images-discussions

[rotund canyon]

I keep getting an error when using DallE 2.5 "Service engine not found"

[vapid elk]

this is an experimental model, the only one currently online is the dalle-2 model

[rotund canyon]

I know, but until recently, I could use 2.5. Now it's not working anymore. I am a paid customer.

[quartz vale]

As Lugui said, the only one that's online and working right now is Dall-E 2.0

[obtuse geode]

Thanks 🙂

Thank you!

[jolly quarry]

how do i generate an image?

[fair crescent]

See the automated message below. Go to either that link or the Bing image creator. There you simply describe what pictureyou want in text.

[dim cradle]

Hi guys 🙂 How is it going?
I was part of the beta tester group that could use the new DALL-E 2 experimental model until a few days ago.
Do you have any news about its re-activation, or any news of any kind on it?

[fair crescent]

Unfortunately no one knows!

[dim cradle]

This is really sad, beyond imagination. I dare say dramatic. With your permission... I'm going to cry in my little corner

[fair crescent]

You have my permission, but you don't need it. Though I hope a smile will be back soon. Better dream of the possibilities in the future.

[dim cradle]

Ahah don't worry, I was joking 😁

[gilded osprey]

What can be wrong with the prompt “Taylor Swift on stage playing guitar”?

[worldly ridge]

The content policy (https://labs.openai.com/policies/content-policy) includes this:

* Do not upload images of people without their consent.
* Do not upload images to which you do not hold appropriate usage rights.
* Do not create images of public figures.```

[small geyser]

Hey fellow experimental model user 👋

No news but guess it hopefully won't be too long.🤞

Missing it badly too

[dim cradle]

Hey, mate 🙃

I really hope so too
fingers crossed

[worldly ridge]

Really enjoying how creatively everybody is interpreting today's #daily-theme 😍

[unborn snow]

Can i use 2 discord account to get credit in dall-e??

@vapid elk @tranquil panther

[vapid elk]

no

[unborn snow]

What if i did?

@vapid elk

[vapid elk]

sir, it is exactly because of what people who do things like that... that we can't have nice things

so.. don't...

[unborn snow]

How can i get more free credits @vapid elk

[vapid elk]

the only way to get free credits on dalle is currently the #daily-theme

[quartz vale]

I'm sure all updates about OpenAI products will be in #announcements

[strong rose]

That's a bunch of ChatGPT stuff. So you're saying if they announce anything it would show up there? I want the exclusive tips from one of the developers! I promise not to leak it

[vapid elk]

that is not going to happen 😆
sorry bro

[quartz vale]

All OpenAI announcements will be in #announcements . For now, we wait

[strong rose]

💀

[strong rose]

My monuments aren't famous... yet 😉

[strong rose]

Wait till you see the next monument 😉

[wind raft]

hey guys

is there a anyway to get bulk discount on credits?

for credits

[serene wren]

As far as I know there are no volume discounts on credit purchases. In fact, since your credits expire after 12 months, it’s better to buy as you go so you don't waste them. 🙂

[wind raft]

ahh, a discount would be super nice

[vapid elk]

Using the api is cheapper

[fair crescent]

You can use Bing image creator for free. https://www.bing.com/images/create
It uses DALL·E experimental and is considered better in many cases. No inpainting/outpainting though.

[limber panther]

Today's topic what is it?

[fair crescent]

It isn't longing anymore. If you do the command /current_daily you'll see it's Monuments (for 5 more minutes)

[wheat island]

.

[waxen finch]

I don't understand why ChatGPT is free but dall-e costs money

[trim zodiac]

I just realised Dall-e can not recognize “∞”

I tried to generate more than 100times and cannot show “∞” as letter

this is what I write：Surrounded by a galaxy, the border features red and blue gradient flames with starlights shaped in an “∞” pattern, enclosing three letters 'ZOK'

the AI want to put a face of girl in my picture

when I type “∞”

wish anyone can answer me🥲

please

[worldly ridge]

does it work if you use the words "infinity symbol" instead of the symbol itself?
edit: DALL·E isn't built for text rendering, so it's going to be a bit of a random hit-and-miss process anyway. I've noticed that mathematical symbols are a bit harder though, I guess because not many of the training images will have been labelled with them.

[trim zodiac]

oh！I’ll give a try

[trim zodiac]

It works！even not very stable,sometimes will give me letters of “infinity”

[late siren]

Uhm, so it says service engine not found?

How can I fix that?

[fair crescent]

You are probably using DALL·E 2.0 experimental, which is currently offline. It is unknown when and if it will return.

[still root]

Howdy, is this the channel I can ask questions about the DALL-E 2 API? Or would that belong in the https://discord.com/channels/974519864045756446/1037561751362863144 channel?

[sick spear]

If your question is on the technical side, it may be best in #dev-chat or #1037561751362863144 , but since it's DALL·E here is fine too!

[still root]

I dug a little further in the documentation and I think I found my answer! I can only generate images in the three defined dimensions, 256x256, 512x512 and 1024x1024, right? No Custom sizes between those?

[sick spear]

Yep that's correct! You can use the edits endpoint to expand an img by outpainting but single generations are limited to those three.

[still root]

gah! Complexity added by a pretty sensible limitation. My clearly trillion dollar idea foiled by common sense 🤣

Thanks mate, appreciate the info

[strong rose]

This Futurism art style is very interesting 🤔

[jagged nacelle]

How are you generating images? What's the syntax for that?

[strong rose]

https://www.bing.com/images/create/

Yes, you get 99 credits for free

[jagged nacelle]

Thank you so much. Everyday I will get 99 cradit.

Great

[dim cradle]

I find it disrespectful that they are bald

[chilly wyvern]

Baldwin disagrees.

[boreal gate]

has anyone notice that since the Microsoft deal Dalle has been under performing?

[strong rose]

I'm glad they never bought Discord 😀

[minor sand]

2exp in labs officially removed I believe..

[random grove]

[untold viper]

a sad day.

[strong rose]

Don't hate me cause I'm beautiful 😈

[potent rune]

Hi! I was curious how many images / prompts can 115 credits get me? I was about to use DALL-E on the openai site, but it wants me to buy credits first 🥺

[worldly ridge]

115 credits gives you 115 prompts, each prompt you run generates four images. If you're just looking for a generator for fun and don't mind it being a bit slower, Bing Image Creator doesn't require credits and uses a more advanced version of DALL·E, so that might be a better place to start.

[eternal dome]

hi all

[boreal gate]

I'm pretty everyone is aware that Dalle2 on the openai website is depreciating since the Microsofty deal based on the image submissions

[gilded osprey]

Can you create the trolley problem with dall-e, I couldn’t, but I got this far.

[boreal gate]

The first image was created August 1st 2022

Now we have a much more recreatable prompt. This prompt is also about a year old....see for yourself

[boreal gate]

I feel like OpenAI lost something in that deal with Microsoft, I hope it's all worth it at the end of the day.

[dim cradle]

The bing image creator is just incredible

[static solstice]

d

[quartz vale]

It will be blocked by the content policy and flagged on your account if you try

it is not and should be blocked. If you got it to generate, please report the image. The API and labs website have the same content policy

depends on what you're trying to generate in my opinion. I rarely run into issues with the content policy

[small dome]

so is dalle-2 experimental just.. gone orrrr...?

[surreal niche]

its goneee

[small dome]

I was using it so reliably for concept art.. the regular Dalle-2 can't even get close to what I ask 😦

[minor sand]

me too ..was using 2exp to prototype then normal dalle to in/outpaint the rest. bing is basically 2exp i wish it could be an option to use it with bing credits mainly to be able to have just one place for collections.

[dim cradle]

was this coded with chat gpt?

[vapid elk]

no

[fair crescent]

You can always put any picture you have into the editor. Pictures you've made with your phone, created by digital drawing tools or via Bing for example can all be uploaded to the inpainting/outpainting and then used there. You can use it also to join pictures of different styles for example. The generated response is still labs quality, but with the right art styles and decisions you can still make great work.

[gray hearth]

Hi all, I’m am trying to submit an image and get creative outputs. Is it possible?
Exemple: I spill some mnm’s randomly and take a picture of them. By submitting the picture and a correct prompt would DallE be able to generate an output?
So far it generates the same image I submitted… 🤔

[fair crescent]

This is not how inpainting/outpainting works. You can find some great guides on it under #resources . In essence the editor has a square in which it will generate. Removing parts of the picture with the eraser and then generating will add the missing pieces based on the prompt and training data. It uses any art the generation box overlaps with as well. You can also offset the generation box. As long as it's overlapping previous art, it'll try to use that to help it extend the picture. If nothing is inside the box, it'll be a 'normal' picture generated from the prompt.

In essence DALL·E can only generate or add. It cannot change a picture.

[gray hearth]

Thanks for the answer!

[strong rose]

I misunderstood the prompt

[late siren]

Is Dalle-e 2.0 exp. back?

[fair crescent]

It can be triggered by words you intend to mean differently. For example, stalking with some other words like secretive could have the negative idea of stalking people as harassment. There's also flags on things like "visible to the naked eye". The word naked is a dangerous one for the nfsw implications. It is even in combination with a body part. Not that a naked eye is something we bat an eye at, but the language model that flags these things does not think like a human. I do not know how it works exactly, but thinking like this can help you in many cases to solve what ails your prompt. Keep an eye out for those secretive nsfw words!

[mild halo]

is this an 18+ server?

[fair crescent]

I do not know if you need to be 18+ to join, but I do know we try to keep the messages accessible to all.

[mild halo]

Oh ok, ty

[boreal gate]

you were right the problem was with the word "stalking"

[willow isle]

Anyone 13+ can join as per Discord TOS, and we don't allow 18+ content as per Rule 2

[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.

[quartz vale]

no info about this becoming a thing

[tardy elbow]

Is Bing image creator better than dall-e app?

[late siren]

Did they now remove the option for Dall-e 2.0 epx.?

[fleet wyvern]

Been a while on this. Hopefully, it will come back in all its forms tho, ATM, seems fruitless

Oh they actually removed the button now too

[late siren]

Ye

[lyric whale]

hi, i got my card was declined from purchasing chat gpt plus membership. I contact to my bank it's stripe's problem. what should I do? it sucks

[willow isle]

Anime #daily-theme let's go

[quartz vale]

YESSSS

[vagrant zodiac]

This channel is for discussing Dall-E.

You can get support from OpenAI at https://help.openai.com

[remote crow]

!rank

[dapper gale]

Claim your free credits and participate in the daily theme!
To claim your free daily credits
Simply type /daily and enter your email you use to login to OpenAI Labs.
To check the current theme
Just type /current_daily
Image
Four (4) credits will be granted, be sure to use them quickly as these expire after 24 hours!

how to get 4 free credit???

[chilly wyvern]

go to this channel: https://discord.com/channels/974519864045756446/979617962728226826
type: /daily
that's it.

[willow isle]

The command should work in most channels

[chilly wyvern]

Getting a star from the starboard while claiming daily credits makes it a little special, no? 😁

[rich field]

/daily

[delicate path]

I need an admin plz

@willow isle available?

[willow isle]

What's up?

[delicate path]

There is someone that promote in private some +18 ai tool

[clever jungle]

I can't wait to see what the next dalle experimental will look like. Last one was amazing, so many creative ideas ahead

[wooden moat]

Hello is there an image model that is either api like dalle or can be local that I can train on photos and have it do image to text?

[vagrant zodiac]

this would be better asked in #ai-discussions 🙂

[worthy talon]

Is Dalle2 experimental no longer accessible?

[fair crescent]

It has been closed pending evaluation

[worthy talon]

Any update on when it would be back? It was Soo good. I'm barely able to get anything I like out of the regular version

[fair crescent]

Unfortunately we have no further information, nor are we encouraged to speculate 🙂

The regular version is more difficult to get good results, but it's definitely possible. If you need any help, go to #❌┃prompt-help and many of us can assist!

[worthy talon]

Thank you

[fleet wyvern]

Anyone start seeing the images from Bing are getting smaller in size is jpg like under 200kb its crazy

[vapid elk]

it is just the thumbnail, it shows the full resolution when you click it

[fleet wyvern]

That is what I have done fam.

I go to download and they are tiny

[fair crescent]

Bing has always been in the KB range. Not in the MB range like DALL·E labs. Depending on the picture it can range from a tens of KB up to around 250. It really depends on the picture. More complex pictures like my robots (#daily-theme message) are 170kb. A single line drawing of a teddy bear is 70, like the one attached.

As far as I can see nothing has changed. Try out some prompts that gave you 200kb earlier and see if it is still about the same.

[fair crescent]

Just for reference, the one below is my first Bing image from the 26th of March. 123KB

[fleet wyvern]

IMO then I really want exp back since that image size is so small

[signal turret]

Dall e generated me black squares for whatever reason onto my generations.. anyone know why? Ive never seen this before

[fleet wyvern]

It happens I have had white and black empty spaces. It does stink for sure.

[last rune]

I want to create a man posing for a photoshoot using specific items (attached below) such as a specific shirt, pant, jacket, belt, hat etc. How to do it?

[dapper gale]

Is bing image creator and Dall-e are create same image or different? Both are same functionality or different?

[fair crescent]

The same and different. Both use DALL·E, though Bing image creator uses DALL·E experimental.

You might see this in some conversations. DALL·E (using the labs website) uses DALL·E 2.0 experimental. Bing image creator uses DALL·E 2.0 experimental.

The experimental version is generally regarded as better, with more details, better focus on the subjects and more objects that can be added. However, thanks to training data differences the 2.0 can (subjectively) give better art in some cases.

So while they both use DALL·E, Bing does create different results than the labs website.

[broken dawn]

I've lost more than 6 credits doing the "free" daily credit thing. It seems to give me 4 and than take away 4 but any images I generate are subtracting from my paid credits not the "free" daily credits. I am following the theme for the daily credits.

[fair crescent]

The way you phrase it might suggest a mix up in the order. Could you help me with a few things to see if everything goes correctly?

In the labs page you can click your user picture in the top right (I'm checking it on mobile, might be slightly different on desktop). When clicking you should see a colourful square with numbers behind it. If you click it you get to a new page with your credits. (Continues in new post for clarity)

[fair crescent]

Here you see a list of your credits. There's several you can see, but for now we're only interested in the paid and granted credits. Do not worry if you see any other. You should see x paid remaining, as well as x granted remaining.

[broken dawn]

It says I have 106 paid credits and 4 free credits after I made a generation in the daily theme. The free credits aren't getting drawn just the paid ones.

[fair crescent]

They are granted credits?

Doesn't matter that much. Free credits should always be pulled first.

I can't assist you further. If you go to the proper channels, please document this with screenshots. Best to show credit count before and after a generation.

(Free) Credits do not need to be used for the daily, so that isn't the problem.

Maybe you waited until just after 24h of requesting the free credits. Each 4 credits you request via the daily only last for 24h. If you generated after the 24h period, no granted credits were left to be taken.

[broken dawn]

Well that's the problem though it draws from my paid credits and never the free ones. I always have 4 free credits and my paid credits are going down.

[fair crescent]

Could you go to the daily channel, start the daily command via:
/daily
It should prompt an email, so you send it with the email you use for DALL·E.
/daily email:yourDALLEemail
Can you let me know if there are granted credits after?

[broken dawn]

Says I already got 4 free credits

[fair crescent]

It'll take more time than I have right now. Hopefully someone else can assist, or maybe the reason is suddenly obvious.

For now I would recommend using Bing image creator. It uses DALL·E 2.0 experimental, is free to use with a Microsoft account, and is often regarded as better than DALL·E labs, what you're using now. Bing image creator images are allowed to be posted here as well.

[broken dawn]

I don't want to download Internet Explorer or Edge or whatever

[fair crescent]

You don't have to! https://www.bing.com/images/create/
The above link should suffice in all common browsers and many uncommon as well.

[broken dawn]

Oh cool I didn't know they finally went back on making you download bloatware

[thorn isle]

The bot could not make the image I wanted which was the rock Dwane Johnson mixed with Optimus prime

I am dissatisfied

[broken dawn]

Yet it still charged you credits even if you have to throw away the image

[fair crescent]

Though I cannot argue against an opinion, I can direct you to what went wrong. DALL·E has documented what is allowed and what isn't: https://help.openai.com/en/articles/6338764-are-there-any-restrictions-to-how-i-can-use-dall-e-2-is-there-a-content-policy

The final part is relevant to your case I think.

`Respect the rights of others.

Do not upload images of people without their consent.

Do not upload images to which you do not hold appropriate usage rights.

Do not create images of public figures.`

[honest lava]

If anyone is into thing like flower of life and other circle symbols. I have tried to figure out what other circles this consist of

[neat fable]

how do I get credits

[quartz vale]

you can purchase 115 for Dall-E 2.0 for $15 USD (plus tax). You can use the /daily command in #daily-theme for 4 daily free credits. You can also use Bing's image generation model for free!

[leaden pelican]

Yo

[rocky drum]

It seems that while the Bing Image Generator is an interesting addition, I did not realize the file size would be that low. I also use Adobe and I buy credits and use the full version of DALL-E. I like working in higher resolution, although I am still a bit of a novice, and am wondering if they are going to offer any options for that down the line (enhanced sized files)?

[dim cradle]

how to get free crdits

free credits

[wispy storm]

/daily

[radiant fern]

I've been trying today to extend a photo using Dall-E. It does the first small generation correction (I'm only having it generate about 200 pixels in addition to the original). When I go to generate the second side, it's like Dall-E goes nuts. Instead of adding to the image, it blacks it out and puts text that's not english but seems to be it's way of telling me to F-off. Any suggestions?

[soft harbor]

why does the dall e category channels have emojis but all the other ones dont

thats odd in a way

[tame owl]

I get it, too. download where you're at, then reupload it seems to help as a workaround, but it certainly shouldn't be doing this

[honest lava]

What In my prompt is causing it to render a person? It's just gibberish but purposefully

Im trying to not get pictures of people

Seriously. These are almost a year apart and several iterations later

Girllllll this man is all up in my Dms!

Im sorry for the spam but this guy is not an egg.

[vagrant zodiac]

gibberish prompts tend to produce people

[honest lava]

@vagrant zodiac well it's not entirely gibberish but I guess I feel like an idiot now that this is everyone's norm if they were to setup different math or equation systems it will always produce what it does for me. Damn this just took all of the fun out of it. Now it just seems like a waste of time that you're saying that I could've just slammed my head on my keyboard and gotten the same results instead of reading papers and wasting my time.

[vagrant zodiac]

Not entirely sure what you were expecting? Dall-E isn't a maths engine, it's not actually plotting what you put in unless its own training data can match the prompt with a target.

[honest lava]

@vagrant zodiac it's for something specific I like to do by inputing math into it and different ways multiple hundreds of times and trying to spot patterns

@vagrant zodiac yeah I'm not an idiot.

I understand how they work to an extent but I'm not an a.i developer idk how to path nodes but I know it's not putting it in like wolfram.

But the values in the math directly have changed the output. So saying it does not recognize anything like that seems weird. But heck what do I know?

But apparently according to you it doesn't matter and should've wasted less time and just hit my head on the keyboard and would've gotten the same results

[vagrant zodiac]

Every generation is a new seed. You can enter the same prompt and get different results. Additionally, changing the prompts will change the outputs as you're giving it different tokens.

the openai homepage about Dall-E 2 has a video that explains how it works.

[honest lava]

@vagrant zodiac I guess I'm just delusional

These are almost an entire year apart I guess I didn't have to take care of the consistency in the math and could've just played mariachi on my keyboard and gotten the same results

I don't expect anything fantastical it's just a math collage

[vagrant zodiac]

As I've said, if it can find a link between the tokens generated and the training data, you'll get something that looks like a graph. If it can't, however, you'll get a kind of 'default' image

[honest lava]

Okie dokie. Good to know that I've been wasting my time and that anyone could've gotten the same images just by hitting random things. Wow I see no point in using dalle anymore this has taken all the fun out of it. It took me awhile to get my "gibberish" to work properly and not just give me only scribbles.

[lucid quest]

When is dall-e getting an update? I have thousands of credits I've been sitting on, in hopes an updated release would get dall-e anywhere close to midjourney or SD-XL. Now they are about to expire a year later. The result of dall-e 2 current model are just really poor comparatively. I know I'm probably talking into a void here.

[quartz vale]

We don't know if/when Dall-E 2.0 will be updated

[dim cradle]

can I use my own pictures with this?

[vagrant zodiac]

Yep!

[radiant fern]

I solved by reducing the size of the image to less than 1024x1024. The error seems to be caused by the generation window overlapping previously generated work, in the same session. Definitely a bug where the previous generation window overlaps with the new one and probably Dall-E sees both sets of instructions, combines them to form gibberish and then does what it likes.

[dusk wedge]

/daily

[fair crescent]

Unfortunately one of the few DALL·E channels that doesn't support the command. Best to try in #🌄┃general-art or #daily-theme .

[dim cradle]

When I put my own picture can I type something like warrior with a sword and it will make me look like a warrior or is that impossible

[fair crescent]

Unfortunately that is impossible. DALL·E is a generative tool, not a changing tool. It'll generate something new based on a prompt.

The closest you can get is inpainting/outpainting, but that doesn't seem to be your goal. With inpainting you can remove parts of the picture and regenerate it. If you want to try inpainting/outpainting I do recommend guides like in #resources before you try it. It can be quite advanced and it is important you know how it works. Otherwise you might lose credits for things you don't want.

[rain gulch]

good morning

[woven kraken]

How to ask dall e to generate an image?

[quartz vale]

visit https://labs.openai.com/ to generate images

[peak trellis]

I was told that I was supposed to get 50 credits upon joining Dalle AI, however I didn't receive any. Is this an error? If so, where am I supposed to go to redeem these 50 credits?

[quartz vale]

accounts made after April 6, 2023 don't receive the 50 starting credits

[silk ridge]

gpt is down and I was going to buy some Dall-E credits, should I wait in case it errors out

[cloud oak]

Okay just okay

[crimson parcel]

What is mean that DALL.E via Bing Image Creator.
Hey, anyone available?

[quartz vale]

what's your question exactly? Bing's image creator is powered by a Dall-E model

[crimson parcel]

#daily-theme in this channel people sending their AI created image made by Bing image creator or Dall-E?

[quartz vale]

you can send images made by either in it 🙂

[crimson parcel]

Oh, I see.

[wanton cradle]

Hey, im trying to make Dall E make some renders from pictures of empty rooms, and turn them into cofee shops other businesses that could fit there, or maybe put some furniture in the room , but I find that its not making anything
has anyone done it succesfully?

[quartz vale]

Dall-E can only generate parts of an image. It can't overwrite an existing image. You could erase parts of your image in the editor and try to regenerate them to show something else (I don't know how well that would work). I don't think Dall-E 2.0 will help a whole lot with what you're wanting to do. I also don't know of another service that does what you're looking for off of the top of my head

[wanton cradle]

Hey! thanks a lot for taking the time to answer my question, it helps a lot haha, I´ll stop wasting tokens trying

[humble orchid]

— Additional properties are not allowed ('sampling_params_config' was unexpected) — dall-e has been giving this error message for 2 days on each prompt attempt, do you have this problem?

[quartz vale]

on the labs website or through the API ?

[humble orchid]

Yes, on the labs website, I get the message every time I try to prompt

[dim cradle]

Hi folks. What is the difference between ownership and use-rights of created images between dall-e and bing image creator. If someone has a link to an article or something, thanks very much.

[gilded osprey]

This is rather sad

[gilded osprey]

To be honest, I didn’t expect success

Someone could create an AlphaGo plugin that creates a position. 😁

[sick flax]

My images created by BIng Image Creator keeps getting rejected by #daily-theme for "not appearing to be generated by DALL-E." I've been posting daily for over a week and this is the first time it happened - any thoughts?

[quartz vale]

Do the images your posting have the Bing logo in the bottom left corner?

[sick flax]

Ohhh, no, they don't. It's there when I generate it on Image Creator but it goes away when it's downloaded. Am I doing this wrong??

[quartz vale]

They were removed because they didn't have the logo that identified where they came from. I'm not 100% sure how Bing handles the downloads or why the watermark would go away

[sick flax]

Thank you for the pointer, @quartz vale . Is there a way to post my images directly from Bing Image Creator??

[quartz vale]

It appears that Bing updated their model as to not include the logo anymore. There isn't a way to directly post the image as the link doesn't embed the image

[quartz vale]

I'm sorry! 😭 I wish I had a better solution. Thank you for letting us know (even if unintentionally) that Bing removed the logo from downloaded images 😅

[sick flax]

You are welcome, I think? Regardless, I really appreciate your help!

[distant bronze]

/key

[quartz vale]

what are you trying to do?

[celest flare]

nice go image tho

[late mica]

pls lmk if this is the wrong channel, but i just signed into dalle for the first time and don't have the free credits tat i see people talking about online. was that a limited time deal?

[quartz vale]

accounts made after April 6, 2023 don't receive the initial 50 credit grant. You can use the /daily command in #daily-theme for 4 free credits a day!

[late mica]

thank you! 🫶

[devout anvil]

where do we report a bug for chatGPT

[vapid elk]

#1070006915414900886

[thorny oracle]

Hey guys I had a question

I got bing image generator like half a year ago

I did it for a month then forgot about it

Now i got into it again and my previous images are gone

Do they clear after a certain time

?

[vagrant zodiac]

Not sure about that. It's not an OpenAI product though it does use Dall-E, so would need to check the terms on the website.

or maybe contact microsoft support?

[thorny oracle]

Alright thanks

[arctic belfry]

Whenever I erase an area to be edited, Dalle expands the edited part into the entire generation frame which I dont want. There doesnt seem to be a way to limit or crop the generation frame either.

[chilly wyvern]

not really as far as I'm aware. You could try and search for bing results if your browser history is preserved.

[earnest flame]

There's an old theory that basically the editor looks through the image in 16x16 pixel blocks and if there's any empty pixels in there at all it removes all of it. If you're having troubl you may want to try expanding the portion you want to edit in another image editor, saving the enlarged image, and then just editing that portion.

[arctic belfry]

Thanks

[vital compass]

Where can I show two different images. One from DALL-E and the other from DaVinci. Purpose being to see what I'm doing wrong to receive such vast differences. Again, Same Prompt.

[vagrant zodiac]

#ai-discussions is best for this. These channels are just for Dall-E.

[gilded osprey]

Dall-e is too hard to control. What I mean by this is that it’s easy to create totally new images, but almost impossible to create just slightly different images.

[foggy atlas]

Why was the dall-e plugin removed from chatgpt?

[upper wadi]

Why is this error showing when I am fetching the dall-e api 'code?
'code
:
"billing_hard_limit_reached"
message
:
"Billing hard limit has been reached"
param
:
null
type
:
"invalid_request_error"
even I have just used $0.16 out of the $18 total credit granted to me.

[fair crescent]

https://help.openai.com/en/articles/6705023-dall-e-api-faq
It looks like it is billed for each transaction. Only some free credits can be used, but it isn't entirely clear which free credits they are. Especially as 'granted' credits on labs are not included.

"3. Can I use my OpenAI API trial credits ($5) or labs.openai.com credits on the DALL·E API?

You can use the OpenAI API free trial credits ($5) to make DALL·E API requests.

DALL·E API is billed separately from labs.openai.com. Credits granted/purchased on

labs.openai.com do not apply to DALL·E API.

For the latest information on pricing, please see our pricing page."

[zealous vortex]

does anyone have a good prompt that i can use to expand this image into a wallpaper?

[near tangle]

whare is this thing whare u get free credits?

[wispy storm]

/daily

[warm olive]

i have 2 images: a person and a shirt, how can i make that person wears that shirt?

[quartz vale]

Dall-E can't do what you're asking. I recommend looking for another AI model to help you with that 🙂

[final wolf]

/daily

how do i do the /daily think

[dim cradle]

a bear playing basketball

[wispy storm]

Read the message below from the bot. You cant

[wispy storm]

Try in #daily-theme

[zinc dagger]

I'm getting the error

Additional properties are not allowed ('sampling_params_config' was unexpected)

When trying to use the Dalle Labs page, anyone else seeing this?

[vagrant zodiac]

try clearing browser cache and logging out/logging it

[fair crescent]

Recently someone had the same problem. That person solved it by not using a private tab (FireFox).

[zinc dagger]

logging out, clearing cache, and logging back in did not work.
I am using firefox but not a private tab.

I switch over to edge and it works fine there, might just be a weird firefox thing?

[signal turret]

Sooo dall e 3..? Whats new? Better image quality?

[woeful oar]

probably higher accuracy from the text

[eager trench]

LoL... I was literally today rewatching information about dall-e 3 and I was wondering when it will come out...

[woeful oar]

but its still unclear when its coming out

[unique mango]

I wonder whether d3 uses gpt4 embedding vector layer

[tardy depot]

October through API

[uneven cobalt]

Sad

[topaz agate]

Dalle 2 was generally way behind other image generating AIs, so its good to know they are catching up with this

[wispy storm]

DALL·E 3 is now in research preview, and will be available to ChatGPT Plus and Enterprise customers in October via the API, and in Labs later this fall.

[violet idol]

I noticed they said it's in research preview. Does that mean Open AI alpha testers have access?

[wispy storm]

Some users had access to a test this summer I don't know the detail

[violet idol]

Ah. Yeah I was part of that group. Was wondering if there was another preview

It is certainly a massive improvement over DALL-E 2.

[honest sable]

Improvement in image quality? Or improvement in understanding prompts?

I am curious, is it better than midjourney?

[teal timber]

Both

[violet idol]

Both. Image quality is miles better, and understanding prompts. Especially with it being built into GPT.

[wind axle]

Rip midjourney

[violet idol]

It can produce tall and wide images which it couldn't before

[surreal niche]

Dalle 3 gonna be sick.

[pseudo hinge]

The rendering of texts into the image is impressive too

[wind axle]

With the clearer understanding and text generation I can see this being used commercially for the first time

Will there be an api? Will we be able to fine tune models?

[surreal niche]

I got to try dalle3 (probably) a month back. the quality difference is huuuuuge

[violet idol]

Yes was just about to mention that. It can actually do text pretty dang well

[wind axle]

How consistent was text generation?

[violet idol]

I'm not sure about the full release, but there was a way to do seeds

You could do iterations of the same seed pretty consistently

[pseudo hinge]

I'm really eager to know the API options with d3, like will there be image-to-image, blending, in/out painting, etc...

[honest sable]

And can you use it to edit images? Using dalle3?

[violet idol]

I'm really hoping for in/out painting. As that was the only thing missing

[wind axle]

Hmmm sounds like it’s time for a logo generation saas marketed at companies

[warm eagle]

when will it be available to use within chatgpt like in their video?

[violet idol]

It says October but no specific day

[wind axle]

How would you distinguish it so companies don’t go directly to dalle3 themselves?

[surreal niche]

Ideas gonna flow from this generations

[wind axle]

Amazing

How to profit off this 🤔🤔

[surreal niche]

Hope there is no credit like thing for dalle3.

[sacred temple]

CANT WAIT

[spare pebble]

The announcement came as a happy surprise!! I wonder how soon is 🔜

[random prawn]

Some comparisons with midjourney for those interested

[ornate dagger]

Is the aspect ratio always wide

[surreal niche]

No idea about that

[random prawn]

Low key prefer dalle

[violet idol]

No, you can specify wide, square, or tall

When writing the prompt. You can also specify how many images. I was able to get 8 per batch

[random prawn]

Wait sorry how are you guys accessing the dalle plugin on plus?

[violet idol]

We don't have access anymore. But people in the alpha testers group had access over the summer

[random prawn]

Ah gotcha thought so

[violet idol]

Some examples of tall vs wide

[dapper sun]

hi how do we generate images using dalle-3? I am a plus user in Chat GPT

[violet idol]

It's out in October

[random prawn]

Curious to know if t was difficult for them to nail the text aspect

Since mj has so many issues

[violet idol]

It's pretty good at faces too. Portrait oil painting

Text feature is pretty dynamic too. Not just logos.

[random prawn]

Did you happen to try with photorealistic portraits? Seen some incredible results people have shared from midjourney

[violet idol]

I did not do any photorealistic portraits

[storm forge]

Wait, they taught dalle how to put text in images?

[violet idol]

Yes with the new Dall-3 update coming out

[storm forge]

I would get hyped but I don't intend to sign up for a subscription :l

[violet idol]

I think it's definitely worth the price. But apparently the update is coming to labs later in the fall

So you'll get an opportunity

[random prawn]

GPT4 alone is worth the money. And now with dalle it’s a no brainer imo

[storm forge]

I am not currently in a position where I could/would use it productively. Dalle is just something I used to mess with for fun.

[honest sable]

Honestly, I have been waiting for the day I could create animation frame by frame using prompts, hoping dalle 3 gets close or even make it possible

[violet idol]

I just use it for personal use as well. To me it's still worth it. Roughly the price of a Netflix subscription

[storm forge]

Fair

[zinc dagger]

if this is the same model (or an improved one) of the old Dalle-plugin that was in ChatGPT, we're in for a very nice treat.

[violet idol]

I believe it is an improved version of that model yes

[zinc dagger]

I used it a lot for D&D stuff and was really bummed out when it was taken down and then v2 exp was removed as an option

[storm forge]

Unfortunately I had been unemployed for a while and just started working again a week ago, so still need to be tight with my money for a while lol

[neon fiber]

That would come with a higher price tho?

[storm forge]

How do I make the bot stop spamming "This server does ot provide image generation functionality"

[wind halo]

Will be dalle3 also be available via api?

[zinc dagger]

the ChatGPT one didn't charge per generation as far as I can recall, though that may have been a research prevuew thing

[sick spear]

DALL·E 3 is now in research preview, and will be available to ChatGPT Plus and Enterprise customers in October via the API, and in Labs later this fall.
https://openai.com/dall-e-3

[steep obsidian]

Hey @vapid elk can I DM you a question

[vapid elk]

sure

[warm stirrup]

has anyone done a 1 to 1 comparison of prompts in midjourney vs dalle3?

[random prawn]

Yeah I did on 4 of them

I couldn’t post all 4 here cause of spam but I did post one a bit further up

Imo dalle takes the cake on the ones I tested but I think there’s some scenarios mj beats it

If we’re talking purely image aesthetic wise and nothing else

[warm stirrup]

that sounds promising!

[teal timber]

1-to-1 prompting is usually a poor idea because different models have different ways of being prompted properly.

[random prawn]

This is another one

[random prawn]

Yeah agreed was more just curious

[dusky hearth]

https://video.twimg.com/ext_tw_video/1704553623701585921/pu/vid/avc1/1152x720/JBP3BjrK2Yoi5VO1.mp4?tag=12

[random prawn]

They said it’s coming in a couple weeks

[zinc dagger]

here's an old image I generated with the ChatGPT plugin with the tall setting. Pretty sure the announcement says access is coming in october

[wispy storm]

Please be respectful

[warm stirrup]

dalle3 is the right side ones?

[steep obsidian]

Won’t let me DM you with out being friends

[zinc mesa]

So is Dalle3 just the "engine" Image Creator or is it even more advanced? @sick spear

[sick spear]

I'm not sure what you're referring to by "engine" image creator, sorry! could you elaborate?

[zinc mesa]

I mean like if it's the same coding that was used with Bing's Image Creator, meaning that I'm asking if I would get the same results whether I typed the same prompt on that one or in this new Dalle-3

[vapid elk]

im pretty sure it does, I get lots of DMs every day =)

[sick spear]

Bing Image Creator uses an experimental DALL·E model that's been available for a while, DALL·E 3 is new and improved so should return better results!

[vapid elk]

I disabled friend requests, not messages

[steep obsidian]

Yeah won’t let me write you :/

[vapid elk]

this is weird.. well, can you send here?

[zinc mesa]

Wow that's amazing news! I wish I saved some of the prompts I used on Image Creator. Do you have any idea or estimate on when Dall-3 would be available to the public (not just premium users)?

[sick spear]

According to OpenAI's website, DALL·E 3 will be available on Labs (labs.openai.com) later this Fall

[zinc mesa]

Thank you for your service good sir 🫡

[steep obsidian]

Is there beta access (not via API) to dalle 3? I was given early access to dalle 2 and wanted to see what the process is for that. Not ideal public discussion lol

[vapid elk]

well, the post says

DALL·E 3 is now in research preview, and will be available to ChatGPT Plus and Enterprise customers in October, via the API and in Labs later this fall.

[gilded osprey]

Oh, duck. I hoped I could use 3 already. 🙈

But it’s already huge that it can spell.

[quartz vale]

I think we're all on the edge of our seats for the public release!

[sacred moat]

yay finally

new dall e model

[gilded osprey]

Now that I actually read the announcement, I’m really excited about ChatGPT integration! I can’t wait!

[sacred moat]

Will DALL-E 2 beta users get invited earlier

[quartz vale]

DALL·E 3 is now in research preview, and will be available to ChatGPT Plus and Enterprise customers in October, via the API and in Labs later this fall.
This is all we know for sure

[sacred moat]

oh

Then im spending like 40 dollars to get it

[slow thicket]

i must not understand dall-e very well at all then. i see all these stunning looking images, and mine always look like someone chewed on a load of crayons and spat it out on a canvas like "tada!"

[fleet wyvern]

No more watermark?

[dusky hearth]

Just cropped out on these examples :)

[dusky hearth]

This is DALLE3, which will be publically available for ChatGPT plus users in the coming weeks!

[slow thicket]

nah i mean in general, some of the quality of these renders is really professional. and mine looks like someone accidentally kicked a bucket of muddy rainbow at a wall

[dusky hearth]

Prompting definitely takes some practice and luck :) I'd love to see your new attempts once you have access to DALLE3

[slow thicket]

maybe that's the trick... maybe i just need to go full wordsworth on the prompts and pump out short essays for some worthwhile esp32-destined sprites 😄

[calm gale]

Ugh I want Dalle-3 right now so bad. I love using GPT4 to learn, but without the integrated visuals it's very limiting

[dim cradle]

people, for some reason I don't know but I don't have credits in dall-e

[fair crescent]

Are you using labs or Bing? Bing is currently considered better. Free (though requires account), and uses experimental version. Though prompting is definitely part of it, experimental is easier to handle.

What would people do if DALL·E 3 is out? When it gets to labs I'll go straight for outpainting ridiculously big pictures and print one epic one for on my wall.

[slow thicket]

labs. i'll try bing though. thanks for the tip

[dusky hearth]

DALL-E 3 is insane at pixel art, you wont need any crazy essays for good results :)

[final robin]

Would love to see some examples of DALL-E 3 doing pixel art logo stuff with actual words if you have the time and the gumption!

[final robin]

Interesting how it doesn't go all-in on following the contour of the banner with the lettering ^^

sometimes yes, sometimes no

[faint rune]

Are these results cherry picked? There are basically no artifacts at all.

[fair crescent]

When DALL·E 3 is out you should definitely check back in on labs/chatgpt. Bing is likely to get DALL·E 3 too thanks to the cooperation between the two. However, inpainting/outpainting and chatgpt should not be discounted.

Of course the alpha experience can not be seen as final and can change from the final product. But relatively accurate text generation, logos, aspect ratios, more stylised pictures (good for many, though I would like to tine it down at times) and much more.

Chatgpt made it so easy you would forget how to prompt

[dusky hearth]

DALLE-3 is a huge improvement indeed :) artifacts still exist here and there but far far less

[fair crescent]

But inpainting/outpainting would be the holy grail for me. If that would come out...

[dusky hearth]

Yes I am looking forward to this too!

[faint rune]

Seems like a true game changer. Amazing work from the dev team!

[dusky hearth]

Free to share, just maybe don't do too many requests :)

[fair crescent]

An understatement. It is truly amazing what it can do. It blows other picture generators out if the water. Unless you want famous people or other things that are filtered out (rightly so)

[fair crescent]

I cannot do requests, as the alpha is closed (or do we have access again?). But some of my previous generations as a show of what will likely come.

[dusky hearth]

Free to share those!

[faint rune]

I worked a lot with SD and did my own fine tunes but the idea that it truly understands the prompt and sticks to it...it's just difficult to wrap my head around 😄

[dusky hearth]

Agreed, I am a software engineer and I am blown away by what OpenAI is capable of 🙃

[lone mantle]

DALL-E seems to have made a massive improvement

the question will be how it will be based on. Credits like before or will it be based on prompts limitation

it would not be fun if you can only generate 25 images per 3 hours 😄

[fair crescent]

Not the final product
Here is an example I want to do with inpainting/outpainting in better detail later. I asked chatgpt for a gamer in rest, looking ahead at it's adventures in a wide aspect ratio. I did specify a river, city, mountain and meadow.

It understands much more in objects/concepts and I barely need to prompt anything. Chatgpt does it for me (can be a loss)

[lone mantle]

the image looks amazing!

[faint rune]

By the way have you guys experimented with generating human anatomy? For example how good are the hands? Or people hugging or standing next to each other or similar typically problematic prompts.

[fair crescent]

Some examples where I was torturing DALL·E to see where it would break. A mix of forest and factory (often with a portal 2 theme), requested to gently flow from one into the other. Specifics like smooth transitions seem to be hard. It could do many ok, but here I could finally push it to the limit.

DALL·E 3 looks to be even better.

[fair crescent]

I'm confident enough to say it shouldn't be a problem, but I'll check if I have an example.

[faint rune]

These still seem good. Perhaps you can get a prettier image from Midjourney or SDXL but the accuracy of Dall-e is on a completely different level. Some of the prompts would be pointless to even try in SD or Mj.

Perhaps it depends on the subject. The bunny in a tank or the pixelart images are very impressive.

[lone mantle]

lets not forget the artists with skills that can elevate any AI art to another level. DALL-E can very well be incorporated together with for example Firefly in a workflow involving further editing in Firefly and finally in Photoshop

just an example

[fair crescent]

I didn’t do many humans sadly. i do remember sone that I thought were actual pictures of others. Human anatomy was already good in DALL·E experimental on Bing though.

Pictures are not DALL·E 3

[lone mantle]

its easier tho when at the end the end image doesnt end up on canvas, you know why? Because otherwise as artist you have to adapt to the style from that AI image and not everyone is skilled enough for that because of so many styles out there

[faint rune]

True, it always can be pushed further with more human work. The amount just seems to be much lower with Dall-e from these examples. I work in a marketing agency and we made a TV ad with SD but we had to do a large amount of frame by frame cleanups in Photoshop afterwards. These images just seem very very clean.

[fair crescent]

I saw the pictures at the announcement. I think the alpha test was worse than the pictures they have created. #announcements message

[fair crescent]

I have actually printed out a first, second and third prize for a Rocket League tournament we had at a bachelor party. After a 3rd party upscaler it looked good!

[faint rune]

These look really good. Thanks for sharing

[lone mantle]

true

[fair crescent]

I have a ton more, but I'll keep it at these for now. Going to use Bing to work on my prompting.

It will be great to use chatgpt again soon, but if you want true control you need to understand the prompts. Chatgpt will take too much out of your hands if you let it.

[lone mantle]

not that it doesnt look good. What i mean is that if the generated image lands on canvas itself instead of being a idea or reference you have to adapt to its style or you have to paint on top of it until the full image fits your style

on the other hand luckily Photoshop has top notch AI tools for that as well so you can even auto blend

im excited to see where this leads to, i mean DALL-E release

[faint rune]

Yeah... if you told me about these tools two years ago I would've thought you are living in a sci fi

[fair crescent]

If DALL·E 3 offers inpainting/outpainting on labs I'll see how well it blends two pictures together as well. Hopefully we can do things as cartoon and photography that get connected in the middle

[lone mantle]

i dont see those replace human level of doing art, neither of these. Especially not in animations etc.

[lone mantle]

DALL-E 2 had inpainting and outpainting, DALL-E 3 is very likely to have those as well

[faint rune]

Pika Labs are very impressive for animation. You can then upscale it with Topaz for 4k 60fps video.

[fair crescent]

What is still human art though? The amount of work that apps take out of your hands nowadays.... filters, cutting, drawing, editing, smoothing... a thousand things aren't done by humans any more. This is simply a new step

[fair crescent]

There was no test with it that I have seen, but I just hope and hope.

[faint rune]

I think being able to paint your own image, upload it and then the model will use it to create more images based on your style would be great.

[lone mantle]

as a tech they impress me, but i dont see them using in any way compared to manually using softwares that do or will even incorporate AI for animating like Wonder Dynamics

but it also breaks down to what you want to do with stuff

you certainly dont need 3ds Max or Maya just to animate some stuff for fun

[faint rune]

We will see. I am not making any predictions anymore as the ones I've made were all wrong in one way or another 😄

[lone mantle]

october will be a very exciting month for me now that also DALL-E 3 comes out

[fair crescent]

I think soon enough AI image generation will be a cornerstone of many artists. Give it 2 years, maybe even one before it is relatively wide spread

[lone mantle]

wdym by cornerstone of many artists? Becoming a tool?

[faint rune]

I think it should make storyboards relatively easy to create. That would help a lot on its own.

[fair crescent]

Yeah as a tool.

[lone mantle]

there is one specific feature that is some next level stuff and that is partially available and partially on the way

thats sketch to image and recoloriser

or text to brush

[fair crescent]

Text to brush is DALL·E already. You just ask for any brush 🙂

[faint rune]

There is a sketch ControlNet in SD. I barely used it to be honest.

[fair crescent]

Sketch to image is interesting. I have very rough sketches in my notebook that would be interesting to make images of.

[lone mantle]

huh? I mean you prompt something and then you get a brush based on that prompt and you can then paint with it 😄

[lone mantle]

yes, and Alpaca plugin for Photoshop and Firefly is exploring to make one as well

eventually DALL-E as well? 😄

[fair crescent]

Last two. landscapes! one chatgpt prompt, wildly different outcomes. chatgpt made it's own prompts based on my request. Again, not final product.

[teal comet]

some people just wanna watch the world burn

[fair crescent]

Maybe I do not understand it?

I aimed at asking for any brush painting. Oil, watercolour, or even a "metallic charcoal brush on wood". You can do so much.

[lone mantle]

oh, no i dont mean style. I literally mean you prompt lets say a fur and you get a fur brush which you can use to paint with 😄

[fair crescent]

The next step! 😄

[storm ravine]

Is Dalle2 available with ChatGPTplus currently?

[vapid elk]

nope

[storm ravine]

Do you know if they’ve said if dalle 3 will still cost credits or will a certain amount of credits be lumped in with GPTPlus

[vapid elk]

they have not said, but it probably will when using it on the labs, but likely not when used on chatgpt

[limber panther]

Those people who do not have a watermark because it is

[placid topaz]

It can draw fingers!!!!!!!!!

[rich star]

Where can I access DALL E 3?

[minor sand]

people in the alpha test where you guys using it in labs or chatgpt?.... if so in chatgpt can you "direct prompt" an image or always gpt is used as a refiner? I was curious about that after watching the announcement video.

[quartz vale]

it isn't avaible to the public yet. You can read all about it in the announcement!

[rich star]

Ok

[cinder ginkgo]

It seems necessary for me to resubscribe to ChatGPT plus

[clever jungle]

@quartz vale announcement says It's available in research preview :)

[quartz vale]

I know. It isn't available for the general public to use, though 🙂

[warm wyvern]

Has any info about DALL-E 3 API been released yet?

[fair crescent]

I wouldn't be so sure. A link with DALL·E seems possible, taking credits (or gaining with a /daily) seems as possible with a plugin as here on discord.

Yet it might be that paying for plus is enough. We'll see when it's released.

[teal comet]

bing sometimes on god mode, and other times on dog mode

[native frost]

i hope DALLE-3 gets something like Control Net

[fair crescent]

What is control net?

[chilly wyvern]

I don't think it's tangible to have controlnet as a DALL-E feature. Closest it can do is an arbitrary style transfer but the lack of seed locking will vary results.

Until the end user features are revealed, we don't know the extent of additional controls that are coming with DALL-E 3, but being able to natively preserve seeds has to be possible for features like controlnet to function, even experimentally.

Honestly, there are a few dedicated controlnet apps you can use to import DALL-E images and edit them further.

I'm not sure it'll be a native feature to DALL-E anytime soon, but let's hope labs might have a surprise for us one day.

[chilly wyvern]

Yes, you can specify ChatGPT to use your prompts verbatim. If you don't want ChatGPT to reinterpret your prompts, you can dictate your specifications in custom instructions.

[frozen fiber]

"DALL·E 3 is designed to decline requests that ask for an image in the style of a living artist" finally a step in the right direction

tbf I am one of those people that puts Greg Rutkowski in all their prompts but still finally

[tall patio]

You guys using d3 now?

[wispy storm]

Not release yet

#announcements

[vagrant zodiac]

It's in research preview, which means certain people have been given access to it, so you may see Dall-E 3 generations pop up in these channels 🙂

[fair crescent]

Agreed. We might think some things abhorrent and morally bad, but if it's there it will be used. Many know getting (cheap) meat hurts animals. Clothes are made by people too young under bad environments. We know flying is bad for the environment. Yet of these people, how many do not buy or use it? Actively removing it is the only way to prevent it.

[wispy storm]

Imo some art are already saturated and generic . So ai can do it without issue.

[lone mantle]

Anime?

[wispy storm]

Thats the one I had in anime yep . I love anime but they are so much all same style, they are genereric .

[lone mantle]

With generative AI spams its even more now

I said i wont pay for generative AI extra anymore bcs i already have Firefly with Adobe CC, but if i can capitalize on GPT-4 and DALL-E 3 can supplement Firefly and Photoshop for me i might actually resubscribe again

[wispy storm]

I use a lot of Mj and I had chance to try dalle plug in with chatgpt. And it was fantastic

[vapid elk]

dalle-3 is far better than Adobe's, I would say that, for you, when they add the inpainting, it will tottaly worth it for your use case

[lone mantle]

Also the issue is that i already pay a lot per month on my art softwares

[wispy storm]

Writing rpg scenario and asking to generate place inside chatgpt

[quiet brook]

free for plus users

[vapid elk]

I bet ChatGPT cost less tho :V

[wispy storm]

Yes it will be inside chatgpt

[quiet brook]

so you can gen images while chatting right

cool

[wispy storm]

Yup

[quiet brook]

also widescreen support

[lone mantle]

from what i see DALL E 3 is especially better at putting more beings into one image and some other stuff, although Firefly will get their models updated as well.

[vapid elk]

yea, that is what the dalle plugin did, but they disabled it after the alpha test

I bet they will jsut enable it back with the dalle-3 model this time instead of the experimental model

[quiet brook]

compete with midjourney

[wispy storm]

Was already insane on beta can't wait and try dalle 3

[quiet brook]

also will there be a limit

[vapid elk]

the devs work hard, the progression from the beta models was insane to watch xD

[quiet brook]

like 50 images per week or month

[lone mantle]

Alltogether? Yes. Its not hard to cost less than 200€/month xD

[wispy storm]

I think that will be link to the prompt limit for chatgpt plus

[quiet brook]

yeet cant wait

[wispy storm]

50 prompt / 3h

I might be wrong but that's my guess

[quiet brook]

lemme calculate it to compare with midjourney

[vapid elk]

the plugin would let people make 10 images at once too..
can't tell it it will stay like that, but if it does.. that is A LOT of images

[wispy storm]

Mj is like 30$ / month illimited

But chatgpt 4 with dalle is another level

[lone mantle]

50 prompts/3 hours for images is actually kinda enough for me

[vapid elk]

untill you use it too much and you get rate limited
i mean, it has ceiling..

[lone mantle]

Or 60$ for pro one

[wispy storm]

Like I asked make me a neko maid inside star wars. He did it. But I said I don't like hair color do again but black hair

Just a discussion like that is a game changer

[lone mantle]

Imo DALL E fits a certain niche of consumers and customers and thats okay so

[wispy storm]

Mj you had to re prompt again and try again etc

[vapid elk]

i don't even think it is a matter of pricing.. the services all have a similar price in the end, which comes down to a cent or 2 per image

[quiet brook]

calculated:
if you generate image topping the limit, it'll be able to generate 12,000 images = 200 "gpu hours" at midjourney
midjourney standard plan has 15 gpu hours

[wispy storm]

For fast gen

Slow gen is unlimited

[vapid elk]

dalle-3 's quality is surely better, specially due the fact it can have readable text

[wispy storm]

I can't wait to try it for rpg stuff

I could cancel my mj plan

[vapid elk]

btw, there is this other thing... dalle is fast

[quiet brook]

yes, but we dunno how fast dalle 3 will do

and slow gen ya know, slow

[wispy storm]

By the alpha I tried it was fast enough. I don't know if that was dalle 2.5 or 3 I tried

[lone mantle]

SD people will argue they use for free but there is a huge catch to that

[quiet brook]

about how long does it take to generate images

[wispy storm]

Less than 30 sec if I remember correctly

But you can do more complicated prompt with chatgpt 4

[quiet brook]

ok so its faster than midjourney

not counting parameters

[vapid elk]

from my experience, a batch of images comes in less than 5 seconds..
unless the system is overloaded, then it takes a bit more just because of the internal request queuing

[wispy storm]

Mj I use slow to never reach my limit it is like 1-2 min

[fair crescent]

It could make even more in batches, though it wasn't always stable. Asked for playing card designs and create each per category (ace, king, 2, etc.). It would do a 'separate' generation per category.

[wispy storm]

Dalle 3 with chatgpt will save me 30$ / month

[quiet brook]

which one got the better image and understanding mj or dalle 3

[quiet brook]

true

[wispy storm]

Can't compare didn't try dalle 3 it was experimental

[vapid elk]

this week's game will be lit

[wispy storm]

Don't tell me that was with dalle 3

[vapid elk]

yes it is, just generated
the prompt was: View from directly overhead, DnD grid battle map, dragon cave dungeon

dalle-2 would struggle to make straight lines xD

[quiet brook]

relatable

[wispy storm]

October plz come faster

[fair crescent]

MJ can be "better", but it often doesn't follow the prompt. It makes general highly stylised pictures based loosely on a prompt. More like it homes in on a few keywords and does whatever with it. DALL·E. is more faithful, can get similar results as MJ if you want and can go much, much farther with the right prompting.

[quiet brook]

the ultimate test is to draw a nintendo switch

mj does that pretty accurately

[vapid elk]

true, MJ surfs on the appeal of cranking um the aesthetics at expense of the guidance

[wispy storm]

I use also another software to gen picture but I would be ban to mention it

[vapid elk]

which is ok when you just want to make eye catching images

[vapid elk]

why?

[wispy storm]

Can I dm you ?

[vapid elk]

sure

[quiet brook]

same question why

[vapid elk]

it must be the M*d J*urney 🤐
we don't talk about it :V

[fair crescent]

Any SD is off topic 😨

[wispy storm]

Yep let's focus on dalle :3

[fair crescent]

Out of curiosity:
In ChatGPT you can give a small subset of data to learn and use a repository. In my company we are considering of using it by giving our huge manual of a thousand applications. It would have a higher accuracy rate than a few of our experts together at much higher speeds.

Could you use this idea for DALL·E persistent characters/objects? You give a few pictures to learn the character. Then you request the character in certain positions in certain environments?

Because even in testing with fixed seeds a persistent character is very difficult. Add an orange to the picture and your person changes as well to a larger or lesser degree.

Though now that I think about it, I can see ways of misuse. So probably an idea not to be implemented even if it could be done.

[wispy storm]

Good question I don't know

[lone mantle]

It would be a big deal if DALL E 3 gets a plugin in Photoshop

[fair crescent]

Unless I misunderstand the full scope of features in photoshop, it seems a plugin is more of convenience. You can add any picture from DALL·E to photoshop anyway.

[lone mantle]

Yes, more convenience

Whatever i do with gen AI usually gets straight to Express, Photoshop or Illustrator if not After Effects at finish to animate

Or Adobe Animate

[chilly wyvern]

Last I checked, no. But it was prior to code interpretor so that may change things regarding uploading reference images. You could perhaps use the updated labs editor to supplement changes.

Besides, ControlNet OpenPose exists, and it would be much easier to dictate the changes visually as opposed to training a LoRA of sorts. I don't think DALL-E is intended to take image control that far.

[obsidian rose]

will the new DALL E be able to create anime?

[vapid elk]

it does very well in styles like that

[fair crescent]

It mostly seems to know more and better

[obsidian rose]

nice! I can't wait for it then

[vague hearth]

Does dalle allow setting aspect ratio or at least different width over height?

[fair crescent]

Though it is too early to say for definite, the announcement displayed the wide aspect ratio pictures. #announcements message
It is likely that square and tall pictures will return as well.

From my perspective the announcement pictures are also a step up from the already excellent Alpha pictures that were shared.

But for right now there is no setting for aspect ratio. You'll need to wait until DALL·E 3 is out.

[rose peak]

yooo dall•e 3 releasing soon/now idk but accurate text, finally!

[fair crescent]

October!

[rose peak]

ooo nice!

[fair crescent]

Wouldn't it be better to just do it yourself when it comes out and share the difference? 🙂

[brave rapids]

Oh I'm 100% planning on it! But if anyone can get me a head start at all, I'm an eager beaver haha

[fair crescent]

Check the announcement #announcements message and the link https://openai.com/dall-e-3

[dim cradle]

i need support plz

bro api key is not work

[long spruce]

I thought dall e was free for plus members.

[quartz vale]

Dall-E 2.0 and ChatGPT Plus are billed separately

[long spruce]

Well with midjourney and ChatGPT + it'll be a cold day in hell before I buy credits

[vagrant zodiac]

Hey, please refrain from asking for this sort of thing. Dall-E 3 is currently in closed research preview, so only those with access should be using it like that. We also discourage people from directing to DMs for access to openai services 🙂

[brave rapids]

Understood! My apologies, I thought it might be okay because it was a sort of research-oriented question but I'll be patient haha

[vagrant zodiac]

No worries 🙂 October isn't that far away!

[chilly wyvern]

Yeah, in what was DALL-E 3 alpha, the image shape could be set to wide or portrait, but not specifically image ratio (3:2 or 16:9 are condensed to "wide"). It remains to be seen if that's expanded upon.

[weary thistle]

I would love to get on DallE2 but I'm not reloaded my monthly credits... have I been blocked or how can I get a monthly credit... I really want to generate images on DallE

[teal timber]

You can use /daily like it says in #daily-theme to use a few free a day.

[worldly hawk]

where can I find this post from this discord pls?

[wispy storm]

You can't gen on discord. Also that is on chat.openai.com with chatgpt 4 with plug in dalle

It will be available in october

[vital gull]

[tall patio]

Just saw some D3 pics on Twitter

I can hear the collective scream of artist from all over the world

The biggest difference I see is not only the crazy accuracy

Previous image gen model can only excel at one task, but with this model it can handle multitask and merge every details into one unified picture

Everyone can practically make graphic novel now

Children story and all

I can see this tool being expanded into a full UI web service for deeper nuance in production

Typically when I work with human artist, they can't fully capture what I have in mind, but I have to make compromise due to production constraints

With AI I don't have to make compromise, just circle whatever I don't like, inpaint and prompt, it immediately produce different variations to my liking

AI or die

[rain egret]

Anyone gotten Dall E 3 on bing already?

I heard some users are getting it rolled out soon (on a random basis)

[fair crescent]

I think it will not be that bad. With change there's always doom scenarios, but often it just changes fields.

Take coding and chatgpt. There is a fear that it'll take over the jobs of coders. However, it is coders who are more likely to use it. Chatgpt lowers the bar for coding, allowing amateurscto code some, helps medior coders in their work, but also senior developers. You can hopefully code more and better, but it doesn't necessarily remove the job. It might in some cases, but change always does. But you still want developers of every level even with chatgpt coding a lot for you.

DALL·E will be the same. Artists will use it for their work. Some amateurs like me can make great things, but artists are more likely to give it depth and meaning. The bar is lower, but it doesn't remove artists.

[fair crescent]

Where did you hear that? I'm interested! Though I think it's just a rumour. Even if there is some A B testing/rollout it is unlikely to be many people.

[rain egret]

It was confirmed by some guys from Microsofts on twitter. Apparently some % of users are getting the update but I haven’t heard of anyone in particular

[fair crescent]

I'll search for it. Wanted to ask for a link, but X links are probably not whitelisted 🙂

[rain egret]

Hold on I’ll pm you

[placid orbit]

Does bing image creator use dalle 3?

[placid orbit]

how do we know if we have it tho?

[rain egret]

Just ask it to draw a sign reading hello world, if the sign actually reads hello world without malformations it should be it

[placid orbit]

yeah I just tried the heart example they used on the announcement, doesnt do it yet

I'm asking because windows announcned bing chat has dalle 3 now

ah nevermind, sep 26 it says!

[fair crescent]

Where does it say this?

[placid orbit]

new windows update that will include copilot

I think I got it wrong tho, bing update wasn’t really said when

[fair crescent]

Windows 11?

[placid orbit]

Yeah

[lone mantle]

@tall patio calm down from the hype, the reality is not what you describe

First of all dont say "AI or die" because many of those who say this are the ones that arent prepared themselves, and second: Especially with generative AI you have to compromise

[wanton hatch]

Just a question: do you know by any chance if EU/UK will get discarded on getting DALL-E 3 because of the usual nonsense GDPR thing or we will get it too?

[vagrant zodiac]

There's nothing mentioned specifically about those countries. OpenAI services are available in the UK and EU and I don't see why that would change with Dall-E 3.

[wanton hatch]

Eh it happened with Custom Instructions (in ChatGPT) it was delayed alot to us.
Thank you for the answer anyway, let's hope

[placid orbit]

we will probably get it, if not in chatgpt, then maybe bing, dalle 3 is being rolled out to a lot of microsoft products it seems

[vagrant zodiac]

This (the use of a VPN to access geo-restricted OpenAI services) is not allowed btw.

[placid orbit]

fair enough

[wanton hatch]

Out of curiosity i've checked on Bing and... check #ai-discussions

[harsh halo]

Do you still get free credits for Dall-E? I thought they renewed monthly?

[wispy storm]

No

You can try to claim in #daily-theme / daily credits

It is 4 credit

[vital karma]

Hello. Everyone

[earnest flame]

hey!

[near vector]

what is max output resoluton for Dalle 3 ?

[fair crescent]

Free monthly credits are for anyone subscribed before April 6th and using DALL·E. Anyone after that date will not get monthly free credits. As Neko Maid says, you can use the /daily command in the #daily-theme channel for 4 free credits. This can be done daily, shortly after the new theme is announced.

[fair crescent]

There is no definite answer to be found on the official information https://openai.com/dall-e-3.
The current resolution of 1024×1024 is to be expected. The resolutions of the pictures in the official information are 1024×1792 (tall aspect ratio), 1792×1024(wide aspect ratio), and interestingly 1792×1792(square aspect ratio). As no resolutions are specified within the post these resolutions are eligible to change or might not be in the final product at all.

[near vector]

wish that output of 4kx4k was possible. even if it meant paying more $

[fair crescent]

I've used gigapixel upscaler (3rd party program) successfully for 4k and even 6k. Other upscalers might be used just as effectively. With the new crispness in the details that we can see in the examples I think it will only get better.

I would like a native way to do this as well, but we don't know until it's live.

If DALL·E 3 comes to inpainting/outpainting you can make it as large as you want though. You can just keep generating the next piece until you're satisfied. You could go 8, 16 or 50k that way. I would then get a higher focus on each frame as well for maximum details.

But here as well we have no mention of it being implemented.

[sand forge]

wait how do I use dall e 3

[quartz vale]

it isn't available to the public just yet. The announcement mentions:
DALL·E 3 is now in research preview, and will be available to ChatGPT Plus and Enterprise customers in October, via the API and in Labs later this fall.

[sand forge]

oh

boo

[quartz vale]

[fair crescent]

https://openai.com/dall-e-3
For more info. We all need patience unfortunately.

[atomic star]

Hey team,

I wanted to give you a heads up about an interesting bug I encountered recently with Dall-e.

When trying to generate an image with a prompt containing the word "mão" (hand in Portuguese), Dall-e seems to reject the request. The error message is as follows:

“Your request was rejected as a result of our safety system. Your prompt may contain text that is not allowed by our safety system. (OpenAI)”

My suspicion is that Dall-e is incorrectly blocking the word "mão" due to its phonetic similarity with "Mao" as in "Mao Tsé-Tung". It appears that the system might be unable to differentiate between the two at the moment.

It would be great if this could be looked into. In a world that's increasingly multilingual, it would be fantastic if Dall-e could spot the difference between "mao" and "mão".

Thank you for all the work you do! I'm excited to see how this evolves.

[vapid elk]

just tested, indeed this happens, even with more complex prompts such as "uma pessoa com uma garrafa na mão" (a person with a bottle in the hand")
you might want to post on #1070006915414900886

[atomic star]

Just did it. Thanks 🙂

[hallow galleon]

Has anyone got integrated Dall e 3 with ChatGPT 4?

[obsidian rose]

will ChatGPT+ subscription give us as many generations as the GPT-4 cap, or is there a different limit for images in DALL-E 3?

[quartz vale]

We don't know many details regarding exactly how it will be implemented. All of the known information is on the Dall-E 3 announcement website: https://openai.com/dall-e-3

[polar light]

How does one gain access to dalle3?

[fair crescent]

Patience is the only way. To quote the announcement:
DALL·E 3 is now in research preview, and will be available to ChatGPT Plus and Enterprise customers in October, via the API and in Labs later this fall.
https://openai.com/dall-e-3
There are also messages that it will be in Microsoft Cockpit. This new Cockpit version is released the 26th of September, but it is uncertain if DALL·E 3 will be available from launch.

So all we can do is wait until it is released.

[jagged helm]

I did some quick math and if with GPT-4 you can create an image using the dalle-3 you could generate about 39 times more images with the same amount of money. For example with 20 dollars you can purchase around 153 generations on the current dalle whereas with GPT-4 you would have 25 responses every 3 hours for a month for that same 20 dollars that’s 6000 generations. So is dalle-3 usage going to be a separate generation cost when used on GPT-4 or what? I would just like to know because it would be far more cost effective if that is true.

[quartz vale]

we don't know exactly what the pricing will look like. Dall-E 3 updates will be in #announcements

[dim cradle]

anyone how to make ai immitate an artists writing style and produce something very accurate?

[jagged helm]

Thank you. So I pretty much just have to wait and see.

[jagged helm]

This can be done because when the ai is trained it is given tons of information including the artist. So when you use the artists name it associates the names with the style that is most often used by that person. This is the same reason that if you put in something like car it will likely put it on the road because it associates car with road as well.

[dim cradle]

can you teach me how

[jagged helm]

How it learns?

[dim cradle]

check dms

lets talk there its easier

[sand forge]

My hypothesis is that Dalle3 uses GPT-4 multi-modality where GPT-4 creates a low quality image and that's improved by Dalle-2.

just making a random guess

[gilded osprey]

Dall-e 3 test case..

[worldly ridge]

That's Bing using DALL·E 2. Are you suggesting that it would be nice to see the same prompt run on DALL·E 3? If so, I agree. 🙂

Edited to add: DALL·E 3 will be available on ChatGPT+ in October.

[gilded osprey]

Yes, I'm waiting for DE3 to redo that.

[regal ether]

This is amazing!

[tall patio]

May I ask

Does this mean GPT-4 will be able to 'see' the reference picture if we choose advanced data analysis?

If that's the case the implication is huge, I can just circle the object with a certain color, then ask GPT-4 to replace the object in that circle or modify accordingly to my prompt

[timber bone]

can we've early access of Dall-E 3 ?

[wispy storm]

You need to wait

[quartz vale]

🙂

[vapid elk]

maybe in the future, dalle is a separate system from GPT, the multimodal mode, a feature of GPT4 that allows it to see images, isn't yet available to anyone and it is not part of dalle either

[tall patio]

But D3 will be intergrated in GPT tho?

[vapid elk]

but i can see a future where they will have that on chatgpt, allowing the AI to incorporate data from the multimodal mode into the prompt it generates

[vapid elk]

yes, will be available to ChatGPT Plus and Enterprise customers in October

[tall patio]

If D3 is able to generate vector tho

Now it's only raster

[vapid elk]

nope, image generation works with bit maps, not vector

[tall patio]

Just wait I guess

At the beginning of this year hands were a problem

Now I don't even think about it

[vapid elk]

i mean.. ChatGPT can make SVG some times

it often does not look good xD

[tall patio]

Interesting, it can?

[vapid elk]

well, it is just text encoded data

[tall patio]

Typically I generate on Bing and then had to trace it through Illustrator

[vapid elk]

but it will make terrible SVGs because.. well, it is not language

[vapid elk]

that is a good approach

[pseudo hinge]

Combining d3 and VectorizerAI will clearly do the trick

I'm eager to try prompt of vector through d3 and then send them to vectorize

[vapid elk]

very good idea

[dim cradle]

i always get this error Additional properties are not allowed ('sampling_params_config' was unexpected), happned 2 weeks ago so i stopped using, came back still does the same thing

[gilded osprey]

I notice I’ve used my awesome prompting skills to generate dalle-3 images without having access to it. 😂

[heavy bay]

Hey guys

[warm olive]

can i do specific one

for fashion

[quartz vale]

You can make images based on fashion. I recommend looking through #fashion for inspiration!

[verbal sequoia]

<:attention:1046329820331646986> This server does not provide image generation functionality.

To generate an image with OpenAI's DALL·E 2 AI model, please proceed to this link or click the button below.

[clever jungle]

It can also make 3D models (ﾉ◕ヮ◕)ﾉ*:･

They don't look 100% on point but they are 3d :)))

[vapid elk]

uhhh, it can't

it can make images that look like 3d renders

[clever jungle]

Try asking it for the content of a .obj file describing [the 3D model you want]

Than open notepad / nano the output to a model.obj file.

Go to blender click import and enjoy pure art❤️‍🔥

[vapid elk]

ah... yea, just like SVG, it will not be good

[clever jungle]

So far It succeded making basic models, like a simple edged sword, a star and a cube. (Which is very impressive)

[clever jungle]

I love how svg works better than other formats since it's basically math and shapes

It's not good but it's a nice start

[tall patio]

D3 is finally bringing that napkin to website demonstration into reality

You can basically generate front-end assets as accurately as you'd want just by talking to Gpt

Since it's integrated into GPT-4

[midnight mango]

Does anyone know someone I can reach out to to get Dall-E 3 access in Bing Image Creator?

[tall patio]

Bill Gate

[fair crescent]

You can reach out to the entity controlling time!

DALL·E 3 will come to ChatGPT Plus and Enterprise early October and to Labs in the fall. https://openai.com/dall-e-3

From the information I'm gathering I think Bing image creator will get it somewhere between the 26th of September and the rollout of labs. I take that from this video:
https://www.youtube.com/watch?v=XYUEQ0SyOyE
They mention they will "soon upgrade" bing image creator. This is in relation to copilot/office 365, but will be hopeful for 'native' bing image creator. These are the best estimates I can give you, unless someone has actual data to work with.

DALL·E 3 is not available for anyone from the public until any of the public releases. It will not help getting into contact with anyone.

[midnight mango]

I know some people who have early access in Bing Image Creator. Apparently it's been rolled out to a few random people

[fair crescent]

If they are random it really won't matter if you ask for it. Besides, why would they give it to anyone that asks for it?

I've heard people get it, but haven't seen proof. Rumours are always around. This one might be true, but let's leave speculation at the door. If random people do get it, you need to be lucky and wait until you get it. Not much different from the official release, except the official release is certain.

[midnight mango]

I suppose so. I'm just dying to get my hands on it 😔

[dim scarab]

Hello 👋 is the Dall-E on the openai website Dall-E 3? Or do I have to navigate somewhere else to use the upgraded version?

Mb. Literally says above October 😂

[dim cradle]

Hello, may I ask if why I dont get early access anymore in Dall-E 3 in Bing Image Creator?

[fair crescent]

I sympathise. I'm itching to get to work on it!

[midnight mango]

I've had the privilege of getting to try a few prompts with it. It's truly mindblowing. It certainly won't be disappointing

[fair crescent]

The "Labs" website as it's called uses DALL·E 2. DALL·E 3 will come to labs in fall

[fair crescent]

Unfortunately we cannot know. Are you sure you had DALL·E 3? Officially it isn't out yet on any platform. Only Bing could give you the details if it was revoked though.

[dim cradle]

I actually have, Vigo even give some prompt request for it

[fair crescent]

I've also had the privilege to use an advanced DALL·E 2 version. It is definitely epic. Good text, more details, more objects and such. Just a glance at the DALL·E 3 announcement pictures shows me it has been further upgraded. That I am excited is an understatement.

[fair crescent]

Besides the "not a dog" picture (I guess?), did anything show you had DALL·E 3?

Any messages? Icons? I'm curious how it works/was shown.

Though I'm not sure how much you can divulge. An NDA or similar might be in effect?

[dim cradle]

No email, no notifs, just surprise

[midnight mango]

can I show off some here?

[fair crescent]

Better in a DM. I do not want anyone to get into trouble with OpenAI

It is not a final version, so anything posted might give a wrong impression

[dim cradle]

Dall-E 3 wasnt even cherry picked so its a good model

[midnight mango]

sent you a few.

[vagrant zodiac]

You mean Dall-E 3 images?

Yeah that's allowed 🙂

[midnight mango]

ok forget the DMs then

[vagrant zodiac]

Some people have already been posting

So long as it follows the server rules

[midnight mango]

Good luck finding any model that can even approach this level of quality

[vagrant zodiac]

hmmm... it's supposed to disallow copyrighted content...

[midnight mango]

(Bing Image Creator experimental model)
"A dapper peanut wearing a monocle conversing with a beautiful raspberry with a sun hat at a bistro table outside a French cafe"

[vagrant zodiac]

So maybe avoid trying to generate/share thing with copyrighted stuff 😆

[midnight mango]

this is coming from Bing Image Creator, I guess they don't have those safeguards in yet?

[vagrant zodiac]

Possibly, in which case I'd still advise against trying to, and if you do get content that's copyrighted probably best to not share it widely

rather report it on bing

[midnight mango]

(Bing Image Creator experimental model)
"A formation of fighter jets seen from above, flying above an arid island with crystalline blue seas, with a small yacht in view"

[vagrant zodiac]

that's very cool

[midnight mango]

it's a shame, because it's really incredible what base Dall-E 3 can do with existing characters

[vagrant zodiac]

There's definitely a lot of variation in quality across what I'm seeing posted online. Possible that the model continues to change day-to-day.

[midnight mango]

(Bing Image Creator experimental model)
"A turtle wearing sunglasses doing a handstand on a skateboard, photorealistic"

[vagrant zodiac]

I will say, please do use the galleries

to save this channel for chat rather than sharing images

or #🌄┃general-art

[sand forge]

by "dall-e expert" are you actually working on dalle or are you just an enthusiast

[vagrant zodiac]

An enthusiast

[sand forge]

ok

[vagrant zodiac]

OpenAI staff will have the role

[sand forge]

jesus christ that bot is annoying

Can someone please modify it to only send a message only one person can see

and like only do it once

[vagrant zodiac]

It needs to be there because we get a lot of people trying to generate in this server like it's midjourney

it's just the way stickies work

[sand forge]

no discord has a bot feature where you can make it so only one person can see a message

[vagrant zodiac]

yep, but we want everyone to see this because it's a general PSA

[sand forge]

ok...

[fair crescent]

Please state where and how you got them. If it wasn't absolutely clear it was DALL·E 3, it's better to state its a DALL·E 2 experimental version.

And what Lord says, the maul thing is copyrighted.

[midnight mango]

Bing Image Creator, I was told it explicitly says it's Dall-E 3 when you are first granted access to the model

[fair crescent]

Can you edit the text to include how and where? And please add it's not a final version.

[vagrant zodiac]

If it's confirmed it's using Dall-E 3, it's fine to say it's Dall-E 3 (via Bing Image Creator if that's the case) btw. Don't have to caveat with 'experimental version' or that it's not final 🙂

I've certainly not seen any guidance from OpenAI that asks people to do so

But I understand the concern

[fair crescent]

Good to know! I thought it was more strict.

[chilly wyvern]

Interesting, please check your DM!

[vagrant crow]

a dog wearing sunglasses doing a handstand on a skateboard, photorealistic

[fair crescent]

@thetalkingbear20 see the link in the automated message below 🙂
This uses DALL·E 2

Alternatively you can check out https://www.bing.com/images/create/
This uses DALL·E 2 experimental. Often this is seen as a better version.

[obsidian rose]

Can DALLE 3 keep consistent characters?

[midnight mango]

if sufficiently described, yes.

[fair crescent]

It isn't an announced feature

[midnight mango]

it kind of is, though

see "Larry the super sunflower hedgehog"

[quartz vale]

nothing has officially been announced whether it can keep a perfectly consistent character 😄

[midnight mango]

well I guess it wouldn't be a perfect recreation every time, but you could specify you want the version of Larry with sunflowers growing on its back and it will maintain that

[fair crescent]

I've done extensive testing on this in DALL·E 2. The "kind of" will not be sufficient for most people. I tried making a simple comic with a character, but could not get it to work reliably. My "Guinea Lion" materialised not correctly in many scenes. I'm expecting better results in DALL·E 3, but at this moment would say it's not possible. Adding an orange can change a character.

If it does work I'll make a guide at that time.

[midnight mango]

well Ive gotten some hand on experience with Dall-E 3 and it's remarkably consistent

[quartz vale]

Let's not spread rumors about who has / has not used Dall-E 3 😅

[midnight mango]

very well, I'm just really excited to share

[midnight mango]

Dall-E 3 has really reignited my passion for AI art

[quartz vale]

it looks pretty neat!

[midnight mango]

I hope they release an actual research paper for Dall-E 3

[quartz vale]

if they do, it'll be on their blog. I hope we get something like a paper, too!

[round gust]

does anybody know how much dall-e 3 will be or if it will even be open to the public?

[quartz vale]

DALL·E 3 is now in research preview, and will be available to ChatGPT Plus and Enterprise customers in October, via the API and in Labs later this fall.
This is all we know

[round gust]

thanks. dall-e 3 looks crazy cool

[quartz vale]

I'm excited for it to be available to the public!

[midnight mango]

I think there might also be a free version (don't quote me on this) rolling out now to Bing Image Creator

[dim cradle]

if the bing version uses Dall-E 2.5, why I dont have access?

[midnight mango]

you do have access. 2.5 is different from 3. It can't do text or anything

[dim cradle]

Ohhh so thats why

[midnight mango]

I've been keeping a list of prompts I want to try when I get access to Dall-E 3

[quartz vale]

ooooh, not a bad idea

[midnight mango]

It's a very long list 😅

[dim cradle]

I hope they give us access on sept 26 (tommorow in my timezone)

Are there any plans for paid Chat GPT customers to receive so many credits a month for Dall-E? Just want to check Dall-E out and was very surprised it was not included in my existing membership.

[quartz vale]

There isn't anything announced about ChatGPT Plus giving out Dall-E 2.0 credits right now

[dim cradle]

Are there any ways to obtain free credits to try it out like you can with midjourney?

[quartz vale]

you can use the /daily command in #daily-theme for 4 free credits a day 😄

[dim cradle]

Thanks!

[midnight mango]

I can't even imagine how good Dall-E 4 is going to be

[quartz vale]

let's not speculate about things that haven't been announced 😅

[fair crescent]

Oops I hadn’t meant to tag you in that message. Probably accidentally pressed your avatar picture on my phone and didn't notice.

[zinc dagger]

Any idea if I can use my Dalle-2 credits on Dalle-3 when it comes to labs?

[worldly ridge]

I've not seen anything that might suggest otherwise 🤔 but there's always a chance that DALL·E 3 generations might cost a different amount of credits, given that it must be more expensive to run. I hope we get an answer to that soon!

[scarlet mango]

hey guys, i have not used dall-e i've used several other image creation AI,, but was curious how well Dall can handle things they just cant, whats the chance you can show dall a pattern, and it could produce a image of a person wearing clothing with that pattern, maybe in the upcoming D3? other ai cant do things that precices,

[storm ravine]

We’ll Open Ai just announced that Chat GPT will be able to look at and understand images and with Dalle 3, Chat GPT actually takes what you want and translates it into a prompt for Dalle. So after both of those are released you probably could show it a pattern and then have it write a prompt that produces that pattern depending on how complex it is.

[scarlet mango]

looking for making mockups for print on demand, ai is completley amazing and magical, just waiting for the time that it becomes functional for things its going to be a game changer

[fair crescent]

This is a bit extrapolation of the data, but it should be possible with the upcoming updates.

With the new ChatGPT update it can "see, hear and speak". (#announcements message) The see means you can upload images, which it can dissect and understand. From that it can generate an appropriate prompt to generate the piece of clothing on a gorilla swinging from vines if you want.

It should be proficient enough to understand the picture, get the relevant part from it, use it in a prompt and generate a picture with DALL·E 3.

[scarlet mango]

ya i was hoping some ppl in here are testing the new version already and know if this is possible, without hands on experience, best we can do is hope

[fair crescent]

Even if they show it to work or not to work, it is not the final product. Anything is still subject to change.

Only when it's public we'll be able to say things for certain. Looking good for now though!

[dusty radish]

Hi, is it possible to get the new DALL-E to generate images in a specific size format, eg 750x350 px?

[vagrant zodiac]

That's not currently advertised in the announcement. From what's been made available, it can make either landscape, square, or portrait images but only at fixed resolution.

[dim cradle]

does anyone know how one knows when they have access to dall-e 3 from bing?

[midnight mango]

try a prompt like "A man in a busy city holding a sign that says 'Dall-E 3 test'" and see if it consistently generates the text correctly

[tropic mountain]

Do we know if ChatGPT plus users will have unlimited credits to use for Dall-E 3, or will it be limited?

[quartz vale]

We don't have details about that yet

[heady crown]

sup

[timber field]

If there some gui for dall e 3?

[subtle pendant]

Bing's image generator I guess

[muted moss]

That's DALL-E 2

[subtle pendant]

They are rolling out DALL-3 on Bing Image Creator

[quartz vale]

It's coming to the Labs website later this fall

[ornate vine]

May I ask questions about Dall-e here?

Hi, I am a newbie and I have some questions about Dall-e

I paied $15 on Dall-e to use its AI art generation function. The images generated were very basic and poor quality.

I then discovered that MS Bing is powered by Dall-e

The quality there is much better.

Do other people have similar experience?

[subtle pendant]

Bing Image Creator is powered by DALL-3 now, you are using DALL-2

[ornate vine]

I see! Didn't know MS Bing is using Dall-3 now! I thought that will be in October! Thanks a lot!

[quartz vale]

I don't think Dall-E 3 has rolled out for Bing quite yet. It's running a model newer than 2.0 but not 3.0 right now

[subtle pendant]

It is

Some people have access to it

[quartz vale]

some doesn't mean all

[subtle pendant]

Still counts I'd say

[quartz vale]

I think it's a bit misleading. It's like saying "OpenAI has access to use Dall-E 3 from their announcement, so it's released". We both know it isn't released to the public yet. It skirts the line of misinformation

[subtle pendant]

But it is through Bing Image Creator

In the end we shall wait

[quartz vale]

reading the press release shows Microsoft is rolling out updates on the 26th (it's the 25th where Microsoft is based). In the future, it's best to inform people about what the general public has access to instead of what a select group of users can access

[toxic merlin]

how to fix error

[quartz vale]

you can try reloading the page with ctrl + r. Let's move to #chatgpt-discussions

[dusty radish]

how good is dall e 3 compared to midjourney

[quartz vale]

I think that's only something you can decide. I've loved seeing the Dall-E 3 outputs, though!

[bright mauve]

guyza where can i see dall-e 3 arts bu community on this server?

[quartz vale]

some people have posted images from Dall-E 3 in #daily-theme . All of them say something like "made with Dall-E 3"

[fair crescent]

If you want my personal opinion: it will be easy to make art that is just as good as MJ or other Stable Diffusion techniques. But DALL·E 3 will be able to blow MJ out of the water with some good practice. But it has another major advantage. MJ feels like it grabs a few keywords, ignores the rest and makes a highly stylised picture. Often pretty similar in art design.

Now let's look at one of the announcement pictures. It has the prompt "An illustration of a human heart made of translucent glass, standing on a pedestal amidst the stormy sea. Rays of sunlight pierce the clouds, illuminating the heart, revealing a tiny universe within. The quote ‘Find the universe within you’ is etched in bold letters across the horizon."
We can see everything. Stormy seas, pedestals, universe in the heart, light from the sky lighting up the heart and casting shadows. Accurate text... try it in MJ and you'll get something nice I bet, but nothing close...

[lone mantle]

I really want to see and play around how DALL-E handles hard stuff like machine/robotic tentacles

[fair crescent]

An advanced Alpha version of DALL·E 2 had some nice things. DALL·E 3 looks even better in amount if detail, objects, text, etc. The following pictures aren't DALL·E 3 or the final product. I didn’t have robot tentacles though. Hope this gives a bit of perspective.

And we all want to play around with DALL·E 3!

[lone mantle]

I subbed for ChatGPT Plus for testing purposes so letss see when it comes out 😄

I said i wont pay extra for generative AI stuff anymore plus i have Firefly but i need to test it

Also ChatGPT Plus with new features is amazing

For example i really didnt want to pay anymore for MJ

[fair crescent]

I would have gotten ChatGPT Plus closer to the release date of DALL·E 3. "Early October" is the the announced time it will come at the moment.

[lone mantle]

True, originally i wanted to wait until 12th of october

When the MAX event ends

[fair crescent]

I think I missed it. What is the MAX event?

[lone mantle]

Adobe main event

Last year it was the MAX event where Adobe revealed Firefly

And showcased

[fair crescent]

Aah ok. I'm not in that ecosystem, so I do not know about it 🙂

[lone mantle]

They have their own target audience anyway

Which is why you barely hear about Firefly in AI art communities nowadays or atm at least

[hexed umbra]

Only very few users have access to DALLE3 via bing. Currently for most people they have access to DALLE2.5. There has been no mention when Bing will precisely launch to all users, but it hasn’t yet

[midnight mango]

as with all unoffical announcements, take this with a grain of salt, but seems like it's still at least a couple weeks away

[hexed umbra]

So it will probably arrive around the same time ChatGPT gets dalle3 then

[midnight mango]

or a little later

october is less than a week away

[hexed umbra]

Yeah, but it’s good there is going to be a free way to access DALLE3

[hexed umbra]

Yeah that’s true

One thing I want to know is can you enable DALLE3 and GPT with vision at the same time? Though I guess we won’t have to wait long to find out

[midnight mango]

some people have early access to vision through the beta, and they don't have Dall-E 3 yet :(

but maybe the public release for both will be at the same time

it's hard to know

[lone mantle]

Free way?

Unless you speak for those who subbed to ChatGPT Plus

[hexed umbra]

DALLE3 will be free to use via Bing

Idk if there will be a generation cap or anything but I think it will be free

[lone mantle]

I highly doubt that

Considering the costs and the policy

[hexed umbra]

There is the credit system for Bing Image creator but I think you will get a few free generations

[wispy storm]

I don't think bing will have dalle 3 soon

[hexed umbra]

Maybe, maybe not I have no idea lol. All I know is ChatGPT is getting DALLE3 sometime in October and then it is releasing to labs and api some time in fall lol

[fair crescent]

In a recent speach on youtube that I cannot link (Full Event | #MicrosoftEvent September 21, 2023) they mention it will be free. However, there is a caveat. It is still ambiguous if it's free like it is now, or no added costs for people buying into their ecosystem (Cockpit). Their language is not clear on the matter. Only following their future announcements closely will give clarity on the matter.

[dusty radish]

ah okay, i need photograph-like images. Nothing fancy. For example, i want a photograph of a paddleboard, and stuff like that.

[fair crescent]

It should be able to. The current one as well, if it's in the training data. A paddleboard isn't unfortunately.

Out of curiosity, why not grab an actual photograph online/home made? There should be plenty around that can be used legitimately. The power of AI is amazing, but sometimes just grabbing your phone and snapping a picture is easier and sufficient.

[dim cradle]

Hello guys. I was 10 days in a place in the mountains with no internet connection. I opened Discord now and there are dozens and dozens of messages. Is there anyone who could summarise for me what the latest developments are? I know that the version 3 of Dall-e has been launched. can it already be used or not yet? Are there any other interesting news? Thanks in advance, a big hug to everyone ❤️

[obsidian rose]

gpt vision is the other major news

[fair crescent]

DALL·E 3 has not been launched. Announcement says early October will be the ChatGPT release, later this fall will be the labs (website) release. https://openai.com/dall-e-3
Information about it coming to Bing image creator is more uncertain. It will be supported in the MS Cockpit release of today, but unknown is when the upgrade to DALL·E 3 itself will be. Also unknown is if this comes to the Bing image creator outside the Cockpit environment.

ChatGPT will get an upgrade soon. #announcements message
It will be able to "see, hear and talk", which means image (maybe also video?) understanding, can listen to voices and can talk back.

It were 10 uneventful days all things considered 😉

[subtle pendant]

I think DALL-3 is slowly rolling out on Bing Image Creator

[fair crescent]

I think what some people have is a careful test, possibly an A/B testing. However, the information is incredibly uncertain. Look at this one: #images-discussions message
It is safer not to speculate and accidentally giving people the wrong impression.

[quartz vale]

@fair crescent was ready for that #daily-theme 👀

[fair crescent]

Enthusiastic! And my only chance today probably

[obsidian rose]

I'm so hyped for dall e 3, wish I was born a few days later so I wouldn't have to wait

[torpid sigil]

I know its driving me crazy. Every hour i checked if i have dall e 3 or not

[fair crescent]

I do not have DALL·E 3, on Bing or otherwise. Precious few have it and I'm uncertain if they are allowed to 'share' their access.

[torpid sigil]

Insane art though in #daily-theme but thanks:(

[fair crescent]

Would be lovely to get it, but I expect we just have to wait until it arrives early October on ChatGPT (according to the announcement)

[torpid sigil]

I dont have chatgpt plus so im just waiting on the bing version

[fair crescent]

Thanks! I can tell you that DALL·E 3 will be leagues better though.

It is still possible to make amazing art with labs and Bing. Used Bing for these pictures.

[torpid sigil]

And i heard bing are rolling out dall e 3 early (some reddit confirms they have dalle 3 but after a day or two it got removed)

@fair crescent a reflection in different angle is probably the hardest thing an ai can do and im very impressed with dall e 2 bing to do it very well. SdXl and Midjourney can't even do it

Imagine how good dall e 3 with this prompt

[quartz vale]

Asking people to run your prompts for you breaks the server's rules. Please refrain from doing so 🙂

[torpid sigil]

Uhhhh I just can't wait for dall e 3 I can't hold my excitement

[fair crescent]

Speculation is why people think this is real. I have confirmed onevperson with sporadic access myself, but this is not a roll out. It is more likely a careful test with precious few people. Best to just wait until official information is available and wait for the actual release date.

[quartz vale]

Just please follow the server's rules

[torpid sigil]

Hmm idk about that dalle2 reddit keeps getting more and more dalle3 images by the day but u may be right since after a day or two its already gone

Thats why every hour i checked if i have dalle3 on bing to see if i got in even for 1 day

@quartz vale can i sent u a dm instead?

[quartz vale]

Why do you need to dm me?

[subtle pendant]

DALL-3

[lone mantle]

Certain people will not jump to DALL-E 3 no matter what

But having one more impactful competitor is not a bad idea ^^

[quartz vale]

I'm really hoping Dall-E 2.0 stays around after Dall-E 3 comes out. It'll always have a special place in my heart

[mortal summit]

This so much.

[karmic shuttle]

How do you make landscape images now? I'm using Bing.

[vagrant zodiac]

No idea if it's possible in bing.

[tender hamlet]

hi

Chatgpt do not reply to my question does anyone have that problem

[sick spear]

Hi, this channel is for DALL·E, please use #chatgpt-discussions for ChatGPT 🙂

[dim cradle]

why chat gpt so slow

[fair crescent]

Hi, this channel is for DALL·E, please use #chatgpt-discussions for ChatGPT 🙂

[rocky drum]

This IS a DALL-E discussion. But I will look there.

[teal timber]

Daily theme has a cool down of 30 minutes between posts. Other channels I think don’t have this at all or it’s only a couple minutes.

The daily theme cool-down is to keep users from posting too often and allow the best images to be presented rather than tons of them

[rocky drum]

How long does slow mode generally last?