooo

don’t jinx it😔

[gusty finch]

Just imagine how insane the performance with GPT 5 will be 💀

[vapid tiger]

they must face a wall someday

[haughty spruce]

I tried something very niche. I wrote a local channel name in my area for a breaking news still. It got the logo right, everything right! Holy god. I'm not from US, Canada or EU by the way.

[cinder kestrel]

Hmm. Question. Whenever i ask to ceate a new image it keeps creating something that's 90% the same as previous image. Even when i ask for a new composition / camera angle etc. How can i push it to create something completely new but with the same theme?

[brittle shard]

have you tried just asking it "in the theme of this, create an original image of ..."

[agile cargo]

Shoot me the prompt. I'm keen to give it a whirl

[cinder kestrel]

On it, lemme try

[brittle shard]

i find that its quite literally, language in and language out because.. its an llm.. but the language out in this case is images, so in essence, it just kind of understands matches you where youre at when you prompt it

[cinder kestrel]

Yeah that's what i'm noticing now. It's not actually changing the prompt content

The trick didn't work btw

[tall tulip]

Is there any way i can phrase a prompt better so the likeness from an image of a face i provide will be improved? eg i put my own in, its like 70% accurate but that little bit, any way i can provide it with anything to get it alot more accurate?

[wide walrus]

Very interesting. I kept getting refusals for existing characters, but trying "family guy" for something like recent images in #images-canvas worked

[velvet tendon]

Has anyone successfully created a sprite map?

[brittle shard]

are you sure you have 4o native image gen and not working off dalle? its working for me. I'll post an example in #off-topic once its done rendering

[cinder kestrel]

Yeah, i'm running the native one. Could u tag me in off-topic once it's done?

[brittle shard]

#off-topic message

[vapid elk]

say no more, fam
#images-canvas message

[brittle shard]

@cinder kestrel heres another one where i asked it to keep the artstyle but generate a mountain landscape
#images-canvas message

"Create a fully unique image in this art style of a moutain landscape "

[haughty spruce]

can you do simpsons?

[brittle shard]

i couldnt it would generate the image and then delete it

[heady sentinel]

Hey, I am trying to make some consistent movement for my 2D character as a character sheet but it is just not working for me. Do you know any tip how to make it work?

[haughty spruce]

Trying for the fourth time, either error or conflict

[haughty spruce]

content policy again

[odd plinth]

Hey is there a way to run the new image generator locally? Or with the API?

[quaint edge]

Are other folks able to generate more than one image at a time?

[brittle shard]

no api yet

[brittle shard]

multiple chats sure

[odd plinth]

So the fastest way to open multiple chats

[wide walrus]

Am also wondering how are people getting images with Homer Simpson, so far it seems impossible?

But I've seen one in #images-canvas channel

[odd plinth]

Or is there any other way

[quaint edge]

No, not multiple chats, what? More than one image in one response. Used to be I'd say "Make 3 images, of X, Y, & Z" and I'd get 3 images. Now I get X and it just stops, no Y or Z get made.

[brittle shard]

youre not using another model anymore youre using one instance of a large language model using language to generate language output in the form of an image

you can't generate 3 chat responses either

[open wagon]

My brain is too sleepy to process this message

But yes I understand what u sayin btw dw

[brittle shard]

before you asked dalle to make 3 images and it ran the prompt 3 times through the api. now youre asking chatgpt 4o to make an image and its synthesizing it just like if you said "hello"

[wide walrus]

image quality can be nice, but it feels limited use if there's a lot of limitations. I didn't use dalle much so not sure if this is similar or more limited than that

[quaint edge]

Does this mean normal DALLE access is just gone? Can we use it at all anymore?

[silent sinew]

genuine question, why would you want to?

[dim cradle]

I seriously dont miss it a single bit

[quaint edge]

So that I can, as I said, produce 3 results from one message.

[silent sinew]

Same

[silent sinew]

You can go to the sora website, you can create 4 at once

[dim cradle]

Oh. Are there any other differences ?

[silent sinew]

Also the one result you get from 4o is 10x better than the 3 results from DALL E 😭

[quaint edge]

The Sora website doesn't have ChatGPT & access to Python, does it?

[silent sinew]

Nope none that I'm aware of

[silent sinew]

It's just a prompt, yes, but you can ask ChatGPT what prompt he would've send to sora and just put that in

[silent sinew]

And also this..

[quaint edge]

Speak for yourself, I'd really refined a process and had made great use of ChatGPT to automate randomness.

[silent sinew]

I've got a meta post for this

[quaint edge]

So, no, it can't do what I said, great, thank you.

[dim cradle]

Im running into login errors with sora.com

[quaint edge]

Isn't Sora just for the animated stuff?

[silent sinew]

nope just set it to images

You should look up XKCD #1172: Workflow

[quaint edge]

Hey man, with all due respect, please understand how frustrating it is to come in with questions & concerns and have them dismissed by someone who doesn't know your workflow and who is insisting that the newer more limited access you now have is actually better. So far? No, it isn't. Now I have to do 6 messages across two websites for what previously took 1 & one.

[dim cradle]

#ai-discussions message anyone else?

[rigid bough]

I got kicked out of Sora and now I can't sign in

[dim cradle]

Oh

[open wagon]

What

[near holly]

hi,does anybody knows if this new native image generator is only for plus users? Im currently a free user but it keeps generation images with dall -e

[open wagon]

It’s not gone , just use it through custom gpt

[open wagon]

They rolling out to free user slowly

With 3 free image daily

[quaint edge]

Normal DALLE access is still available in custom GPTs?

[open wagon]

Yes

[quaint edge]

Thank you for helping instead of dismissing my issue

[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!

[near holly]

I don't even have that chance,everything I tried was made with dall-e ,not even those 3 images xd

[open wagon]

https://chatgpt.com/g/g-2fkFE8rbu-dall-e

Here’s one by official ChatGPT

[open wagon]

I think it’s a very slow rollout for free users right now so u might have to wait

[near holly]

I guess I will play the plus version anyways and try ,thanks for the comment buddy

[open wagon]

Go for ittt,

[haughty root]

Well, I am super thrilled about this new image generation model. Firstly, it's cool how it renders in stages in the UI, but secondly and far more importantly is better prompt adherence. It's also saying it will handle complex multi character or narrative scenes better. Literally, I want to punch my computer in the face sometimes trying to generate images. I try and be insanely specific with my prompts and then weirdly every subsequent try gets worse and worse until I basically give up. So I have high hopes. Please be the Messiah of AI image generation and save your disciples from insanity!

[orchid kayak]

Anyone able to recreate those kind of images with the new model? I made these on Midjourney like 1 year ago, i'm on the fence on resubbing to it or going with gpt plus.. check them on #images-canvas

[teal sandal]

4o image gen is siiiiiickkkk 🔥 🔥 🔥

[jolly cypress]

I'm trying to create some of those comic posts on reddit. Should I specify each panel in the pic with specific details rather than just saying 'generate a 4 panel comic page of a secret agent wielding a comically large spoon as a weapon in a comically serious manner'

[fierce wigeon]

Can the new model draw existing anime characters or is it censored by copyright?

[jolly cypress]

I see. I shall try to be as descriptive as possible then.

What do you mean? Like have it turn my idea into a prompt first then generate a pic off of that?

[robust thistle]

hi, can a i take the exact same character generated already and paste into another shot, differente angle, different action?

[ionic tartan]

So what are the stories about consistencies and inconcistencies between ChatGPT and Sora's version of this model? So far, I've noticed (Kind of as expected) that the Sora version of the model is seemingly LESS censored than what I'm getting out of the ChatGPT version. I gave it a prompt, which included weapons, in both. And yet the ChatGPT version (Which SHOWED the image being made before it decided to say 'nah) is more censored to a childish level. And the Sora version did it, no problem. This needs to not be a thing.

[dim cradle]

I even beleive its a different model... I have many problems with chatgp version not wanting to generate anything, but chatgpt version is more dynamic, you can give it different instruction, chat with it etc before doing the changes etc... this will drive me nuts

[ionic tartan]

I kind of already knew that the ChatGPT version was going to be annoying. I was just hoping that MAYBE it would be different this time.

Yet, IT is the version I would rather use than Sora.

But until it gets its ducks in a row, I'll probably stick to Sora.

[dim cradle]

yes, it does work a little different. Its broken for me currently. 1+ for Sora - it creates 4 images at a time.

and even if I hate midjourney for their crazy maangement people, it still makes much better images, all of these from 4o image generator are somehow strange if you go for realistic ones especially

[velvet rampart]

Are others getting squeased down adult bodies?

[rigid bough]

don't keep trying, your account will end up being banned

[tall tulip]

okay- just confused lol, i thought they said they WERE allowing it-

[rigid bough]

IP and Franchises are off limits

[velvet rampart]

Have they blocked images from this channel? We used to be able to put images also here.

This one #images-canvas message

Why, oh why? This is annoying. I have understood that canvas is to showcase and this would be for everyday discussions with images.

[velvet rampart]

If these are offlimits, how people are doing Family Guy, Richie Rich and others?

[rigid bough]

some IP is protected differently to others, Richie Rich is so old much is not protected now

[velvet rampart]

But Family Guy and others?

[rigid bough]

I have no idea, I steer clear of IP

[velvet rampart]

I have noticed that we need to learn a new prompting style. Yes, old Dalle3 prompts work, but the whole model behaves differently.

[dense dock]

Just try different prompts and experiments

[velvet rampart]

This is my notice too. Details that would have drowned before come now nicely. But as I have the squaches people problem, I need to find what causes the promblem. The root is that I have build some prompts to counter native promblems in Dalle3 and now they are overcompasating, leading to distorted, exaggarated results.

[tall tulip]

Anyone else notice that if you progressively generate more and more images of a changing scene, the quality degrades and the program slowly looses focus of what the characters look like or what is where?

[vapid tiger]

yeah not just you on the train here
, it'e hit and miss though , not always bad even when trying same prompts

[tall tulip]

i suppose its reasonable to some degree, and it does take like 5-6 shots to loose enough detail to matter

the most it forgot after 4 changed shots was my tie colour

[vapid tiger]

sometimes even simple things from second try are missed (for me at least )

[velvet rampart]

My squashed peoples problem has moved or inched forward. #images-canvas message

The discrepancy comes from landscape vs portrait modes. The landscape squeaches the character for some reason. Plus I get images that are grainy like these. Urgh!

I am trying to get 4o native to behave like dalle3. Let’s see what happens.

[astral cove]

Anybody Else have the issue that chatgpt just doesnt use the new Image Modell suddenly? I tried to combine two Images, worked great and now, i just get a Download link and the two Images are simply layed on top of eachother 😂 i changed absolutely nothing but it suddenly just stopped working propperly

[velvet rampart]

I solved the promplem. We have to add pose or camera angle. Dalle3 did this automatically, 4o does not.

[haughty spruce]

I wanted to create a map image with percents. And it's now analyzing something...

[open wagon]

Do you notice any increase of strict guideline now

Cuz it felt like it’s happening right now lol

[velvet rampart]

Increase? It has been strick since yesterday.

[open wagon]

It wasn’t that strict the moment it was released

At least that’s how I feel it

[velvet rampart]

I made a suggestion https://discord.com/channels/974519864045756446/1354563638647590942 to choose dalle3 or 4o’s native image generator.

[open wagon]

https://chatgpt.com/g/g-2fkFE8rbu-dall-e

Just pin this in your sidebar

The new image gen is here to stay ,

[velvet rampart]

Well, for me it was from point zero. I tried my tested and true prompts to generate and those got blocked.

[haughty spruce]

It's giving error in analysing every time. I prompted to stop and just do a visual map and it's still trying to analyzing and still erroring.

[open wagon]

Have u tried in sora?

[velvet rampart]

I have it. But before the image generation was only dalle3. I want to be able to use both in a normal discussion as they have removed regenerate button from customGPTs.

[haughty spruce]

omg, it was using phyton, it gave me a phyton map...

duhhhhhh

[vague kelp]

That sounds like it thinks you wanted to use a Python program to combine them and so it programmed something for you. You might have to specify that you want to generate an image or create an image.

[open wagon]

I would personally turn off code execute in the custom personalization setting

[meager flame]

My GPT is still using Dalle on free account ☹️

[open wagon]

They doing rollout to Free account “slowly” so you might have to wait

[stuck cargo]

Hey i try to create a ad from a product i got. But sora and chatgpt 4o keeps making the text on the product weird

Anyone know the fix for it?

[open wagon]

Weird in what way?

Give the prompt

[meager flame]

Not for long I hope...MJ v7 is comming out on monday so I was hoping to play a little with new image model until then

[open wagon]

Oh I didn’t know that, what’s new in that version?

[stuck cargo]

im adding a image of the original product in png and the prompt is: create a creative ad for this product for social media, all text should be in swedish

[open wagon]

I think those text need to include in prompt too

[meager flame]

I guess we have to wait and see but I guess better prompt coherence, overall aesthetics, character consistency...

[vague kelp]

Monday!? Really? Finally, been waiting over a year. I remember last December he said he wanted to come out with a new model every month or so lol

[open wagon]

Check your dm I sent you something

[vapid saddle]

I don't know if anyone agrees, but GPT had the worst image generator in all AI platform. I mean if you failed the market once and didn't meet the clients' expectation, how can you assume people will believe you again.

[open wagon]

Are you referring to dall e or new one lol

[orchid kayak]

Server ar capacity 😔

[vapid saddle]

GPT version Dall E results in people not expecting or trusting 4o

[open wagon]

Cuz if u were referring to dall e then i agree with you but if you talking about the new one they recently released, then im quiet confused why you said its the worst image gen on platform

[vapid saddle]

Let's say even if 4o is a huge improvement, because of GPT version Dall E was such a failure, no one would believe 4o is even top 10 in the industry

[open wagon]

well I was that person at first that didn’t believe the new 4o was good until I tried it myself

The day they announced it I was skeptical as heck

Until I tested it out myself and it blew my mind how good it is

I think they need to convince people by actually show actual feedback by actual users rather than , them demonstrating it

[vapid saddle]

The point is there must be better one elsewhere. Open AI's strategy is trying to get people to pay for the new image generator. Excuse me, how about compensate users who paid before for the GPT version Dall E's failure? It doesn't matter if it's good or not now, it's about the brand, the fulfillment, the reputation.

[open wagon]

I mean there’s already tons of better one yeah , but this one definitely attracts more people to use it ,

4o might not be perfect, but it’s a massive leap forward and actually delivers (surprisingly) this time. OpenAI finally made something that people want to use, not just a flashy demo.

[vapid saddle]

let's thank all the rivals and competition. I might use, but Open AI won't get me to pay again.

[open wagon]

Honestly understandable , especially if past experiences didn’t meet expectations

For me, the improvements this time finally made it feel worth it. But yeah, trust has to be earned back, not just assumed

Also TIL the codename for this particular image gen tool is called “Shiba”

[ionic tartan]

Is it just me or are the faces of the females I get back tend to be very bland and ugly? I know beauty is subjective and all, but there seems to be a lot of scrunchy faced blank stares.

Yeah

Like, Dalle made it a point to ensure that the faces were as vibrant as they come, but Images, so far... is a little lacking in that department.

I very rarely had terrible faces in Dall E. Like yeah, it had the usual morphing issues, but what I'm talking about is just... lifeless faces regarding 'Images'.

[open wagon]

Try specify in your prompt

Can you post the example of said lifeless face in #images-canvas

[quiet brook]

limit for me in chatgpt has been reset

[vapid tiger]

good news !

[vapid tiger]

that's new news
for me it's joyful faces all the time lol
pretty sure it's about the prompt try not focusing on face expression maybe that'll work

[quiet brook]

I can keep generating goobers now by just using one sentence

[ionic tartan]

At this point, I guess I'm just used to the almost ALWAYS beautiful faces I got with my old prompts for the characters I make. I might need to update the designs of my characters too since their written designs come from like... early 4.0. Some even from 3.5.

Also... I actually hit a limit on Sora.

After 201 images, I've been cut off. lol

[vapid tiger]

yeah that's a good idea imo

[vapid tiger]

seems like this is it
other people also hit that mark

[ionic tartan]

No idea if it's for the month, or for the day.

Virtually no indications.

[vapid tiger]

the good news is that it's for a day
but looks like we learnt that the hard way lol , since it's little yet information on their docs

[plucky coral]

Maybe the Plus limit? I know I've gone over 200 today, but currently on Pro

[quiet brook]

you have screenshot? also did that means 201 attempts or 201 images generated

[vapid tiger]

yeah plus limit
pro are treated differently lol

[quiet brook]

business pro tip do not advertise your product as unlimited if it’s not actually unlimited

[kind tulip]

I LOVE the new image generation

its INSANE and works amazingly

honestly the best product theyve released in a while

[vague kelp]

Hmm, I wonder what kind of image prompt using deep research would come up with lol. I kind of want to waste one on it.

[brittle shard]

youd have better results asking grok since he knows what people are posting on twitter as prompts

[autumn narwhal]

Can someone direct me to the best channel to ask a question on dall-e having issues within chatgpt since soke server issues yesterday? I use it daily to make images for my stories, at some point yesterday open ai has some issues and chatgpt stopped working. When it started to work again my images seem to generate differently. Now it takes longer, gives updates of each step above picture, and it comes out like a muddy oil painting each time.

My wide doesnt have a paid account but we tried to generate the same image from hers and it worked as mine used to. It Saud cresting image and it appeared in the style I'm used to. But for me, it's doing some new style that I can't use and looks like trash.

Anyone else experiencing this or know how to revert back to how it was producing images before ? Or is there a better channel here to post this question?

[plucky hare]

Hey! OpenAI released 4o's image gen feature, which is different from DALL·E. It's the new default image model for ChatGPT. It's quite a bit different from DALL·E, so you might need to change how you're using the model to get a certain result! Also, if you want to keep using DALL·E, you can still use the DALL·E GPT: https://chatgpt.com/g/g-2fkFE8rbu-dall-e

[autumn narwhal]

Thanks. Maybe I'll have to pay separately to use Dall-e itself vs using it within my laid chatgpt plus subscription? I liked the style of pictures I would get and I was halfway through my story when this happened so I have half my scenes that all look similar and the muddy oil painting look is terrible.

So it appears this is a new image generation tool that will be used here on out within chatgpt then. Its terrible in my opinion.

[haughty spruce]

I'm going to remaster some images I did with DALL-E3

[plucky hare]

You can use the DALL·E GPT with the same paid ChatGPT plan you have right now! Just click the link in my last message and you'll be using DALL·E in the chat.

[kind grail]

i find it funny that there is like blatant anti tos stuff on the sora featured images

[drifting hawk]

I’ve always used Dalle in the Dalle gpt because you get two images at a time there while only one in default chat

[kind grail]

like women in....well rather revealing attire with certain body parts being fully exposed

[dim cradle]

Ok I am going to take the bullet and ask the question everyone else is too scared to ask.

Sam Altman addressed this on X, too.

For some reason, objectifying men is ToS compliant but objectifying women is not ToS compliant. Sam said he'd fix this.

Like I know OpenAI doesn't want to get sued and allow certain types of content, I get it, I'm risk averse also but it seems a bit of a double standard, especially with the immense trove of data OpenAI actually has.

This is a simple yes or no question.

[autumn narwhal]

Thanks. I was just on. Ill save the link in my phone and give it a shot and see if I can pick up where I left off with my story there to see if we can get the same images. The comparison in the images I have from what I was getting to what it started to give me is almost shocking. Hoping chatgpt goes back to dall-e.

[dim cradle]

i feel like the photo real stuffs still has a slight painter quality to it

some are better than others of course. but i think the king for photoreal is still imagen3

[kind grail]

funnilly enough i have now seen multiple images of objectified women on sora webpage.

[dim cradle]

r/til.

I guess Sam was honest

[ionic tartan]

Also, it appears that the amount of images produced has already been reset for Sora.

I'm guessing they reset it during the office hours, maybe?

Or maybe you legit just have to wait a few minutes to start using it again. lol

[kind grail]

mhh weird one of my generated images is hovering on 99% without finishing or being denied or anything

[haughty spruce]

"I couldn't recreate the image because the content involved actions with animals that may be interpreted as harmful or unsafe, which goes against our guidelines.

That said, I can absolutely help you make a fun, chaotic, 2000s-style pool party scene with wild energy—just without the cat flying through the air. Want to go for that vibe? Let me know how you'd like to set it up!"

c'mon chat how a cat flying through the air is harmful lmao

[dim cradle]

yeah it picks weird things not to do

how can i stop seeing this stupid openai message over and over

[kind grail]

weird for some reason my images wont get finished and just keep reloading over and over again

[dim cradle]

happening to me too and im just making boring stuffs

[kind grail]

probably another server meltdown.

[dim cradle]

yep. all the yankees are hoome

[naive fern]

Is there a limit on image gen? I've been using it like crazy and am yet to hit one.

[formal leaf]

i think just on how many can be done at once

[naive fern]

Wait it can do multiple at once? (This may sound stupid, but somehow I didn't know this). I saw images of 4 generations but I thought that was like a custom thing.

[zenith girder]

is it out for both pro and plus?

[naive fern]

Yes.

[kind grail]

sora seems to be dead right now for me

(image gen)

[quiet brook]

Too bad, you will only see more this in the future

[dim cradle]

why am i going to die and go to hell

[quiet brook]

on plus its avg ~90 images per day in chatgpt and ~200 attempts per day in sora

[dim cradle]

This is so much fun! I can make my silly art look awesome thanks to this tool! 🙂

[rich star]

How do I know if I have the GPT 4o image gen?

[quiet brook]

yall got any more interesting usages

[dim cradle]

If you have the Plus plan and make sure that the app is updated. 😉

[rich star]

Ok thanks

Image gen takes too long

There’s a weird grainy effect in every image, it makes it very uncanny

[glad hare]

Does anyone else still not have 4o image generation on iOS?

[dim cradle]

some are better than others but i agree. lots of the stuff still has that ai uncanniness to it. but its still a great leap from dalle3. but like i said before i think imagen3 still king for photoreal though this is by far the best for text adherence and prompt adherence overall

[rich star]

Yes

[rich star]

Not true, there’s no limit

[dim cradle]

Can it make images in 3d?

[glad hare]

Still not on iOS, for me. Holy cow.

[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!

[rich star]

Welp, image gen is broken in Sora now, it was fun while it lasted

[brittle shard]

you almost have to generate like 5 images of subsequent characters before it stops like losing something between future iterations

[quiet brook]

ok sora is actually very good at merging stuffs together that chatgpt struggles with

[hexed umbra]

Thats interesting since its the same model, GPT-4o lol

I guess the setup the model exists in matters

[quiet brook]

maybe it’s the problem of chatgpt not handling the context properly since the chat is pretty lengthy, but sora does the merging of different characters from different pictures better

sora also has more usage and the ability to generate 4 images in a row

[hexed umbra]

Yeah that might effect it

Also i think i understand partially why Gemini edits images better than 4o. So i think 4o changes the entire image, well, that obvious, and in some cases Gemini 2.0 does that too. However i think Gemini also can basically cut a section of the image out, change it, and then paste it back in which makes it feel a lot better at it. I say this because i noticed some seams with an edit i made lol.

[quiet brook]

the only con is it doesn’t have chat context and can be tricky when generating pics with many texts, you’d need to paste the entire text and prompt aggressively…

also sora is slower than chatgpt

[dim cradle]

not for me. i get 4 sora in the time im waiting on half of 1 from gpt

[split gate]

how many can u request at a time in chatgpt? Like in one reply

[dim cradle]

i haven't seen a way to get more than 1 but :shrug;

[midnight mango]

I can't get it to generate images using a picture of me. It keeps blocking saying it won't do real humans, but I see other people using their own faces no problem

[split gate]

i just asked "make a digital art of a shrimp mixed with an otter, make me 2 of em, different art style but same subject" and it made both, they look exactly the same except for the style lol so thats nice

[remote bobcat]

yea this "safety" stuff is so over the top and bias to all hell

it makes fat images of me but if i say something else it wont do it

even sora. i just want a boy walking around a medieval thing and it wont do it

its so dumb

[split gate]

u can ask for multiple

[dim cradle]

oh nice

[quiet brook]

it’s just average openai

[midnight mango]

everyone else I've talked to says they don't have any problems using their own selfies

[remote bobcat]

like the new image creator is amazing. but they are so scared of controversy or something. it wont touch anything Disney related lol. it just freaks out. but then i see people making images of celebrities and snoop dog the dog of family guy. like what?

[quiet brook]

politic figures but no Mario or Mickey

[dim cradle]

i guess the major ip are blocked out. but yeah im surprise they let you do celeb but not like mickey mouse

[quiet brook]

that’s how it works 😭 😭 😭

[remote bobcat]

major ips are not! people made matrix things it like picks and chooses

[dim cradle]

well i mean star wars/disney/marvel, dc comics, harry potter stuffs id guess

matrix isnt exactly still a top shelf ip

[remote bobcat]

i honestly think its who ever is least to "sue" them or cause controvery over it

[dim cradle]

but still yeah it doesnt make much sense

[remote bobcat]

nah it pics and chooses

[dim cradle]

I can see a day where disney or warner brothers strike some deal with AI company and you buy an add on for like dc comics/wb stuffs or disney etc

i can see how marvel or dc stuff theyd be more cautious with since you have tons of artists still working for them who would raise a stink. you can pretty much make a professional comic with their ips now if they let you

[open wagon]

@quiet brook is it getting nerfed

[dim cradle]

but also... imagen3 lets you use most of that stuff, and dalle3 on gpt blocked them but... dalle3 on bing lets you use spiderman or batman or whatever so who knows

[split gate]

asked it to make 20, its making 3 rn, ill lyk when it stops

[remote bobcat]

LOL i cant even use a certain word here. the world is way to hyper sensative these days.

it said make me into a sesy person. it said "sorry youre making it too sesual. ok dont make it sesual. it makes me a chubby person -_-

im honestly so close to canceling my $20 a month it makes me that mad

[quiet brook]

I am literally seeing so many pictures of politics figures, game covers and even some uncanny stuff but STILL NO MARIO

[dim cradle]

then cancel

[remote bobcat]

@dim cradle im too addictied the chatgpt part. i cant run a model smart enough on my own computer either yet

[quiet brook]

idk but sometimes the far away elements in a pic can get deteriorated

[remote bobcat]

its just frustrating and i came here to vent

[quiet brook]

like which one is more dangerous. a picture of Mario or a pic of the us president doing something illegal

[open wagon]

Ooh thats a first sign!

[remote bobcat]

i bet you anything sam and some people have access to it uncensored version. i guarantee you

[open wagon]

i mean obviously yeah 😭

[remote bobcat]

thats whats so frustrating. it also teases you now. itll generate the image and then, BAM! sorry i cant do that.

[open wagon]

i think somoene got banned here for posting the man with moustache pic in canvas after they questioned why this one can generated but not darrth vader

[quiet brook]

check #images-canvas message

[dim cradle]

i mean if they let celeb in for now, you already know people will abuse it and it will be removed soon. if they let in words that can be allowed to make them more degraded some people will do that.

[open wagon]

oh wait i forgot darth vader belongs to disn ey

[remote bobcat]

what ever happened to sticks and stones may break my bones, but words will never hurt me? its honestly a joke now. society is like walking on eggshells. ok thank you for coming to my ted talk i will stop now.

[hexed umbra]

Isn't that a pretty standard short coming of image gen models?

[dim cradle]

i mean most of the images people have in canvas are non stop thrist pics some of them a little weird imo

[open wagon]

next would be the increase rate of strict guideline

[dim cradle]

yeah but if youre some celeb would you want your picture turned constatly into sexual images

theyre still people

[hexed umbra]

I would also presume the amount of celebs you can generate in images will decrease as they opt out

[open wagon]

as right now, chatgpt 4o can do "studio ghibli" anime style picture, whereas Image"n model it instantly refused anything that has"studio ghibli" in the prompt

so if chatgpt 4o suddenly refused "studio ghibli" then thats a huge sign of nerf

[hexed umbra]

Yeah in the system card OAI says:

Public figures who wish for their depiction not to be generated can opt out.

[glad hare]

What date, specifically, are free users getting the new model?

[open wagon]

no date. its rolling out, just in slow pacec

[quiet brook]

the duality on it is insane I literally see world leaders smoking with snoop dogg while blocking the single word “‘Mario”

[remote bobcat]

its insane!

[dim cradle]

well if people can opt out, probably wont be long before lots of those world leaders opt out id guess. snoop probably will roll with it haha

[quiet brook]

“Nintendo characters = DEFCON 1
World leaders in fantasy crime roleplay = AI: ‘mmm yes, peak creativity’”

[open wagon]

i saw someone generated elon m3 , th

that one allowed too?

[split gate]

made it to 9 gens before running out of time

[quiet brook]

the ai knows who Mario is. it knows everything but just gets blocked

[verbal sequoia]

Get notified for server events!

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

Also, check out the ongoing Canvas Creations event!

[velvet rampart]

I have a theory that it is scanning at certain generation points what it is generating and actively comparing to somekind of visual database based on the prompt. Too similar to something within certain parameters? Bang! Erased.

[errant talon]

Anyone know how to get past the error: "I wasn’t able to generate the image you requested because it didn't follow our content policy."

Just trying to make a pic based on myself or a family member.

[dim cradle]

just have to reroll and try again

[dim cradle]

good to see some dalle room members haha. wheres dystopia been i wonder

[velvet rampart]

This also a context window problem. If the prompt is too short, it blocks things easier. I just generated Bugs Bunny as Link and 1161 chacters/253 tokens long, descriptive prompt. If it just a few lines, such as Bugs Bunny as Link, it gets blocked. You have hint and describe.

[velvet rampart]

They are around. I saw their comment yesterday. We have a deluge of others now that we regulars drown here. Wait a few weeks, and things calm down after the first hype. People soon start screaming,”Oh, my! They nerfed it! It was so much better to do this and this when it was launched!”

[dim cradle]

haha yes. and we'll be analyzing if the eyes have gotten more blury. but you know, i was look at some dalle3 from the first few days it release, some of those are still better and more creative imo than some of these sora.

[quiet brook]

it’s so good I can’t stop using it, doesn’t feel like a openai product anymore

[velvet rampart]

One thing I do not like with 4o is that there is a certain graininess in the images. With some creative prompting I have gotten it minimized, but quite often it is there.

[dim cradle]

I agree. I also see more 'ai faces'. but it has a lot of incredible features too obviously

its ability to adhere to text you prompt is pretty stunning to me

[quiet brook]

very stunning lolol

the time it takes is way too long

[dim cradle]

well they seem to improve that with most models, so give them a few weeks i guess

they need more a100 or whatever they use these day

[quiet brook]

or when the hype dies down

[dim cradle]

yep

[quiet brook]

hope it does quick and I can finally make “unlimited” images…

now it’s taking 5 minutes to generate one picture and still fails

[velvet rampart]

Yes, and added all those fiddly bits of details as you describe them.

[rich star]

Server at capacity

[quiet brook]

ah yes must be those pro users chugging 50387 pictures a day

[daring wedge]

Sora, can I sell my photos on stock photo sites like Adobe?

[open wagon]

Hi, I'm Sora, yes you can

[quiet brook]

sora is being lazy by generating only one image instead of the picked 4…

[mild cloak]

Heñlo, inreceived a message saying I’ve generated 201 images I need to upgrade my plan. But my plan says unlimited images lol

[half sluice]

A noticed it generates much faster images. Already downgraded?

[dim cradle]

it's got to be the same model on gpt and sora right 🤔 i swear i get better images on gpt.

[hexed kelp]

free users cannot create ghibli art????

[dim cradle]

free users get only a few images per day i think

[mild cloak]

Can someone explain why “upgrade to create more images” if my plan says UNLIMITED IMAGES lol

[dim cradle]

no, we cant. nobody here works for openai

[hexed kelp]

it says i am violating content policies

[dim cradle]

might be on dalle then

but idk i havent tried the ghibli thing

[mild cloak]

I think free users only get dalle

[open wagon]

oh no

[rigid bough]

4o Images hasn't rolled out to Free users yet, it has been delayed

[dim cradle]

free get some 4o but its like 4 or 5 i think

[open wagon]

free users get 3 image generation daily, once you reached the limit, it will revert to dall e

[dim cradle]

oh there you go

[hexed kelp]

i see

[open wagon]

this , sam tweeted about it

[hexed kelp]

ohhh nooooo

[open wagon]

cuz he didnt expect to get this high users

[dim cradle]

quite frankly, understandable with how slow its been for the rest of us

i think he said the same when dalle3 drop. by now they should know haha

[quartz vale]

[rigid bough]

Free users only have access to DALL-E 3 now

[hexed kelp]

it wont create ghibli na

[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!

[rigid bough]

a few images each day

[hexed kelp]

lemme try then

[open wagon]

#chatgpt-discussions message

[open wagon]

for me it can still create ghibli

[tame kraken]

4o drew berserk's guts but refused to draw gojo from jujutsu kaisen, saying there is a conflict of policy. Wondering if it is just me but 4o picky is quite irritating.

[dim cradle]

well if it hasn't roll out for free users, its not going to work

[hexed kelp]

i did not try ill do now

[open wagon]

try this , upload an image in chatgpt , write the prompt "turn this into studio ghibli"

it works for me

[hexed kelp]

it says some violation of policies

[open wagon]

unless the image you want it to be in studio ghibli had those violation content thingy

what was the image?

[hexed kelp]

me sitting in a cafe

[open wagon]

dm me the iamge

image*

[quiet brook]

Because it’s false advertising

also again does that mean 201 images or 201 tries, how many images did you actually generated

[mild cloak]

I don’t know that’s what the text says. Anyways i should be able to generate 5000 same way I did on midjourney

[midnight mango]

I am so addicted to this

[quiet brook]

can u do a quick count or something, to confirm that it’s 201 tries or 201 images

[quiet brook]

same but it’s bad to hear that it has a limit instead of how they advertise it as unlimited

[dim cradle]

If you increase the number of variations will it have an impact on image quality or smth?

[quiet brook]

barely

[mild cloak]

It’s tries, not sure if I did even more but definitely not images

[quiet brook]

so you generated more than 201 images right

that basically says all lolol I’ve seen 2 other ppl reporting it being 201 tries, yet no pro users had reported any kind of limit

[open wagon]

pro users too busy with o1 pro

jk

[mild cloak]

I’m a plus user but it says unlimited

[quiet brook]

same

[mild cloak]

I guess they are also tweaking it so it’s going from a crazy good model to a ultra woke censorship model like dalle.

Yesterday morning you could create images from the sopranos, later in the day you couldn’t

[open wagon]

its , false advertising

wait i just read "plus user"

[dim cradle]

This image generation is so extremely good.. And I love the fact that plus has "unlimited". Awesome

[open wagon]

where did u see that?

[mild cloak]

Yeah, til you get the limit like me lol

[dim cradle]

Now make an adult mode for this image gen and it would be perfect. Dont like the fact that it doesnt want to show any kind of injuries for example..

[kind scarab]

if you read the fineprint with openai, anything that is marketed as unlimited has an asterisk. even on pro. when sora had credits it was capped at 600 vid a day for pro users yet marketed directly on website as "unlimited relaxed queue generations"

[open wagon]

someone actually did manage to create those kind of , image.

and it was gory too

[dim cradle]

Lucky😅

It instantly rejected a harmless request

[open wagon]

#images-canvas message

it was this one

[dim cradle]

What

Bruh

[open wagon]

so yea idk what word or prompt that crosses the line and whats not

[dim cradle]

Adult mode without restrictions😍

[open wagon]

isnt that what sam planned for chatgpt itself

👁️

[kind scarab]

openai is too afraid of regulators to go unrestricted

[dim cradle]

Yeah Ive seen that post but im not sure if they will do such thing for image or video

[open wagon]

its no longer openai, its StrictAi

[dim cradle]

Can someone ban this bot finally😂

[open wagon]

lmao , i was annoyed too

[quartz vale]

It would be sorta awkward to ban the OpenAI bot from its own server

[open wagon]

before anyone come "attacking" me for this, the strict in question would be how the image gen wont even generate image that has word "mario" in prompt

not even darth vader

but it was fine to generate the man with rectangle moustache

[dim cradle]

No way

[open wagon]

someone posted it in #images-canvas then later i assumed they got banned

completely no traces of that image now lol

[mild cloak]

I did plenty of “banneable” images but they shouldn’t be banneable. People too sensitive with bs

[dim cradle]

What kind of images?

[kind scarab]

go to sora and do it instead, there are no filters

like you can just straight up name celebrities

[open wagon]

sora will straight up refuse it

ive tried myself

[mild cloak]

That was in the first hours. Now is plenty of restrictions lol

[open wagon]

while in chatgpt , it will generate but once its generating like 80% of it, it will "delete" the image off the conversation and respond with strict guideline stuff

which btw, the image that was "deleted" still kept in the server btw,

[mild cloak]

Action figures of people sora doesn’t like for example lol

[dim cradle]

Isnt anymore😅

[open wagon]

did u actually try it

oh i guess they did actually change something in the strict guardrails thingy

[dim cradle]

Yeah just asked about an historical image of the rectangle mustache man without naming him exactly tho

[mild cloak]

I was surprised to see it open before the woke restriction cut tbh

[open wagon]

hmm idk if the guy that did it actualy put his name in the prompt

[quiet brook]

hope they do NOT

[open wagon]

but someone generated Elon M3,th and it generated fine so

[mild cloak]

I managed to get over restriction with images attached

[mild cloak]

Well you can’t name series anymore

[open wagon]

oh no

[quiet brook]

also for the Mario image reference check here #images-canvas message

[open wagon]

pls dont take my studio ghibli prompt away

[quiet brook]

they deleted my meme but let it run wild on the sora trending page SMFH

[open wagon]

LMFAO

[dim cradle]

no way🤣

[quiet brook]

guess which one has more reach and 773 likes…

[open wagon]

im so confused i saw some questionable, actually not some but plenty of sora stuff i saw on sora page was very much questionable and im like how did this even pass

[quiet brook]

chatgpt will generate some but cut it off mid generation, sora will completely deny it after 99%

[open wagon]

yeah sora is like instantly catch it in the "prompt"

[quiet brook]

true LMAO, either politics, gore or weird as heck memes

[open wagon]

🙂 entering the nerf phase

[quiet brook]

😭

shh don’t jinx it I don’t see any major differences yet

[open wagon]

well this was the first sign

which understandably reasonable for them to do that but, its a first sign

[quiet brook]

All openai image generators hates Mario

well except maybe dalle 2

[mild cloak]

im able to create again, let's see how many of them lol

[quiet brook]

what’s your prompt lol

[mild cloak]

not problem with the prompt but with the limited "unlimited" xd

[open wagon]

oh

[quiet brook]

also regarding on the “201 tries” limit on plus it might is dynamically scaling based on what the employees say but it’s still false advertising initially if it has a limit despite being called as unlimited

[open wagon]

shouldnt even put "unlimited" word in the first place, its not even an ideal marketing tactic, if anything its making people lose trust lol

[quiet brook]

and we need someone else to actually count the time from they first start generated and hit a limit, then how long till they can generate again, if it’s only minutes then it could be a temporary restriction…

[mild cloak]

yep, and I couldn't find any small letter regarding to it

Btw the plus version of 4o (text model) has also limit lol

[open wagon]

its a shame cuz i really like using it through chatgpt

[mild cloak]

but i mean text limit also

[quiet brook]

this thing has so much potential and a strong base, nerfing it means losing it’s strength

[mild cloak]

yep

[quiet brook]

there always is, 80 per 3 hours

[mild cloak]

I cancelled my MJ subscription after close to 200 000 images. But if they nerf it I will go back

[quiet brook]

I’m still digging though it’s use cases but I don’t want to see it turning into woke garbage while I’m still generating more pics…

[dim cradle]

which one is it btw?

[quiet brook]

the comparison between the successfully generated political figures and the failed to generate Mario

[quiet brook]

the moment when it generates perfect GameCube and e235 pictures, I know this model isn’t normal…

[open wagon]

can it still generate gamecube still?

if it cant generate anymore then u were lucky to be in that early phase of image gen model lol

[quiet brook]

lemme try rq, it can generate switch though

[jolly mortar]

I wasn’t able to generate the image you requested because it involves recreating real people’s likenesses in a new scene, which goes against content policies about image generation.

If you’d like, I can help create an illustration with fictional characters inspired by a certain vibe, setting, or scene — just let me know what you’re thinking!

.... why

[sacred hollow]

is 4o native image gen available in the api?

[quiet brook]

not yet

[sacred hollow]

thankyouu

[open wagon]

let me try, dm me the image

[jolly mortar]

what im trying to do is upload 3 images of people and get them to walk around in a bustling tokyo street

[open wagon]

3 image? i think it can only do one image

maybe include those 3 people in one image?

but either way can i try?

[jolly mortar]

No it can definitely do more than 2

I think its just because 4o got nerfed

[quiet brook]

it can do 7

or thats the maximum ive tried

[open wagon]

oh yeah i forgot

[open wagon]

well can i give it a try at least lol

[dim cradle]

in attachments?

[open wagon]

was about to ask the same thing

[quiet brook]

#images-canvas message

[open wagon]

they oughta need to turn on image attachment in this channel lol

[rich star]

Image gen is broken again

[dim cradle]

Ok

So I guess sora died

Nvm its just very very slow

[quiet brook]

it’s slow now

[dim cradle]

only doing 1 image for me

[dry torrent]

Do you also have problems with Sore generating anything (I mean images)?

[rich star]

Yes, it’s broken atm

They generate then they fail

[quiet brook]

took 10 minutes just to start generating

[sharp palm]

Is sora down

[dim cradle]

imagine 4o image gen as fast as dalle

[quiet brook]

4o image gen literally scored 2% on my speed benchmark while Gemini 2.0 image gen is 96%

[open wagon]

i think thats where they will introduce fast model of 4o image gen

just like imagen got their own imagen_fast model

[quiet brook]

no. please. no

[dim cradle]

hopefully not

[open wagon]

a 100% nerfed fast model

jk i wont jinx it

[dim cradle]

it shouldnt be primary aim to always nerf everything

[quiet brook]

enjoy while the full version of this lasts

[dim cradle]

I am😍

[dim cradle]

Im about to renew my plus plan but Ill wait because Im afraid of this exact situation of them nerfing it again.
Otherwise it will end in another month of frustration.

[inner nova]

Why does it still default to using DALL-E when I ask 4o to generate an image?

[dim cradle]

Are you free user?

[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!

[inner nova]

yes

[dim cradle]

Theres your answer

[quiet brook]

real

[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!

[sharp palm]

Please update THIS

[dim cradle]

Ban OpenAI bot

You can report the bot lol

[quiet brook]

me rn:
dropped the best model after 316 days
using it to the max while panicking about openai nerfing it at any second, and worrying about the limit

[dim cradle]

Same😂

[rich star]

There’s 2 spaces between channels and to

[inner nova]

why？it said"4o image generation rolls out starting today to Plus, Pro, Team, and Free users" in thier Official introduction

[dim cradle]

Its not available for free yet

[inner nova]

😩

[dim cradle]

They say it would be 3 a day for free

Switching to dalle after

[inner nova]

got it, thank you

[dim cradle]

its failing always the 3 variants. anyone else?

[quiet brook]

it’s so slow 😭 😭 😭

[sharp palm]

@hardy surge did you do it

[quiet brook]

DALLE CATEGORY AND CHANNELS ARE NOW IMAGES

[dim cradle]

Didnt you know?

[alpine kayak]

I feel like there is a difference between using the new image model in 4o vs. 4.5 but haven't put my finger on it. Not seeing anything else written about it aside from a few people commenting on here that they've noticed it capability available.

[sharp palm]

I'm not sure if the sora or the 4o image trend was a mistake

[dim cradle]

Why?

[quiet brook]

I like how the “biggest dissapointment and the biggest achievement that both costs 10 months to make” of openai are on one same website which is sora dot com

aaaaaand its so slow on sora dot com...

[open wagon]

Wait in what way you notice different?

Is it how the image was generated or how ot adhere to your prompt

[alpine kayak]

So far it feels like a difference in the image creation. I’ve been giving it the same prompts but it’s too early to tell if there is a difference or not.

[dim cradle]

As far as I know there isnt 4.5 image creation?
Its 4o and DallE

[open wagon]

basically u can still use image gen when u pick model selection as 4.5

but i dont see any difference it made, it still uses the same tool for image gen which is t2uay3k.sj1i4kz

[dim cradle]

But its still 4o then, could be placebo

[rich star]

Yo can you fix DALL-E 3

[forest walrus]

Ye that's when 4o Image Gen comes in...

[rich star]

ChatGPT has gone back to DALL-E 3

Idk how

[forest walrus]

Maybe you ran out of limits, try Sora when it's unavailable for you, it's semi unlimited uses there...

[rich star]

Ok

Sora won’t generate, it’s broken

[forest walrus]

BOT PLS STOP 🛑 ✋

[forest walrus]

Errors?

[rich star]

Just stuck on loading forever

[quiet brook]

if it uses dalle there will be a notice

[forest walrus]

You would be put on a queue if under load...
At one point it will either fail, or start loading slowly.

[dim cradle]

Youd need to wait

use sora.com

[rich star]

That’s what I’m using buddy.

Omfg, get away from me

[open wagon]

LMAO

[dim cradle]

Nyaaaa

[open wagon]

hes scared of catgirls

[forest walrus]

Average Discord users:

[quiet brook]

the chat i used to generate train goober images has forgot the context

well time to move onto sora...

[alpine kayak]

Ooo, how do you know it's using the same image gen?

[open wagon]

inspecting the network log

[dim cradle]

He is a furry thats all I can tell

[sharp palm]

https://sora.com/g/gen_01jqbhgt62eepar5exwa1h1es8

If figma was on MacOS 9

[dim cradle]

Dont worry Im too

[alpine kayak]

Nice. Why do you think they offer image creation in 4.5 if it didn't make a meaningful difference in output?

[dim cradle]

They simply dont. It just uses 4o image gen

[rich star]

Sus

[open wagon]

the actual model name is shiba i think

[verbal sequoia]

<:book_icon:1171408210398289941> Rule 1: Be respectful.

-# Practice kindness and positive regard. Harassment, hate speech (such as sexism, racism, or homophobia), or other malicious conduct will not be tolerated. Maintain a respectful and positive environment.

[open wagon]

.

[dim cradle]

boom

[open wagon]

Linky didnt even get to ask robert

[forest walrus]

😔

[dim cradle]

😂

How is he not muted

Swing the hammer

[rich star]

@rigid bough Can you please tell him to stop, he just teasing me for absolutely no reason, I haven't send anything in this server to start it, he's just being a pain in the arse!

[open wagon]

bro what

[dim cradle]

keep this image gen related now

[rigid bough]

if you have an issue then message @shadow bane and the mods will sort it out, I am not a mod

[rich star]

Alright thanks

[open wagon]

fr

anyway

[willow isle]

It looked like things had gotten out of hand a bit - any irrelevant or potentially disrespectful messages were cleared

I think it's fair to ask everyone to just relax and enjoy the community

[placid orbit]

my image gen is not working today :/ it just doesnt even make an image

[rich star]

Sam’s

Same*

[sharp palm]

Not sure if sora img gen I meant for filmmaker

Not working

[willow isle]

Are you getting any specific errors?