images-discussions

[vagrant zodiac]

I'm certainly hoping it comes to api soon

and labs

so we don't have to keep fighting ChatGPT's prompt changes

[hardy heath]

Something went wrong. If this issue persists please contact us through our help center at help.openai.com

does anyone know how to fix this ?

[rigid bough]

do you have a product called NordVPN installed? If so disable it's web protection feature and then uninstall it

[hardy heath]

no, i don't have it

[rigid bough]

try from a fresh install of Firefox

[hardy heath]

google

[rigid bough]

yes, so try from a fresh install of Firefox

[hardy heath]

company laptop only has google

can't install anything else

[rigid bough]

try from incognito mode then

[hardy heath]

tried, same

[rigid bough]

that message means that some page resources are being prevented from loading. If you can't modify your system and can't try with a non Chromium browser then I don't know what can be done to help you

[hardy heath]

i assume Edge not helping either ?

[rigid bough]

Edge is Chromium based, as is Brave

[hardy heath]

firefox should be good ?

[rigid bough]

it may help, that's why I asked you to test with it

[hardy heath]

installing it

you are champion mate

this workaround works

thanks robert

[rigid bough]

great, have fun 🙂

[hardy heath]

great helps

[white skiff]

oops sorry, in my language there is distinction between so i sometimes forget to check before saying 😅

[final compass]

Does anyone know a list of visual styles you can ask for, like main categories? EDIT: Yes I can just ask ChatGPT to produce the list 💡

[agile peak]

Dalle. Allowed to create disney/pixar and certain anime/manga. Seems heavily restricted with studio ghibli style.

With same seed and subject only slightly changes.

[glossy scroll]

How do I get access to post on the Dall-E Gallaries channels?

[sacred helm]

I don't remember having to pass some requirements

[glossy scroll]

It won't let me post anything there, let alone images.

[sacred helm]

ping the mods
they can help you

[glossy scroll]

I don't want to wake people up

It's still quite early

[sacred helm]

pick some mods in online

[wispy storm]

Have you take the role ?

[wispy storm]

Avoid ping the staff please

[glossy scroll]

where is it?

[wispy storm]

If you have any issue contact modmail in dm

[glossy scroll]

No I haven't taken the role I don't think

Which channel is the role?

[wispy storm]

In channel and role

I can't link but second in list at left side in every channels name

[glossy scroll]

OH right

lol

of course

haha. thank u neko

[wispy storm]

My pleasure

[glossy scroll]

wait a minute. which role is it?

I'm in the role menu and I don't see any that lets me access that area

[wispy storm]

My bad they edit it.
Can you send a screenshot of what do you see in #1154829862171844679 ?

[glossy scroll]

One moment. I'm having some weird problems.

[glossy scroll]

Ok, back.

[wispy storm]

Can you click on dalle gallery instead of guideline?

[glossy scroll]

What you mean?

[glossy scroll]

Oh wait, I think we're lost in translation. I don't mean that gallery. I mean the Dall-E Gallaries channels.

[wispy storm]

This channel is now locked and will be archived

That's the message when you click on them

#animals message

Last message from the bot

[glossy scroll]

I can see the channels, I just can't post anything. says "You do not have permission to post in this channel"

[wispy storm]

The channels listed you shown will be archives you can't post in them

[glossy scroll]

Ah, gotcha.

Sorry to waste your time @wispy storm
Thanks for the help

[sacred helm]

any ideas for custom commands for dalle?
I'm trying to achieve easier dalle interaction with seeds
here's what I have currently
!{seed} : user wants to use the provided number as a fixed seed for image generation. !random: use a randomly chosen seed for the next image generation, ensuring variability in the output. !series: user is interested in creating a series of images and would like guidance or suggestions on maintaining consistency. !explore: generate multiple variations of an image for the given prompt, allowing users to explore different outcomes. !lastseed: request ChatGPT to recall and display the last seed value used in the previous image generation. !defaultseed {seed value}: set a default seed value that will be used for subsequent image generations unless overridden. !help: When users are unsure about how to use seed values or need a refresher, this command will provide a brief explanation and guidance on using seed values with DALL-E. !debug: makes gpt return verbatim response body from dalle after a trial of image generation. toggle option.

[wispy storm]

Not wasting , asking question is important

[glossy scroll]

That's a nice set of instructions.

[sacred helm]

thanks I'm planned to share complete one on the use case channel or smth

[glossy scroll]

I don't go that detailed to make super prompts like that, but I do have some ways I stick to, which I find to be really helpful.

Such as using like a subsequent set of commands that are led into each other by a colon, such as Put this in front of every image prompt you write: "3D Animation: Octane Render: 16 by 9 Aspect Ratio: Technical Lines:"
And I find it works quite well to create some really powerful images.

[sacred helm]

What images you were tryna share in the galleries btw?

[glossy scroll]

Not sure if I can share them here, but one won't hurt right?

[glossy scroll]

I was gonna flood the #characters channel

[sacred helm]

yeah just don't spam it

[glossy scroll]

No, not spam.

I meant that as a joke

Like use it a lot, is what I meant.

[wispy storm]

Can always create your place in #1154829862171844679 and post your creation

[glossy scroll]

Yeah, I've done that for one set of images already. Not sure if getting rid of the Dall-E Gallaries channels was a good idea though.

Cuz like...it's nice to have categories for things.

[wispy storm]

#1070006151938314300 leave a feedback always interesting

[glossy scroll]

You know what, I will actually.

There. Done.

[astral pond]

im trying to recreate a background like this in DALLE-3. I want a white background with sparse and subtle patterns but it doesn't deliver anything relevant

any idea?

[sacred helm]

have you ever tried to make a description about the image with gpt-4V?

[white skiff]

i got to close on first try, did you use abstract wallpaper as starting point?

[tall patio]

ripple gradient from one concentric point

or anything that is relevant to concentric point, basic geometry rule for designing effects

you can further specify 'gradient ease' between each circle

Back in the 90s my friend told me no matter how powerful computer might become, it can't beat a pen and paper when you want to get things done

After this AI boom I have a simple benchmark rule

Just ask AI to generate one single circle on white background

So far D3 arrived to the correct answer the fastest

The future is bright

[forest seal]

Definitely so. I can only begin to imagine how better the models coming soon will be

[tall patio]

But pen and paper...still supreme

[thick smelt]

since matthew Lillard is playing purple guy, scooby animatronic for the shaggy afton using bing

[near tiger]

Anyone have tips for talking with Dalle 3? I'm having trouble getting it to alter the images how I want it to.

For example: I'm trying to create a spaceship, but even if I specify for the spaceship to not have wings, Dalle 3 is putting wings and saying in the descriptions that there are none

I want things like this first image, but it keeps giving me things like the other images

[plucky hare]

There is no negative prompting built into DALLE currently, so your best bet is to focus on your positive descriptions and reduce/eliminate your negative descriptions, as unfortunately that will only confuse it as it exists today.

[agile peak]

Should try "Wingless" word. Since dalle not understanding negative prompt.

[near tiger]

Ahh ok, that makes sense, thanks I'll try it!

[verbal sequoia]

<:book_icon:1046291503158399107> `` Rule 6 `` No political and religious discussions or content.

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

[agile peak]

My most consistent generated from Dalle. Let it generate 4 then focus on the good one and customize the prompt — with same seed.

[plucky hare]

What language are you using in your text prompts to indicate seed?

[agile peak]

First. You just input the prompt as usual and then once it’s done. You can ask it for the seed for each images.

[plucky hare]

I'm sorry, I meant to ask what you're saying to it to get it to use the seeds in future generations. Are you just saying like "Please use seed 46883 for all four images"?

[agile peak]

Oh, right. I just say “I like the third panel. Please use same seed 5,535,665” then proceed to say what you like to customize, for instance: “Adjust the boy hairs to be black colored and wears navy coat with comfy black turtleneck”.

You can also just say “Make the image using manga style from 1980s”

And the last one, I also used Disney classic animation.

[near tiger]

Very nice!

[cobalt hull]

lol saw the answer above, how are you getting the seed, just asking it?

[agile peak]

Simply like this

[cobalt hull]

ah gotcha, so you are just providig some random seed to it?

and then using the same seed for other prompts if it fits your visio

vision*

[agile peak]

You gotta ask first after dalle done generated the images.

Then use the seed from the images you prefers to focus and customize it. For the next generate.

[silent thorn]

funny enough bing suddenly says it cannot list seeds - yesterday it did

[agile peak]

Seems to be issue with dalle now.

Also ChatGPT just got new update on the app a slightly change with history bar similar like desktop version.

[near tiger]

Oh it did? I don't think I'm seeing that

Any tips for specifying when you want something added/changed on a specific side of a character/object?

For example, I want a giant cannon on the right side of the ship in the first image, but in the 2nd image Dalle keeps placing the cannon in the center and in the description saying it is on the right

The prompt I used: "Please use the same seed 3790475397, and add one giant cannon onto the right side of the ship" (I also tried "FAR RIGHT")

[crystal badger]

i hope we can use reference images with chatgpt + dalle soon

[near tiger]

That would be super nice

At least for now we can have GPT4 describe the image

[sacred temple]

what are you fav prompts to get realistic photo quality? thanks!

[plucky hare]

One pretty funny thing I've been seeing people trying is including fake camera/photo metadata at the end of the prompt. Like the kind of stuff you'd see in Flickr tags to indicate camera model, lens, aperture, other camera stuff I know nothing about. I've seen some people say it seems to create more photo-like results, but I'm just sharing hearsay.

[near tiger]

Hah, that'd be pretty cool if that actually works

[calm ether]

Ios chatgpt dalle3, dont create image only json

[quartz vale]

Dall-E 3 can only generate .png files to my knowledge

[plucky hare]

I think they're saying they're getting a glitch where the model is displaying JSON-style formatted text instead of initializing the actual image generation.

[calm ether]

same error

[twin canyon]

Yep, I just tried using it. Displayed prompts in text only. It tried sending me links afterwards but they were completely blank, each of them labeled “Zero KB”

[naive lake]

Guys, anyone else is noticing super slow image generations and fast rate limit hits?

[quartz vale]

it's likely due to how many people are using it

[naive lake]

I assume 1 GPU takes over one submission

are there gonna be any speed increase? Or should I open the second account?

[quartz vale]

I don't work for or represent OpenAI. I do not know what they have planned

[naive lake]

You're violet, you should know something haha

[glossy scroll]

Is there an issue with generating images for other people?

I'm having errors from dall-e in chatgpt

[twin canyon]

Yes, but I’ve found that it can generate images on old chats, like from yesterday. Just not able to generate on new chats. (on the mobile app.)

[glossy scroll]

I tried in old chats just now and it comes up with the same error message.

[twin canyon]

Using browser or mobile app?

[glossy scroll]

Browser

[dim cradle]

Yes

It is down

Im pretty sure they already know

Same with bing search

[boreal coral]

why is dalle3 down

[charred sonnet]

it works on my pc but not on mobile

[peak trellis]

what are seeds?

[charred sonnet]

If you use the same seed, and same prompt, you get the exact same image. Changing the seed randomizes the result.

[neon sequoia]

You can essentially tell Dalle to create variations of an output you like, keeping the prompt the same, but changing only the seed value. It’ll keep the general aesthetic (typically), but changes the output a bit in composition. You can later refer to that seed to use for future generations, however, changing prompts even by adding a single random word itself can vastly change the output too.

[naive lake]

I can't live like this 😦

[neon sequoia]

I think there’s just problems in general right now, some weird errors and outputs and things like that

[remote stone]

Can I ask for help with prompts over here?

[zenith girder]

hello

[sacred helm]

yep, but also #1155772063596953642

[remote stone]

I'm trying to generate different anthropomorphic animals but depending on the species I'll get results like this, totally normal human body with just the animal head, and I dunno if there's a way to avoid that

Or even more ridiculously this

[sacred helm]

have you turned on the beta features on the settings? maybe that got reset

[sacred helm]

have you tried seed control?

[remote stone]

No idea how to do that

[neon sequoia]

There may be easier ways but if you tell it to make variations of an image and keep the prompt the same, changing only the seed, that’ll do it

[sacred helm]

try this

give me 4 iteration, I'm tryna find the seed that can meet my requirement well. randomize seed by not inserting seed value on your dalle request, and give me seed value of each images after generation. we'll repeat generating until I find the one I like

[still zinc]

Where'd you learn about seed control @sacred helm ?

[sacred helm]

I found it myself
you can learn about them in my reddit post

want the link?

[still zinc]

sure, always interested in learning new techniques

[neon sequoia]

It’s been a big thing in MidJourney for a while too, if you look online for seeds in MidJourney there’s a good article too (syntax with Dalle is different of course)

[remote stone]

Got it, what to do once I find one seed that does it well?

[sacred helm]

thats simple

please fix the seed to {the one you like}

and have fun generating

for additional tip, keep your prompt related with anthropomorph same throughout the conversation. as seed is related how dalle interpret them

[robust sonnet]

Could you please send me the link to the seed control?

[sacred helm]

just sent

[remote stone]

Yup, finally found designs I'm happy with, only problem is it'll generate every character using the same pose too

[sacred helm]

glad to know you found it, I think it needs to be determined the pose you like

[remote stone]

At least I finally got to here which is muuuuch better

Thanks a lot!

[naive lake]

trying to generate in 2 different chats by accident? Don't... it died on me 😦

[runic prairie]

that looks great

I'm doing the same thing. Longer prompts make a big difference

[paper jolt]

They're not in Settings -> Beta Features?

[knotty olive]

can y'all try to generate an arrow that points left please? Edit: Context https://discord.com/channels/974519864045756446/1167263855391539220

[sacred helm]

have you tried seed control?

thats interesting let me try

[knotty olive]

it's such a simple instruction, would that matter?

[sacred helm]

I think you're right, it can't point left

[knotty olive]

Huh, how strange

[quartz vale]

it didn't work for me either 🤔

that's incredibly bizarre

[knotty olive]

Btw my test was with 3, was y'alls?

[karmic ridge]

Hi, If I upgrade to Chatgpt Plus, will I be able to use Dall E with no limit or do I have to buy more tokens to use it?

[quartz vale]

you'll still hit rate limits like the 50 msg/3 hour limit

[willow isle]

There's definitely something weird going on here

Got a semi-successful image: Photo of a man in casual clothing pointing left with his index finger, standing next to a metallic arrow sign pointing left. Above them, a neon sign shines brightly with a left-pointing arrow. In the background, a brick wall is visible, and on the pavement, there's a clear arrow painted pointing to the left.

[lime fern]

Hello, I hope you’re doing well. I’ve been facing an issue with your website on both desktop and mobile. When I click on the ‘buy credit’ option, the page doesn’t load, and I’m met with a blank black screen. I’ve attempted multiple troubleshooting methods, but none seem to work. I’d appreciate any assistance or insights on this. Down for everyone or just me? Thank you for your time!

[acoustic bay]

I am in shambles boys.

[knotty olive]

Some other fun ones that don't work: right arrow turned 180 degrees, mirror image of right arrow (thinking maybe it could do right and then make it left, but no). I tried prompting in spanish, but that didn't work either, so it's not the work "left" itself I think it's the concept

I ask gpt-4 to make me a prompt with the most left-pointing arrows as would be reasonable.

In a brightly lit room, there's a large wooden table with various objects all pointing to the left. On the table, there's a compass with its needle pointing west, a digital clock with an animated left-pointing arrow blinking, a hand-drawn map with a big arrow indicating a direction to the left, and a laptop displaying an icon of a left-pointing arrow. Above the table, hanging on the wall, is a framed painting of a medieval knight holding a shield with a left-pointing arrow emblem. Beside the table on the floor, there's a toy arrow sign, again pointing left.

Yet trying to generate "a laptop displaying an icon of a left-pointing arrow" on its own fails. So interesting and so strange

[knotty olive]

Payment screen loads for me. I didn't actually check out, but it pulled in my info and whatnot. :/

[lime fern]

Thank you for letting me know. It’s strange because I’ve tried on multiple devices and browsers, and I consistently encounter the blank screen issue. I’m wondering if it’s a regional issue or something specific to my account. I’ll continue to troubleshoot, but if anyone has had a similar issue and found a solution, I’d appreciate the guidance.

[knotty olive]

perhaps regional. Just to follow up, I did buy credits successfully. Hopefully it resolves soon

[knotty olive]

bought credits for dall e 2 because now I'm going down this rabbit hole. Used your exact prompt and got one image of left arrow and three images of right arrow.

Anyway, that was my final test, but would be curious if anyone else tries or comes up with a reliable left prompt for 3. I really would like to generate some left-facing characters haha

[lilac obsidian]

Sometimes it will generate 4 images and other times 2. Is there any reason for this?

[sacred helm]

seperate results under seed control

[agile peak]

Ever since the last update. GPT seems to change my seed without permission 🙈

[agile peak]

Cel-shaded and Pixel animation. But was kinda difficult for these due to content policy -- eventhough the style was suggested by chatGPT.

[tall patio]

I'm wondering when will Dalle-3 be able to generate vector graphic

Or export svg, png files

[lapis fractal]

I generate vector style graphics in Dall-e and then download the image to process it in Photo Shop where I 'vectorize' the image.

[tall patio]

Direct vectorize is better...

[glossy scroll]

My Dall-e prompts are still not working.

Unfortunately

[tall patio]

If only OpenAI would hire me to prompt crack...

[glossy scroll]

Let chatgpt make you a resume

[agile peak]

Using the ios app is easy to make the logo without background imo

I can click on the logo inside of GPT and save/copy it.

[loud field]

Hey

I need help:

see this thing, the French tarasque?

I need it to be super-stylized, like a sly company logo

any advice to tell Dall3 to make it far more stylized?

[sacred helm]

flat design or semi-flat design

[loud field]

idk, I would say flat. This is supposed to be the logo of the french mafia in a cyberpunk game I play with friends

[loud field]

aaand I need a mug shot of a menacing cyberpunk french criminal, but dall3 refuses that as it's "unsafe"

sigh

[agile peak]

Using film scenes as references

[crystal badger]

Try Vector, that usually helps, alongside flat design.

[loud field]

vector wasnt working but I didnt try with flat design, ty

[crystal badger]

if neither works try to post your actual prompt here, so we can try it.

[loud field]

stylized vector art of tarasque with six legs, a scorpion's tail, a tortoise's, shell, a lion's face, a dragon's body, full figure, centered in the image

[crystal badger]

do you need any text on it?

[loud field]

if possible, Milleau

but dall3 doesnt understand french

[sacred helm]

but the word is in alphabet

[loud field]

no, what am I saying? Milieu

[loud field]

yeah but I've already checked - it makes a word salad if it's not an english word, even a single one

[crystal badger]

so it is Milieu or Milleau

[loud field]

Milieu

[crystal badger]

aight ill give it a shot.

[loud field]

oh we're getting there, these look cool

what's the prompt

[crystal badger]

for one of them? or a specific one

[loud field]

they all look good

[crystal badger]

this is the prompt for the 2nd one Flat vector logo of 'Milieu' using black and gold palette with the golden text 'Milieu' placed below. Create a central tarasque figure having a lion's face, dragon's body, six legs, a scorpion's tail, and a tortoise's shell. Aim for a modern and clean design.

[sacred helm]

prompt challenge : can you generate POV of a head pat reciever?

[crystal badger]

ohh no

i can try

[sacred helm]

I've been doing for so long, even prompt is written by gpt in very fine detail

all I got is POV of a head pat giver

[plain beacon]

Hello

[crystal badger]

i got a hiccup,

this aint what i wanted at all, not even close

[crystal badger]

same

[sacred helm]

that's unfortunate

[crystal badger]

i refuse to give up

[loud field]

@crystal badger prompt for the Milieu logos?

[crystal badger]

all of them?

[loud field]

Yess

[crystal badger]

@loud field should be extremely similar to this Flat vector logo of 'Milieu' using black and gold palette with the golden text 'Milieu' placed below. Create a central tarasque figure having a lion's face, dragon's body, six legs, a scorpion's tail, and a tortoise's shell. Aim for a modern and clean design.

[loud field]

Niiice

[agile peak]

Saw your post about seed in dalle gallery? Did you share about the custom prompt or interactions somewhere? 😁

[sacred helm]

`By default:
You'll ALWAYS ask for my approval on the prompts.
You'll ALWAYS share the seed value with me after the image is generated.

Commands I can use:

!imagine: You craft ONE dalle prompt based on my input, adding details I didn't mention is essential. prompt approval is needed.
!prompt: I provide the exact prompt for dalle. No changes allowed and skips prompt approval.
!careless: Skips prompt approval. Toggle option, off by default.
!modify: Adjust the current prompt based on my feedback. the modified ones are also need approval
!defaultprompt: Indicates the default prompt format that must be included in the prompt.
!seedcheck : starts a session to create four unique image variations, using different seeds, based on one approved prompt. nested with !imagine or !prompt, you'll send the same prompt in four times in the "prompts" array(the one you see in the api request body), and NEVER send request for nested ones. randomize seed value by not having seeds param in the request body
!fix: Use my provided number as a fixed seed.
!random: Random seed for next image.
!explore: Create multiple image variations (4 by default) with tweaked details in the prompt, all using the same seed. when being used with !defaultprompt, keep the default prompt intact as much as possible
!goback : invalidate previous command(s), state what commands you currently have after this`

[crystal badger]

various Prompts and CI can be found in #1019652163640762428 and #1155775326253756456 Also @sacred helm why dont u post the full prompt in prompt labs so ppl have access, easier than pasting it all the time.

[sacred helm]

why dont u post the full prompt in prompt labs so ppl have access
it's still WIP

[crystal badger]

ohh

highly recommended using Data Analytics for CI creation. and prompt creation. by feeding it Guidebooks for prompting for DALLE and etc.

[stone scroll]

hand-drawn portrait of Emperor Marcus Aurelius

who hand-drawn portrait of Emperor Marcus Aurelius ?

[plucky hare]

Image generation is not available in this server.

[hexed mountain]

hey guys! feel free to submit some generations for the next #spotlight with the word “spotlight” 👀 get creative and it just might be spotlit as well! #1154829862171844679

[wind crane]

"You never reach the limit, most people don't reach it, they always have too many unused chips."

[fallen mulch]

is there any option to edit images like fotor, canva, pixlr, picsart etc

[agile peak]

Thank you appreciate it so much, mate 🙏

[plucky hare]

50msgs/3hrs is 1msg/3min 36 sec, if that helps pacing!

[tall patio]

[wind crane]

It doesn't work when you've reached the limit.
I really have to wait 3 hours real.

[plucky hare]

You'd only have to wait 3 hours if you used all 50 messages at once! You get every message "back" 3hrs after you use it.

[scenic cloud]

Anybody know if img/img will ever be a thing with DALLE3

[sacred helm]

I cracked some frontend code and the response from dalle api was denoted as txt2img
so I guess img2img is possibly planned

[scenic cloud]

Awesome thanks

[crystal badger]

in that case i haver a great offer for you, should you be interested.

[sacred helm]

yeah?

[crystal badger]

ill dm you the information as itll take up too much space here.

[lilac obsidian]

Why does it sometimes generate 4 images and other times only 2?

[plucky hare]

Probably just server load things! You can always try asking for four to encourage it to try to get four each time.

[lilac obsidian]

oh. Okay! Thank you. 😊

[crystal badger]

yeah that or Some of them was flagged.

seemingly innocent prompts get flagged sometimes.

[sullen fox]

Is the bing image creator not working for anyone else?

[lilac obsidian]

Me

[sullen fox]

Ok just wondering if it was just me

[lilac obsidian]

It usually says “sorry for the inconvenience…. Was only able to load 2.” But often it’s just saying here are the images and shows 2. When it shows 3 it says “sorry for inconvenience…. Was only able to load 3.”

So I don’t think it’s because they got flagged.

[sacred helm]

It often doesn't tell you it has been encountered with content policy

I had had that blocking with generation of Spider-Gwen, which is an existing art you can't generate

[lilac obsidian]

What the heck is spider gwen?

Oh, I see. I just googled her.

[sacred helm]

Gwen Stacy spiderman that belongs in spiderman franchaise, basically spiderwoman

[lilac obsidian]

👍

Yeah I don’t think it should be flagged.

Then again the AI often shows me images that I personally feel should’ve been flagged.

I think it’s really inconsistent

[severe cliff]

Hello,
Sorry in advance if the question is a bit repetitive and/or has already been answered.

By doing a "default" chat with GPT4, I can send her any picture, and she can describe me correctly.
I took the random ship/character "Cheshire" from Azur lane. GPT4 was able to describe to me exactly who it was, where it came from, all the details, etc, etc.

However, when I copy and paste the promt that GPT4 has just given me (of what it described on this image) within it's Dall-E 3 plugin, it systematically gives me images with totally "random" characters. In no case did I get the character, only (very) vague resemblances.

Why is that? Does DALL-E 3 have a rule forbidding the creation of characters that already exist?

[sacred helm]

Does DALL-E 3 have a rule forbidding the creation of characters that already exist?
yes

[plucky hare]

There are limits in place on DALLE regarding copyrighted characters, yes!

[severe cliff]

Oh dear! Too bad for me, I guess. It's still a very impressive tool.
Does anyone have the links to all the Dall-e 3 rules? I can't find the limitations and/or concert rules.

And thank you both so much for your answers :)

[plucky hare]

You can ask the DALLE model itself, it seems to have a good understanding of most of its rules and limitations. There's a bit more info on the original announcement post too:

https://openai.com/dall-e-3

[severe cliff]

It's amazing that despite these rules, well-known characters like Peach or Dark Vader can be created quite easily...
A bit weird now that I think about it. Shouldn't DALLE3 just be more precise about "censorship" given that there are hundreds of millions of images of these people?

[sacred helm]

how dalle did manage to deal with hands?
it's so perfect most of the time

[severe cliff]

Yeah indeed, there's a huge jump in the drawn hands between DALLE2 and 3

[trail vault]

No more strange gestures.

[final compass]

I wish there was a bulk download feature for chats. You could just download it all when you are done with exploring an idea.

[sullen basin]

Wait, how do you send a picture to chatgpt?

[main bronze]

Good Afternoon All, today is my first time attempting to use the DALL-E 3 portion of ChatGPT. I am having a horrible experience in trying to generate images. I am not sure if this is the correct place to ask or discuss this. This issue is the content policy and its sensitivity in what is being asked of it. Is this the correct place for this discussion?

[plucky hare]

On ChatGPT Plus, the default GPT-4 model includes an image uploading/vision feature now.

[plucky hare]

Sure! The content policy is in place largely to prevent images of public figures and copyrighted imagery. This is presumably to minimize legal risks

ETA: oh yeah and violence and other adult content is also restricted! This is just part of OpenAI's general policy for all its AI models though.

[main bronze]

I fully comprehend. I comply with the content policy because I don't want to see the images it is saying it won't produce. That isn't the problem. The problem is that it rewrote the prompts multiple times to fall within the content policy and no matter how many times it rewrote it, it still would not produce the images saying that it went against the content policy. I questioned it. It stated that although it rewrote the image prompt and it fell within content policy, it couldn't tell me what part of the policy was being violated. I then went to the online browser and asked the same. It talks about sensitivity, which I also comprehend. However, no matter how simple I took the prompt down, it still would not generate the image, even when it wrote the prompt itself.

[wintry dune]

is anyone else having the problem where DALLE 3 is only making 2 images instead of 4

[sacred helm]

same

[wintry dune]

damn i was hoping it was just me

[sacred helm]

I even instructed like this
you'll send the same prompt in four times in the "prompts" array(the one you see in the api request body

[haughty bough]

how do i get

free prompts

[wintry dune]

that sucks, i really hope this is a bug and not how its gonna be from now on

[sacred helm]

I suspect it to be as a resource saving
saving resource by making gpt send two separeate requests in two response, so dalle api can have a time to rest

[agile peak]

Yeah. 4 images keep ended up 2 images max and without any message telling about problem in generating or content policy.

But I get much better and consistent results today in compare to yesterday.

[sacred helm]

I made a reddit post about it, let's see how many people are struggling

[zinc dagger]

I've got that on my end as well, requested 4 images, only got 2

[plucky hare]

I don't think it's capable of making that determination comprehensively yet. That is, I don't think it can write prompts that it's sure fall within the guidelines. You might need to take a more active role as prompter in this context. Also, it's possible that sometimes a prompt can get through with compliant text, but still generate prohibited imagery inadvertently. At least to me, it seems like there's some kind of behind the scenes image evaluation that is trying to check whether the images it's made follow the content guidelines visually, not just according to their prompts.

[zinc dagger]

what exactly is the prompt you're trying to send through? there may be some meaning or association tied to it your unaware of.

One of the content filters looks at the intent of the prompt as a whole and not individual words, so if the tone of your prompt is suggestive of something that is against the filter that will sometimes trigger it.

[spring eagle]

getting 2 images too

they silently changed that param

[zinc dagger]

it is a beta after all, I seem to recall that the Dalle V2.E plugin had the default generations as one wide image.
though it would be nice if Dalle acknowledged that instead of saying "Here's what you asked for! 2 Images, just as you requested"

[misty crater]

is image generation down ?

[agile peak]

It also seems dalle can now understand -stylize -chaos similarly like Mj.

I think we could see big improvement by next month especially since the OpenAI dev day are coming. I heard.

[serene ether]

Me too

[agile peak]

OpenAI DevDay, on November 6, 2023 in San Francisco. The one-day event will bring hundreds of developers from around the world together with the team at OpenAI to preview new tools and exchange ideas. According to site. Would be a big day.

[scarlet urchin]

why did my art in gen art get deleted?

do i post something wrong?

[dim cradle]

After the update images look different... style is altered. It get worse...

[main bronze]

Can I share the chat link here, I feel like it would be easier than trying to explain it all. Also, I had this conversation multiple ways adjusting what I was asking.

[zinc dagger]

sure, or send it in #1155772063596953642

[void musk]

well, seems we cant make 4 images now

sadly it's not just me as i can see

[main bronze]

It would not let me share the initial conversation, so I am inputting a link with the second one. It is a rewrite that it came up with on its own. https://chat.openai.com/share/fb399b5d-cd3a-44fc-b957-4898ec4f6fc0

[knotty olive]

TIL that the GPT-4 dall-e prompt can actually interpet images as well

Wonder if can just ask it to iterate on its generated image automatically Edit: yup. cool

[zinc dagger]

yeah I can see why it's being rejected.

I think the term "Black Panther" is why you're getting the error. I tried the prompt on my end with just saying "panther" and it gave the expected image results on my end.
As for the reason why that is, it could be that the no politics filter is interpreting your request as the 1960's political movement of the same name, or the copyright material filter is thinking it's the marvel movie of the same name.

[main bronze]

I see. Yes, it makes sense when you phrase it that way because even when I attempted to do it myself with my own wording it rejected it. I comprehend how certain word pairings could be taken out of or put into a different context based on what it knows. Thank you for helping me figure that out. What did you change in the prompt that produced an image? Can you say the same for this one with the wording changed? https://chat.openai.com/share/88afce3d-13b0-460d-971d-7906d585234c

[wary stump]

Its not a bug. There is a meta-prompt you can use to get the instructions that chatgpt has for dalle. In those instructions it says to generate 2 images

Hopefully its not permanent but its definitely intended

https://chat.openai.com/share/28b455c2-40bc-4b67-b5c2-645ca8e0a22c (these are the instructions)

[knotty olive]

I apologize for the oversight. You're right; I should have provided four images in one response based on your request. Let me generate two more images using variations of the road sign arrow prompt to provide you with a total of four options.

It did give me three in a response just now, but it was iterating, so it was one after the other. Might be able to ask it to iterate and get two more automatically?

[wintry dune]

wow thats so lame, good things never last

[plain brook]

for dall e i cant buy credit

why ?

[zinc dagger]

I think the issue is Dalle trying to modify the prompt to specify it's a black panther. I got this image by using your prompt and telling dalle to not describe the panther:

[crystal badger]

or use its scientific name instead of "panther"

use genus Panthera

its all wordplay

[main bronze]

Thank you so very much! I will definitely work on my prompting skills with regard to "negative" prompting if I am not able to produce the initial images based on the prompt. This just helps me mentally expand. Again, I appreciate the time you took to help me. May you have an absolutely Beautiful evening/rest of your day! 💜

[main bronze]

Thank you for that!!

[crystal badger]

np, just doing my part.

[void musk]

yo why is dalle saying he cant generate what im asking?

this is getting worse and worse

[plucky hare]

There are content policies in place to restrict the generation of imagery of: public figures, copyrighted imagery, art from artists alive within the last 100 years, violence, and other adult imagery.

[void musk]

so frustrating

bro, im generating a kid playing nintendo on his room

censorship is killing this industry. Hopefully we get uncesored ones soon

[plucky hare]

"Nintendo" is the sticking point there I'd guess! Copyrighted name and logo. Try "video game console from [year]"?

[knotty olive]

try tiny human playing a retro 64 bit console?

[void musk]

godamn

true

[crystal badger]

playing a slime in my dnd campaign here we are

Cosmic Plasmoid from space lol xD, DALLE sure did capture it perfectly.

[agile peak]

Just use game or movie reference rather than mentioning its studios. For instance.

[crystal badger]

gives me treasure planet vibes.

[agile peak]

I tried madagascar and toy story as reference today, dalle got it 100% matched and even let you combine both into one.

Madagascar and Toy Story fusion together. And Rexy from toy story.

[dim cradle]

Is so funny all image seem darker today

[hollow patio]

Hi,
where can I try dall e 3 for free except Bing image creator?

[plucky hare]

That's currently the only free option! DALLE 3 is on available there and with ChatGPT Plus as of today.

[fallow quail]

ask chatgpt how to change the prompt or how to describe the thing you want to generate without violating the terms and conditions. works like a charm when you are to lazy to think about rewording your prompt.

[crystal badger]

Ask for Dalle core guidelines and use that as a rule set for gpt when prompting

[mental creek]

Craft a detailed narrative about the grand opening of a new business where a well-dressed gentleman arrives in style, and a large crowd gathers to celebrate his presence. Describe the anticipation, the interactions, and the atmosphere of the event.

[ionic dome]

is there any official word on why is only 2 images now not 4?

[quartz vale]

[ionic dome]

hopefully this is temporary. half as many images is half as much fun

[tepid adder]

honestly I would even mind it if Bing's servers weren't in a constant state of death

[crystal badger]

i still think mine is better, i tailored it to you personally.
||this was a joke in case in doubt||

[wary trench]

are we only getting 2 images now? what's up with that?

[plucky hare]

No word yet as to why, probably just server load stuff!

[remote stone]

Just two images feels awful

[fallow quail]

chatgpt says it's to ensure services. "I apologize for the confusion. You're right that there has been a change in functionality. This decision by OpenAI is to ensure the efficiency and stability of the services and to prevent potential overloads. I understand this can be confusing, especially when things change. I apologize for any inconvenience and thank you for your understanding."

[pliant spindle]

Why is bing image creator not working? It says cannot generate images right now.

[forest seal]

I dont even think chatgpt itself knows dall e 3 even exists

This seems to be a hallucination

[tepid adder]

because it never works

unless you intentionally bomb your sleep schedule

[pliant spindle]

Now that isn't true.

[fallow quail]

it lists two images as the maximum for generating pictures when asked. some days ago the answer was 4 images in one request. even if it thinks dalle3 is not exisisting as of january 2022. there has to be some custom instructions in the module, so there is the possibility that this is no hallucination.

[remote stone]

Dall E 3 does not have image reference compatibility yet right?

[subtle ferry]

what?

[knotty olive]

yeah

[fallow quail]

they did the same with dalle2 last year. from 6 to 4.

[vagrant zodiac]

Response rates have always been variable on ChatGPT

[verbal sequoia]

<:book_icon:1046291503158399107> `` Rule 2 `` Keep it G-rated.

Sharing NSFW or violent/disturbing content through messages or within Discord profiles is strictly prohibited.

[subtle ferry]

Probably a ton of strain on the servers to do 4 images per chat.

API will be here sooner than you know it and you can use that instead.

[quartz vale]

you can ask for multiple batches of 2 images as well 👀

[vagrant zodiac]

Please tone it down.

[subtle ferry]

This'll probably make the waiting time to use DALL-E per each message shorter if anything, since less images are being generated.

Makes life easier for me when I try to get the perfect gen and keep getting blocked

[quartz vale]

what constitutes the perfect gen

[knotty olive]

when you see it you know

[quartz vale]

that's so true, though 😆

[subtle ferry]

Exactly what Astro said, it's the one that feels right, y'know?

I make a bunch of anime stuff but every now and then the gen is off the mark just a tiny bit

Sometimes it takes me like 20 minutes just for one image I like.

[quartz vale]

I know that pain. I've spent over an hour on 1 image with Dall-E 2.0 outpainting/inpainting before 🥲

[subtle ferry]

Huh, I gotta try outpainting one of these days.

[quartz vale]

I highly recommend it! It can be a bit finicky at times, but I think it's pretty good overall

[cobalt hull]

Is dalle only generating 2 images for anyone else? Ive explicitly asked for 4 and only getting 2 still

[quartz vale]

I'm getting 2 images per prompt now as well

[cobalt hull]

🙃 k good to know it's not just me

[drifting hawk]

dalle2 launched with 10. i was so upset when they decreased it 😆 i get it though, not enough gpus

[velvet nexus]

For folks curious about the 2 image limit: can confirm the maximum setting has in fact changed.

[drifting hawk]

can confirm, its 2

[final compass]

Noticed i am no longer getting seed numbers at times.

[vapid elk]

you can request 10 images at once if you are on dalle2 API =P

be prepared to a hefty bill

[wintry dune]

DALLE 3 is giving me a lot of "Error creating images" today

[quartz vale]

I just got one of those 😆

[fair crescent]

Feels like a short time 😉
If DALL·E 3 comes to inpainting/outpainting,you'll not see me here for a few days as I hunt that perfect picture

[drifting hawk]

if you got error creating images can you share convo id? (its in the url, at the end)

[pseudo agate]

Hello.

I have a problem with chatGPT login safari

It used the work before but now it won’t let me I even refreshed but it still goes like this for days.

[drifting hawk]

restart your phone

[pseudo agate]

How?

[drifting hawk]

https://support.apple.com/en-us/HT201559

[pseudo agate]

Oh you mean turning off

[drifting hawk]

also try using data instead of wifi

[pseudo agate]

Bruh why the button still won’t work

I have data turned on not the wifi

Why is it working on PC and not longer on my phone?

Instead of preview it show me this

[drifting hawk]

i am confused what that screen is from. it is not from our website because of the bottom bar. it looks like youre in safariVC, so you're in an app? whcih app?

[pseudo agate]

Safari on my iPhone

[drifting hawk]

huh safari shouldnt have OK in the top left

[pseudo agate]

Discord but Safari preview site

[drifting hawk]

ok thats why...seems like a discord issue

[pseudo agate]

No, even in safari itself watch

[drifting hawk]

does login in safari browser work?

[drifting hawk]

hmmm that's weird... do you know how to read the console errors?

[pseudo agate]

It never happened before I don’t know why it is not letting me in now.

[pseudo agate]

No

[deft musk]

Is this happening to you on the web as well as the app?

[drifting hawk]

do you perhaps have javascript disabled or some sort of ad blocker

[pseudo agate]

I never used the app ever only the web site

[pseudo agate]

I don’t know, how can I check that?

[drifting hawk]

like did you install any safari extensions

[waxen shard]

Is this a permanent change?

[drifting hawk]

for now it is, depends on capacity

[forest seal]

Getting vibes to the gpt-4 downgrade from like 100 messages per 1 hours, to 50 per 3, to 25 to 3, and then to 50 per 3 hours again…

Lawl

[waxen shard]

Ok. Too bad. Just as midjourney is rumored to be increasing to 16

[drifting hawk]

if you wanna donate gpus we can increase it!

[forest seal]

Lmfao

[pseudo agate]

I don’t know

[waxen shard]

Yep, I'm sure another $20 a month will solve the problem 🙄

[drifting hawk]

money cant solve unfortunately

[forest seal]

Let me spend a couple million on some h100s rq

[drifting hawk]

plssssss

[forest seal]

Then agi will be achieved internally…hopefully

[pseudo agate]

Just saw JavaScript turned on

Does it has to be turned off?

[forest seal]

I would say to use firefox for chatgpt. No funny business usually happens on chatgpt there

Or any website, reallty

[drifting hawk]

i dont know but i forwarded your issue to the auth team

for now why dont you try the iOS app?

[forest seal]

^

[drifting hawk]

(i worked on it :D)

[forest seal]

True

[waxen shard]

It can buy happiness though. "Because I'm happy when I can buy what I want"

[forest seal]

Money can solve it when some billionaire donates a bunch of h100s

[waxen shard]

Perhaps Mr. Gates? Or can't go back to the same well?

[pseudo agate]

I need to update but I have too much space

[void musk]

im having considerably more difficulty generating images today. always drops the error message

is it the censoring or whats going on? im even writing on the prompt to adhere to content policy

[drifting hawk]

What’s the error message

[boreal coral]

woww you just gonna remove my image like that

ok i see how it is

[sacred helm]

same here, the prompt indeed adheres to the content policy, I know this because I'm using the same prompt that worked perfectly fine

the error message is just but it seems there were issues generating

[drowsy bane]

Really struggling with exploratory promoting/exploration of ideas with just the 2 images being generated. Hope this can change back to the 4 soon. 🤞

[cobalt hull]

Agreed, I loved asking for 4 different styles and then working off of those, it's so much more work now with just 2 images

[normal loom]

So there is a daily cap on images in gpt+ now.

"I apologize for the inconvenience, but you've hit your daily maximum number of images. To ensure the best experience for everyone, we have rate limits in place. Please wait for the next day before generating more images. Your daily maximum will reset in 18 hours and 54 minutes."

[wintry dune]

damn, the fun never lasts does it

[cobalt hull]

lol damn that's a new one, wonder what the cap is

[drifting hawk]

Can you share convo id? It’s in the url

[sacred helm]

Oh forgot to mention, I used custom instruction on the session

[wintry dune]

I was getting the "error creating images" then i realized i hit my daily limit hours ago and it only told me just now

I apologize for the inconvenience, but it seems you've hit your daily maximum number of images. To ensure the best experience for everyone, we have rate limits in place. Please wait for the next day before generating more images. Your daily maximum will reset in 18 hours and 6 minutes.

[drifting hawk]

Yeah we just fixed the messaging lol

Like ten minutes ago

Yay it works

[sacred helm]

but I got an image on the other session, tho it was only 1 instead of 2, I didnt hit the limit yet

[drifting hawk]

Yeah it must be something else, will look once I’m home

[sacred helm]

can you tell us how much is the daily cap?

[drifting hawk]

300 right now but may change depending on gpu capacity

[sacred helm]

300?????
that' a lot than I expected

[drifting hawk]

Some people are going ham 😂

[quartz vale]

👀 😳 I wonder who that could be...

[cobalt hull]

lol me

[ruby maple]

I'm impressed but at the same time it's kind of what I expected lol

[cobalt hull]

Takes me like 20-30 attempts to try and get the details of my dog correctly

[ruby maple]

Sometimes the images turn out phenomenal, others it's completely off the mark

[wary stump]

PSA they added seed support but didnt announce it

[drifting hawk]

moderation error, since we randomize the seed now its possible a non-blocked prompt generates an image that gets moderated

[sacred helm]

Is that an 'error'?
isnt it just a content policy violation that detacted after generation?

due to the randomized seeds

[drifting hawk]

it is a content policy violation, the error messages were broken, it should be working now!

[sacred helm]

thanks for your service
btw, is it really possible to give dalle gpu some rest by requesting max of 2 instead of 4? I mean most of us will just go straight to further generation

[drifting hawk]

we have logs and it has decreased overall usage

you're free to keep generating

[sacred helm]

that's great then

[ruby maple]

Wait, does that mean it has no limit currently?

[cobalt hull]

Nope, there's most definitely still limits lol

[rain flower]

So how many images per day is the limit now? It used to be about an hour or two max wait and now it's like 20 hours..

[cobalt hull]

I think Moxi said 300 images

[wary stump]

does anyone have the magic sauce words to make the faces less smooth?

[outer compass]

impossibie

[wary stump]

thats definitely what it seems like

[outer compass]

exactly

[somber coral]

Photo of a community hall where an award ceremony is taking place. The individual from the previous giving scene is being honored with a correctly spelled 'Humanitarian of the Year' award. Applause fills the room, and banners reading 'Celebrating Acts of Kindness' hang prominently.

How to correct spelling mistakes?

[wary stump]

ask it to do the same image with a new seed, and just keep doing that until you get lucky with a perfect one

[outer compass]

just one A

[dim cradle]

get a new seed is new chat?

[wary stump]

if you are using chatgpt then they have generative seeds for the images now. a seed is just a number that can be changed to make a slightly different image using the same prompt. (sometimes the change is more than slight)

[somber coral]

limit will reach

[wary stump]

your only other option is to use photoshop

its not making spelling mistakes on purpose. it tries its best

[wintry dune]

i pressed regenerate on the rate limit message and the wait time went up two hours 💀

[somber coral]

U can just tell the image number itself. Like edit the first image with ...

[rain flower]

OMG

[sinful crane]

Can DALL-E3 only generate 2 images at a time now?

[lethal lake]

yes

[subtle creek]

Please give a pro plan. I need less limits and more images at once and omg just take my money it’s not that hard

[quartz vale]

there's already ChatGPT Plus

[subtle creek]

I said pro for a reason

[quartz vale]

that's what ChatGPT Plus is... is it not?

[subtle creek]

I’m asking for better than plus homie

[lethal lake]

How to maintain the consistency of painting？Does anyone know?

[quartz vale]

so you want the API basically. Good news for you, it should be released later this fall!

[blissful mango]

How can we find out the font name that dalle 3 used

[quartz vale]

like text style that it generated in an image?

[blissful mango]

Nope. Actual font used

Eg, times new to man

Times new Roman

[shut niche]

Guilty as charged. 😅 But any advance notice ⚠️ of a fixed limit, before allowing users to incur a 24 ban ⏰ , would be nice... 😑 It felt like a "surprise gotcha" moment.

[blissful mango]

We might request modern font style, dalle then allocates one. Does it ever create its own font, or are they all pre existing

When it makes spelling mistakes I want to match the font and make manual adjustments

Don’t know which one to use

The font matching sites don’t really work for me

[quartz vale]

Dall-E doesn't generate text with any real font because it generates new content in every output. You might be able to find a similar font to what it generates, but it doens't generate any "real" font

[blissful mango]

Aagh ok, great. I do t need to worry about license issues with the fonts

Is that correct ?

[quartz vale]

I don't work for or represent OpenAI and can't provide legal advice on behalf of the company

[remote sphinx]

Anyone else getting only 2 images?

[quartz vale]

yes, it is outputting 2 images at a time

[remote sphinx]

Why

[rain flower]

2 images has been default for almost 6-7 hrs..

[quartz vale]

#images-discussions message

[remote sphinx]

Well that stinks

[rain flower]

2 images is not bad as 24 hr ban for hitting use cap..

[blissful mango]

Sorry dino, just joined. Thought you did work for them ! You seem so knowledgeable!

[quartz vale]

I think it's more of a rate limit, not a ban 😅

[quartz vale]

all good! I just wanted to make it clear that I can't answer every question 😄

[rain flower]

Well, yah but.. 24 hr unable t o acceess = literally ban in gaming world

[shut niche]

It's only a rate limit if it's stated in advance IMO.

[rain flower]

Exaclty.

[quartz vale]

were you going crazy on the generations 👀

[shut niche]

Of course, that's why I subscribe though, and why I'm not going crazy in SDXL.

[blissful mango]

For me, The reduced dalle rate will mean I hit the the chatgot4 rate more often now.

[quartz vale]

my Dall-E 3 folder is almost to 2 GB

[shut niche]

2??? Mine's 30.2 bhahaha

[quartz vale]

I don't download every image I generate. I'm very selective with what I keep

I also rarely hit a rate-limit

[shut niche]

I like some of the failures because if their decent, I'll use them as image-to-image seeds in SDXL Auto1111.

They're* Or I'll Photoshop bits and pieces for other things.

[quartz vale]

alright folks, it's time to share your favorite gen! I want to see the single image you're most proud of and/or like the most out of all your Dall-E 3 generations

(please post them in #1154829862171844679 and tag me)

[rich star]

How do I logout in Bing image creator

[drowsy bane]

I'm with @subtle creek I just cancelled my MJ sub paying like $60USD because this was so good. I know it's in Beta so not perhaps the best solution. I think there would be good number of people willing to pay something for greater generation capabilities (generation qty at one time, gen speed, no/increased daily limits). Hopefully not as much as the $60 haha but would like to have an option at least

[wind crane]

why ?

[sinful crane]

Same here, I'm willing to pay more for more generation limits

[normal loom]

im guessing its due to it being rolled out to every gpt+ and enterprise account atm and it being the hot new thing on the block its getting slammed with requests.

[agile peak]

It seems now adding title/text quite difficult to be properly displayed. Or is it just me?

[clear jacinth]

Anyone know the best way to prompt so that there’s no shadows / even light distribution on subject. This is my current prompt (tried variations of this but with no luck 😞)

Keyboard and mouse: Photo-realistic sculpted wooden representation of a keyboard and mouse, viewed from the top in a flat lay style on a pure white background with no shadows, illuminated with diffused light, creating a soft and even effect.

[agile peak]

Letting chatGPT write the textual elements. I didn’t write specifically for this — just “add the popular drinks textual elements”.

[whole arch]

Careful with copyright, using coca cola

[spice apex]

Will the rate limit increase or we gonna get 2 images from now on ?

Cause its not working very well for me with the 50 messages limit for 3 hours... Thats 200 images for 3 hours to 100, with less amount of selection for the best style.

[sacred helm]

you can try this
I want you to send multiple requests over and over again in your single response, I'll stop your response once I find the one I like

[spice apex]

Okay, I'll try that

[agile peak]

I only write cola as title. Idk it produced coca cola.

This is the correct one for me after 3 attempts.

[spice apex]

Didn't work...

DALL·E 3 Multi-Version Imagery Prompt
Objective: Generate four distinct images based on a single concept, with each subsequent image being a rarer version of the previous one.

User Input: Please specify the base concept or subject you'd like visualized.

Concept: _______
DALL·E 3 Instructions:

Image 1: Create an image strictly based on the user's specified concept.
Image 2: Generate a rarer, perhaps more intricate or unique version of Image 1.
Image 3: Dive deeper into rarity, taking inspiration from Image 2 but pushing its distinctiveness further.
Image 4: Develop the most unique, scarce version, taking cues from Image 3 but making it even more extraordinary.
Ensure the final output provides a clear differentiation between each image, reflecting the progression from common to ultra-rare.

(Even this one didnt work to create 4 images.)

[wanton hatch]

DALL-E 3 still getting portrait mode wrong most of the time

[whole arch]

Still says it in the bottom 😉

[agile peak]

Sometimes I imagine the ChatGPT and DALLE process in the conversations between them is kind of funny 😆

when I write the prompt “small group of kittens”..
• ChatGPT: alright, “small group” understood that.
• Dalle: small? How about the whole scene should be flooded with more kittens 😂

[vapid elk]

dalle likes to repeat some patterns, when you ask for things that comes in "multiples" dalle usually exagerates the quantity a bit

[glossy scroll]

I'm being limited to only to images at a time. Does anyone know why?

[vapid elk]

this is the current amount of images dalle makes on chatgpt per reqeust

[glossy scroll]

Sorry, I meant to say two images at a time.

Instead of the normal four.

[vapid elk]

this is correct

[rigid bough]

it's changed, it now just does 2 images

[glossy scroll]

Pity.

[agile peak]

I tried the same prompt template from OpenAI website. One of the image seems to produced an inappropriate symbol that resembling german waffen logo.

[vapid elk]

this is something that highly depends on context, if you think it should have not been generated it is a good idea to use the 👎 button at the page to report that result

[agile peak]

I literally copy pasted from OpenAI websites.

[vapid elk]

please, don't send it here

[agile peak]

Alright. I will do feedback.

[wind crane]

So the generations of 4 images are over?

[sacred helm]

I'd say it's not forever-gone

[wind crane]

Is this the case for everyone?
No more than 2 images maximum whereas yesterday it was 4.

[sacred helm]

Is this the case for everyone?
probably
No more than 2 images maximum whereas yesterday it was 4.
same

[wind crane]

why doesn't anyone say anything? Haven't the Reddit mourners come forward about this?

[sacred helm]

that's why I said 'probably'

If not, it MUST be the case, except for enterprise plan

[final compass]

The two image limit has been rather constant topic since yesterday at least in here. When that changed i also noticed, that if i liked a generation it had done, i would ask for the seed number in a follow-up message and it is no longer working. Asking for seed numbers upfront still works.

[wind crane]

It's frustrating, I can only request 2 variations instead of 4 and I have to limit my requests otherwise it's chaos.

[wintry dune]

Yeah it’s kinda terrible, plus there’s a 24 hour time out that somehow goes up every time you try to check. They said the limit was 300 images but you need to press regenerate like 5 times before you get an image with no visible issues so you hit that limit really quickly

[final compass]

I guess the demand has surprised or something. At the Bing side of things i saw a 3hrs wait at some point per image. To be fair DALL-E 3 is the first ai image generator i have been rather impressed and interested in using.

[wind crane]

I'm even more surprised that they limit bing in addition to GPT.

[subtle terrace]

Do OpenAI's policies permit the same person to have multiple ChatGPT Plus subscriptions? If so that's another possibility for someone desperate for more usage.

[rigid bough]

you can create a maximum of two accounts per phone number

[fickle pumice]

question, dall-e only works with gpt-4?

[rigid bough]

yes

[fickle pumice]

ok, thanks 🙂

[wind crane]

Dude......

image is dead

[vapid elk]

you might have skipped this button

[agile peak]

I see there an improvement with producing vintage comic now on dalle (chatgpt).

Especially for wide screen.

[sacred helm]

I asked for new emojis for feelings or things we don't yet have symbols for
can you guess the meanings behind?

[rain flower]

The one with a gun and needle worries me LOL

[small scroll]

how did you do that ?

[agile peak]

I am letting chatGPT to produce poster for animated film first. Then customize it.

[cloud pawn]

is the server too full right now

i can't get any prompts through right now

[agile peak]

Look comfy. This entirely made by chatgpt without initial prompt.

[small scroll]

nice ok thanks

[brave barn]

This two image limit sucks🤦‍♂️

[spice apex]

Bring back 4 image output!

[quartz vale]

donate GPUs and maybe it'll happen

[gilded osprey]

Optimize the algorithm!

[spice apex]

I'm already paying open ai for those GPU's. 20 bucks a month

[wary stump]

Their GPUs cost $30,000 each. Unfortunately that means it would take you 1500 months to pay for a single GPU

[young forge]

Is dall e still with one prompt one seed ?

[drifting hawk]

Share the error message

[drifting hawk]

Trying…

[burnt vessel]

Does anybody know how get results that have consistency with each other for example I want to make 5 pixelated animated images of dogs, this was done via bing-ai

[sacred helm]

use seed'

[sacred helm]

https://discord.com/channels/974519864045756446/1167273182386991105

[burnt vessel]

These are close ups of two from my prompt they both look nice but I like the top one better because it's more blocky, I want each of the images generated to have the same blockiness but with a different species of dog for example the same pixelated blockiness but with a dalmatian

oh its dalle-3 by the way

[sacred helm]

#images-discussions message

[burnt vessel]

thx, i have watched some yt videos on image generators and prompts and there is a whole world when it comes to prompts apparently you can get very specific down where the shadow is and light

[wary stump]

No, they added randomized seeds

[burnt vessel]

When you make an image on chatgpt 4 using Dalle-3 do you actually own the rights to the image or

[plucky hare]

https://help.openai.com/en/articles/6425277-can-i-sell-images-i-create-with-dall-e

[burnt vessel]

thanks for that, so would I be able to copyright my image without problems?

[plucky hare]

You just can't make it seem like you made it yourself without the assistance of AI

https://labs.openai.com/policies/content-policy

[burnt vessel]

Obviously I would mention that AI made the art/image, but there must be some way that artists use ai even by companies like open ai and then copyright right it partially, because it is there's they mention that the image(s) can only be used with the permission of the owner

[crystal badger]

bet if you took out a loan and made this actual thing u could sell it for a load of cash

[plucky hare]

Yeah they clearly indicate that you can use the content for merchandising. I don't know the full implications on your claim to copyright in that case! But there is language about how others who generate content similar to yours have the same claim to the content as you, so basically anything you make with OpenAI's service can be made by anyone else with the same services, and all those people have equal claims to the content.

[burnt vessel]

But it's never the exact same image so how do they get ownership to the image created by myself

[plucky hare]

It could be the exact same image. If someone uses your exact image prompt and uses the same exact seed, they'd get the same result.

[agile peak]

Perfect for next bull run 😁

[crystal badger]

indeed

just start making AI furniture irl sell it at a good price

[burnt vessel]

where do you get the actual seed for the images generated

[crystal badger]

ask thats all u oughta do

[plucky hare]

I'm not certain. I think you can ask the model to tell you. Someone else also figured out a way to find it using some console/debugging tools in browser.

[crystal badger]

or have GPT state the seed each time or per image.

[burnt vessel]

does that work

[crystal badger]

ofc it does

otherwise i wouldn't tell you

[burnt vessel]

So back to this, if I took the image created using dalle-3 and ran through stable diffusion and changed it up then how would ythat change things in terms of ownership and individuality of the image

[plucky hare]

You'd have to check with the content policies and terms of whatever SD model you're using.

[crystal badger]

the usage polcy says for the GPT version that is, that every image u make belongs to you.

but its still watermarked.

so u cant claim u made it, nor can u remove the watermark.

of course this includes commercial use.

[plucky hare]

I think you're allowed to remove the watermark, but you still can't claim you made it without AI. https://labs.openai.com/policies/content-policy

[burnt vessel]

I'm ok with but what happens if you run it through other platforms like stable-diffusion and ended up with a different end product

[crystal badger]

issue is Solbus

u cannot see the watermark

its embedded in the code of the actual image

[agile peak]

Wait a minutes. Dalle got new rules today?

[plucky hare]

I see, it still sounds technically allowable though right? Even if not currently possible.

[agile peak]

No more rate limit per-minutes? But now daily?

[crystal badger]

its set up as such as if u manage to remove the watermark, the image is useless

great right?

[burnt vessel]

So technically I could take a screenshot of the image and run it via another platform

[crystal badger]

like any other image yeah

but if u claim ownership and i made it, i have a legal right to sue.

[plucky hare]

How much have you generated today?

[crystal badger]

various other factors as well.

[burnt vessel]

lol

[crystal badger]

what? if u copy songs its copyright, its the same with ai art, to an extent.

varies by where your from.

[burnt vessel]

I'm not talking about that there is a famous nft artist who uses ai to make art and sell it for millions

[crystal badger]

ohh waste of time and energy.

dont bother

the big "players" have their place u cant get there, even if u tried.

[agile peak]

Probably more than 50 times. Cause I had to do multiple seeds change to get the text right.

[crystal badger]

NFTS are a fixed game

[plucky hare]

More than 50 batches of 2?

[burnt vessel]

It's an example, and people do it and make money out of it

[crystal badger]

yeah i see your point but dont go that route.

its riddled in the legal grey area.

not worth anyones time.

you're better off utilizing GPT V and Dalle for website creation, way more profitable.

[agile peak]

Indeed. 50 times per batches of 2.

[crystal badger]

the idea to make money in the "AI" age of art and etc is to never do what others do under any circumstances.

[burnt vessel]

No not true, there are projects that take off, there was a project that might not have a high floor but it took off doen a eth volume of over 116 eth it began on the 17th of of this month that's over $200,000 for that one individuals project

[vapid elk]

OpenAI attributes any rights of images you generate to you, however it has been decided that AI generated images cant have coppyrights

[crystal badger]

exactly.

to copyright your AI art it must be 20% you and 80% AI

by bad reverse those numbers. 80, you 20, ai.

only then can u copyright it.

[burnt vessel]

I see what you mean but in order to invent, make and be a conniseur you need funds/ money

[crystal badger]

A.T do you know the amount of time you need to get up there?

i know for a fact you cant. if one is just casually uploading AI art.

its a full-time thing, u dont have time for another job, plus the subcategories of research one needs to do to even Qualify.

as someone who has previously been there, in that market, i'd say dont waste your time, but if ur burning for it, go ahead.

[burnt vessel]

I have plenty of time on my hands but I'm not looking to "get up there" or get rich off of any of this

[burnt vessel]

what did you try

[crystal badger]

i wont specify, that will reveal who i am, just saying NFTS, is dying. and if ur gonna go for AI art, Do it properly. meaning dont just hop on the big wave, find a spot within that wave that folks arent using, and start dominating that.

i can even give u a few examples if you'd like.

[burnt vessel]

Ok what did your nft look like and what was the concept very vaguely

[crystal badger]

• Website UI
• Preprogrammed Buttons sliders Art with UI
• Untapped market such as Roleplay.
• etc etc.

Those are the niche things one would go for now. and just dominate them and once u dominate them more discover this and they go for it, but at this point its ur turf so they will have a hard time getting in. meaning more dough for you.

its quite simple.

[burnt vessel]

what did your nft look like and what was the concept very vaguely

[crystal badger]

Lets stick to Art.

inb4 Lugui slaps me.

[burnt vessel]

ok was it partially successful and then it fell of or not at all

[crystal badger]

Successful? ohh yeah most def, living comfortably.

[burnt vessel]

what we talking over 100eth in volume?

[crystal badger]

no comment

Like i said, if you intend to go AI art route and selling it, just do it properly if u wanna make it big

[burnt vessel]

what will me knowing very vaguely do lol

[burnt vessel]

I don't wanna go big making ai art, normal art will always be better than ai art no offence

[crystal badger]

but u wanna earn money doing it.

Well "Better" is subjective.

in due time Ordinary Artists will lose even more hope.

approx 1½ years maybe 2, worst case 3.

[burnt vessel]

Yh by simply making something aesthetically pleasing and selling it, as many as possible, like a capital rasing thing, I would take that capital and do something usefull

[crystal badger]

if u wanna earn dough selling AI art you most likely wont be noticed unless you are different.

thats the main issue at hand

because 2-80 million ppl are thinking the same as you.

hence why u gotta stand out.

[burnt vessel]

Again not ai art, if I was doing ai art iwould put a whioe heap of money into it, promo all sorts

[crystal badger]

Ordinary Art?

[agile peak]

Looks like mentioned logo twice in the prompt, work much better for Dalle to recognize it.

[crystal badger]

Looks great, it has this Anime vibe that i like.

[plucky hare]

Did it change its mind on the rate limit it told you about?

[crystal badger]

Excited for when DALLE gets Vision in the future so it can see what its doing

rn DALLE is drawing blindly.

[burnt vessel]

You see starlight take this image for example this would make for a good nft to mint because the art looks decent, obviously you mention that ai was involved during the making process and yiou make a variety of image that look the same as this with different traits.

[crystal badger]

AT the issue with NFTS is that its useless now, its a dying brand even more so than ever.

Now a Physical NFT. i'd buy. with its own integrated identifier.

[burnt vessel]

I wouldn't say so there is a lot of money in it, when crypto does well once more Nfts will follow, even if only for a short period of time,.

Look at The Sandbox

[crystal badger]

absolutely true A.T, but i know that ordinary folks wont ever get to hold that money.

thats how Hierarchical it is.

and with AI art Delving into NFTs, it''ll drop rapidly in value. even more so than ever.

its like a fight between pepsi and cola, Cola always wins.

[burnt vessel]

Of course not but my generations knows about cryoto and nft's, I heard that online banking with mobile apps isn't much of a thing i the US, idk know how trye that is, but over here with revolt, and Paypal average people are getting some exposure to crypto.

[crystal badger]

Personally with the money ive made off NFTS, i'd rather do stocks. more profitable. easier to get in.

Well Created any Great art recently?, as we all know "Art" is in the eye of the beholder.

[burnt vessel]

I can see what you mean but when crypto takes off that's when that heard mentality get exposed every flocks to crypto, for example next year is the bitcoin halving when good things happen to bitcoin good things happen to crypto, personally I think it will be a shorter window than many are actaully predicting

[crystal badger]

issue with NFTs is that i know it will die and soon.

It's ill suited for this "era" i would say, but it was created too early. but a huge boom will come, no doubt just not in the first 1-4 years.

[burnt vessel]

Haha nahh there's too much money in it to die it might crash drastically but it will pick up or evolve into something very different alongside crypto like the crash of 07-08

[crystal badger]

but hey u wanna do NFTs go do it, i have too much experience in there. i wont stop you. just telling you the truth even if its uncomfortable.

[burnt vessel]

Did you make something doing nft's

[crystal badger]

did i make money? yes, did i make a lot of it? yes but that was in its prime.

[burnt vessel]

I would love to know who I'm talking with what's your twitter

[crystal badger]

Im called StarlightSyntax everywhere, but u wont find my earlier profiles they've been Erased under GDPR Law right to be forgotten. all my previous data is erased.

[burnt vessel]

ok

[crystal badger]

quite easy to do tbh, just get lawyer in on it and ur good.

[crystal badger]

But do you know what i am actually trying to tell you? and explain to you.

[burnt vessel]

Is this you

[crystal badger]

yup you found my Twitter, Congrats.

[burnt vessel]

Is it a burner, r you a well known crypto or ai person but go by Starlightsyntax

[crystal badger]

If everyone is doing x and getting in good on it, means you should not do x since its too risky now that everyone does it.

thats the gist.

[burnt vessel]

is that a reply

[crystal badger]

a burner?, what do you mean?

Temporary Alias?

[burnt vessel]

an social media account you use to go under the radar, burner is london slang, i'm from london

[crystal badger]

ahh slang, No.

I merely DEMANDED companies delete my data with proof or i sue them

thats all

but back to DALLE generation.

Have you generated any recent great images? if so can i see?

[vapid elk]

guys, this is completely #off-topic

[crystal badger]

tried to divert the convo Lugui.

[burnt vessel]

yh we'll have this convo there I didn't see there was an off-topic lol apologies

[crystal badger]

if you have any further questions about "me" u can ask in #off-topic you might have to ping me.

But have you generated any good images, A.T? at the moment?

[burnt vessel]

I depends what one would call good and the purpose as well, what I created before is something that too me looks aesthetically pleasing

[crystal badger]

This one turned out great

[burnt vessel]

see that looks good but it's kitsch no offence or anything

kitsch is the word that means art that was made for novelty to look good

but yes it does look great

art usually has meaning

and sometimes even utility

take the mona lisa it has all three it looks great, it has some meaning behind it bridges, why it was painted

and utility back then as statement piece but also to show her likeness they didn't have camera back then

[crystal badger]

Indeed

[burnt vessel]

If I was ever going to ai art it would be a collage mad e of ai images, but also a very long prompt and somthing that says something about the time we are in but also something that looks very good

[crystal badger]

I use DALLE for website UI creation

And sell those components I make for a small profit after making them functional

[burnt vessel]

Like page design, so for example apple most likely has a website ui team, that why their website is so sleek

[small scroll]

do you have a example to show ?

or a typical prompt

[crystal badger]

I do but won’t, you can easily do this in under 3 minutes utilizing GPT V and Dalle

[burnt vessel]

good question

an example

[crystal badger]

#off-topic

[burnt vessel]

and would you ever consider teaching or onboarding people

there aren't enogh people willing to teach

[remote stone]

Does DALL·E 3 understand hex color codes?

[crystal badger]

in my Experience yes. i've had successful runs with it.

[remote stone]

Okay yeah it does seem to work to a degree, thanks!

[pearl silo]

I have the impression that it's more likely that gpt-4 understands hexa well, but not dalle-3 itself.

[crystal badger]

ChatGPT is the Interpreter for DALLE, Dalle cannot see anything you write, so assuming ChatGPT writes it so dalle can understand there wouldnt or shouldnt be an issue. However what people fail to realize is that DALLE cant see anything it makes, its making the image Blind.

[pearl silo]

I don't understand why you're saying this to me

[earnest flame]

Hey! what exactly do you want to know?

[drowsy bane]

Can't wait for the vision stuff to then check what it actually made aligns with the original request. Rinse and repeat

[earnest flame]

ChatGPT seems to send a prompt to Dall-e. You should, however, be able to ask it to send exactly some prompt over.

[crystal badger]

indeed.

[earnest flame]

E.g.

Test for Hex

[crystal badger]

if u dont specify for keeping the exact prompt it wont. itll interpret instead.

was what i meant.

[earnest flame]

I think it's an important point so it's a good one to raise.

[crystal badger]

being specific is good

its like asking gpt for a story about a dragon then to scold gpt for not giving it a hat when never specified.

[earnest flame]

It's honestly a little hard to say right now what exactly is the most helpful thing to do. Dall-e 3.0 seems to do VERY well at being specific. Here for instance I asked for a cat illustration and it ONLY gave me a cat.

Some people may actually mean they want a pretty picture of a cat.

[lilac obsidian]

Dall•E 3 is very bad with writing and text isn’t it?

[earnest flame]

I tend to side with the specifc though.

[crystal badger]

agreed

[plucky hare]

Better than many (all?) alternatives!

[lilac obsidian]

Like I can tell it do a particular minimalistic image, and it will do it phenomenally well. However it can’t spell basic words.

But I guess you’re right.

[crystal badger]

also depends on your image detail

thats what ive found out.

or rather complexity.

[lilac obsidian]

No detail really. Very minimalist. Black background, white snake logo.

[naive lake]

It's so sloowwwwww

[lilac obsidian]

Mine isn’t so slow

[plucky hare]

Yes, they're different "problems" for the model, really. I bet ChatGPT could give a better explanation than me but script is a largely static "thing" -- letters always look the same. Meanwhile DALLE is designed to recreate images that are much more dynamic. So throwing a command at it like "Put the word 'hello' there" is giving it a very specific task, when it's used to being able to "come up with its own answer" most of the rest of the time.

[naive lake]

Ot

It's getting worse every day

[crystal badger]

its better than nothing

[naive lake]

All I want is to pay more to get more

[crystal badger]

also understand that DALLE cannot SEE what its creating.

its blindly Creating the image

it dosent know whats there

[paper jolt]

I wouldn't say "can't"

[lilac obsidian]

It’s not doing a bad job. 60% of the images it’s giving me have the word spelt correctly. But it is pretty shocking I can’t lie. So much development into this and it can’t always spell the basic 6 letter word properly.

[lilac obsidian]

Why not? Lol?

[crystal badger]

because it can

[naive lake]

for every generation I'mm waiting 2-3 mminutes

anyone else encountering this?

[crystal badger]

Not on my end.

[paper jolt]

Mine took about 30 seonds

[crystal badger]

How are you prompting?

[naive lake]

ughhh

[crystal badger]

prompting is 99% of the deal

Good tip dont prompt DALLE like you prompt MJ its two entirely different things

two Entirely different creation methods

[lilac obsidian]

The prompt is basic and clear to understand. The main meat of what I’m telling it is:

Black background

Minimalistic white snake head

Word

[crystal badger]

is that it?

[lilac obsidian]

Obviously not in that format

[lilac obsidian]

No

[lilac obsidian]

What’s mj?

[crystal badger]

tried something like this?
a minimalistic white head of a snake on a black background above the snake head it says "Word"

u gotta talk not give snippets of words. needs the full context.

To enhance your result u can tell your general idea to ChatGPT, and it'll give you an optimized prompt.

[lilac obsidian]

This is pretty much exactly what I’ve told it

Exactly how I’ve said it

[naive lake]

what are you trying to achieve @lilac obsidian ?

[lilac obsidian]

I got what I’ve wanted from the ai, thanks.

[crystal badger]

hol on ill try generating an image.

[lilac obsidian]

It’s okay it’s doing a decent job

[naive lake]

and what was it?

[lilac obsidian]

It’s just the only problem is that it’s not always spelling the word properly

[lilac obsidian]

Just an image bro

[plucky hare]

It's trained to generate visuals. When it comes to text: letters have a consistent and specific appearance, and any deviation is immediately noticeable. It's like an artist who can create detailed paintings but can't do calligraphy. So it basically has to learn and integrate how text is "different" from normal visual imagery.

[naive lake]

🙂

[lilac obsidian]

It’s fine don’t worry about it

It’s just a little tedious to work with sometimes

I’m sure it’ll get better later 👍

[naive lake]

I just wanted to do some tests on my end

[crystal badger]

i used "Minimalistic white king cobra head on a black background with the word 'Here' above it."

although text is placed below its just the beginning of the convo another iteration and its spot on.

[plucky hare]

I know! I'm just trying to clarify the point that it being good at making art, complex or minimal, are both just separate challenges for the model than text generation.

[lilac obsidian]

Yeah it’s giving me similar but the word I am using isn’t actually a word in the English language. Perhaps that has something to do with it.

[lilac obsidian]

Thank you 🥰

[crystal badger]

whats the word?

now u got me curious

[lilac obsidian]

Sorry, it is a word in the English language

Xd

[crystal badger]

but whats the word

[lilac obsidian]

Apathy

[crystal badger]

Apathy

ahh

[lilac obsidian]

I literally just didn’t know it was a word until I googled it now

XDD

Lack of emotion

[crystal badger]

What are you trying to create?

[crystal badger]

well other than your prompt, what are you aiming for.

Looks great.

[naive lake]

ewww

it's getting the word correct each time for some reason

[crystal badger]

becuase DALLE is a Distinguished Gentleman

only knows hard words

this is my 2nd iteration

Black and gold themed

[ruby maple]

Is it just me or the aspect ratio of the images decreased as well?

[crystal badger]

we can ask DALLE hold on

still only these three

• Square: 1024x1024
• Wide: 1792x1024
• Tall (Full-body Portrait): 1024x1792

[lilac obsidian]

Mine is giving me a lot of incorrect ones.

“Alan”

[crystal badger]

i like the 1st one.

looks wicked

[naive lake]

me too, the first one is dope

[lilac obsidian]

These are the best two thus far.

[crystal badger]

Also DALLe can make Transparent Backgrounds

i did not know that

[lilac obsidian]

But we use black backgrounds