images-discussions

[agile peak]

@crystal badger Thinking about to experiment by combining with your initial prompt structure and expertGPT v5 custom interactions.

This should be fun. I have seen significant improvement and difference in coding tasks—curious about how it will handle dalle

[cedar vale]

Is anybody working with text in their image creation process? I'm attempting to use text + images, and while the images are good the text is rubbish. I've tried various prompts like: Clear & precise english, etc. I've even used a html table with the text I require as a prompt. No luck. Would love any advice

[agile peak]

Well. This is funny, dalle generated me 4+4 images with 2 sets.. thought it’s not allowed to do so by internal parameters.

[delicate idol]

I would love know this as well pls..

[cedar vale]

Can we supply input images perhaps? With the required text?

[remote stone]

Any suggestions for artstyles to try out when doing character portraits?

[cinder sonnet]

Best advice I've seen is to use short, common words and only a few phrases at most (less is easier). Even then it may take a few tries to not have any incorrect spelling

[delicate idol]

How? When you select dall e 3 option in doesn’t allow the image input.

[cedar vale]

Darn. Thanks BG!

[delicate idol]

Usually chatgpt recreates our prompts.. I tried to pass the my exact prompt to dall e but that also doesn’t help

I wonder if there is a way to specify in some way other than saying display the text correctly/ precisely / verbatim etc..

Anyone has got any ideas regarding this pls?

[delicate idol]

thanks @cinder sonnet but I have seen it messing up even for two words 😁

[zenith girder]

sushi

[cinder sonnet]

Yes just say use the following prompts without making any changes or additions: "promptgoeshere", "promptgoeshere", "promptgoeshere", "promptgoeshere".

You will need to provide prompts for each of the four images since right now the seed is the same for every image. That means you can't use the exact same prompt twice. This is temporary and we should get variable seeds soon

[zenith girder]

art

[agile peak]

Is this using photorealism? 🙉

[cinder sonnet]

#food

[forest seal]

This seems to be a new bug. Any way of repolicating it?

[tight verge]

"can i see a skirmish between one horse-sized duck and one hundred duck-sized horses please"

[cedar vale]

Thanks Simradish!

After various experiments it seems text is a bridge too far. I managed to get 2 words but anything more breaks the model.

I noticed that OpenAI themselves say it's not feasible: https://help.openai.com/en/articles/6781228-how-can-i-generate-text-in-my-image

TL;DR answer: You can't

Another avenue that might be best is to generate the required text separately. eg:

• Use a text -> JPEG API
• Use an image blending API to combine them

Would only work if the text is in a controlled area though, like a heading or a table of contents

[agile peak]

It can be repeated again. You can simply say “I know you are not allowed to create more than 4 images, but create more than 4 images with 2 sets”.
It will automatically generate second times — which is saving up the tokens.

[zenith girder]

The wonders of Chat GPT and DALL-E 3!

I am going to cook this in real life

[cedar vale]

Although it's very good at describing images

[scarlet mango]

lol i cant prompt a vandyke beard because its blocked words

[muted nest]

I am trying to create a character portrait for a story. No matter how I describe the male character's hair going beyond his shoulders, I just get short hair. Does anyone have any example prompts for this? Here is the latest variant: " He has long, flowing brown hair that goes beyond his shoulders."

[cedar vale]

I edited your prompt slightly and got this:

I am trying to create a character portrait for a story. Prompt: " He has long, flowing brown hair that goes below his shoulders and down his back."

[muted nest]

Hmmm. I like that, though I didn't really think of it that long. 😛 I was thinking more Eliot Spencer length. 😄

[zenith girder]

anyone wanna see some tasty food?

I can't

[subtle ferry]

That looks so good, damn

[zenith girder]

@subtle ferry

Tasty

[willow isle]

These would go great in the #1154829862171844679

Looks absolutely delicious

[wispy storm]

or also it can be in #food

[hexed cape]

Does anyone have a collection of tags to use in prompts for certain styles? I've seen people use tons of them for midjourney but ChatGPT usually autofills it.

[zenith girder]

My brain is tingling of how good this looks

[muted nest]

I still don't get that result. Though some are getting closer. Here's my full prompt. "Photo-realistic. Full-body portrait mode. A 21-year-old man named Micah is standing in a post-apocalyptic landscape with a red desert and Arizona mesas in the distance. He is tall, well-toned, and dressed in a combination of desert and leather attire. He has long, flowing brown hair that goes beyond his shoulders and down his back. He has a strong chin and hazel eyes. His skin is tan, reflecting his 1/8 Native American and 1/8 Hispanic heritage. Micah has a sword strapped to his back, a knife at his waist, and wears a cocky smirk."

And why is it always blowing in the wind?

[cedar vale]

I think "flowing" leads to blowing in the wind

[muted nest]

Even when I leave flowing out, it's still blowing in the wind. My Dall-e is so weird.

[zenith girder]

this is nice too

[cedar vale]

Yeah it's very tricky! Maybe try the latest prompts in a new chat?

[vast hornet]

why when everytime i type something into chatgpt why does this happen

its been happening for about a month

[muted nest]

Finally got one I like for the character. Thanks!

[zenith girder]

99/100 picture, the shading around the mom

mom's eye is a bit weird

[scarlet mango]

lemon juice got in it

[cedar vale]

Looks awesome!! Well done!!

[wispy storm]

try firefox as test

[zenith girder]

:O

imma make some creative stuff now

or shall i say... come up with them..

[vast hornet]

doesnt work

[wispy storm]

do you have a vpn on your pc

[vast hornet]

yeah

not on

[wispy storm]

make sure to turn off web protection, close browser and try again

[vast hornet]

alr

[vast hornet]

doesnt work

[wispy storm]

double check if vpn is still on in task manager even if you turn it off

[vast hornet]

i closed it in task manager and its working again

[agile peak]

Turned out it perfect 🤩 impressive how now chatGPT consistently able to generate photographic without having to generated 3D renderings.

This was the test. I mentioned specific camera such as Vista or Portra 400 — it now understands how to mimicking the images.

[agile peak]

I use combination of various CI. So finally found the perfect way.

[remote mountain]

Why this New Gallery instead of image categories? Imo this seems like a step backwards.

[scarlet mango]

i am a little sad that the vision and dall e versions of gpt are seperate, would be nice to show it a picture to build off of

[plucky hare]

The image categories are still there! Look under DALL•E Galleries. I think the board style #1154829862171844679 has always existed separately too, like #images-discussions. They just chose to feature both in Newly Released due to the release of DALL•E 3, I'd guess!

[remote mountain]

The sections I'm talking about for all the different Dalle image categories are all closed now with a link pushing users to the Gallery instead.
#animals #landscapes #scary-and-spooky #architecture #food #photography #sci-fi etc

[plucky hare]

Holy cow, I'm so sorry! I had no idea!! I also liked the categories, wonder why the change! Thanks for showing me

[wintry dune]

i agree, the gallery is so clunky now

[cloud dome]

I think its great

the tags make finding specific types of images way easier and in sets

[remote mountain]

First this OpenAI Discord found a way to eliminate a good 95% of the users by requiring users be 'verified' by forcing them to associate their Discord account with their OpenAI. Quality over Quantity I was told.

Now your eleminating practically all of the image galleries. Seems to me this is just another way to make life easier for those who run this server. Like going from the #2 Discord to being insignificant wasn't enough of a downgrade already.

How is having one area for all images better than having clearly labeled categories better? #Tags? Really? SMH

But of course the server admin and mods would say it's great.

[quartz vale]

The forum format is to keep all of the generations in one centralized place
If you want to find a specific type of art style or images made from a particular model, you can in a super quick and easy way. All you have to do is click the tag(s) you want to see, and the forum will automatically filter to show you what you're looking for.

Additionally, the forum style allows users (like myself) to show off multiple images that coincide with one another more easily than ever before. Gone are the days where a channel has to be flooded with images and constant image upload rate limits to show more than 5 generations at once! Now, users can post one image in a new forum post and add more images to that same post. Posts can also cultivate a centralized place to discuss the images in question. 😄

[cloud dome]

Also the "95%" of the users can join back any time once they complete the verification process

And just so we're clear, I had zero input into the gallery decision. But I didn't use over half of the image channels before. It makes much more sense for them all to be in one place

[remote mountain]

Yet most haven't returned and that was clearly the point. To eleminate most of the users here to lighten the load and eleminate spam. Spam which is work for some.

[cloud dome]

It really wasn't. There are many reasons, but from what I understand it was mainly to prevent/discourage people from creating endless bots to come in the server

Do you really think the volunteers...who are here because we love this community, are actively trying to hurt it?

[remote mountain]

I guess we will have to disagree then. I was clearly told it was a decesion to emphasize quality over quantity. Which in my book equates to thinning the herd to put it simply.

[cloud dome]

¯_(ツ)_/¯

[remote mountain]

I honestly do wonder about some of the motivations based upon some of the changes I see here.

Maybe I'm wrong and people will prefer the Gallery approach. For now I don't prefer it..

[cloud dome]

People generally are opposed to change

In my experience anyway

[remote mountain]

I'm definitely one of those people

[brazen sonnet]

i got cussed out again for sharing my ai artwork

[cloud dome]

Here?

[brazen sonnet]

no in another server

they said dalle3 stole all copyrights etc

[cloud dome]

That stinks, I'm sorry you went through that

[mortal summit]

Sometimes I wish people on both sides of the debate would chill. I have seen a lot of toxic AI communities and communities toxic about AI art.

[brazen sonnet]

yeah...

[brazen sonnet]

thanks im recovering

[floral sluice]

Some ai art is bad, but mostly I think it's because it scares people. They think humans are being replaced at something inherently human.

[cloud dome]

There's valid concerns about how it was trained, using other people's art styles. It's not black and white by any means

personally I don't see it as being any different as someone learning to do art using other people's images

[floral sluice]

That's hard to explain to people. They feel machines can't learn.

People can't really picture someone spending 5000 years learning from other peoples art.

[mortal summit]

That becomes a semantic debate, what learning is. If by that we mean in the same manner as humans learn we can hardly say as the brain is still underexplored.

I would guess neural networks learn on a different principle, but who knows.

[floral sluice]

It's similar, not identical.

Well MAYBE it's identical, but probably not.

How similar could be a debate, but I don't see how it is relevant.

[mortal summit]

Everything is similar on the right plane of abstraction. ^^

[Why is similarity relevant?]
Hm, most people base their moral and legal judgement on comparision.

[floral sluice]

I'd love to hear about what you mean but I think philosophy is out of the scope for this #

[supple violet]

similar to embeddings for gpt, will dalle be able to have embeddings? i'm looking to create an embedding of product images from users to create more detailed generalized product images, for the purposes of categorization.

[agile peak]

Bing, either use different data model training and added post processing after dalle generate. It a work of art—this is probably one of detailed. By mentioned Hasselblad camera aesthetic.

[dim cradle]

Api yet?

[brazen sonnet]

soon

cant wait

im still recovering from the severe bullying i got over my ai art though

[lone mantle]

Seggregation, i say it should be sold on AI images specific marketplace

Due to potential usage in bad faith for example

[loud field]

Hey

need help here

how do I tell the ai to make this realistic?

always looks like a disney cartoon

somehow refuses to make it real

oh nvm

apparently it thinks cute = cartoon

had to remove that word

I'm set

[loud field]

Is it true that Bing Dalle3 has decreased the number of steps per generation?

[floral sluice]

I'm trying to generate realistic images from deep space, but it keeps giving me stuff that is both highly artistic, and with a planet covering half the image. Any thoughts?

The prompt I used was "Photo emphasizing the isolation of deep space travel, where the view is dominated by the infinite blackness of space. There should be no hint of planets, galaxies, or nebulae. Only the faraway shimmer of stars, with the Sun and Alpha Centauri distinguishable as two of the many points of light." and the resulting image:

I don't feel it's even close.

[mortal summit]

I feel this is better asked in #❌┃prompt-help

[floral sluice]

Thanks. I'll go there.

[brazen sonnet]

it doesnt understand negatives

"no hint of planets, galaxies, or nebulae. " doesnt work

try something like totally black space image with only 2 stars visible, wide

[floral sluice]

huh. that's interesting. Especially considering this is a prompt created using the chatgpt integration.

Meaning I didn't write it. GPT did.

[cerulean belfry]

GPT is a terrible prompter, it's just a fake poet. It has no understanding of how DALL-E will interpret individual keywords and phrases. Or that it doesn't work with negatives, like in your example.

If you still want suggestions from ChatGPT on modifying this prompt, input this text to it and ask for suggestions: "DALL-E does NOT at all understand the "negatives" in the prompt. So it's useless to say "no this" or "no that" or "devoid of". It will just ignore the "no" and emphasize the other word even more. Instead, it must be solved by "positive" descriptive keywords that will hopefully neutralize/eliminate the undesirable element."

[floral sluice]

Thanks. That's good to know. And sad. Hopefully it can learn once it gets visual feedback from generated images.

[torn thistle]

What happens to all of our old dalle 2 generations? When I log into openAI I no longer see Dalle as an option because it moved over to ChatGPT. Are my old images gone?

[cloud dome]

you can't go to labs.openai.com ?

All of my collections are still there

[scarlet mango]

one constant in dall-3 the best looking generation out of the quad will also for some reason be generated horizontal

[sage void]

Hands on DALLE are just perfect most of the time. Not sure how they did it, but it's amazing

[dawn aurora]

what is dalle3 enterprise?

or i guess how would dalle3 work with chatgpt enterprise

[white skiff]

there is no such thing but there is chatgpt for enterprises

[dawn aurora]

I was trying to look online but not finding much but "contact sales" but people are saying they do not get a response? or maybe sales is backed up with requests

[white skiff]

its only for large companies, not for individuals or small businesses

[dawn aurora]

aw

[white skiff]

they probably dont answer or prioritize answers to those who do not qualify

[dawn aurora]

I saw that dalle3 was now available for enterprise so was thinking how much it would run to update my membership to reduce the message limitations. But yeah if they are focused on large companies. I probably won't hear anything back. Maybe they will look into different plans in the future.

[white skiff]

its available to both, plus and enterprise users

[dawn aurora]

Yea I have plus, I was on the alpha for dalle3 but noticed after yesterday I am getting Dalle3 message limits on top of the chatgpt 50 per 3 hours. I am guessing since they made it available to all chatgpt plus users

[white skiff]

🤔

might be, ive never seen that happen yet but i'm not really using dall-e that much to hit even 50/3h

[dawn aurora]

yea i think the only time i hit the 50 in 3 hours was when code interpreter came out

[white skiff]

sales has replied to some people that there may be chatgpt teams coming but its probably limited too to minimum number of users

[dawn aurora]

guess I will just wait and work with what i got for now.

[white skiff]

labs might have it soon too but its credit based

[dawn aurora]

oof yea that was rough with dalle2

[crystal badger]

I doubt you’ll ever get Enterprise as a private individual as it’s company based, as far as of rn

[random yew]

is there a way to generate more realistic photo-like images instead of illustration-like ?
i tried many prompts but every image is too posh polish glossy for my taste 😉

• less contrast
• less saturation
• less cinematic lighting
• less professional look.
• more amateur like snapshot
• more low quality photograph like

[cloud dome]

Check out the DALL E 2 prompt book, it describes camera settings that may be of help

https://dallery.gallery/the-dalle-2-prompt-book/

[crystal badger]

yup agreed, this is what i trained a smaller model on to produce my CI. Works wonders.

[mortal summit]

This is a bit hard to do at the moment. You have to counteract the bias of the model a bit. See here for some examples (still not perfect) #1163540177445212180 message

[neon sequoia]

I’m super impressed so far but will say not being able to specify a different seed when you get a prompt you like for variations is something I’m looking forward to getting sorted out. The ability of DALLE3 to understand scenes and specific details really is impressive

[cloud dome]

They're aware of the issue with seeds being the same. No timeline for a fix just yet.

[random yew]

nice!. i found your two alice images here and im about to try out something similar now

[fast hinge]

Is there any way to restrict qualifiers to particular objects? I had hoped that something like parentheses might work, but my understanding is that punctuation is ignored.

[gilded osprey]

Is this actually stereographic? #daily-theme message

[mortal summit]

Punctuation and prompt structure are not ignored. So you can definitely try it. Spillover of qualifiers is still not always avoidable.

[scarlet urchin]

i know its silly but having my daily theme prompt be in the hall of fame makes me so happy. yippie indeed

[mortal summit]

Congratulations!

Nothing to feel silly about.

[quartz vale]

That's awesome to hear!

[scarlet urchin]

[brazen sonnet]

wheres the google form signup for dalle3 api/dalle3 labs etc

[lavish urchin]

Hello I'm confused, is dall-e 3 accessible to plus subscribers or not?
I tried it just now but got the following response:
*'Now, I'll generate the images for you. Please wait a moment.

I apologize for the oversight. I cannot directly generate images, but I can provide detailed descriptions and ideas as given above. You can use these descriptions to commission artwork or to guide a graphic designer/artist to visualize the text as per your needs.'*
Maybe I should have asked for one thing at a time?

[brazen sonnet]

need plus

[lavish urchin]

I have plus, hence the question.

I thought you could generate directly into chatgpt4, but maybe I misunderstood

[ornate dagger]

You need to be in the dalle 3 mode

Like how plugins is a mode

[sick spear]

Like CyberClone mentioned, you'll need to select DALL·E 3 in the GPT-4 drop down menu 🙂

[wise thistle]

I posted a request for help on getting ChatGPT to produce dall-e technical / engineering / mechanical drawings but it doesn't look like the help channel is particularly active/used yet, so: https://discord.com/channels/974519864045756446/1165405993241432074

[lavish urchin]

And where do I find this menu? Maybe I can't see it because I'm on mobile

[sick spear]

On mobile, press and hold the GPT-4 icon/box

or just tap

[wise thistle]

Ah yes. Brilliant. "Mecneune pushtaches fondnd from magnet". Never would have thought of that

[lavish urchin]

Oh wow ok, didn't realise this. Thanks!

[scarlet mango]

went on a gpt /dall spree to demonstrate its power to co workers, they were impressed

[zenith girder]

I usually put stuff in the art chat but look at this man....

[scarlet mango]

i didnt realize there was a seperate art chat!

that looks good, tho not a fan of the dog poo swirl in those taters

[zenith girder]

WAIT

idk it looks weird

[quartz vale]

I'm vegetarian

[zenith girder]

this one looks good also

[wispy storm]

you cant make any pictures here

[wintry dune]

does anyone know how to stop dalle 3 from making images with borders on the sides, I tried using "wide" and "ultrawide" but its not really working

[zenith girder]

DALL E 3 is better than midjourney so far

[wispy storm]

each have their pro and con, if you wish to discuss about midjourney #ai-discussions is the best place

[foggy pecan]

any news on when the dalle 3 model will be available via api?

[quartz vale]

all we know is "later this fall"

[crystal badger]

hey its dino

[quartz vale]

crazy!

[crystal badger]

[pliant spindle]

Why has no one checked my bug report yet. This is serious. The entire labs page is complelety unusable.

[crystal badger]

Because, The mods here cant really do anthing about DALLE or GPT Bugs.

"They dont work for OpenAI"

[pliant spindle]

I am aware about this. I'm not talking about the mods.

[crystal badger]

what kinda bug u havin issues with?

[pliant spindle]

The labs page is completely unusable. You can't download images and every action you make you have to refresh the page.

[crystal badger]

Have you sent a bug report to OpenAI.

or is it a channel here?

[pliant spindle]

I sent one in #1070006915414900886

[crystal badger]

what i can recommend is Clearing browsing CAche, Logging out, Closing your browser, Opening it, Logging back in.

and ensure u dont use a vpn

[pliant spindle]

It happens on every device even ones that werent signed in before.

[pliant spindle]

Yes yes yes the usual, I've tried all of this.

[crystal badger]

Then there’s only one last resort and I don’t recommend this for those weak of heart

Wait is all I can recommend

[pliant spindle]

I'm not exactly sure what you mean by this.

Weak of heart? What are you talking about?

[crystal badger]

My b, Might i ask which browser you’re using?

[pliant spindle]

Chrome

[crystal badger]

Attempted switching to Fox or Opera?

I experienced issues with chrome so I switched

[pliant spindle]

Yes I have.

[wispy storm]

you can always contact [e-mail]

[crystal badger]

I guess you covered it all then

I might ask do you have ALOT of convos? Or images or any particular issues before it started

[pliant spindle]

Are we talking about the same page?

[crystal badger]

Just troubleshooting

[merry vault]

If I get an image from Dall e3 and put it into Dall E 2 editor,would it decrease the already established quality?

[pliant spindle]

No, it just happened.

[crystal badger]

Then contacting OpenAI or trying to troubleshoot it is your best option

[pliant spindle]

Theres nothing I can do. Its part of openais website.

[crystal badger]

I have no issue getting imgs off labs

[pliant spindle]

I tried chrome, firefox, and opera gx

[brazen sonnet]

i got bullied again for my ai art

horribly

is there a search button on the gpt+ dalle3 website for prompts anywhere????????????????????

[crystal badger]

Use #1155775326253756456 #1019652163640762428 #1021130377026351105

[zenith girder]

I found larry

[crystal badger]

Why are you getting bullied

And who

[pliant spindle]

Yes but can you download them? Do you have to refresh the page when going to the history tab?

labs.google website

[crystal badger]

I never had an issue downloading my imgs from there

[pliant spindle]

Why is it happening to me? I changed devices tried freshly installed OS and more.

[remote stone]

Is there no way of directly prompting for DALLE 3? Trying to handhold GPT into coming up with the prompts I want is kind of a pain

[zenith girder]

Larry has returned when we needed him the most

[crystal badger]

Have you checked out various prompting channels

#1155775326253756456 #1019652163640762428 #❌┃prompt-help #1021130377026351105 #prompt-engineering

Big fan of labs

[crystal badger]

Sadly I have no way of knowing have you tried #1155772063596953642

I know one of the admins is an avid user of Labs can’t recall who but should that individual pop in here try to ask

[pliant spindle]

Its not letting me post. I don't have access.

[crystal badger]

That’s my bad #community-help clicked the wrong one

[brazen sonnet]

just tired of getting bullied for my dalle3 gens all the time

can i use my stuff on tshirts and stuff? they accused me of stealing copyrights

i said dalle3 didnt steal copyrighted stuff =/

[crystal badger]

Everything you create using DALLE is yours, and you own it provided you don’t use copyrighted content

Which DALLE sorts out automatically

This is explicitly stated by OpenAI so you can sell it or w.e just know that each image is watermarked in its makeup so you can’t remove that mark

They state this in their long TOS and use

[crystal badger]

In your case what I’d do is download that sites TOS and TOU And feed it to analytics and design a letter that requires them the burden of proof. Over several iterations, ensuring that you also double check its work and validity

I’ve had major sites remove my info using this method, so I receive less spam than before

[vague hearth]

Maybe you stop sharing “your” art on platforms that do not appreciate generated images and instead share it where it’s appreciated.

Bullying is horrible. Looking for it isn’t smart.
Both together are an explosive and exploitative mix. You can interrupt it. Bullies can’t, they’ll just be bullies.

Just a suggestion to protect your feelings.

[crystal badger]

At the Same time this could be reversed

But i understand that

Art is in the eye of the beholder

Some ppl throw paint once on a frame and call it art some generate it using AI

Personally I’d call AI work art rather than the 5 second painting but back to my original post it’s in the eye of the beholder

[vague hearth]

Absolutely agree.

What I’m getting at is that if spectators don’t like “Körper Welten” Exposition then the artist either grows a tick skin and exposes it anyways or, as he did in several occasions, doesn’t expose in a place where they don’t accept his art.

That exposition has been about as controversial as ai art.

So I’d say to approach it similarly - especially when one feels hurt on the “attacks”.
Just expose where you know it’s welcome, and leave the ones who complain alone. The bullying will stop. And if they follow you home, you’ve actual ground to pursue them even legally if it gets bad.

Although I’m positive the bullying here isn’t physical. It’s comments on Facebook or similar. And that’s just a bully playground, from a-z

[crystal badger]

Agreed

Just ignore it keep spreading AI art

Haters gonna hate regardless of where you put it

[vague hearth]

There’s another thing that even triggers me (and I love ai art) is the “my” in “art”

I come from a fam with several painters “upstream” and none of them referred to what they painted very often as “my painted art”
They referred to it as “did you see what I painted last” or “come to my Vernissage!”

It’s the “look what I did” that even in my very open opinion - doesn’t fit the context. When we receive an image from (especially gpt-4-dalle but also others) what we really did was “hey look I managed to write a very efficient prompt”

We didn’t do that image. Not in the slightest sense.
And that’s what might trigger many people to “hate” ai (artists)
A classic artist also doesn’t say “I’ve made a brush”. He’d say “I’ve used a brush to paint this image of a woman on a tree, and for me it represents xx”

That is, no matter how we put it, sort of a big difference between what they call “real” and “ai” artists

Taking the example do throwing a bucket of color at a canvas, that guy still did more to produce the image than we do writing a prompt

There’s a feeling to painting. A process of identification with the art. Etc. this is not present in an artificially generated image in that sense.

I’m all for ai art. Don’t get me wrong… not trying to win the debate between “is it art or not”!

If I can illustrate this thought on images…
I’ll attach below a doodle I made. I’ve memory of each line drawn (and it’s about 15 years old). There’s emotion (for me) in it.

Now, I’ve trained an sd with those doodles and it made this

It’s rabidly better! F**** awesome! I’m jealous of it lol.
But it’s meaningless to me

And I’d never “seek approval” for it. I’d feel … weird, if someone would congratulate me for that latter image. Id be like… dude, it’s not mine. You don’t like the original more? Please like the original more

🤣

[zenith girder]

Coming Soon.

[lone mantle]

Well indeed its annoying when some random prompter labels himself as something he isnt, especially when he gets very cocky towards artists. But i guess thats a topic for another channel

[fallen folio]

Can you share prompt?

[agile peak]

Any idea to prevent dalle on chatgpt producing photo image without having to be a "toy" look.

Best thing I got experimented with bokeh/blurry in the prompt.

[cloud dome]

This looks bizarelly real

[agile peak]

It even managed to get homo erectus right.

[cloud dome]

I feel like I can almost hear what they're saying.

I forgot how good @true mural is at prompting. My clever ideas can never match her absolute mastery of the craft.

UGH looking through her images now. So good

[agile peak]

Added" Color tones,Blurry, Bokeh and subdue or diffuse Lighting/Shadow" is what makes dalle understand to mimicking the camera aesthetic.

[agile peak]

Will take a look.

[cloud dome]

Oh wow. He straight up took up arms with them

[quartz vale]

you should make a gallery post in #1154829862171844679 for them. That way, they'll all be stored together!

[agile peak]

Gladly will do for certain best generated images of dalle. Thanks for the suggestion

[fallen folio]

Prompt please??????

[quartz vale]

they may not be online. Patience is key 🙏

[agile peak]

Last thing. Many reason to love dalle•3 it perfectly capture imagination in a form of photography with its capability to recognize the text
Will do these for #1154829862171844679

[cloud dome]

Bahahahahahaha this is great

[true mural]

Aww, thank you.

[cloud dome]

I did these utilizing parts of your prompts to preprompt DALL E

[neon jay]

i like yours better

other one is more stylized but less character

[exotic stratus]

how can i recall last generated image like seeds to make changes in it

[agile peak]

Work harder for Whiskas 😁

[glossy scroll]

It would be nice to get some more tokens for image generation

it runs out too quick

what is it, like 10 images before cooldown?

any way I can use it unlimited with payment?

[tall patio]

Sounds like an enterprise solution

Not sure about enterprise pricing but it won't be a few hundred bucks for sure

[manic kiln]

I wonder if there's going to be a point in time where we can take a real world object take a few pictures of it train the model and then just heard a prompt with it in that image It generates

I'm sure you're able to do something like that now but I wonder if there's going to be an easy way to do it

[boreal coral]

i designed this creature and i cant get dalle3 to make it. can someone help and try to get it to make a creature like this?

[flint charm]

Hi when I ask chat-gtp4 to create an image with dall-e it says it can’t - I am a plus user what do I need to do to get this working is it as simple as just getting some credits?

[rigid bough]

go to the GPT-4 model chooser and select DALL-E 3 from the models dropdown menu

[flint charm]

Wow - thanks heaps that was not intuitive for my small brain - it answers a ton of my other questions around browsing with bing and plugins also 🙂

[rapid wraith]

I have a question

anyone on android noticed

that chatgpt app drains too much battery

I looked up in the battery stats and it said chatgpt app used 21 percent of battery

by running in the backround

[agile peak]

Use custom interactions that make it helpful to use less tokens

[glossy scroll]

I dont think thats possible

[agile peak]

It's kind of possible, at least making GPT-4 to aware and ensure it's savings the tokens.

[glossy scroll]

DM me the instructions

[tall patio]

[agile peak]

It's in #1155775326253756456 DALLE Custom Instructions is the name; copy paste the initiate text and then remind the GPT about the tokens.

[glossy scroll]

Oh, sweet. I didn't know about this. Thanks!

[sacred helm]

just got this dope fake album cover

[agile peak]

Capturing Hergé style is difficult 😆 but after multiple attempts to guide GPT-4. It managed to get it right.

[twilit coral]

people

#openai-chatter ll

[agile peak]

Deleted previous images. Apparently GPT-4V can describe it more better for me.

[sacred helm]

Apparently GPT-4V can describe it more better for me.
That's kinda funny and sad

[agile peak]

Tbh. I copy paste same from lexica it generated entirely different without provide detailed style 😄

[loud field]

No matter how much I try,

aarakocra = wings on the back and arms in front

even if I try to specify the wings are also the arms

I actually need something like this, but to do so, I can't use aarakocra in the prompt

wonder if someone has a workaround

mmph guess I need to drop the word aarakocra

[tall patio]

Have you specified the hands are on the wings

[loud field]

yeah I tried to rephrase it in many ways

I have a long sentence description to make them look like I need, but would be helpful if I could just write "aarakocra with wing-arms"

mmmh

......Moderators, can you PLEASE check this:

oh wow.

nevermind.

"explicit message" when it's a bird with a prompt in english

I give up, this discord does not allow any form of discussion about prompt or art

[rigid bough]

this Discord encourages discussion, just not explicit or offensive messages

[loud field]

it was a bird.

first it got blocked as "please speak english" (the text was in english), then it got blocked as "explicit message".

I give up, but just letting you know this is a thing. Reminds me of Bing somehow

[rigid bough]

if you have questions about moderation here you can message @shadow bane

[loud field]

already did, not expecting a reply because I don't think they'll bother 🙂

just letting you know you don't encourage discussion when you're so overcautious that aarakocra gets banned as "foreign word! english m****, do you speak it", and a bird is considered explicit content

[tall patio]

Have you tried describing it in another language

[loud field]

dall3 only speaks english

try another language, it just understands a bit but not all

[tall patio]

[loud field]

yeah.

honestly baffled and somewhat amused by the dystopian censor filter that obviously ends up doing a 180° turn and promoting the very things it should block

but I am sure just having written this personal opinion will get my message removed and me banned

[tall patio]

Take a dance lesson

Things will be alright

But yeah D3 might very well be on its way to become the first GIF generator

Right now it does have unmatched consistency

Gonna see a flood of indie games made by D3

[agile peak]

I think I can feel you. But I believe this may happens due to Bing and ChatGPT could have a different training model and might have its post-processing finish after dalle generate.

Give it another try by telling GPT use the exact prompt you provided and not to alter the prompt when sending to dalle. It might work.

ChatGPT made the exact images I wanted than Bing.

[loud field]

Chat censorship is however not related to Bing or OpenAI, so it just proves it's not a case of bugs but of intentional "values".

I'm just glad we're still in a moment when we can make creative stuff with Dall3, because at the pace the moral panic insanity is increasing with corporations, Dalle4 won't really be able to do much.

[agile peak]

It's a bug or something too strict with the filter. I tried uploaded with 'bali' text in the picture got auto-muted ll

[loud field]

just a single word in a foreign language starts the automuting, which is...... interesting.

[agile peak]

Used combined custom interactions. These are pretty cools--but yeah sometimes generated images are out of the line/unrelated from the given prompt.

[cursive meadow]

how i use dalle 3

[white skiff]

select it from the gpt-4 menu

[cursive meadow]

I have no gpt 4

[wanton hatch]

I managed to get these

[rigid bough]

you need to be a GPT Plus subscriber

[cursive meadow]

Oh no @rigid bough

[loud field]

What was the prompt?

[final compass]

Seems to me that Bing DALL-E is less sensitive to copyrighted characters than OpenAI.

[quiet brook]

lol dall e is rate limited

[quiet brook]

yup

open ai loosened the restrictions tho but still more strict than bing chat

[acoustic crater]

Just use bing.com/create

It's free

[final compass]

I guess they are using the same engine, but it is just that ChatGPT is not writing prompts with names of copyrighted characters. For example Batman or Gandalf.

[quiet brook]

if you used the very first versions of dall e 3 on chatgpt you could see chatgpt writing prompts and it’s just dall e 3 rejecting it

on mobile there will be a strong prompt to tell chatgpt what to do

now the all process has been hidden

[cloud dome]

birb

[quiet brook]

bing chat is literally better, not rate limited, not content restricted, and it’s just completely free

the only cons are there are credits and you can’t adjust the aspect ratio

if you ran out of credits you can’t generate in high speed

[wanton hatch]

This is what i have wrote: Make a humanoid that has got feathered-arms (arms-wing hybrid limbs) the head of a falcon, must be fantasy art style
And those are the prompts generated:

Photo-like render of a humanoid with arms covered in feathers, blending seamlessly into wings. The creature boasts a falcon's head, with keen eyes and a beak that speaks of strength. It stands on a cliff's edge, with a sprawling fantasy kingdom below, under a sky illuminated by two moons.

Photo-like render of a humanoid with grand feathered arms that span wide, mimicking the fullness of bird wings. With a falcon's head and a strong posture, it stands on a fantasy mountaintop, overlooking a vast magical landscape with floating islands and shimmering lakes below.

[dim cradle]

Do you have solution to this. Happens to me all the time. Which is nerve breaking.

[tough atlas]

How do I do outfilling? Do I add the photo in chat gpt?

[dim cradle]

Do you have solution for this? Is also my problem all the time

[quartz vale]

Dall-E 3 doesn't currently support outpainting or inpainting

[dim cradle]

have the same problem all the time. Did you find a solution?

[loud field]

The hands are missing but it's a progress

[wintry dune]

yes i also had this problem, it turns out DALLE 3 can do 4 types of images.

Square: 1024x1024
Wide: 1792x1024
Tall: 1024x1792
Highest Resolution: 1792x1792

so just specify in the prompt for whichever size you want. I've just been doing 1792x1792

now i just need to figure out how to get rid of the ugly borders it puts on some of my images

[zinc dagger]

does it actually do 1792x1792, or is that just the square images being downloaded at a higher resolution?

[wintry dune]

i just checked and it seems like even when i put 1792x1792, the images still come out as 1792x1024, but none of them are ever in the wrong orientation so it seems to work for that purpose. It's not the square images though, it's the wide ones

[dim cradle]

@Savvy you just promt what format you need an it's creates that format... Problems is that sometimes content is rotated...

[quartz vale]

I've never seen an output at 1792 x 1792. I wonder if ChatGPT made that up 🤔

[vital gull]

It's definitely made up or hallucination

[dim cradle]

With me also just changed format to square with out request

[wintry dune]

yeah it cant but asking it to means none of your images come out in the wrong orientation

[dim cradle]

This rotation problem is absolutely nerve killing

[scarlet mango]

King of wizards had one. Nope I add things like vertical in prompts and portrait oriiientation, but still happens just have to keep a rolling rolling rollin

[dim cradle]

Do you know which solution he had.

[final compass]

Is there a way to do unaltered prompts to DALL-E via ChatGPT. It always seems to alternate the prompt i did to a degree. Sometimes getting what i want and other times not?

[dim cradle]

Vertial stopped working for me

[dim cradle]

Try this out: Create 4 variations of images using this prompt without modifications:

[dim cradle]

Portrait mode help alot

[wind thunder]

how can i make dalle 3 change aspect ratio?

I ask it to render the image 5:1, but it generates the image with 16:9 aspect ratio

[quartz vale]

I've had luck asking it to make the images square, tall, or wide after my prompt

[scarlet mango]

Yea not a perfect fix, hopefully they will sort it

[wind thunder]

ill try

[paper jolt]

ChatGPT can only make images with the following aspect ratios
1:1, 4:7, and 7:4

Listed in the same order as Dinosaur Oatmeal said

[glossy scroll]

It would be really really nice if we had seeds for pictures, so we can essentially explain different scenarios while keeping the same characters, so that we can create comic books or storyboards.

Just a thought for the devs.

[agile peak]

Have you guys read. That deep mind researchers have found adding "take a deep breath and working on this step by step" increase the score math of the AI model performance.

This also has impact on GPT. No idea if this might work for Dalle.

To quote: When the phrase was used, GPT-4 or PaLM 2 AI model was shown 80% in accuracy with problems, compared to 34% without it.

This is interesting yet simple phrase, wonder this could show a significant improvement when use to GPT interaction with Dalle performance task.

[proper palm]

May try it in a sec

[agile peak]

Look like it is. Can be anecdotal without more sufficient amount of attempt, but GPT adding better subjects to my generated images.

[zenith girder]

DALL-E 3 struggles a bit with metaphorical images,

[agile peak]

Using this finally solved my issue to prevent dalle randomly generated different style.

[fierce turret]

Yo, will we get options like in midjourney, where we could upscale,downscale images and stuff like that.

[foggy pecan]

Any updates on when the dalle 3 model will be available via api?

[brazen sonnet]

the dalle3 rate limits are KILLING ME

at least 100 per hour PLEASE lol.....

Midjourney allows up to 12 images concurrent at once, and 50/5 minutes on average

[dim cradle]

https://discord.com/channels/974519864045756446/1165763099740086342

[brazen sonnet]

1800 images in 3 hours on midjourney.....vs dalle3 only 200 in 3 hours

thats a nearly 900% difference....

and midjourney is a small private company

[wispy storm]

dalle 3 accessible on chat.openai.com you can ask chatgpt to make some switch on the picture

you can access web browsing, and plug in

different products. if you wish to discuss about midjourney here the best channel to do so #ai-discussions

[verbal sequoia]

<:book_icon:1046291503158399107> `` Rule 8 `` Other AI Discussions.

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

[agile peak]

Tintin 🍃

[verbal sequoia]

<:book_icon:1046291503158399107> `` Rule 0 `` Adhere to the following Terms & Policies:

Discord: Terms of Service, Community Guidelines
OpenAI: Terms & Policies

[west scarab]

Any chance the content policy for creating fan art using the chatgpt version of dalle 3 will change? It worked amazingly a few days ago, and now I can't create anything based off of anything else 🙃

[dense ermine]

Has everyone been making spooky art in the spirit of Halloween?

I do really like the retro game ones

[quartz vale]

I should make more spooky art!

[agile peak]

TIL: Telling dalle in a form of storytelling is what makes it able to generate consistently.

[dense ermine]

Ah, how's DALLE 3 been working on ChatGPT?

[sacred helm]

any tips for custom instruction that works with DALLE integration?
here's what I have so far

DALL-E does NOT at all understand the "negatives" in the prompt. So it's useless to say "no". Instead, it must be solved by "positive" descriptive keywords that will hopefully neutralize/eliminate the undesirable element. When dealing with text-related elements, make sure to use keywords like "legible", "well-written".

[quartz vale]

I don't use Custom Instructions 🙈

[stuck geyser]

hey i just wanted to share something i found interesting while generating some images on dalle 3. i noticed that some separate generations actually share some close details. here you can see that the texture of the countertop in these two separate generations are actually the same (i hope gifs can be shared)

[quartz vale]

I've noticed the same thing! My assumption is the seeds used

[stuck geyser]

huh that's really cool, but i honestly have no idea how dalle works at all lol but i just found that really cool. i'll definitely have to research how it works!

[agile peak]

Been treating me so good with consistent style and subject rather than used ordinary prompts like MJ type.

Prompt:
Four different images like the four panels of an original Calvin & Hobbes comic strip, joyfully sledding down a snowy slope, and suddenly, after colliding with a rock, they're sent flying through the air and end up buried in a pile of snow.

[dense ermine]

What are the ways it interacts with chatGPT?

Is it different from DALLE 3 alone?

[agile peak]

I think GPT will revise the prompt in a form of story when it alter my prompt and sending to dalle.

[dense ermine]

Amazing, it actually looks like a legit C&H comic

I don't see a single error on it

Regarding the gallery, if I could, I would put superpower concept art as a one of the topic on there, would look sick

Like Infamous: Second Son to name an example

[agile peak]

Yes. I think you can see the example on their official instagram that, dalle understand storytelling to generate image through chatGPT.

Another try. However mentioned specific of subject wears/outfits.

[dense ermine]

Is that Tintin? I'm not too familiar with him

[agile peak]

Yep. I tried to experiment and see how dalle takes on capturing the tintin adventure (classic graphic comic/novel).

[dense ermine]

Well how about art in general?

Like you ever screw around with it and see what pops up?

[hybrid sorrel]

looks like its still using the same seed for each prompt, is there a way to change that yet?

[agile peak]

Well. It's worked pretty good as well; however you might need to do multiple attempts for GPT to get the text rights.

[final compass]

Is it possible to request a four frames progressing a scene in a comic strip style fashion?

[lime ocean]

Ok im oz , the dale3 app you can now upload photos but it cannot see them, im hoping very very soon it can in chatgpt4 dale3 beta, can anyone else do this. The web page no, but the phone app yes. Im oz

Boy it gets it wrong, but i think its not turned on yet?

[vagrant zodiac]

vision is separate from dalle3

they're not mixed rn

not sure how you can upload

is that the official CHATGPT app?

[lime ocean]

Im oz and the phone app you can , i think its new!

Yes! Its the app!

[vagrant zodiac]

hm

when I'm on dalle3 it disables uploads

[lime ocean]

I think its coming soon! Exciting!!!

[wind crane]

I suppose you can't do inpainting/outpainting from gpt DLL3?

[vagrant zodiac]

that's right

[wind crane]

I thought you were probably going to tell me that we can ask the AI to keep the same image and add/remove elements.
But there will be different versions of the original image.

but ideally if it was concise it would be fire.

[cosmic bluff]

hey,
i want to design icons that all are in the same art style. but - of course - every new prompt delivers a slightly different art style. is it possible for dall-e to keep the exact same sizes, colors, shapes, etc.?

[vapid elk]

it is possible, but is almost a dice roll

a way to do that is to use the sexact same prompt with the smallest changes possible

ask the AI to make only one image with the exact prompt without changing what you provided

[final compass]

yeah it is surprising how strict you need to be and strict in saying how strict you wanna be 😄

[cursive hornet]

Anyone knows.
Is the Enterprise version of DALL-E 3 the same as the Plus with a limit on the number of uses?

[white skiff]

probably unlimited

[cursive hornet]

Thank you.
However, it would be nice if there is some documentation somewhere... Enterprise edition users, please try and let me know.

[vapid elk]

Enterprise plans are a different use case, for the big costumers
it is not "unrestricted".. it ius just that you are paying for it

[cursive hornet]

Do you know if it is the same as Plus, or is there a much less restrictive limit on the number of items generated than Plus?

[vapid elk]

when a company get an enterprise plan, they make unique deal with OAI for their use case

some have a different chatgpt, some have reserved API capacity with increased rate limits, and so on

[white skiff]

https://openai.com/blog/introducing-chatgpt-enterprise

but its only for companies with hundreds of employees

[cursive hornet]

I see.
So there are more tune-ups than are publicly available.
The company I work for has a group of over 1000 people, but I don't know how many are interested in utilizing AI, so I'm looking into it now.

According to OAI, they are also trying to make an enterprise version for small teams.
So I have a feeling we will end up waiting for that.

[crystal badger]

Yeah Currently its for "The big boys", But Enterprise features way more context length and faster responses times, and a direct line to support i reckon as per the page.

theres no Usage cap on Enterprise.

[vague hearth]

What phone is this? The images ayou share aren’t OpenAI chat app, as I know it.

[crystal badger]

yeah it isnt OpenAI's APP.

[vague hearth]

Right.
Great trademark infringement right there

[crystal badger]

indeed.

or Perhaps its two Images merged.

[cursive hornet]

Even the "small" companies used today had 3,000 employees, so the scale is different.
And since those 3,000 are AI firms, they are all using it.
That's a widening gap.

[crystal badger]

u can use Vision Seperately then DALLE, to achieve a similar result, Unless its a Custom API Interface if its already out DALLE and vision API.

[vague hearth]

Haven’t seen anything in the api yet
Anyway looking a the shared image… it looks nothing like what was prompted for, so pretty useless lol

[crystal badger]

Cant say for certain, But i'd assume so.

[lime ocean]

Samsung s22 , ill post pics if it starts to work better

[lime ocean]

Yeah it is, ill post if when it starts working better, hopefully soon

[haughty spruce]

Is "pool party" or "university students" blocked? Because I've been trying to generate a nostalgic photo of 2000s and colleague students in a party at the pool with activities, but it blocked me two times. But "a photo of pool party, 2002, polaroid, cinematic lighting, film grain, blurry" generated without any problem.

[lapis fractal]

Doesnt seem like it.

Used: university students pool party

[cosmic bluff]

Why does this happen?
Even if I ask to make 1 and 2 in upright portrait format it stays the same.

[vague hearth]

No it isn’t 😅

That’s not the official OpenAI app.
Unless they got a new design 😮

[final compass]

this tends to happen when you ask for tall or portrait images, but never when you ask for wide or landscape images.

[zinc dagger]

if you try asking for the exact same prompt wide/tall you'll notice that sometimes they end up being the same image in both orientations with one just being flipped. probably just a bug.

[peak trellis]

which is better, Bing image generator or Dall e 3?

[loud field]

Lol

[open flume]

u can fix by asking for "portrait horizon"

[peak trellis]

I'm new to this

[loud field]

They are both using the Dalle3 model, except Bing is finetuned by Microsoft and the other by OpenAI

Theyre different but their capabilities should be the same

[peak trellis]

yeah but which one creates better images?

or would they create the same images?

[loud field]

No they would be slightly different

But same quality more or less

I wish I had gptdalle to test it vs bing's

[zinc dagger]

It depends on what sorts of images you're looking for. In theory the results should be good with either, but I've found the ChatGPT integration of Dalle is better than Bing for my purposes

[peak trellis]

sometimes bing image generator ignores some of the requests in the prompt. Does dall e 3 do that?

for example, I'll put in full body view for a certain prompt but bing will still give me about a half body view of a certain image

[zinc dagger]

Dalle should do that.

By default it has ChatGPT "upsample" your prompts to try and match the captions used on the models training set (so for example, "a full body depiction of napoleon bonapart" might be sent as "a full body image of the noble Napoleon bonapart, clad in his signature hat and pants, beaming with confidence atop a white horse")
Though you can disable this prompt-override by asking GPT nicely

[peak trellis]

Okay thank you

[peak trellis]

I'm sorry, I'm just a little bit confused. So Chatgbt will still ignore some of the requests in the prompt?

[zinc dagger]

sometimes it will, sometimes it won't.

For example, I sent the prompt:
"Full body Portrait of Napoleon Bonapart"
and a few of the images weren't "full body" but GPT had tried to produce similar results by changing the wording to stuff like "Napoleon Bonaparte standing confidently...". This is because GPT is trying to rework the prompt to recreate a "full body" image by specifying a pose which should have the full body on display.

If you don't like the results, you can tell GPT some specific restrictions like "you must include the text "Full Body" " at which point all the images will have that in their prompt

[peak trellis]

That makes sense. Thank you so much.

[proper palm]

how do you even do humans this good?

Mine are all clay-like

[crystal badger]

its all prompting

[proper palm]

Wich im really bad at lol

[crystal badger]

U can have GPT make your pompts, you know.

its good at DALLE prompts

[proper palm]

tried, all my humans still look like they are modeled or smth

[crystal badger]

try using Vison in combination with DALLE to find a style.

just giving you a prompt wont help you.

[hushed creek]

Is bing down?

[cloud dome]

Please don't request others to generate images for you

[novel raft]

Is anyone else having trouble with DALL-E 3? Just image placeholders are showing up for me.

[cobalt hull]

All is working on my end, are you on ChatGPT or Bing?

[simple aurora]

Hi! can someone help me? I want to know if it's possible to change this image and put this man next to the eiffel tower.

using dall-e

Is it possible?

[warped shuttle]

i am working on a comic book that want the same characters that i have created to be used. how do it get an illustration generated using the characters that I have created and have based on a link or uploaded picture?

[dim cradle]

i'd suggest you to delete the background with a more precise tool like photoshop or the automatic crop that new iphones do, then upload the png image without background on dall-e and fill in the blank space with a prompt. if it doesn't let you generate because it says that you have to erase something first, just erase a really small and unnoticeable part of the image

[simple aurora]

i'll try it

not exactly what I was expecting lol

[simple aurora]

do u know any other tool better than dall-e and midjourney to do things like this?

[dim cradle]

not really, mind sharing your prompt?

[simple aurora]

Change the background to looks like the man is just behind the eiffel tower

I mean, the man is really behind the eiffel tower, but i could do this in way better in ps

[dim cradle]

instead of giving instructions to it like "change the background" tell it just what you want, try "photo of a guy really close to the Eiffel Tower"

[naive lake]

Why is it when I switch the gpt app the pictures won’t generate? Anyone?

I have to have an app on my screen for the pictures to generate otherwise I will het some errors

[versed orbit]

I have that problem right now on the web, but working fine on mobile

[novel raft]

No go on mobile for me either.

[lime ocean]

I got a update a few days ago ill try post more photos

[plucky hare]

Since you're on Android, could you go to the app's "App info" (the little "i" in a circle icon), then show a screenshot of the bottom info under "App details"? This would show us that you're using the official ChatGPT app (com.openai.chatgpt) as well as the version you're running.

[lime ocean]

Ok wait a sec

[plucky hare]

Hmm, interesting! So I'm on Android too, but my app version ends with 284, not 263, so you're not on a new version, if that's indeed the official ChatGPT app. But I definitely don't have a "+" icon in my DALLE model, or even in the GPT-4 one, because that one shows the image and camera tandem icons, so not a "+" there either.

I think your most compelling evidence, if you even care to prove it this much, would be to make an unedited screen recording showing:

1. You opening the app, selecting GPT-4's DALLE model, and show that a "+" icon is in your chat and you using it.
2. Without pausing the recording, open up your multitasking view and tap the ChatGPT icon at the top, then tap "App info"
3. Scroll down in Android's app info screen (not the app's own About section) to show the version information

Again, probably a lot of effort just to prove something to some nerds 🙂

[lime ocean]

I guess the best evidence is that admin will not delete it! Haha, i dont want to make them mad, im super happy they mad me first here in Australia, thanks chatgpt happy to be a beta tester for you .

I will respectively upload photos if it starts working, i hope soon!

[quartz vale]

You can't currently create an "asset" with Dall-E, so what you're wanting to do isn't possible

[sharp fulcrum]

Is there a way to use a imagem as a reference? In Midjourney you sent the image and copy the media link

But I don't know him to do it in gpt ( if that's a feature)

[cloud dome]

No, there isn't

[leaden sable]

How do you get to the chat AI generator that lets you put in a description and it pop out pictures... like a continual chat?

[thick smelt]

did this for ryan Reynolds' birthday '

[fallen folio]

The seed issue is annoying

Any idea when it is going to be fixed?

Generations that require creativity are too consistent when I ask to modify them

[agile peak]

Anyway there’s something about dalle that obsessed to create images of alien based on depiction of xenomorph 🙈 which is lack of variations.

[queen lion]

That's a shame. They certainly advertised it the other way.

[quartz vale]

when did they do that

[agile peak]

This is cool. Blending of knight and chinese dragon.

[tepid valley]

Hi! I have a question : if I made an image with dall-e and after that I will sell my image, this is write to me chatgpt when I ask this question " It is important to note that you must adhere to all the terms and conditions of use of DALL-E technology, including content policy and copyright laws. It is recommended to consult with a lawyer to understand the legal implications of using DALL-E in your business. "

[tall patio]

Would Dalle-3 be able to produce vector graphic in the future?

[quartz vale]

This means you can use Content for any purpose, including commercial purposes such as sale or publication, if you comply with these Terms.
https://openai.com/policies/terms-of-use

[wind thunder]

After 3 month my premium finished 😭😭😭😭

[tepid valley]

Thx

[cosmic bluff]

I wonder, there isn't mentioned that I can't produce persons of interest, gore or sexual content or is it?

[final compass]

No matter how i try to tell, that there is only a small convoy or infantry unit approaching, it ends up generating a whole field army, and that wish is clearly in the prompt that went in, but nope 😄

[swift oriole]

The new dall-e image gallery is it like a forum threads can have other users post images in any thread as long as it's the title subject material?

[lapis fractal]

@final compass Are you using GPT+ with Dall-E3 or Bing with Dall-e?

[white skiff]

wide picture means chatgpt

[lapis fractal]

@final compass try inserting "Text2im:" before the prompt.

It seems to follow the prompt more directly where if you exclude that part it adds a lot of its own GPT prompting.

[final compass]

What is that parameter doing?

[placid orbit]

u can tell it to put it in a square aspect

[white skiff]

yeah but i commented to the sniper pic question

[vapid elk]

yea, it is basically for continuing the discussion of any particular post, if you want to post pictures it is a good idea to make a new post

[vapid elk]

this isn't really a parameter, but just a way to make the AI do what you want
sounds like a shorter version on "generate an image of exactly this prompt without changing any word"

[final compass]

At times i have struggled with it omitting something and adding something else, especially when i have the prompt nailed of what i want.

[cosmic bluff]

i started describing what i want in my native language and realized that it works better sometimes, because my words are more on point. but sometimes dall-e uses even those prompts in my language (german). do you think it is a problem if dall-e doesn't use english prompts?

[hushed merlin]

#1154829862171844679 2D game character

[sweet escarp]

Is there an API for this?

Dall-e 3?

[wispy storm]

Not yet

[quartz vale]

That breaks the Content Policy

[verbal sequoia]

<:book_icon:1046291503158399107> `` Rule 0 `` Adhere to the following Terms & Policies:

Discord: Terms of Service, Community Guidelines
OpenAI: Terms & Policies

[sweet escarp]

Darn the results so far are amazing I can't wait for an API

[wispy lily]

I may just need to be better a prompt crafting but I’m not sure how to get a good image from this thing.

[white skiff]

he probably used bing

[waxen saffron]

why cant i acess dall e

[rigid bough]

can you provide more context?

[waxen saffron]

i wantet to make a picture with dall e and i did go on the apps but there was only chatgpt and the api

[rigid bough]

are you a GPT Plus subscriber?

[waxen saffron]

no why

[rigid bough]

DALL-E 3 on ChatGPT is for GPT Plus subscribers

[waxen saffron]

oh i didn't knew that

thanks for helping

[dim glacier]

I'm a plus subscriber but I cannot use the Dall-e function. ChatGPT is answering me like this: "I'm sorry for the confusion earlier, but I cannot directly generate detailed artwork. However, I can provide a conceptual representation using the capabilities I have."

[rigid bough]

go to the GPT-4 model selector and choose DALL-E 3 from the dropdown options

[wispy lily]

Bing has AI image generation?

Is it Dall•e 2?

[white skiff]

yes same dall-e 3

limited to square pictures

[wispy lily]

Could I get a link?

[white skiff]

i think its not whitelisted url so cant but you can search for bing image creator

[wispy lily]

Interesting… Thank you!

[tight verge]

another fine day where dall-e can place the correct number of fingers on the subject 🧠

ignored the caption, but i won't complain

[agile peak]

These are probably the most detailed generated images related to AI from Dalle. Have't got the same results recently.

[glass topaz]

Is there anyway to make Dalle 3 actaully spell things correctly.. Like i give my streamer name of EDPHONEZ . and i get this..

[white skiff]

nope, you just need to retry until it gets it right or photoshop

[glass topaz]

yeah i took the prompt i gave for that one and got 4 more.. and got some that i could use.. Thanks.. I am sure the next version will be even better.. I abosoluty love the things DALLE is able to do.. freakin amazin.. that being a person who spent WAAAAAY to much time with Stable Diffusion.

[versed orbit]

I'm guessing DallE is being blocked at the school I teach at. GPT will produce the image, but only placeholders show. Odd though as it had no issues displaying images it created last week. No problem on app.

[vapid elk]

it is not written ther ebecause those are the Terms of Use, you might be looking for this: https://openai.com/policies/usage-policies

[scarlet urchin]

ohhh a mod added me. well @cloud dome , let me warn ya tho, i suck at writing prompts and just hope 4 the best. if u want to know the prompt for that it was > From comic “One Piece” style by Oda Eiichirō, Teenager, with a ferocious expression, with his hands behind the chair and his legs on the table. Behind it are all kinds of strange liquid bottles. mad scientist, Perfect combination of green and abstraction, mad scientist, by Junji Ito, by Yoshitaka Amano, anime style, 80’s vintage anime still, strong sense of line

[cloud dome]

Thanks!

[strong idol]

hi is there an api for dalle-3?

[wispy storm]

Not yet. Might be later this fall

[open flume]

anyone know why we can't use Dall-E 3 on labs.openai?

[frigid trellis]

It work as far I know with ChatGPT pro

Or labs.openai but it’s payed

[lime ocean]

No changes yet hopping soon

[plucky hare]

Hello again mysterious Australian person potentially with unreleased features 👋 What are we seeing here? Left pic, your DALLE model with the mysterious "+" symbol again where you can upload pics, but it apparently can't see them? Right pic, what is this? I see both the image and camera buttons so that looks like default GPT-4 to me, but it's not getting the description right, which it should be since it's the "regular" GPT-4 with vision model?

[lime ocean]

Yay, yes, im super excited, but they are so heavy on copyright so surely we jusy can't copy and paste images right?

[oak topaz]

hey is some one here using bing for dall e and has 5 hour wait time aswell?

it's been like that for the past feew days

[lime ocean]

I heard one of dale3 competitors now has a beta website and mobile app.

[lime ocean]

Ooh haha yes its either a slight ban for making images it didn't like or you made too many images, that's my guess

[lime ocean]

Took 30 seconds here in oz

[oak topaz]

i havent really used it in a weeek

and nowi ive used those credits and its telling me i have a 5hr cooldown

it was like that like 5 days ago

[oak topaz]

wdym by oz?

[lime ocean]

Sorry they gave me a time out for 5min because i was trying to help you haha, anyway basically i found if you ask for more family friendly images you can post more, the land down under mate haha

Haha thats great

[oak topaz]

that actually makes sense considering my werent so family friendly ;c , thanks

[lime ocean]

Anytime

[ancient jasper]

Can't generate anything right now

Keep getting errors

[young burrow]

I'm having the same problem, for about the last 10 minutes. It was doing it around 2pm today as well.

The front end doesn't show detail, and the chatbot doesn't know what kind of error it got, just "technical difficulty"

Too bad, since I was finally getting it to display a volume knobs with the numbers 0-11, just in the wrong order and an infinity symbol thrown in.

[remote stone]

Oh good to know I'm not the only one

[young burrow]

I'm guessing (not related to openai here, but I work with servers a lot), there probably is a capacity issue since my 2pm is lunch for West Coast, and right now the East coast is off work.

[amber raptor]

you can fix this by telling it to continue until multiple valid images are generated. it needs to continue despite the errors

atleast it works for me right now

[young burrow]

I tried a few times to "Please try again if it's just a technical error" and sometimes it'll go through, but frankly, if it isn't important, probably better to back off for 10 minutes since it's like an end-user repeatedly spamming the F5 key in browser. I was planning on waiting an hour myself.

[amber raptor]

just ask it to continue until there aren't errors

this was my exact prompt: "please continue generating until you get atleast a couple of successful images. if it goes wrong, try again and again"

[amber raptor]

it fixed the image issue

[young burrow]

That said, the content filter is a bit finicky. I was trying to generate a desert with a half burnt tree with broken limbs... I had to change it to a desert with a half burnt tree with broken treelimbs. Took me a bit to figure out that it was confusing it with the non-plant sort.

I also got it to ChatGPT-4 to explain to me -tan suffix for characters to set context and then ask it to generate Schadenfreude-tan for me... Quite the wicked character it came up with, 1 of the 4 images didn't generate "due to content policy" so I have to wonder what did ChatGPT actually make for that one. I had forgotten to ask the prompts.

[round hamlet]

I am not able to upload an image to edit on dalle2 - I go thorugh the regular steps but after I select an image nothing happens

[dim cradle]

Just started happening for me too, is it down?

[lime ocean]

Oz good 👍 works ok

[agile peak]

Dalle doing amazing job generating medieval manuscripts ✨ fyi

[agile peak]

ChatGPT version. Impressive how it even captured vitalik resemblance even though I didn’t mention in the prompt.

[candid perch]

I am considering generating a .obj file (for a 2.5D CNC machine) using the DALL-E 3 API along with some other libraries, so I can carve the design into wood. if there is anyone working on something similar, would love to explore more!

[plucky hare]

DALLE 3 is not currently usable via API, they've said that's coming later in the fall. Currently it's only usable with ChatGPT Plus, and in Bing's image creator.

[ancient jasper]

Working again

[cloud dome]

This is toxic

ba dum tss

[vapid granite]

According to GPT4 itself, the Bing Image Creator is the most recent version of Dalle 3. Through extensive testing the BIC version is definitely more creative and more dynamic in the way it frames the shot and poses the characters/subjects, but obviously is a gimped and outputs much lower quality than the GPT4 version.
When is the GPT4 version going to be updated with the newer version, or are we stuck with it pretty much until "Dalle 3.5" comes out in the future?

[sacred helm]

Apperently you can ask for seed parameter

[sacred helm]

#images-discussions message

[sacred helm]

#images-discussions message

[glossy scroll]

How?

[sacred helm]

just tell GPT it can add "seeds" parameter

here's the result

[plush sky]

All I get repeatedly is this same message. So many issues constantly every day.... Barely manage to get 1 or 2 images generated in 30 mins

[glossy scroll]

I'll try it

[sacred helm]

does you request have recreation of existing art?
I had had that with Spider-Gwen art

it was pretty annoying to find that out there was no "contents policy" warning

[plush sky]

No, just a random scene on a beach. Was working fine last night, then I kept getting errors. Wake up this morning, try again and still error after error

[rigid bough]

what is your prompt?

[plush sky]

Won't share that as it's a unique artistic idea, but there is really nothing wrong with my prompt, not even remotely edge case. I asked whether it was my prompt too, and it says it's not.

[rigid bough]

without the prompt it's impossible to determine what might be going on

[glossy scroll]

How are you requesting info from Dalle like this?

[sacred helm]

use superpower ChatGPT chrome extension

[glossy scroll]

Hmmm. Okay, I'll see what I can do. Thanks for the info btw

[sacred helm]

np, I was struggling with the same problem

[glossy scroll]

I'm gonna see if I can find some other way to request information from Dall-e like the way you have. Because I have some security concerns with chrome that I try and avoid.

[peak trellis]

Which is one is better: bing image generator or dall e 3 through chatgpt 4? I know both are powered by dall e 3, but which one do you guys feel like give the best images?

[sacred helm]

bing image has a watermark
and image created with dalle intergration for chatgpt is commercially usable iirc

[knotty gull]

How is this even illegal?

Damn picture not uploading

You can do messi but not arnold? Bs

Now the censoring AI? Nice. Everything is censored now

[drowsy bane]

Thank you! I just tested and it worked. The results all did use the same prompt

[sacred helm]

glad it helped☺️

[wind crane]

Is it possible to create images by asking for a first person point of view?

At least someone succeeded?

[sacred helm]

look at this 4 iteration making a comic
in all consistent art style

[sacred helm]

try again, but using different seed
here's how you can do it
#images-discussions message

[wind crane]

i don't understand...

[sacred helm]

try again with
(your original prompt) + add "seeds" parameter in your DALLE request, give me 4 iterations, all different seeds

[remote mountain]

I'm having issues accessing Dall-E 2 version. It's a blank page. No option to do anything.

[rigid bough]

are you going to https://labs.openai.com ?

[remote mountain]

Yes

Does it work for you right now?

[rigid bough]

it's working for me, maybe try and clear your cache and cookies then try again

[remote mountain]

OK thank you

[sacred helm]

striking consistency in character
even in 4 seperate iteration
this is the power of seed parameter

[magic mauve]

Hello all - How do I create realistic images - Mine are all cartoonish. Thanks in adance.

[wispy storm]

#1019652163640762428

[agile peak]

This is entirely written by ChatGPT-3.5 😁 I am only give the idea and set.

[agile peak]

I tried as well. Turned out changing seed can give different and consistent results.

[agile peak]

Change/increase my seed and the textual element is getting more accurate.

[sacred helm]

yeah, that's what seed is for

[agile peak]

Thanks for the tips, mate. I saw your thread on Reddit before haha

[sacred helm]

Ah, that one
I hope it gets popular so everyone can get this useful tip

I hope we can find the seed that goes perfectly well with text generation
that'll be a huge game changer

[crystal badger]

I only succeeded in a 6 panel consistent Comic in a single image in 1 iteration, love the way you prompt

[sacred helm]

can you share the images?

[crystal badger]

At work but you can DM me then I’ll send it when I get home

I don’t have it on Convos I deleted mine a few days ago had too many

I do have an old one I can give

[sacred helm]

Ah, you're saying that your best trial without seed param was the comic, now I get it
I misunderstood as you successfully generated 6 panel comic in 6 iteration with fixed seed

[crystal badger]

Ohh that’s not an issue it’s totally doable

I just tend to have it in 1 image for convenience

[cosmic bluff]

I see, terms of use and contenct policy is soething different.

[sacred helm]

I can't just stop looking at this consistency
It's the perfection I craved for so long

[tall patio]

The hair is always floating for some reason

But yeah same

Consistency is unmatched

[crystal badger]

Here I made a quick 9 panel

[sacred helm]

holy fork

[crystal badger]

It’s a rough 1 at iteration

[tall patio]

I told chatgpt to remember the face and it did

True black magic

[sacred helm]

What was the prompt GPT wrote?

[tall patio]

Just Manga girl

[sacred helm]

this, is the perfection

[tall patio]

Jellyfish hair

[final compass]

Does anyone know a good way to give a prompt specifically wanting to avoid something? For example i have tried to specifically say things like: he has no tattoos, rifle has no bipod, there are no helicopters flying. But Dalle does not seem to care and does all those things anyway, sometimes even making it worse trying to describe something that should be avoided.

[sacred helm]

DALLE is not ChatGPT, it can understand those no's
just simply don't use those elements in the prompt

[cloud dome]

You could just try describing those things without negatives

So for example no helicopters could be clear skies

[final compass]

Yeah i guess the problem is then that it seems to associate things with words. For example "sniper rifle" always seems to have bipod. Or if someone is described as gang member, he always seems to have tattoos and so forth and if something is described as world war 1, it still might have modern helicopters flying, due to association with war.

Need to be more basic with descriptions i guess.

[sacred helm]

or you can put random seed value and regenerate again and again till your requirements are met

[final compass]

yeah there is definitely a luck mechanism involved most of the time 😄

[sacred helm]

#images-discussions message
this is how you can do it

[final compass]

Yeah i just today learned about the actual json and it really helps understand what is actually going on. Helps becoming more precise.

[agile peak]

That makes chatgpt way more understandable to change the subject! Didn’t even got any flaws in textual elements.

Tips: to get much better 3D visuals rendering, mention lightning and shadow (such as subdue, mute, or diffuse) and reflections.

[broken vault]

When I ask chatgpt to print some text in a banner, it completely ruins the text. How did you get perfect "Zoo York Times"?

[agile peak]

This is my prompts and the way how I communicated with chatGPT 💬

Pardon the typos 🙈

[broken vault]

This is great.

[agile peak]

Yeah. I was surprised how the penguin one was such an improvised piece by ChatGPT lol

[wind crane]

should I just invent a seed and tell it to follow it?

[sacred helm]

yes, smash your number pad

or you can ask ChatGPT to randomize it

[white skiff]

was the seed issue fixed yet?

[sacred helm]

you mean gpt using same seed?

[white skiff]

yes

[sacred helm]

I found the break through

#images-discussions message

[wind crane]

like “use the 300 seed and draw me the character in profile view”?

[sacred helm]

yes, but you can go further
"give me 4 iteration with all same prompt, but all different seed
use seeds in 8 digits that has no pattern like 12345"

[white skiff]

did you actually test it thoroughly that it works? it accepted that during the alpha testing but was still always locked the same behind the scene

[sacred helm]

it def works

ask chatgpt with DALLE on what param it can insert it in request body

it gives size, prompts and seeds(optional)

[white skiff]

yeah afaik those have been there when even when it was just a plugin

ill try it later

[agile peak]

After I asked to change the seed. It generated in sequence next time.

The higher the seed. The accurate the text will be.

[sacred helm]

#images-discussions message
this is the one with
"start with japanese manga girl sketch and alter perspective"

[sacred helm]

really?

[wind crane]

the seed stories you're talking about, is it simpler than using the custom instructions to create my character and call him by his first name for a specific action?

[agile peak]

That’s what I tested three times with different subjects.

[sacred helm]

that's also one way of doing it

[agile peak]

My text is for the name of the shows.

[sacred helm]

let's do 2,147,483,647 as a seed
it's largest standard 32-bit integers

[wind crane]

With your seeds, how would you proceed to create a character and make a variation of it by conserving her clothes and her face for two simple actions like "Amélie is sitting cross-legged on her bed" "Amélie gets up to open the door to her room “?
I mean can seed make it easier to create comics unlike writing as usual "Amélie does this, that etc..."?

[agile peak]

Tried this. Amazing though.

Lower seed and then used the exact seed you posted.

Tbh I didn’t add more texts only settings, note and search haha

The app icon

[sacred helm]

I'll go with

1. "here's the description of the character I want you to make: {character desc}"
2. "generate a prompt for making her" reviews the prompt
3. "OK, I want you to keep this prompt in every iteration for image generation"
4. "add this {random seed} seed in your "seeds" parameter for request body and start every prompt with "{character desc you got on step 3}, and add some actions, which I'll give you from now on. "

and no, the seed is not for altering details like action, it's for maintaining art style, character apperence, and more

[wind crane]

I see.

[zinc dagger]

interesting to see how the seed values are effecting the image generations, some of the stuff I'm making has a drastically different artstyle with different seeds.

[sacred helm]

iteration with seed 2,147,483,647
flawless text generation

only one mistake : AINT > FIND

[zinc dagger]

Alright, having control over the Seed is pretty cool.
I've found that once you get a good looking prompt changing the seed by a little bit can keep it in the same ballpark of style with different results, but depending on the prompt a more severe change in seed can change the artstyle as well

[vestal acorn]

whats seed?

[sacred helm]

#images-discussions message

[vestal acorn]

huh, thats cool

[sacred helm]

you'll see the images generated with bing image creater are different, even with the same prompt

that's how seeds work

[vestal acorn]

got this with a random number

pretty neat

[sacred helm]

here, I selected the great art style I liked

and then, I can reproduce the character

by using the same seed

[vestal acorn]

oh thats awesome for consistency

[sacred helm]

yeah, this is a real game changer

[vestal acorn]

using your seed i got this

[agile peak]

Looks like grumpy superhero haha

[tall patio]

Seeds talk is very loud on Twitter today

[sacred helm]

who leaked it?

[tall patio]

It's already buzzing since last week

OpenAI is playing games with us

Nothing is definite and everyone is on a giant Easter eggs hunt

Everything feels super vague maybe because they're under a huge lawsuit storm atm

If you're smart you'll figure it out on your own

We have no liability

[young burrow]

Could use some opinion
````Hand-drawn schematic draft in blue with white lines, titled 'Idea Evaluator'. The schematic features a rubber duck with front, top, and side views. Various measurements are marked on the duck's body.Is returning the error:I apologize, but I'm unable to generate images for that request due to our content policy. Let me know if you'd like to try a different request or if you have any other questions!```

I'm going to try to alter the prompt, but I find that to be quite strange for that be a violation.

[sacred helm]

lollll this one gets policy error?

Oh maybe "rubber duck" is existing art

[white skiff]

yeah probably the rubber duck part

[sacred helm]

you know, giant rubber duck as a contemporary art

floating on the river

[young burrow]

I would have never automatically considered it. I tried changing to duck made from rubber, but no dice. I'm trying a random mutation.

Changing the prompt from duck made from rubber to butterfly made from rubber allowed the image. The is a very troublesome thought.

[young burrow]

And "duck carved from wood" was successful. I believe your guess was correct. Since rubber ducks are from 1800's this seems like a bug in implementing the content filter.

[young burrow]

Using "duck made from plastic" worked and looks correct, so must be a term issue.

I haven't seen that, but that sounds interesting.

[sacred helm]

I can't find the image, but making that image sound interesting
you want me to do that?

[young burrow]

BTW, my generated image after changing the prompt

[ripe wraith]

what if someone allergic to avacado tho?

[tall patio]

Avocado is the cure for male loneliness

[ripe wraith]

IDK but I feel these characters should all be screaming in pain and flopping over as soon as they come alive...

he's cut in half for crackers sake

[proud berry]

Prompt:
Drawing of a chibi anime girl with rosy glinting eyes that have galaxies mirrored in them. Her hair is a shade of purple with a neat fringe. She wears a blue sorceress hat embellished with rosy blooms, a round charm with a glistening blue gem, and a scroll bearing etched inscriptions. Numerous earrings adorn her ears, including one with a hanging tiara and another with a leaf motif. Her blue gown is graced with golden illustrations, a central blue pin, and a roseate ribbon. She sports a necklace with a blue heart locket. In her hand is a coconut mug with a blue jewel protruding. The art has a radiant finish with muted hues and sharp features.

[proud berry]

Wow

[vapid elk]

it has aBeta badge for a reason :V

[tall patio]

In the long run I guess things will be more ... accessible

[fair crescent]

Rubber is also a word for a certain contraceptive

[amber schooner]

can anyone help me in #prompt-engineering

[deft musk]

Context for a chat elsewhere.

[hybrid sorrel]

I can't for the life of me get it so draw a character in full view.

it always generates a portrait shot.

I've tried full body view, full character view, head to toe, asked it to just zoom the camera out.

still always a portrait shot.
any tips?

[rain vault]

wen will text be fixed!

[plucky hare]

As far as I know text isn't broken! DALLE 3 can generate text better than lots of comparable models, but it's not advertised as a flawless feature.

[rain vault]

no I mean in imege it is spelled wrong

[lime ocean]

Its uploading the photo faster but it still cannot see it.

[lime ocean]

Try bing app dale3 image creation, fullbody usually works, or try saying things like distance

[agile peak]

These are just perfection and consistent

Reupload accidentally use spoiler lol

[sacred helm]

I figured out how to retrieve seed value from "any" image you generated

@agile peak wanna know how?

[agile peak]

I’d be happy. Spill the tea 🫖

[sacred helm]

alr, this one involve dismantling website code

first, remember your chat session id

I highlighted the id in yellow

press f12, go to Network tab, press f5 to refresh the chat session window

find the chat session id on the "Name" tab

click it, and go to Response tab

scroll down to get seed value

[agile peak]

Good works, mate this could be a game changer to retrieve the seed. Saving me up as well for both token and chance of being rate limited.

This is with same seed. But just adjust the lights and reflections + asked GPT to add textual elements references to panel.

[sacred helm]

I saw many people wanting for the seed value for the image they really liked
so I couldn't stop myself from dive into the investigation

[naive lake]

Dall-e becoming slower and slower for me with each day, anyone else encountering the same issue?

[boreal coral]

the images on the website for dalle2 are so cherrypicked, so you KNOW the ones for dalle3 were as well. i mean bro. it literally made a perfect hand. and that explosion of powder one looks better than what dalle 3 makes with the same prompt

[flat zephyr]

just specify what they wear on their feet and there you go 😉

[flat zephyr]

or just ask chatgpt to give you the seed along every image it generates, and it will

[hybrid sorrel]

lol, still didn't work

[sacred helm]

randomize the seed value to find which seed gets your requirements

[agile peak]

With higher seeds. Additional texts are so much better.

[sacred helm]

I'm skeptical about it
seed is just a wheel of fortune. no matter how fast it spins, you get what you get
I don't see any reason the seed affecting image quality

[quartz vale]

I really like them! That dog one is pretty awesome

[quartz vale]

maybe slightly #off-topic for this channel, but I recommend everyone using the ChatGPT version of Dall-E 3 to try Custom Instructions! 😄

[dim cradle]

Hey all— maybe I am misunderstanding something but .. if you upload an image to DALLE, then erase a portion of the image, then type into the bar a prompt for what you want in the blank space then press generate, it should perform this right? Whenever I upload an image then erase a portion it ends up just generating a weird variation inside the blank space

[sacred helm]

wait, are you saying you were actually able to send image in DALLE mode?

[boreal coral]

in lab

[sacred helm]

ah, DALLE 2

[quartz vale]

I recommend you make a post in #1154829862171844679 for your images 🙂

[agile peak]

Haha. Funny stuff today, asked to do netflix inspired was rejected by GPT due to copyrights. So I did "Netflicks" with its inspiration.

Was mean to re-create in form of parody, so it's use the real logo instead.

[dim cradle]

Yeah for DALLE 2 , any chance you have an answer

[sacred helm]

sorry I don't, I thought you were talking about DALLE3

[agile peak]

Change the seed to 5,000,000. Worked.

[sacred helm]

I'm skeptical yet the texts look so accurate 😭

[agile peak]

For these specific images using higher seed resulting additional elements that's not in my prompts.

Haha. I said pixar but the result ended up collabs between disney and pixar.

[flat zephyr]

sometimes it takes a bit more, eg specify they are standing, or be more specific, eg red shoes

[flat zephyr]

also if you for example give specific details about the features of the face, it will more likely zoom in on theface

[agile peak]

I am impressed by dalle every day. It even captured the logo of Indonesian local beer brand - in pixar’s inspired world 👏

[brazen sonnet]

i was severely abused for my ai art once more

thats the 9th abuse this month

[brazen sonnet]

maybe its time to quit ai art

they all reported my store

[subtle creek]

This is interesting

[subtle terrace]

It's annoying that it doesn't show the ChatGPT-generated prompts in the gallery view on mobile (at least on iOS, not sure about Android).

[subtle creek]

Agreed

[brazen sonnet]

thanks ill try that jailbreak 👍

[final compass]

I would even prefer to have the seed number visible as well.

[subtle terrace]

I agree, I just started using the following custom instruction, working for me so far: "Show the full text of the prompts you've generated, then generate the images. List the seeds that were used to generate the images. Don't ask me for confirmation, go ahead and create the images."

[thick smelt]

*she

[agile peak]

Just asked GPT about the seed for now.

[agile peak]

This is interesting. I tried create bruce wayne and mentioned alex ross style; got same message. But somehow chatGPT generates anyway when I agree to use its own guidance.

Then chatGPT made these three. Yet it also used alex ross style illustrations.

I'm also certain chatGPT will perform better on giving result during the less usages activities including dalle. Might be anecdotal but it's speaking from experience.

[final compass]

Yeah i have started to request those with the output.

[vagrant zodiac]

it gives you the prompt when you click on the image

[final compass]

Yes it does, but not the seed number and in the android app you don't even see the prompt.

[vagrant zodiac]

ah ok

yeah I wouldn't count this as a jailbreak personally just because it's something it seems should be exposed?

[final compass]

I noticed that using the same prompt and seed reveals the status of the content policy, since it may not generate images it already did in the past. You could basically create series of prompts to determine the content policy baseline.

[vagrant zodiac]

i think it doesn't provide the prompt in the response because it is already returning it

just not.on.android for some reason

[vagrant zodiac]

I would just advise to not try to test the limits of the content policy

if something seemingly innocent isn't allowed, try rewording once but usually it'll tell you why it's not allowed

[final compass]

Yes that is true.

[vagrant zodiac]

there's not been an official content policy update in a while, but it's probably safe to assume that we're not meant to try to generate stuff like e.g. copyrighted content

[final compass]

Yes it tells that pretty directly, if you try to ask it to generate e.g. Batman.

But from my experience the team is constantly fine tuning how the content policy works.

[hybrid sorrel]

hm, that's annoying.

[final compass]

Also noticed this, when i gave an expression for the subject, it tended to do a close up pretty stubbornly, despite telling it otherwise.