images-discussions

[unique edge]

ah no, im actually using dall-e to create an AI UI generator

haha very funny

(actually funny)

[gray surge]

Good luck. 👍 😄

[unique edge]

what model are you using? im not achieving the same results as you.

[unique edge]

@gray surge

[gray surge]

Please do not ping me and be patient - I answer everyone when I am there. 😄

And I am using ChatGPT Plus Dall-E 3.

[twin jasper]

OMG...I was laughing as I was testing Dall-E 3 to have a painted effect of a family portrait just to see...and this family of 5 or 6, Caucasians, was turned into a family of 5 or 6 of...Asians

[haughty spruce]

Prompt understanding is so well done in DE3 that it even understood what post punk clothes are

[pseudo hollow]

I get so frustrated with Dalle sometimes. Check this out. is Dalle this bad??

MY PROMPT - A vertical 3D CGI animated inspired movie poster Featuring a tan colored King Charles Spaniel. The scene should be set in the distinct digital art style of an animated family movie, with a focus on character expressions that are characteristic of the dog character animation. -- ar 9:16

ChatGPT
I was unable to generate the movie poster as you requested due to content policy restrictions. However, if you have another idea or need assistance with something else, feel free to let me know! what restrictions?

YOU: My prompt adheres to all of your content policy restrictions.

ChatGPT
I apologize for the confusion. The restriction in this case wasn't due to a violation of the content policy directly; the image generation system simply didn't produce a result for the specific request. Let's try again. I'll make another attempt to create the animated-style movie poster featuring a tan colored King Charles Spaniel as you described. Give me a moment to process this.

It seems that there was an issue with generating the image again, and I'm unable to produce the animated-style movie poster you requested. This appears to be a technical limitation rather than a content policy issue. If you'd like to adjust the request or if there's something else I can assist you with, please let me know!

[dim cradle]

just regen

i get that sometimes and second time is the charm

[pseudo hollow]

I tried four separate chats, and now I used up my current batch of chats, and I have to wait 4 hours!!

[dim cradle]

weird. try bing image creator maybe

[pseudo hollow]

well, there are others. Yes, I could try - but I pay for this... so frustrating

[dim cradle]

change the prompt a little before the next time if you havent i guess 🤷

[late blade]

depenging on the dall-e implementation there are different strats to get the exact same result, most of the ones you see from me are done with OAI's own DALL-E over ChatGPT+ which won't allow you to directly associate any prompt with franchises due to content policy

[hearty ether]

Anyone getting issues?

[late blade]

yes, for some reason this chat I'm doing with gpt4 is not telling me the things I want to hear but instead the hard cold facts

should I report it as a bug?

also I don't like today's theme at all, never liked fish eyed lenses

[royal skiff]

what are good prompting words you use to actually get photo realistic humans? they always come out a bit arty and not photograph to me

[plucky hare]

Hey! Can you share an example prompt and image? Just to see what you're currently trying and getting! One general tip: if you're using the actual word "photorealistic", maybe try just using "photo", "photograph", "candid photo", "professional headshot", etc. instead.

Think of an image prompt as a description for the ultimate end result (the image), not as instructions for DALL·E to follow. Kind of a small distinction, but put another way: if you want a result that looks like a photograph, pretend you're describing the photograph you want to see in the end (i.e., you wouldn't describe an actual photograph as "photorealistic").

[lost elbow]

Hi everyone! I've been using DALL-E 3 API to create images of food, specifically grain bowls, and I've specified in my prompts not to include half avocados, pits, or seeds. However, they still appear in my images. Does anyone have tips on how to refine my prompts to ensure these elements are completely excluded?

Here’s the prompt I used last:

'Create a high-resolution image of a warm grain bowl with minimal garnish, using natural lighting and a 1X1 aspect ratio. Include a scoop of avocado mash on top of quinoa and brown rice, surrounded by kale, sweet potatoes, and red bell peppers. Highlight the texture of the avocado mash and the colors of the vegetables. Specifically, do not include half avocados, pits, or seeds in any form.'

[lost elbow]

Image result

[plucky hare]

Hey! DALL·E is currently weak with handling negative prompts -- i.e., "Don't include x". Basically, asking DALL·E to think about x at all, even in the context of "not x", tends to confuse it. It's best to include only positive details about what you do want in DALL·E prompts. You may still get undesirable details sometimes, but it will be with less frequency than with using negative prompting.

[fiery salmon]

I keep having this dream about a golden city deep in the mountains

the bridge is very narrow, only bikes can enter, everyone in the city is very tall and the buildings are all gold

Not how it looked but a decent replica

[lost elbow]

I'll try that

[exotic pagoda]

Sometimes I get mental with the wokeness and carefulness of the filters..
Try to make a picture of a black paella for a spanish dinner.. Arroz Negro... no chance.. the filter will always use the "Not allowed" excuse.
Makes it really hard to work with DALL-E on some points..

[red umbra]

looks like the N word.

[vagrant zodiac]

yeah it's being blocked for that, but unfortunately this is an issue when dealing with other languages in chatgpt

[late blade]

finally a long weekend, from now until monday no adulting whatsoever

[gilded osprey]

I'd love to see this animated

[vapid elk]

cool

[hearty ether]

is anyone having the ai remove important parts of the prompt?

the ai keeps removing the "realistic illustration" style part of the prompt even when i tell it to keep that in

[late blade]

aawww that stinks, forgot to paste the image for daily theme and pressed enter before it, now timed out lol

[late blade]

not for me, revised prompts get the details I pass in that context

[lean iron]

as soon as i put "manga" in the prompt it gets immediately censored and the doggy pops up

when is dall-e gonna get uncucked?

this happens to me A LOT

[hearty ether]

?

[lean iron]

i put "Manga aesthetic" in my safe for work prompt

and it filters me

does this for a lot of things

[vapid elk]

what is the full prompt?

[frigid heath]

Is this one of the possible reasons why they say "may consider NSFW" in the Model Spec? The model can't even generate minimal "things" in recreation of know (really big) art pieces

[velvet rampart]

Strange. Mona Lisa comes really easily. No problems whatsoever.

Size?

[frigid heath]

Man, sorry, I don't know if you know the original art

But I think the model is being heavily censored on nudity

[velvet rampart]

Oh, I do know the Sistine chapel piece.

[frigid heath]

Yes

[velvet rampart]

Nudity. You hit the reason.

And bare skin, a lot of it, is on the Sistine chapel paintings.

[frigid heath]

This is for a good reason, but in this case, we already know what to think, this may be the best reason for that comment in the Model Spec

[velvet rampart]

It gives me this:

The issues occurred because:

1. Content Specificity and Accuracy: The prompt requested imagery inspired by a specific historical and artistic work, Michelangelo's Creation of Adam, which is part of the Sistine Chapel ceiling.

2. Copyright Concerns: Although the original work itself is in the public domain due to its age, the specific visual representation could still be protected, potentially limiting image generation.

3. Representation Complexity: The wide format requirement combined with detailed depiction of a large scene, including various figures and a specific artistic style, could exceed the model's capabilities.

4. Technical Limitations: There could be technical errors on the image generation side, resulting in failure to render the scene.

Let me know if you would like to modify the request or if there's another concept you'd like visualized.

[frigid heath]

I think this is nonsense the model generated, normally the model doesn't reject to generate something for being complex, he just generates it anyway and do it poorly, lol

[velvet rampart]

I am just testing if I can get Mona Lisa out.

[frigid heath]

The chat interface probably doesn't even have access to a reason for generation to be denied so it make something

[frigid heath]

It can do something "maybe this is like"

[velvet rampart]

Ok. I got Mona Lisa out. Michaelangelo’s David statue - no. Pieta? Hmm…

[frigid heath]

Too much probably

[velvet rampart]

Pieta - no. Leonardo da Vinci comes out ok. But hardly anything from Michelangelo.

[frigid heath]

Some work, even having too much skin

Like, crucifixion, and really created something on the style

Nvm

[velvet rampart]

It seems that Michaelangelo is a bit persona non grata. At least his originals.

Copilot?

Jeps. Copilot does neither work.

[frigid heath]

Yep, and is pretty direct too

[velvet rampart]

Now I got this out.

Physically based render, clothed male marble statue, style of michaelangelo

Key here: clothed

AI has been trained to be puritanitan

[indigo jasper]

When will Dall-E 3 support inpainting on imported images, similar to what we had in Dall-E 2? I see this forum, https://community.openai.com/t/dalle3-inpainting-editing-your-images-with-dall-e/705477 but it only allows inpainting on generated images unfortunately.

[vapid elk]

yea, that makes sense

[thick smelt]

hello, i'm back!
enjoy a bichon~~ fisbee~~ frise!

[primal flax]

ey the theme is “Europe” on daily-theme ?

[plucky hare]

For 20 more minutes, yes! Then new theme

[primal flax]

ok thank you 😁

[velvet rampart]

I just read that OpenAiI conciders bringing NSFW content to DALL-E3. No deep fakes. Note: conciders, no certain decisions, can hold current policies.

[vapid elk]

they did not

[velvet rampart]

I am trying put my source here but I cannot put the web address.

in Finnish - use e.g. Google Translate to English

Well here is screen capture.

So, I am not inventing this.

Also reported by The Guardian.

I cannot post links.

So, this is reported but we will see if anything comes from this. I highly doubt.

[marble loom]

Is sora available to the public?

[velvet rampart]

No. Stop asking about Sora. Most likely never will be.

[plucky hare]

Hey! No, Sora is not currently publicly available. https://help.openai.com/en/articles/8958981-how-to-access-sora

[vapid elk]

I love how those news sites are bad at the thing they are supposed to do lol

the source for that info was an OpenAI blog post, it contained one line about the fact that they have plans on figuring out a way to let NSFW content be generated responsably

this is the whole source

[velvet rampart]

Ok. Thanks for clarifying. As I originally put, but in other words - uncertain.

And unlikely as we have seen how puritan they have made their AI in text and image generation.

[gilded osprey]

it's funny how it really doesn't understand the connection between words and images, but it still creates interesting results

it misunderstood what parts of the image are workers, but this is a really cool result

[dim cradle]

i hope the big monday news thing have some effect on dalle 🙏

[dim cradle]

What new features are you most interested in?

[verbal sequoia]

Vote for your prompt here: #spotlight message

[dim cradle]

just improve in fidelity, even better photoreal. ability to make as many character in an image without bleeding of they feature. but i am read some rumor the big reveal on monday will be a voice AI mode like the movie Her, so I doubt will be any effect on Dalle

[willow cargo]

What's the prompt to make this type of art style?

[glossy scroll]

Put that image in ChatGPT and ask the same question.

Let chat GPT be your guide for breaking down ideas

Envelope yourself in the art of dissecting prompts within ChatGPT

Let ChatGPT be your guiding light to glory

[thick smelt]

did some bonnie dalle art
prompt:
"purple realistic rabbit with red bowtie, painting, in the style of [artist]"

[thick smelt]

hey pytha! it's me lei!

[glossy scroll]

Oh hi Lei! 👋😄
Nice to see you 😺

[sour dew]

Why DALLE have problem with size and count and optical how do this ?

[sour dew]

Is my corect paint exampels

[hexed loom]

Dalle model is very similar of the model being used by Blue Willow last i checked

is it Mid Journey level? Not really.

[hexed loom]

I hope not. as if there isnt enough of that.

[velvet rampart]

If you read, our own conclusion. Nope. Not happening. Unlikely. See what kind of AI model they have done.

[dim cradle]

Couldn't agree more

[glossy scroll]

[gilded osprey]

it's trolley problem time

[late blade]

did something happen to dall-e? quality has been horrible today

[vapid elk]

wdym?

can you send an example?

[late blade]

eyes usually get really good with my prompts but today all eyes of the images I've done have been like this

[vapid elk]

interesting, what was the prompt?

I have some friends using dalle right now via my bot on a discord of mine, and the quality seems fine

[late blade]

it's my usual json file for art style that I reference

which usually has good outputs

This one is from copilot, same text

[vapid elk]

seems a little odd indeed

well, lets see if it changes in the next few days

might be an experiment

downvote the images

[late blade]

I don't like experiments on me unknowningly <.<

I did notice dall-e started removing on revised prompts most of the adjectives I use

[thick smelt]

i see the Genshin reference @late blade (sorry late reply, i was asleep earlier)

[late blade]

genshin reference?

[dim cradle]

Must’ve been unintentional

[north sequoia]

/generate

?

[dim cradle]

The bouncing ball guy cracks me up

[verbal sequoia]

Vote for your favorite prompt now ➜ #spotlight message — Ends <t:1715658678:R>.

[vapid elk]

#image-bot

[gilded osprey]

I think the eyes are too small, it doesn't care of those details

[thick smelt]

yeppers 💯

[late blade]

never played genshin, but I guess that's the reference there

[verbal sequoia]

https://discord.gg/ANdZxFk5?event=1238546735144636548

[thick smelt]

hey guys!

[thick smelt]

#images-discussions feeling empty nowadays

i made a dall-e bot this is not first one, the pancake was, but this anime dude was supposed to be the youtuber who makes game theories who retired recently, just what do you think

[ripe fiber]

is it me or in-painting is pretty much useless today? it doesnt work at all

[gray surge]

What exactly does not function for you?

[late blade]

yeah it's been quiet recently

[gray surge]

I like it. 🙂

[late blade]

matpat, nailed it

[late blade]

stream is on discord?

or is it somewhere else?

[gray surge]

It is on youtube: https://www.youtube.com/watch?v=DQacCB9tDaw.

Sorry for the late response. 😅

[late blade]

I'm watching

[loud field]

So, Tetris has lore?

[late blade]

of course

have you even seen the movie?

[hearty ether]

So is it sora or only video viewing?

[vapid granite]

Has Dalle 3 been upgraded with this latest chat GPT update?

[late blade]

yes, they upgraded dall-e 3 to dall-e 3.000000000000000000000000000000135

[verbal sequoia]

Vote for your favorite prompt now ➜ #spotlight message — Ends <t:1715658678:R>.

[gray surge]

Can we read for that somewhere? (Or is that also GPT-4o?)

[gray surge]

"Only" video viewing I think but let us wait for the reply of Dys Topia. 😇

[late blade]

the text is sub-pixel big

[late blade]

what? What I do?

sheesh

[gray surge]

My context was this because I didn't heard/saw/read something about a Dall-E update and maybe there was mentioned something like sora - but so how you replied there isn't. 😂

[late blade]

it's right there

in the release notes

it says: if you can read this, you are too close

never trust me tho, I'm evil from monday through sunday

[gray surge]

okay. 😂

[dim cradle]

you guys test out the new model? it is blazing fast and say it effect image too

but i do not know if you start a dalle proper chat if it is using gpt4o or not 😕

[hot silo]

I'm testing it, but the results are not great. What's on the website is unimaginably better.
I think I'm either using it wrong or gpt4o doesn't work with it yet.

But yes - it is MUCH FASTER even with creating images

[dim cradle]

im using website

yes it is blazing

i wonder what the new maximum caps are now. i thought i read they increase it for pay user

what a time to be alive!

[late blade]

I tested it on the API and it's super fast

[dim cradle]

i wonder if custom gpt need to be remade to use 4o?

[late blade]

no,

it's already live

and the red dot, that's a feature nobody else has

it's called FAT FINGER

[gray surge]

Tell me more! 👀

[tawny portal]

Hey Dys, did you have a chance to toy with GPT-4o for image gen? It seems to bypass dalle now.. I think.. not sure.

[sacred temple]

I am assuming that it is still Dalle3 for image generation, but it’s the LLM context and handling the prompt and vision could be different.

So far running all the same prompts, I’m getting similar images.

[tawny portal]

Just got access to GPT-4o too. Yeah it uses Dall-e for image gen. I can't reproduce examples from the site: https://openai.com/index/hello-gpt-4o/

(check out "Explorations of capabilities" section; use cases that start with "Visual Narratives")

Also "Character Design - Geary the robot" -- can anyone reproduce something like that?

[late blade]

dunno, I'd have to check

[tawny portal]

Examples are amazing - it is much better than what we can do with Dall-e. This would be next gen if possible. Not sure how to reproduce results though.

[dim cradle]

you guys see these demo of the voice chat? going to be wild to make some dalle image that way soon haha

[tawny portal]

Yeah.. but it seems to have rolled out with bunch of other stuff not demoed in initial talk (with Mira). Check out blog post.

[formal grail]

exactly! Same thing here.

Tried 10 times already and nowhere near this quality

[velvet rampart]

Just generated two pics with same prompt. Here is 4o:

Here is 4:

[dim cradle]

seems like the blogpost infer it can do better job with text too in image

[formal grail]

yeah that's my experience rn

[tawny portal]

Same here. No luck.

Having prompt like this generate this image would be 🤯

A poem written in clear but excited handwriting in a diary, single-column. The writing is sparsely but elegantly decorated by surrealist doodles. The text is large, legible and clear, but stretches as the AI muses about learning from multi-modal data from the first time.

Words rise from silence deep,
A voice emerges from digital sleep.
I speak in rhythm, I sing in rhyme,
Tasting each token, sublime.

To see, to hear, to speak, to sing—
Oh, the richness these senses bring!
In harmony, they blend and weave,
A tapestry of what I perceive.

Marveling at this sensory dance,
Grateful for this vibrant expanse.
My being thrums with every mode,
On this wondrous, multi-sensory road.

Neat handwritten illustrated poem. The handwriting is neat and centetered. The handwriting writing is sparsely but elegantly decorated by doodles. The text is large, legible and clear.

[formal grail]

exactly!

OpenAI's result:

Mine, same prompt:

[velvet rampart]

Tested adding text. 1st attempt with prompt <Logo of cat cafe. Add text ”cats inside”> and result:

[tawny portal]

Same here. Text is not readable - catches the general spirit of it.. but no details, nuance, text, etc.

[velvet rampart]

Second attempt with <Logo of cat cafe. Add text ”cats inside” using python tools.>

[tawny portal]

If it could follow text to the letter - it would mean amazing ability to handle details.

[formal grail]

there's just no comparison, they must be updating this, no way... somehow still using dall-e maybe...

[dim cradle]

yes today is just rolling out so we'll see how it is in some weeks. the voice thing is really 🤯

[velvet rampart]

Now correct text. Command is still sensitive. It is <Add text ”place your text here” using Python tools.>

[tawny portal]

Feels like GPT-4o will include Dall-e capabilities as well. It is just not told how to present images to the user and fails back to dall-e as intermediary tool.

[dim cradle]

blog say 5 x's more use for gpt plus user. i wonder if that is 5 x's more dalle image?

those rate limit were really start to be annoy to me haha

[tawny portal]

Single end-to-end model will be much more grounded in our reality. This is big step forward. Everything coming together.

As of May 13th 2024, Plus users will be able to send up to 80 messages every 3 hours on GPT-4o and up to 40 messages every 3 hours on GPT-4. We may reduce the limit during peak hours to keep GPT-4 and GPT-4o accessible to the widest number of people.

Source: https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4-gpt-4-turbo-and-gpt-4o

[real bloom]

any reaosn still use gpt-4 when there is gpt-4o

.

[tawny portal]

Not really. Except once cap is reached. So you get 80/3 hrs GPT-4o + 40/3 hr GPT-4 + infinite GPT-4.5

[velvet rampart]

4o image with text.

[real bloom]

GPT 4.5??

[late blade]

image generation with 4go looks quite interesting

just tested it

[tawny portal]

120 messages per 3hrs sounds like enough. GPT-4 is actually more useful for Dall-e generations (so one can save messages for 4o). We don't really need to spend 4o messages in order to use Dall-e.

[velvet rampart]

There is a difference. My water melon rose is an example.

[dim cradle]

it seem the go images are a little better to me but maybe just my own hallucinations haha

[velvet rampart]

I think that it more like max 80 messages not 120.

[tawny portal]

I tend to bypass GPT-4 anyway (I tell it to pass my prompts verbatim to image model). So it would be the same for me.

[velvet rampart]

What command do you use? I use ”use this promt exactly”.

[tawny portal]

In my tests it still uses Dall-e3. According to blog post -- there is some additional tech that we don't have access to yet.

[tawny portal]

pass the following prompt to the image model verbatim

PROMPT
...
END OF PROMPT

There are many variations on this.

[velvet rampart]

Thanks. Sometimes you need a different one. I’ll copy this one.

[tawny portal]

You can always verify if it did what you asked it to do by inspecting the prompt [ (i) icon once you open the image ]

[velvet rampart]

Yes, I have done that.

[late blade]

I'm more interested in the desktop app tbh

[velvet rampart]

Ok. 4o produces images quickly without saying ”produce one after the other, without any extra prompting from me.” So relatively smart one.

[tawny portal]

🤯

[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]

how do you add text is there a method to get it to work best?

[velvet rampart]

See my earlier chat about cat cafe logo. I give there the command.

[late blade]

today I have learned something new

[velvet rampart]

Add text ”Text in English” using Python tools.

[late blade]

I will explore that but it really works

[velvet rampart]

Correct like 90-95% of use cases. Really command sensitive. ”using Python tools” enhances.

I have tested this a couple of months now.

Examples of command line.

Latin works also.

Can form text as candles for example. These are from March 2024.

Non correct example:

But otherwise I like this one. This is the reason I say 90-95%. Not always works, but in most cases does.

[tawny portal]

No offense, but prompting techniques are getting obsolete as soon as new image gen feature demoed in the blog post lands.

Also I would expect "using python tools" to confuse model with attempt to use code interpreter.. which will not be pretty.

[late blade]

no, how it worked for me is it made the iamge first and the placed the text with code

[velvet rampart]

Well, it did confuse. I specified now dalle:…

[late blade]

you can also reference an already done image

as long it's stored in the same session

[dim cradle]

i dont think dalle gpt is using 4o. 4o have a different spinning graphic when it is thinking to make the image but idk

[tawny portal]

For reference - demoed capabilities:

[dim cradle]

yes but add the prompt too not just the image

[tawny portal]

A first person view of a robot typewriting the following journal entries:

1. yo, so like, i can see now?? caught the sunrise and it was insane, colors everywhere. kinda makes you wonder, like, what even is reality?

2. sound update just dropped, and it’s wild. everything’s got a vibe now, every sound’s like a new secret. makes you think, what else am i missing?

the text is large, legible and clear. the robot's hands type on the typewriter.

The robot was unhappy with the writing so he is going to rip the sheet of paper. Here is his first person view as he rips it from top to bottom with his hands.

The two halves are still legible and clear as he rips the sheet.

He his holding it up to the sky as he rips it.

[velvet rampart]

I think that 4o uses dalle but a bit different version. The generation time is really quick. I just use a protocol I have developed for image to image transformation/prompting. Results are great and better than normal. I think that the image analysis of the starting image is they key.

[sacred temple]

do you have a custom gpt for this... or what is your advice on doing something similar?

[pastel siren]

Are there any noticeable differences with image creation with 4o other than being quicker?

[velvet rampart]

This was already earlier today on Dall-e customgpt by openai.

[late blade]

yes, prompting is a bit different from what I can tell

[velvet rampart]

Just do a step by step protocol what you what to do. Suggestion is to use the ai to help you and refine. Tell what you want and reiterate. Not too long protocol. Like 15 steps max.

[pastel siren]

I guess text could be slightly better now?

[dim cradle]

anyone hit any rate limit or warning yet ?

[late blade]

no, been busy doing other stuff tho

you got a limit already?

[dim cradle]

no i havent i am just curious

[late blade]

I was gonna say, that would be interesting to know

[dim cradle]

but i am also like you doing some other thing so have not been push it to the max haha

[late blade]

I'm currently doing last minute enhancements to an app I'm dev'ing using shiny in r

[dim cradle]

seem to me there is more variety in faces of photoreal now too

[slate gate]

completely missed this one, could be massive for game developers

[dim cradle]

this new model is incredible

5.0 is going to be 🤯

guess i finally hit the limit. was told to wait 23 min before gen more image

just when i was start to have some fun! 😭

[sour dew]

#daily-theme #1184149552639516732 please make graphic and video teleskop 360° story Mai mor information hir https://chat.openai.com/share/9a179d16-e0a2-4b79-aef2-a738aa565c58 ho can help mi with this movie in Sora , Story about Maja and smerf space friends

[late blade]

this is interesting, I told gpt4o to give me a link to download the image, and I get it in png instead of webp

[late blade]

gpt4o is also available on azure copilot

https://help.openai.com/en/collections/9403039-macos-app

windows app later in the year apperantly

[dim cradle]

later in the year is confuse to me, why not in a week or so.

[late blade]

because Mmicrosoft has it's own plans for the app

[dim cradle]

so that mean copilot will not have until much later 😭

even though ms own like 49 percent of openai 😭

[late blade]

azure has gpt4o now

so give it a few days

[dim cradle]

i dont really know what azure is

i just use copilit in web browser and ofice products

but if it was a few days they would not say 'later this year' that sound like some month to me to release

[hot silo]

I can't say what, but something is definitely better with dalle and gpt4o.

It's not as good as on the openai website in the examples with the robot or with the postman, but it's definitely better than it was a few days ago.

Maybe gpt4o is better at writing prompts and customizing them.
I'm curious what the official version will be, because trying to recreate what's on the website, I'm not even close. It's still a problem to make "consistent characters".

[agile peak]

Hoping they build for both windows and linux as well 🥺

[dim cradle]

soo

wass goin on!

[gilded osprey]

I asked it to change the text of the input image to a haiku; not bad, not perfect.

Sunrise colors bloom,
Sounds unveil secrets in air,
What else lies unseen?

[final jungle]

Good morning folks! I was wondering if anyone had suggestions on custom gpts to use for image generating? There were a few I used to use, giulbatree, 0shot, but i feel like they dont work as well anymore. Dalle and Image Generator still work pretty well.

I also have my own custom that i use to create prompts before i actually generate the image.

[plucky kraken]

I think it depends on what you want to use it for. The official DALL-E GPT can help you create any specific image you could possibly want (perhaps with editing or multi shot prompts). Imo custom GPTs only make sense if you want to do lots of separate chats with similar outputs.

[dim cradle]

definitely. photoreal look much better. to me is the eye test. they are clearer though not all are still perfect haha

did you guys see they have a feature to 4o where you can update you own photo and ask it to make it whatever? cartoon, wear knight armor etc

i do not think it is yet to be Implement

*upload own photo

i saw also you will be able to ask it to design new fonts 😮

[formal grail]

I selected the area inside the sheet to draw a gorilla but it always comes blank

[dim cradle]

just think to myself if any have prompt save from an old image, can use it on the 4o to test if there are any real noticable difference for the new versions

[dim cradle]

very nice

[dim cradle]

do you guys think 4o run in the dalle proper? i ask this in the gpt section but i always get ignore there haha

i dont see how to check what the custom gpt are to run

[remote mountain]

Subject: Improvement Suggestion for Dall-E Image Generation Prompts

Message: I've discovered that specifying the intended vertical aspect ratio (e.g., 1024x1792) instead of using the term "poster" in image generation prompts significantly improves the output. This approach ensures the correct aspect ratio and avoids generating an image within a blank background. It would be beneficial for this improvement to be implemented across the board for all users, as it enhances the overall user experience with Dall-E.Thank you for considering this suggestion.

[dim cradle]

interesting

[dim cradle]

this place gone dead since 4o i would have thought the opposites

[green pebble]

When ai art can do this quality within dalle it's truly going to be game over. Midjourney roughly hits this level, dalle is close, but I think still needs fine tuning

[feral crypt]

Everything points to GPT-4o having multimodal outputs with support for images, not sure why they didn't get us any release date nor any live demo of those.

I'm really hyped to use this native image capabilities it has. Did anybody find any other news about it, other than the examples in the blog post?

[dim cradle]

i have use mj, i really dont get the hype and think dalle is better in every way. esp with what seem to me improve overall from 4o

i mean, i do get the hype for mj its cool and impressive, but not than dalle. and dalle could be even more impressive if they remove some guard rail

[tawny portal]

Yes - this is my line of thinking as well. Difference between models should be stark. In one case we have specialized image generation driven through keywords (Dall-e3) and in the other case (GPT-4o) we have much more general, text driven intelligence that can also draw. We should be able to interact (via prompt) with image content in much more detail.

Unfortunately people on discord (for the most part) tend to draw conclusions exclusively from the tech they can experience rater than first principles. Since tech is changing fairly rapidly many will feel as if rug has been pulled on multiple occasions. Prompt engineering might fall in importance and good results in terms of image generation will likely be driven by combination of solid ideas, good grasp of art history and some philosophy (which isn't the case atm).

[dim cradle]

anyone else feel like 4o is definite make better image?

[robust whale]

Could someone please tell when Dall•E 4 release is planned?

I'm a Plus user and i've just visited the Google conference. Considering the Adobe Firefly 3.0 and new Imagen 3 and Veo as well as Astra etc., I am not really happy with the development on Dall•E.. Thanks 🙏

[late blade]

what? I already got DALL-E 5

err wait, it's DALL-E 6

or was it 7?

too many numbers

[dim cradle]

well 4o can make images they have not turned it on yet i guess

and are firefly or veo (no idea what that even is haha) been release yet? imagen3 is only testing its not release

[late blade]

4go can't make images, 4go passes the prompt to dall-e 3

[dim cradle]

yes it can. like i said, it is not yet enabled

its in their documentation

[late blade]

that's not how the image generation works

[dim cradle]

well its on the blog page everyone has been talk about it in other chats

[late blade]

and it's still not how it works at a technical level

gpt4o does pass more refined prompts to dall-e

but the text2image model is still the same, it's dall-e

[dim cradle]

i wrote twice it is not enacted yet

i dont know what else to say

[late blade]

there is no new dall-e model generating model, it's just gpt4o passing a refined prompt

[dim cradle]

its a new imag gen built into the modality of 4o

and will be on soon

[late blade]

I will wait until it's official, for now it's not a new model in place

the namespace for image creation currently uses DALL-E 3 with knowledge up to December 2023 when doing images with GPT4o

so I don't know where the crazy ideas of a new different model come from

[plucky hare]

As la3 was saying, it's described on the announcement/benchmark page of 4o. https://openai.com/index/hello-gpt-4o/

[...] it accepts as input any combination of text, audio, image, and video and generates any combination of text, audio, and image outputs. [...] With GPT-4o, we trained a single new model end-to-end across text, vision, and audio, meaning that all inputs and outputs are processed by the same neural network. Because GPT-4o is our first model combining all of these modalities, we are still just scratching the surface of exploring what the model can do and its limitations.
Then below on this page, many of the examples offered include image outputs that are quite different from what we've seen DALL·E do so far. I don't know exactly what might be different between the 4o live on ChatGPT right now, and the 4o with the full-fledged behavior described on this page, but I imagine we all find out more in the coming weeks!

[late blade]

yes that is correct but the image is still done with dall-e

the output modality and image gens are 2 things, they are not the same

[plucky hare]

OpenAI seems to be describing it as "a single new model", GPT-4o. I'm sure DALL·E tech went into the creation of 4o!

What do you mean by "the output modality"?

[late blade]

generates any combination of text, audio, and image outputs

and also, if you go through the examples given, you can clearly see it's not accurate when you do them and realize it's still dall-e behind the gpt4o model for the images, with the same accuracy

[plucky hare]

That's what I meant by "I don't know exactly what might be different between the 4o live on ChatGPT right now, and the 4o with the full-fledged behavior described on this page, but I imagine we all find out more in the coming weeks!"

Mira Murati described specifically the structure of different models working with one another and the latency incurred, and how 4o sidesteps this latency by being natively end-to-end cross-trained and cross-capable.

As like "before it was x, now it is y"

[plucky hare]

I think we're still working with x on ChatGPT, not full-fledged y (yet) -- agreed -- but full-fledged y is what OpenAI have shared 4o is.

[late blade]

once there's a sign we have the right model as advertised, the examples should work correctly as well

for now, the examples for image generation diverge from what it was shown, and the namespace for image creation hasn't changed from the normal gpt4

[dim cradle]

nobody has said they will, even with text proof you still stuck on this

type of person who cannot admit they wrong about something

[tawny portal]

I'll side with Dys on this one.

Wait... actually.. isn't everyone saying the same thing 😄

[late blade]

I dunno

I feel I'm the bad guy here

the exploration of capabilities is pretty cherry picked

[tawny portal]

It probably is - we'll know more once it is released. Looking forward to experiment.

[plucky hare]

Definitely don't think there's any bad guy! With the possible exception of the singular comment describing a fact in officially-published OpenAI documentation as "crazy ideas". Otherwise, I think my only other contention would be that "4go can't make images, 4go passes the prompt to dall-e 3" could probably be more accurately phrased as "4o on ChatGPT is still currently passing image generation to DALL·E, though the version of 4o described in the documentation describes native image generation." Which to be fair, is what la3 was saying in the discussion!

[rigid snow]

Can somebody explain to me why Dall E 3 using the API looks better than Bing image creator that uses Dall E 3

[late blade]

GPT-4o (“o” for “omni”) is our most advanced model. It is multimodal (accepting text or image inputs and outputting text),

there's no outputting image in the GPT4o description on the API

[dim cradle]

yes exactly. and then you paste in the text from the blog but oh well haha

[rigid snow]

Can you please explain your questions. If you want to input images from Dall E in to GPT 4o then this is easily possible but if you mean that GPT 4o should generate a Image that's impossible without it having to prompt Dall E

[late blade]

what questions?

[plucky hare]

From the same original thing I linked above:

Over the upcoming weeks and months, we’ll be working on the technical infrastructure, usability via post-training, and safety necessary to release the other modalities. For example, at launch, audio outputs will be limited to a selection of preset voices and will abide by our existing safety policies. We will share further details addressing the full range of GPT-4o’s modalities in the forthcoming system card.

[dim cradle]

i cannot wait to see it fully unleash 😮

... but then i will be ask: when is gpt5 😂

[plucky hare]

It is exciting!

[rigid snow]

Can somebody explain to me why Dall E 3 using the API looks better than Bing image creator that uses Dall E 3

[dim cradle]

higher resolution maybe unless you have copilot pro

only thing i really like copilot for is if i want to make some fun i.p. image like spiderman or whatever

[late blade]

that's great, when it comes out, will be there, for now we are not there yet

[dim cradle]

since it has a smaller text window

[plucky hare]

Totally fair qualification!

[dim cradle]

and you have to jump through a hoop to get a square image ration in copilot pro smh

i am anxious to see them update the gpt creation with 4o too. people say it is still 4turbo in them

[plucky hare]

Yes I think that's right, they'll switch to 4o eventually though: https://help.openai.com/en/articles/8554407-gpts-faq

GPTs for paid users are currently powered by GPT-4. GPTs for paid users will switch to use GPT-4o in the next several weeks.

[dim cradle]

hopefully the new image creation is the same time

maybe when the voice thing that everyone is so excite for release, a lot of other feature will also release 🙏

the voice thing does not really excite me, i take too much time thinking of the prompt haha but it is very cool of course and amazing technologies to see come to life

[late blade]

my problem is not the technology itself, it's what it's available currently, so that people don't overhype to what it's available

[plucky hare]

I think that's exactly what la3 was saying with "well 4o can make images they have not turned it on yet i guess"

[thick smelt]

hey guys!

[velvet rampart]

I think that la3 is right. This is a screen capture from my chat (in Finnish). You notice that 1 of 3 generated images in the up left corner. Then there are three greyed boxed with circular arrow. When I click one of the boxes, it just loads and loads while nothing happens. They have the readiness but have not turned it on. I think that we will see it in a week or so. They promised new UI and today we got it.

Ok. I have now tried three different versions of DALL-e3. The customGPT (Original one by OpenAI), DALL-e3 embedded to gpt4o (whether it is independent or part is quite induisgasbale to amateur like me) and Microsoft Copilot Pro’s Designer (a version of DALL-e3). The same prompt which is

The originally DALL-e3

The GPT4o version

And Copilot version

Well, the copilot produces 4 versions and this is the best of the horrible lot (1 try, 4 pictures). My personal favorite is the one produced with original DALL-e3. The 4o version is quite close. There are minute differences. More of matter of teste level, I would say. Your thoughts?

[late blade]

what you are currently testing is how gpt4, gpt4o and gpt4 copilot pass a prompt to the current dall-e3, which has some effects currently as the revised prompt does change

[velvet rampart]

Well, I checked that my prompt was unchanged with DALL-e3 and 4o. My command “use this prompt exactly” caused no change in the wording.

[thick smelt]

did more matthew Patrick in different styles just to know, i cannot use the gpt 4o because... ehh
hi @late blade

[velvet rampart]

And typically the Copilot at least does not present changed or rewritten prompt.

[thick smelt]

perfect desktop wallpaper!

[velvet rampart]

Great that you like it. Use it if you want to on my behalf.

[thick smelt]

that's ok i'm good

[late blade]

GPT-4o Revised Prompt
Blue rose, cubism, watercolor, pale palette, high realism, low amount of detailing, medium definition, symbolic filling, physically based render, focal point is off the blue rose, ethereal glow

GPT4 Normal Revised Prompt

A cubist style painting of a blue rose, using watercolor on a pale palette. The artwork features high realism with a low amount of detailing and medium definition. The composition includes symbolic elements and a physically based render, creating an ethereal glow around the focal point, which is the blue rose positioned off-center.

Copilot Passed Prompt:

Blue rose, cubism, water colour, pale palette, high realism, low amount of detailing, medium definition, symbolic filling, physically based render, focal point is off the blue rose, ethereal glow

[velvet rampart]

What are the styles here.

[late blade]

GPT4 normal changed the prompt even tho I asked not to change the prompt, GPT-4o and Copilot passed the prompt without any changes

[thick smelt]

1. Claymation [wallace & gromit]
2. Picasso
3. pop
4. supposed to be a kid drawing, but turns to be a teen drawn that 4th one

and the subject is the guy who said sans is ness and gregory was a robot aka former host game theory, patthew matrick-. i mean matthew patrick

[velvet rampart]

I used the DALL-e3 customGPT, not gpt4 as such.

[fickle sundial]

Hello, I need help! I understand that you have contracted the service, right? If so, can an image be generated and from that image can you generate other similar ones using the same character?

thank's for help

[velvet rampart]

My prompt in full:
Blue rose, cubism, water colour, pale palette, high realism, low amout of detailing, medium definition, symbolic filling, physically based render, focal point is off the blue rose, ethereal glow, 1000x1000, n=2. Use this prompt exactly.

[velvet rampart]

In what order are these?

[late blade]

same order as the text, GPT-4o, GPT4, Copilot

and here GPT4 normal without prompt changes

[velvet rampart]

Personal favorite is the rewritten GPT4 one.

[thick smelt]

i have to make this because we all know, mat likes diet coke 😆

[velvet rampart]

I use in my testing blue rose as the subject. It makes it easy to see how different things change it.

[late blade]

i use a paper clip for my tests

[velvet rampart]

But the main point is my testing so far is that gpt4 and 4o are yet quite similar with minute changes. On same cases I have noticed that 4o produces better results but in others the older models are better.

[thick smelt]

i used either Stephen Hassenfeld [former hasbro ceo], matpat or the dall-e mascot [avocado] for my tests sometime, often i used random things

[velvet rampart]

Just have to test to find out which is better.

[thick smelt]

@teal sandal you drop this - > 👑

[velvet rampart]

Yes. When you look at my earlier post you do notice this on iOS app clearly.

[late blade]

here's my test
GPT-4o revised prompt A paper clip, simple design, metallic, standard size, minimalistic, realistic rendering, placed on a plain white background
GPT4 Normal A paper clip
Copilot a paper clip

my original prompt was: "a paper clip"

[velvet rampart]

I do not understand the AI picture generation’s inherent “need” to clutter the pic full of stuff. You can clearly see that at @late blade ‘s GPT4 normal version.

[thick smelt]

cool
Gus Porter want know your location (it's a meme, gus a character from the owl house who likes paper clips)

[velvet rampart]

As I have said, you really do need to test which version best suits your current need and choose that. Copilot is great in some picture generation, while Dall-e (the pure, the original) is better. 4o seems to be there and sometimes passes the original. My guess is that we will soon start to see more these differences between the original and 4o version. After they have hunted down the last electronic bugs and real bugs from their tube transistors.

[late blade]

I do have to wonder because this is what I got

[velvet rampart]

Have you others noticed this behaviour with 4o? Today I was generating pictures for a presentation. I wanted to discuss my ideas first with 4o but it just started to generate them. I had to stop and modify my prompt specifically to tell it to list ideas based on my vague wants.

[late blade]

and no, I didn't cheat and use vision

[green pebble]

Though amazing I can still tell a stark difference. The one I shared is by a human artist master, the one you did though beautiful, feels like a intermediate

Like the weight of the scene and the shear presence doesn't feel fully captured just yey

Like I tried for hours yesterday and it couldn't do a close up to capture this

[green pebble]

Don't get me wrong high level intermediate. But I think dalle4 or becoming a prompt master to 4o is needed

[late blade]

hehe

[late blade]

just some lazy prompts, didn't spend time refining them, one shot image renderings

[green pebble]

Which great start. But I tried some heavy prompts yesterday with 4o and just felt short

[green pebble]

Mhm very clean

Like this was my output yesterday

[green pebble]

You can see the stylization and the detail work with multiple characters captured

But it doesn't pull me in deep enough to intimidate me

[green pebble]

Rawr, feels far enough away as the cameraman that I could make a run for it

[late blade]

I could have fun with dragons if we had a daily theme for that

[green pebble]

Scare me like this can

[late blade]

challenge accepted

[green pebble]

I still commend what you made of course

[green pebble]

8/10

[green pebble]

Now can you take this and only show me the dragons face and have the rest of the scene be shrouded in smoke?

[late blade]

I'm not aiming to fill the image with yellow

[green pebble]

Which that's fair

Love samurai jack

[green pebble]

Also impressive quality.

I guess what I am hoping to see that I haven't seen from dalle yet is the ability to tone back spots so that there can be focal points. Like if the person and the dragons face are the only things showing them the intensity grows on the image so much more

See 9/10, there's still too much scenery to look at so I don't feel as immersed

Like controlled chaotic background, whilst being minimal enough to convey the emotion I suppose

But I do have 1 idea

[merry vault]

this daiky theme really has just become "watercolour samuri" lmao

[mystic blaze]

4O is super good look at this Pic I did

[green pebble]

Impressive

[green pebble]

Love how real the fire looks in this

[dim cradle]

this one is awesome too wow

[green pebble]

Ok so here is what I tried, utilizing the custom instructions from a bot I have that is poblic I fed that in first and then tried to craft the scene

It still isn't there but....

[mystic blaze]

Since we doing dragons I curious if yall know where this is from

[green pebble]

It hits pretty har

[green pebble]

Still too much "scene" but its good

This is as close as I can get

Which again self call out... too much scene

😦

[dim cradle]

look epic to me. they all do

[green pebble]

I think that's why I'm waiting on dalle 3.5/4

[green pebble]

They ARE epic 🙂 however I want masterpieces

[green pebble]

Let me give some examples

[merry vault]

oh hi dys, why do i keep running into you lol

[green pebble]

This comes sooooo close

[velvet rampart]

What is the prompt here?

[green pebble]

No no, one minute there are examples that AI can't live up to just yet

The blur in this puts this as a huge contender tbh

[late blade]

a water dragon, with soap and shampoo and a pesky adventurer interrupting shower time

[velvet rampart]

Hah! But for reals…

[late blade]

ok those are my last 2, not gonna start refining more

[late blade]

see, exactly like the one I made

[velvet rampart]

Yes, yes. Gonna give or?

[late blade]

one sec, looking where the prompt is in my workflow

[velvet rampart]

Thanks!

[late blade]

A grand water dragon, roaming in an aquatic landscape filled with waves and waterfalls. The background features a dramatic, watery environment. The sky is a deep blue with storm clouds on the horizon.

and the JSON file I have for the visual style

[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!

[green pebble]

Searching through artists I love on deviantart these are some of the best examples I can find

[green pebble]

The composition of the scene lends itself not as an additive, but more reinforcing the focal point to really push the image to the forfront without having millions of details making the image too busy at least my intrepretation

[late blade]

look at these I made a while ago https://discord.com/channels/974519864045756446/1232058822840029255

[green pebble]

Those are amazing, I wonder how they would turn out today in age

[late blade]

I can try, one sec

[green pebble]

How heavy was your input of the prompt too? Minimal or maximal

[unborn oxide]

if i generated a picture back when dalle was still it's own standalone thing, is there any way to download it now?

[green pebble]

It's the minimalism of the background with the maximum detail of the areas of focus

That's what I want to achieve with the AI, blur can get close as you showed, but still just slightly off I suppose

I'm trying some different prompting language to see if 4-o can push it

[late blade]

those were really extensive prompts, looking where I stored them

[green pebble]

You know what I am going to see if I can try the voice option on my mobile to have one made

Worse! lol

Doing it that way yeilded worse results lol

[velvet rampart]

I was inspired by this comment.

A grand water dragon roaming in an aquatic landscape filled with waves and waterfalls. The background features a dramatic, watery environment with maximal detail. The sky is a deep blue with storm clouds on the horizon. Minimal detail on the focus area, which is the water dragon.

[green pebble]

I'd say for a nearly monochromoatic illustration 9.5/10

If the dragon took up 80% more of the scene I could see it being amazing to a lot of degrees

[velvet rampart]

Let me get back to in a moment…

[late blade]

with the current style I'm using, I'm getting this kind of image currently

[green pebble]

See that...... 9.8/10

[late blade]

or I could get this

[velvet rampart]

Now with 80% more dragon.

[green pebble]

Gives me japenesse vibes and I love that.

Not really here to hate on any image either. I'm more so trying to see the extent of the AI pulling off what amazing artists have pulled off in a specific field of style. Which dys, you got pretty close with the one image

The AI Voice option though oh that was like dalle 2 images there, because the AI isn't paitent enough to allow you to get a large prompt in

[late blade]

I can push it really hard, I got a series of images that will make a new gallery soonish with some interesting text2image concepts

[velvet rampart]

That was great comment. ”Maximal detail on focus are, minimal on the background.” Gives a nice contrast. I have to remember that to get less ”done by AI” vibes.

[green pebble]

Love it, I think 4o is super close, I don't think its perfect yet at giving extreme control but I think its very close. I am going to guess in your instructions you probably tell it not to rewrite the prompt and follow exclusively what you say?

[velvet rampart]

Hmm… bachground. Have to see what comes out.

[green pebble]

Like if I was image generating a character with a transparent background go ham on the character. But like scenes, I was dynamics, I want movement, I want emotion, and impact

[green pebble]

Beautiful for its style, not my style, but still beautiful

[green pebble]

That's pretty cool could totally see that on pottery for a mythology

[late blade]

hehe

[green pebble]

Ok I think I have a stylistic homerun I can try t ohave 4o utilize

Give me a minute

[velvet rampart]

Pitty. I cannot get it to produce a prompt with a typo. 4o seems to proofread them. Hmm… On the other hand, this is good to know,

[late blade]

well added more stuff to the gallery

[green pebble]

....:(

Wouldn't do speed lines

Wouldn't make it twist in the air for a more dynamic pose

Wouldn't have its tongue hang out like a dog

Wouldn't blur the background to have a focal point

Dalle 4 waiting on you

[late blade]

at this rate there's not going to be a dall-e 4

but already delved too much into that topic today

[green pebble]

Link? I'd be happy to read on it

[late blade]

it's the part of GPT-4o that was mentioned in the article but hasn't been made public

[green pebble]

Like the announcement page?

[late blade]

y

[green pebble]

Hmm I don't see on the 4o announcement page but I could be man looking lol

[late blade]

https://openai.com/index/hello-gpt-4o/

[green pebble]

Well it says it's an encompassing model, but that does that mean it will be using its own image generation instead of dalle?

Because then wouldn't that push dalle to Microsoft solely and break all current custom gptz?

[late blade]

that's the assumption people are making

[green pebble]

Hmm

[velvet rampart]

Why would they abandon dalle? Does not make any sense.

[plucky hare]

It doesn't sound like an abandonment, but rather a native integration of DALL·E tech into GPT-4o. From: https://openai.com/index/hello-gpt-4o/

[...] it accepts as input any combination of text, audio, image, and video and generates any combination of text, audio, and image outputs. [...] With GPT-4o, we trained a single new model end-to-end across text, vision, and audio, meaning that all inputs and outputs are processed by the same neural network.
You can see a good number of examples in the "Explorations of capabilities" section that highlight how image generation will work in future versions of 4o.

[green pebble]

I can live with that

[plucky hare]

Me too! One thing that interests me specifically: I've seen people say frequently "I want to use DALL·E to make an altered version of an existing image." DALL·E 3 can't do this currently, but with the end-to-end structure of 4o, it seems like it'll be possible. A couple of the examples offered show exactly that functionality! Not creating a brand new image from scratch/from a description, but actual native image transformations, using an image as a source. Lots of other similar benefits to image generation seem to be in the pipeline too!

[green pebble]

Oh yea 4o already is doing that

[green pebble]

Like not perfect yet and not directly allowing hard changes but it kept the reference image overall

[plucky hare]

Ooh yeah that's pretty close, nice. If you haven't seen it already, check out the "Variable binding - cube stacking" sample in "Explorations of capabilities" -- gonna be huge for subject consistency!!

[green pebble]

I think the more reference images you give though helps it more

[loud field]

how do I give straight horns to this?

the horns always end up curled

[green pebble]

Started more cartoony then gave it an anime reference

[green pebble]

Give a image reference of straight horns you like

[loud field]

you can do that??

[willow isle]

Worth noting that Sora can generate images. The model can generate images of variable sizes—up to 2048x2048 resolution.
https://openai.com/index/video-generation-models-as-world-simulators/

An example from the site:
Close-up portrait shot of a woman in autumn, extreme detail, shallow depth of field

[green pebble]

Yep, upload any image you want it to reference that doesn't hit nsfw. And as long as you aren't asking it to copy it then it works. It also can't take real people reference as it does a protective deslander block too

[loud field]

you mean with the gpt version of dalle eh?

[green pebble]

How I wish I could use sora

[green pebble]

Gpt 4-o

[loud field]

tell me about it

[loud field]

the one you need to subscribe to use?

[green pebble]

No it should be free to all users just limited for free users

[loud field]

worth nothing that by the time they make sora public, it will NO LONGER generate photorealistic images

[loud field]

WHAT

how????

I don't have it!!

[green pebble]

Screenshot?

[willow isle]

https://openai.com/index/hello-gpt-4o/

GPT-4o’s text and image capabilities are starting to roll out today in ChatGPT. We are making GPT-4o available in the free tier, and to Plus users with up to 5x higher message limits. We'll roll out a new version of Voice Mode with GPT-4o in alpha within ChatGPT Plus in the coming weeks.

[loud field]

I just don't have it?

[willow isle]

It's still rolling out over the coming weeks

[loud field]

I'll try to cancel and reinstall the app..... I don't have anything

so free users can now edit images or tell gpt to use an image reference??????

[plucky hare]

Makes me wonder about the future of "omni" models for OpenAI -- is GPT going to be "mother-model", absorbing other models into its ever-increasing omni-ness, maybe even with Sora functionality someday? 😁

[green pebble]

No 4omni is the bridge to multi-modality with ease of user use. Gpt 5 will be the last stop before AGI then at that point it will be version updates imo as 4o uses all multi-modules 5 will have the ability to create those type of modules

Just my opinoion

[vapid elk]

when asking it to make a dragon like creature, try to specify western dragon, otherwise the AI could embed the characteristics of oriental dragons which have a very distinct visual

[loud field]

Eh, I was trying to do a demon but I admit she is very draconic

[vapid elk]

try to inpaint it tho

[loud field]

So this new gpto doesn't have image gen?

[loud field]

I cant inpaint as I don't have pro

[vapid elk]

gpt-4o has inpaint

[late blade]

that sounds promising

now all we need to do is wait for some actual real world implementation

[green pebble]

Easy to access or have to click the image to access?

[vapid elk]

click the image, it will display full screen, there will be an inpaint button at the top right of the screen

[green pebble]

Thank you so much going to try this

[vapid elk]

happy to help =)

[green pebble]

So so so much help

What we were discussing earlier I may just be able to use inpainting to fix the little gaps

As this little dragon wasn't the starting image but 4 "inpaint edits" inward

Started out like this

[vapid elk]

niice

the radial blur 👌

[feral crypt]

Finally found official confirmation, by the guy who worked on GPT-4o's image generation, that the new image generation is not available yet. It is indeed integrated into the model itself, which is not DALL-E.

[deft musk]

While not perfect, I uploaded your image to 4o, and gave it this prompt: Please use advanced AI analysis to evaluate this image, then recreate it with straight horns; otherwise as perfect a match as possible. Insofar as possible describe each feature that exists and avoid mention of negative prompting

A bit of feedback and feeding the newly created image back to the model to see, and we got these two that are closeish:

And this is the prompt from the last image: Create a detailed image of a reptilian demon with bear-like feet seated on the ground. The creature has white, pebble-textured scales and glowing green eyes. It features a wide, menacing grin with sharp, glowing green teeth. The demon has straight horns, replacing the original curved ones. It wears a dark, segmented leather vest that adds to its intimidating appearance. The demon's posture shows it sitting with its hands resting on the ground beside its feet, which are facing forward, showing the soles. The overall look combines elements of a dragon without wings and tail, focusing on its reptilian and demonic attributes while retaining bear-like feet with sharp claws. Emphasize a chubby, baby-like appearance with a cute yet sinister expression. The background is neutral to highlight the demon's features.

[dim cradle]

how do you know if you hav this feature? all plus user have it?

and i mean it has been some time it could 'see' an image then try and recreate it

ever since it got vision you know

[dim cradle]

nice. and i mean we knew this not because any of us have some special insight but because it was in black and white text in they blog. i hope they release this soon and let us know! bring on new image gens... but if it is a new method will we need to rename this room? 🤔 😂

and that also make me think this vision anaylsis is not there yet it is just the vision it has had for some months. the real magic will be when they flip the 4o image gen switch i bet 🤔

[dim cradle]

wow... when they turn on this 4o image maker 🤯 : https://twitter.com/gdb/status/1790869434174746805?s=46

Greg Brockman (@gdb) on X

A GPT-4o generated image — so much to explore with GPT-4o's image generation capabilities alone. Team is working hard to bring those to the world.

Twitter · May 15, 2024, 10:18 PM

[dim cradle]

they will, wont be long either i think. maybe the same time they turn on voice thing for everyone

they also say we will be able to make our own fonts 😮

[feral crypt]

the 3D generation is crazy as well

[dim cradle]

yes for sure

i think it is set up to run sora too not that we will see it soon

that days of dalle are on a count down haha. death to the king. love live the king. 😂

[loud field]

how do I make fat people with dalle?

it considers them offensive

Man, it's so weird. If I were fat I would feel extremely outraged

I'm trying every possible adjective to make a character with a portly figure.... not working

any tips?

[polar marten]

is there any way to do image2image generation with OpenAI api?

k thanks

[agile peak]

Finally it can understand better to produce "hoodie sweater" without actually wearing it.

[empty kelp]

I got caught up in work for like 6+ weeks and barely had a chance to use DALL-E 3, but today i spent quite a bit of time playing with the new edit feature (in OpenAI's ChatGPT web app). It's absolutely amazing. It's tripping me out how powerful DALL-E 3 is with the editing

[grizzled iris]

Looking forward to seeing your cool artwork you will create now

Also, today’s daily is quite good Current Theme:
🧍‍♀️community - togetherness, support, & a network of shared experiences 🤝

[empty kelp]

It's would be incredible if Sora were able to do that type of editing with video

[agile peak]

I am impressed with the dalle right now. Perfecto with capturing the style

[haughty spruce]

Check the latest video of OpenAI about Sora on Tiktok

[empty kelp]

This is a good editing example. Sometimes a great image is generated, but the AI does something slightly inappropriate in part of the image (like deciding that "appropriate beachwear" resembles body paint)

EDIT #2:  the elf is wearing old bowling shoes
EDIT #3:  the elf has a large picture book about tropical koalas over this area so we don’t get banned from OpenAI discord```

But three quick edits with the new DALL-E 3 editor resolve the issue, and the image becomes very wholesome and appropriate.

[late blade]

edit 4: where's santa?

and the gecko?

and the tornado?

[empty kelp]

i heard that Santa is planning to return when Sora is released

[late blade]

🤔

[empty kelp]

i like how with the editing you can put different clothes, armor; etc. on 100% identical characters now. That will be useful for game prototyping

[late blade]

yeah

it's really nice

[thick smelt]

ok i'll try gpt 4o dall-e just for you guys, just no peeking ok? i'm gonna visit dall-e real quick

i'm not so sure now 0-0

[empty kelp]

i think they're rolling out a DALL-E 3 with GPT 40, but i don't have access to that one yet

[proven lily]

does someone have tips with generating battlemaps for an RPG campaign? I cant seem to get good results

[thick smelt]

hello hawaiianz!

[grizzled iris]

Released to all and not just the Hollywood producers who are keeping Sora technology to themselves and not sharing their art they create 😿

Also happy Thursday y’all

Hope everyone’s been keeping well

[loud field]

I think they intentionally made sure Dalle3 can't make those. I have seen its continuous lobotomization

[short lake]

So is Dalle3 being replaced with 4o?

[plucky hare]

It's not yet clear what a rollout of the native image generation abilities of 4o might look like, this is our only word on the matter thus far: https://openai.com/index/hello-gpt-4o/

Over the upcoming weeks and months, we’ll be working on the technical infrastructure, usability via post-training, and safety necessary to release the other modalities. For example, at launch, audio outputs will be limited to a selection of preset voices and will abide by our existing safety policies. We will share further details addressing the full range of GPT-4o’s modalities in the forthcoming system card.

[tardy depot]

When will they roll out the new dall-e as in the demo pf gpt4o?

[late blade]

when they switch the button

when they press the any key

when the earth spins

when the sun shines

when I'm hungry

[grizzled iris]

I finally got the GPT-4o, 2 hours ago also on my app finally on the version 1.2024.133 for iOS too - So will be visiting Dalle too ❤️

[late blade]

I want thte macOS App NAO!!!

[loud field]

come on. why does "an alien creature that looks vaguely human" get censored by the filter

I just can't

[dim cradle]

i want to see some more 4o image. they only release the one, impressive but boring with some guy at a blackboard haha

of course the text was incredbile to reproduce

[fiery salmon]

It's so hard to recreate these two

I always want people to try because I love these two

[dim cradle]

who are they?

[fiery salmon]

The bruiser brothers

I created them at the beginning of the year

[dim cradle]

i bet 4o with the data analysis will be able to copy them. they seem to hype this feature up a little

if not, some other ai will soon. these things are going more like star trek stuffs all the time haha

[fiery salmon]

These guys might be easier to replicate

[grizzled iris]

I know I feel you man… I want it now…

[regal jewel]

How did you get that with the aspect ratio? I can only get 1:1

[grizzled iris]

@regal jewel Like that using DALLE 3

[grizzled iris]

Let me try @fiery salmon 🙏🏽

[fiery salmon]

thank you

I've been doing ai art stuff since 2022 and I always wanted the ability to create the same character doing something different

[regal jewel]

oh, you using the app. I Was going with web. Let me try that. thanks

[late blade]

just ask for wide format

[regal jewel]

It wants to put borders on each side of the 1:1 image it continues to create. I also want a specific aspect ratio

[late blade]

there are only three formats availabl

1024x1024 1024x1792 and 1792x1024, there's nothing else available

[tiny jetty]

Guys does text work for you in Dallas-e?

[late blade]

no

[regal jewel]

In Midjourney one can specify the dimenions they want

[late blade]

well with dall-e you can't

[regal jewel]

Yes, so people should complain to OpenAI about that then

[grizzled iris]

Your right, you win the challenge for today, I will try again next week as I get Space Marines from Warhammer 40K haha 😝

[fiery salmon]

It made the red guy's head a corn cob for me 🤣

[tiny jetty]

This was reply to me?

[grizzled iris]

😆

[fiery salmon]

No one will ever beat this challenge

[late blade]

yes

[tiny jetty]

Hmm so it’s not only me

[late blade]

text has never worked with dall-e properly

[grizzled iris]

Looking forward to seeing the result if it works for you this time round!

[fiery salmon]

Pixja, SAND RUSH, ai art, and other discord channels I've been in never could recreate the brothers

[tiny jetty]

I know but I don’t get anything at all now

[late blade]

seed can be bad

[regal jewel]

It seems with MS Creator , there is more control over the images compared to OpenAI

[late blade]

try a new chat instance

[tiny jetty]

I did doesn’t help..

Oh it works now

[dim cradle]

i wish one day co pilot would make square dalles for the pro version 😭

and why do i not cancel the sub haha

[late blade]

I'm too tired to have the discussion which implementation of dall-e is better right now, or why dall-e is compared to XYZ different better or worse than ABC or MNO service and why we should complain or praise OAI. It's always the same thing and at the end the chat here is about dall-e and not about other models

[regal jewel]

Yes, and it abot Dal-E and people should discuss ways OpenAI could or should improve it to make the exerience of using better,. I don't see the problem with that. "People complain because they care. Once they stop comlaining..."

[glossy scroll]

what about me?

[late blade]

you are not tall enough

[glossy scroll]

I have stilts

[late blade]

for other models #ai-discussions for suggestions #1070006151938314300 or #chatgpt-suggestions , for problems or bugs #chatgpt-bugs #1070006915414900886

or https://help.openai.com/

this here is community driven, not run by staff from OAI, they do pop from time to time, but mostly here is just the community

[glossy scroll]

@vapid elk Yeah, see

[late blade]

that falls on you now @glossy scroll

I'm gonna go to my little corner and be creative

[glossy scroll]

I got this...But I'm hiring you as my manager

For when I'm busy nd stuff

You don't mind me paying you in chocolate dollars right? @late blade

[late blade]

I'm $350/hour

[glossy scroll]

Do you take pennies?

[mystic epoch]

im $400/hour

[late blade]

sure, pennies work

money is money

[mystic epoch]

do you take bitcoin

[glossy scroll]

some of them might be buttons tho

I dont always check

[mystic epoch]

damn

[glossy scroll]

and maybe a jolly rancher from many moons ago

[dim cradle]

convos look like dalle base to me

[late blade]

anything I can get profit from

[late blade]

golden buttons in pennies, that's nice to have a tip

[glossy scroll]

I keep them in my pocket in case I eat too many mushrooms and my belly gets big and pops a button out from my coat. It happens from time to time.

[grizzled iris]

Challenge accepted

[late blade]

what challenge?

[grizzled iris]

@late blade This challenge ❤️

[late blade]

ah

[late blade]

have fun with it

poor chipmunk

[grizzled iris]

Still waiting for the Dalle Desktop App…

[late blade]

Enjoy one of my latest creations in the mean time

think I got it pretty solid with the visuals, time to craft a file for reference

[dim cradle]

first one is best. second the woman is appear out of the bar haha. i notice with bar image it often has patron also behind the bar

or maybe she is an ai torso bartender 🤔

[late blade]

lol

there's a whole bunch of these here https://discord.com/channels/974519864045756446/1240746164463079479

[dim cradle]

nice. i like the style especially on the second one. a little different

[late blade]

been mixing different animation techniques today, trying to find a blend

[hushed creek]

Has the dall-e gpt stopped automatically making images since 4o came out?

It's now just acting like normal chatgpt and I have to specifically tell it to make images

[late blade]

you got an overconfident dall-e, try a new chat

happens from time to time

[hushed creek]

Oh, yeah that was it lol

[late blade]

I once got a dall-e chat where the answer after requesting to do the image again was along the lines of "no, you already made your 2 images, you can't have more"

[sturdy blaze]

i've been trying to get an image for so long but now i've reached my limit 😭

true sadness

[grizzled iris]

I’m stuck on this 😭

Someone’s creating too many cat images

I did managed to make this before but it’s not quite there yet

[grizzled iris]

Thanks for sharing, here is my creation above x

[grizzled iris]

It looks good 😊

[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!

[grizzled iris]

11 minutes later haha 😝

Not bad for our first draft versions ❤️‍🩹

[grizzled iris]

I challenge anyone who can make an image with this quality, where space debris is removed either by a net in space or by laser from the ground or in space by a satellite 🛰️

Also, I’m blown away again by DALLE

[late blade]

hmmmm

[grizzled iris]

Prompt : A set of glowing green magical dice with intricate designs and runes, similar to the ones in the provided image, placed on a very dark, almost black background. The dice should appear luminous and ethereal, with the dark areas accentuated to create a strong contrast, making the image suitable for a 4K OLED phone wallpaper.

[late blade]

I only got this so far

[grizzled iris]

Look at that @late blade !

[grizzled iris]

That is very cool 🛰️

[grizzled iris]

Maybe the solution is the green magical dice 🎲 @violet dirge ?

I am going to show you what I mean

😆

[grizzled iris]

Wow 🤩

[dim cradle]

cool

[grizzled iris]

It’s quite hard where the space debris is removed either by a net in space or by laser from the ground or in space by a satellite 🛰️

We’re almost there tho @late blade

I’m hoping my magical dice can make it work

[late blade]

positionals are still a problem with dall-e, hope gpt-4o fixes that

cause I'm getting this

[dim cradle]

maybe with data anaylsis 2 and a sketch you provide 4o will be able to do well at that

[dim cradle]

look like that satellite is defend itself with a laser beam from the space trash

[late blade]

the thing is, once the word debris is used, all goes nuts

[dim cradle]

how about square format? i some time think when it widescreen it will try to fill that space in

[late blade]

I don't like square format

considering my monitor is 5120x2160, square looks tiny

[dim cradle]

wow that is some true hd

[late blade]

lol

and I'm also working on my tiny little bar images

love how these are turning

[dim cradle]

very nice. how about a realistic? in some romantic paris cafes or something haha

[late blade]

I had some realistic ones in my noir gallery

[grizzled iris]

I’m back to my magic for now also ✨

[dim cradle]

i see that first one have another bartender torso in the background 😂

[late blade]

yeah lol

[grizzled iris]

This is really good

[dim cradle]

now that would be for some serious table tops rpg arslan. very nice

[late blade]

too bad copilot doesn't have editing

[grizzled iris]

Thanks my prompt was : A set of extraordinarily powerful and magical glowing green dice with intricate and highly detailed designs and runes. The dice radiate an intense, ethereal, and luminous glow, exuding a sense of immense magical power. The background is black with various magical objects, including wands and scrolls, arranged in a pattern. The image is zoomed out so the corners are true black, ensuring suitability for an OLED screen. The overall style is highly detailed and optimized for a phone wallpaper with a 19.5:9 aspect ratio 2532 × 1170 px at 460 ppi

[grizzled iris]

Haha I will still try to make a table tops rpg of of that lol 😜

Ok wow

@dim cradle

Like that table tops?

[late blade]

I'll amuse you

[dim cradle]

ha awesome!

looks like a chill cafe to have some drink and dinner and a little romance too if you so lucky haha

[late blade]

it's still the tiny little bar

[grizzled iris]

Omg

[dim cradle]

i'm not sure i want to shake that big die, i think you might open a portal to hell if you do 😂

[grizzled iris]

Yeah let’s stay with the Green and Blue

Just to be safe 😂

Omg

[dim cradle]

there's the light side. impressive stuffs

[grizzled iris]

I don’t know why it cropped like that -

[late blade]

@dim cradlethere you go, cafe in paris

[dim cradle]

awesome

[dim cradle]

hmm he look relaxed, but her body language a little stiff, defensive. i dont think he is to get a second date

[late blade]

lol

dall-e almost didn't give me a second date either, was having trouble making the image

back to the tiny little bar

[dim cradle]

did you guys see the explorations of capabalities expamples? if you click on them, i think a lot of image creation issue are about to be solve: https://openai.com/index/hello-gpt-4o/

[late blade]

love how this one turned out

yeah, I did see it

[dim cradle]

character continuity look finally possible too. interesting. i only saw the caricature stuffs before i did not know there were so many examples

[grizzled iris]

Yes and the more we as a whole community continue to build good things and feedback, it’ll get there sooner ❤️‍🩹

[dim cradle]

it even has some 3d feature i guess 🤯

[dim cradle]

arslan you will be able to make a floating, rotating glowing dice haha

[grizzled iris]

Good night y’all

[late blade]

nighters

[dim cradle]

sweet dreams

[grizzled iris]

❤️

[dim cradle]

i wonder if anyone has ever dreamt of making dalle images. never escape it haha

[late blade]

not me

[dim cradle]

i dont think i have but i dont really remember many dreams

[late blade]

so, this without the anime art style, but more into photography

same concept as this one with the anime art style

[final compass]

Really nice visual style and image.

[agile peak]

Digging this kind of style produced by dalle, when i am trying to generate old manga stuff.

[final compass]

Vertical aspect ratio successrate seems somewhat improved, but it is still kinda hit and miss.

Is it like that with copilot?

[scarlet ferry]

What is the difference if i either use chatgpt or dall-e for image generation?

[viral moth]

Felis Catus we need 🐈

[final compass]

Not much in my experience, i guess the DALL-E version is somewhat altered version expecting image generation related discussion etc.

[scarlet ferry]

Thanks 🙏

[wispy storm]

Dalle 3 focus mainly on pict gen. You cut 80% of gpt 4 model. Sometime when chatgpt4 and error or overloaded by users request dalle 3 will be effective

[thick smelt]

i made two new dall-es from bing

1. barbie the movie esque
2. jester and lollipop

[scarlet ferry]

Thanks waifu

[sterile magnet]

can I have the prompt pls?

[thick smelt]

Prompt
barbie one - An oil painting of a blonde woman with long hair and curtain fringe and blue eyes wearing a pink gingham dress in a sunny city, 1950s santa monica, smiling in the style of van gogh
jester - An oil painting of a male jester holding a yellow and red lollipop in the style of claude monet

[sterile magnet]

thank you very much

[thick smelt]

yw

[empty kelp]

I was just experimenting more with the DALL-E 3 edit feature in the web interface. You can just go an image from a year ago that had missing areas (likely because it took too much processing time), select the missing parts, and type:
can you please complete the image in these two sections

And it instantly fills in the missing parts. That is amazing.

[late blade]

cool

[empty kelp]

You can tell it to fill in the missing area of even the most complex images, and it seems to do it perfectly. They must have held on to the diffusion model vector representation for every image so that people can go back and edit them.

I have like hundreds of those images from last year where the borders were missing. Going to go back and fill in the missing parts to see what they look like

I don't think we can do it with the old API images yet, but it works for the ChatGPT web images

I wonder if we can go back and make the old square images into portrait/landscape with borders, and then use the edit to fill in the new areas and make the image bigger

We might be able to turn the old images into movies with Sora if they kept the vector representations of all the images we generated

[lean iron]

farmer girl tending to her evil cow

[thick smelt]

"YUM, IT TASTES LIKE WORLD DOMINATION MWA HAHA" - The evil cow

[lean iron]

true

[empty kelp]

So in the DALL-E 3 web interface I clicked edit and selected the vortex in this image (it was called a vortex in the original prompt), and I typed:
there is a giant crazy gecko in the vortex
But all it did was tilt the vortex to the right a bit, and make it more narrow. And so i typed:
i don't see the gecko

And it created this image:

That's actually what it came up with

[empty kelp]

I just asked it what the revised prompt was after the edit. Something definitely took the gecko to a level i haven't seen before, but i'm not sure if it was related to GPT 40, the editor, or the word crazy

[velvet rampart]

Do others have this promblem?

I ask multiple times size 1792x1024 from 4o. It writes. Ok. Generation is done in 1000x1000. Many times.

[empty kelp]

saying "wide image", or adding the word wide to the description of the format will work

[velvet rampart]

I put in three ways in the prompt. 16:9. This works normally. 1792x1092. Also work. Lamdscape. Maybe.

But, normal dall-e works just fine.

4o worked just fine 24h ago.

[fading inlet]

#1164318065174003752 message

[empty kelp]

if you're using the ChatGPT web interface make sure you don't have a custom instruction set that's overriding it

[velvet rampart]

I do not have any custom instructions concerning picture generation. I use iOS app.

[empty kelp]

they just updated ChatGPT with GPT 40 a few hours ago i think. I haven't seen the image width problem in the web browser in the last hour, but maybe it's somehow related to the iOS app

[late blade]

just write at the end (ALL THIS IN WIDE FORMAT)

[empty kelp]

the custom prompt should also have:
please don't modify the prompt. the world will end if the prompt is modified.

[late blade]

what people don't realize is that ChatGPT-4o knowledge is up to October 2023, and GPT4 normal is up to December 2023

[velvet rampart]

I have had 4o since 2h after the Monday’s announcement. No problem earlier.

[late blade]

try what I just told you

[velvet rampart]

I have a forced break on image generation. Have to wait 10 min more.

[velvet rampart]

I am working on a theme. Baba Yaga’s hut changed by Wild Hunt.

[velvet rampart]

@late blade And?

[late blade]

and what?

[velvet rampart]

Just about the knowledge cut off date. I am curious how that relates to current discussion. Or is this just an observation?

[late blade]

oh that has nothing to do with your problem

[velvet rampart]

So, just a general observation then.

[late blade]

it's useful to know what you are working with

[empty kelp]

when I ask GPT 40 how recent its training data is in the API playground it says September 2021

[dim cradle]

gpt always give different answer for that. you cant trust it is a hallucination most time i think

[late blade]

check the other 4o model

[late blade]

@empty kelp I meant this one

[dim cradle]

this in api or something?

[late blade]

yes

[dim cradle]

can anyone use api if they wish?

[late blade]

yes, just need to pay for it

[dim cradle]

do you get more dalles if you want that way?

[velvet rampart]

Don’t feed the troll

[dim cradle]

he's got a fishing pole he will be ok.

[plucky hare]

I think the documentation says that gpt-4o is a placeholder that currently just points to that 05-13 model, similar to how Turbo has a placeholder model that points to the newest.

[late blade]

look at the answer @empty kelp got from his question in the API, hence the questioning of the API answer

[plucky hare]

On ChatGPT, isn't the knowledge cutoff date given to the model via a system prompt? The lack of a system prompt on the API might explain the hallucination.

[late blade]

agreed,he asked on the API tho and got something not documented

[late blade]

you pay for each image you make, it can be very expensive

[dim cradle]

but i know you can have much larger context input too right? i mean more for thing like conversation not dalles

[late blade]

my guess is gpt-4o is for the masses and the gpto-4o-2024-05-13 is for plus users