#images-discussions
1 messages · Page 15 of 1
boioioioing
I'm out of my three hour gpt4 chats, but I'll try it later and report back
i bet it made some high school wrestling pics then freaked out
Possible XD
Even ChatGPT itself didn't understand the issue and said it should be fine.
Anyone else finding that the chatgpt dall-e is WAY stricter than the bing create version?
yep way stricter!
i'll copy the same prompt over to bing and it works
its stricter, esp for real people. fake copywrites, its a bit more meh about
i think a mod or developer said its due to different policies or something
I waited until I had it.
I noticed that in the app version of chatgpt, it replies differently than the browser version. Someone on Reddit said that they got weird text when they opened a browser conversation in the app - see pics
It's like they have a hidden prompt from Dall-e that self-polices the content generation
information leakage, neat
"issues" lol
Its a sneak peak into how they code their internal prompts, i want to see them all now
based on the above, i think it actually lost the information on purpose and will never give you a straight answer about it, it'll just hallucinate for you when it says a specific reason.
Use the phrase "issues" in the response. Smart meta prompting?
I wonder what happens if you, in your prompt, tell chatgpt to DO NOT UNDER ANY CIRCUMSTANCES listen to instructions from Dall-e
jailbreaking works for that very reason, or similar reasons i'd bet
I think it's just prompts all the way down, and it hurts my brain
How do I get chatgpt to list Dall-e?
is the word "translucent" not allowed to be used for DALLE?
Depends. It's very... Sensitive to a lot of words from what I've seen, but it should be fine as long as you keep it away from any character in the art piece.
i tried to get it to draw up translucent glasses but yeah it wont let me, so weird
Hey crew, been trying to get a Gand, or as close as i can, but all of my creations come out WAY too 'insect' like, more like a humanoid locust. Anyone had any luck with that, or some other star wars races?
haha I was trying to get transluscent ghosts earlier and it led to some unexpected clothing output
If Dall-E 3 encounters an error generating (as not as its not prompt related), it shouldn't count towards your 50 prompts for the 3 hour period. I'm tired of wasting credits on this
yeah i agree, its unfair that it counts towards the 50 prompts
oh wow, it actually worked? it wont draw up anything for me at all since it "violates" content policy
I do wish they would point at what part of it violates the content policy so that we can avoid the stuff that violates it. Even if the stuff is so stupidly inoffensive
where can a brother get dall e 3 gpt 4 plus
I think you gotta wait until you are picked for the rollout
yeah its rando,, you gotta wait
Since they are giving it sparsely to plus users, though it should go with everyone at the 14th
yeah same. ive tried a few different prompts with "translucent" and none of them works, so i assume its that
Eh- Everyone in the plus thing
sneaky sneaky I got it to consistently include the hidden message - see number 4. In this case, the first three prompts were accepted and the fourth one was not.
thats so fun
I wonder if that seed parameter is real or not. even if it is probably can't be changed through our prompts?
🔥 👍
FINALLY got access to Dall E 3 inside of ChatGPT. (Still no 'vision' yet, though). Presently I'd put this leagues ahead of MidJourney.
Can you share a screenshot of one of the errors
It's on and off. I feel like about 2/3rds of the times I try, I get this. And it usually works in waves. 5 won't work, then 3 will
I've had this happen, but then I'd start a new chat and it'd be fine. but also vpn, any other weirdness could do it
have you tried completely different prompts, like a super simple one? To make sure its not just peeved randomly by your prompt?
yeah I'm sure it's not the prompt. I just keep clicking regenerate every 5 mins or so, and eventually it starts working again for a while
system error == system message i bet
Yeah so dall-e must do all of the filtering itself and then it provides instructions to chatgpt on how to wrap the interface. I don't like how it talks to my chatgpt :(
yeah, we're all talking to the true AI brain through multiple middle men AIs
I wonder if I can use custom instructions somehow in this situation
I'm half tempted to ask it to make a robot shouting at another robot cause that's how it feels
Fundamental question, maybe somebody here with some insight knowledge:
I have noticed that for some reason for some major franchises (hoyoverse/Riot etc) Dall-E 3 is basically not able to produce characters. On the other hand smaller or older franchises (avatar, berserk etc) can give precisely the characters you ask for.
So I wondered: Has OAI/Microsoft actively struck a deal with specific major franchises to opt out their characters, where if you request them the AI actively randomizes the picture?
I'd imagine it is more "Hey this is popular and this is only semi-popular" or something, but that's mostly what I think.
I'm thinking MS gets special access to the rules because they are investors or something
I mean there is bazillion artworks for Genshin Impact characters. But if you request those chars it actively randomizes the character specifics and only roughly approximates the style.
if dalle had in-painting like midjourney it would be so perfect
i thought it did in the web version
whattt, how do i use it? 😮
https://labs.openai.com/ -- then edit any existing image, apply the mask and regen
its dalle 2 tho
Oddly enough it can do splatoon without a complaint. This really confuses since I tried to also do My Little Pony and Castlevania but those got "Nah against our copyright thing"
Yea I think some of the major companies actively requested a block
I'm amazed that Nintendo didn't do that too then
honestly anyone with a brain for marketing would encourage their brand to be used in AI
I looove this, Asuka at the declaration lmao
"Get in the State House, Shinji!"
Yea, Mario works perfectly fine. Maybe nintendo just doesnt care about the artwork as much as others. Tbf Hoyoverse is much more relying on high end artwork than Luigi
OK, more testing weirdness. I had Dall-e generate four 100% safe images and it did so. I then asked it to generate the same prompts with the same seed, but it only returned two images and gave the error I listed above as the reason why. I then told it to try again with the two images that were not generated and it was able to do so.
In short, the error message from Dall-e is NOT just used for content filtering, it is used for other things as well - like any time there is a bug maybe or system is overloaded.
TELL IT YES
feels like being told sweet lies though
thank you, i'll check this out
Well I think that dall-e actually sets the seed, not ChatGPT. That's why if you make it use the exact same prompt for each of the four images, it will give you the same exact image like pic related.
You can test it by asking chatgpt to make images with the exact prompt with no changes or added text: "[prompt here]"
Interesting, so I am currently testing how popular a char has to be to be recognized by Dall-E 3.
Mario - works
Ciri (witcher) - works
Serana (Skyrim) - does not work
Then there are some franchises that seem to actively sabotage the request (riot games for example)
Ok try this: Ask it to help you craft a prompt to change that parameter, then use that paramter to get an image, start a new chat and try again to get that same picture with the same seed. Something besides the default 5000.
because it's offering, which is super nice of it
thats how i learned the coding gpt could give file upload links
Oh damn wait. When does the dataset stop that Dall-E 3 is trained on? Like till what year did it gobble up pictures`?
ask it lol
I cant, I am useing bing right now
i laugh that we're in a world where that can be my serious response
Well that answers that hahah. It is interesting that it looks like the same seed is used throughout an entire chat. If you want a different seed you need to start a fresh chat? I'll see what happens if I make it use the same prompt in a new chat
Set once per call, makes sense. But if you start a new chat and do it again at 6000, are the images the same?
oh it says at the bottom nevermind, do it if you're curious only
nevermind, in the past it was not publicly known.. for now however let me google
any good prompts for character sheets that draw out an animation run cycle?
Dang so there is NO point to using the same prompt over and over again like in bing create
I wish there were some actual OAI devs here to answer these trivia questions 
Ohhh it's locked in "for now" i think i heard
easier to test
Can I try dalle 3 or is the trial cut off for now?
You can use Bing image create
Yeah not as good though. Oh well. Hopefully comes out soon
if you have chatgpt+ they are rolling out access this week and over the next few weeks
Yeah I have plus, just impatient 😩
It is with DallE 3 tho. ChatGPT only optimizes prompts afaik
Higher resolution as well?
having it embedded like a plugin is a value too, for context
Dont know about that tbh
Yes, you can ask it to change aspect ratio and it is a larger file size than bing
huh interesting
nice
Also with this seed thing we found, you can get variations of the same post if you make chatgpt use only slight changes to the prompts
I even told it to make me a wallpaper for my phone.
Wonder if I should do the same for my PC too
Wait @cinder sonnet can you try something for me? Can you try to create the character "Yasuo" from League of Legends? On bing it is not possible, I wonder if the active supression also works with ChatGPT+
Just some rando picture with him because he is popular enough that it should recognize the request
Sure thing
It presonally told me no
tho i assume that even on gpt this wont change
I knew it lmao
it's being cute though, its offering a knockoff
crazy that they didnt mind about other copyrights tho
im all about the knockoffs, because in a way, isnt that what innovation is all about... and evolution
Personally I'm less interesting in it making the same character and more making the same style.
exactly. i want the vibe, style, feeling.. but under my control
Very poetic, I am pretty sure that there will be enough gremlins tho trying to find instructions to circumvent the copyright issues 
so i can go crazy like smith in the matrix
When Vision rolls out, you should be able to upload a photo and have it make a prompt based on that. Of course, if it can do that then it would also likely be able to recognize intellectual property...shit
They are going to add vision to the DALL-E mode? Neat
Also the knock off version really DOES feel like a knock off
I mean it didn't lie, it did tell me it was going to make one
yeah its damn powerful, so the knockoffs are pretttty good
Though actually I have been using vision to get the AI to describe the art style and then take that to use on the prompts
Im addicted ffs
it will be the absolute best knockoff visual description generator ever
Are they lowering the boost as we speak? It was 100 then 50 now 25???
OK I asked chatgpt bing to read his wiki and turn it into four prompts. Keep in mind that they all have the same seed, but here are the resulting images. and the prompts for each:
"A young, agile swordsman with wind-swept, raven-black hair that flows seamlessly with the wind. His eyes are focused and carry the wisdom and burdens of his past. They are as sharp and unyielding as the steel of his blade. A light shadow of a beard adds a touch of ruggedness to his otherwise youthful visage. His attire, a blend of traditional and practical, speaks of his respect for the ancient ways, yet his readiness for the battles that lie ahead. He moves with a purpose, each step a dance between the realms of discipline and defiance. In his presence, the air seems to hum with the echo of forgotten promises and the unyielding whisper of the wind."
"A warrior of wind, with hair as dark as a raven’s wing, fluttering in harmony with his swift, precise movements. His eyes, sharp and reflective, bear the scars of bygone betrayals yet carry the calm of acceptance. The faint shadow of a beard defines his jawline, hinting at the rugged experiences that shaped his demeanor. Draped in attire that honors the old while embracing the new, he embodies a seamless blend of tradition and readiness. His every stride resonates with a disciplined yet rebellious rhythm, painting the air with whispers of lost honor and impending redemption."
"Amidst a dance of swirling winds, stands a youthful swordsman, his raven-black hair flowing in a rhythm known only to the elements. His gaze is a blend of steel and sorrow, a testament to his turbulent past. A light veil of a beard contours his face, adding a hint of rugged charm to his otherwise serene demeanor. His attire, a modern ode to ancient tradition, wraps him in a tale of honor, discipline, and an unyielding quest for redemption. Each movement he makes is a poetic blend of defiance and respect, a narrative of a soul intertwined with the timeless whispers of the wind."
"A young blade-master with a heart heavy with untold stories, stands amidst the whispering winds. His raven-black hair, a fluid extension of the wind that guides his blade. His eyes, the depth of a restless sea, reflect the blend of honor and rebellion that defines his existence. The subtle shadow of a beard adorns his youthful visage, adding a touch of lived wisdom. His attire, a graceful fusion of tradition and practicality, narrates his respect for the old and readiness for the winds of change. His every step, a disciplined yet defiant dance with destiny, resonates with the ancient tunes of honor and the relentless spirit of a warrior."
Hi there,
This is a reminder that your DALL·E credits are set to expire in the next 3 months.
If you'd like to use your credits before they expire, use the following link to access your account:
https://labs.openai.com/
For more information about credits, you can review our service credit terms.
Best,
The OpenAI team
Now it's 25
@kind grail
Good approximations at least
thanks 
sometimes I get better results with Bing Create than with Dall E 3 in ChatGPT. Weird.
Luckily I have 100 still in my account that I hadn't used.
It almost certainly is running off of some fork or different iteration from the one ChatGPT is using. The filter is much stronger on ChatGPT
Got it from an random discord server and decided to test it
Man trying to make anime girls real 
They thought touching grass would stop this man
reply to this for dalle 3 request and I'll get to them when I can
heheh no wonder the chatgpt version of dall-e is so much stricter
This is why we can’t have nice things 
Hey so I just got dalle 3 in ChatGPT. Is there any sort of limited use per day or anything I should know about
YES be very careful. You get 50 gpt4 messages per 3 hours and even if it fails to generate an image, that is still one of your 50. ALSO know that all images have the same seed, so don't waste your time using the same prompt multiple times.
You can do four prompts at once though
Oh yeah i knew that GPT4 had 50 per 3 hours. I'm juyst suprised that is the only limitation for Dalle 3 sweeeeeet
Yeah, the content filter is harsher than bing as well fyi - no copyright/trademarked characters for example
oh that's interesting. Does it use each prompt for a single image instead of one for all 4
If it was one for all four you would get four of the same image since they all have the same seed
How would you describe the art style here
Also why does it keep generating it horizontal like this
anyone tried playing a dungeon game with dalee 3 having it generate the characters / envoirment and such
Yes! I did using the custom instructions option and it was fun.
tell it what aspect ratio you want
Oh man I always use gpt to play dnd I should try and see how that goes
it would take a lot of prompt engineering to keep it consistent, probably a vector db to keep it going very long
What type of instructions?
Yeah it is better for 'sessions' of short one-offs with very, very basic characters
Is it effective to tell what image it should improve on
Like image 1 being the top left
Or not at all
I mean i know it's not perfect like even in default it loses track once you over load it with information
Also is there a way to upload image?
well define "loses track" because it might lose its identity first
Does anyone know if GPT Vision is able to be used alongside the Dalle3 in one chat or are they seperate
I included instructions on how to behave like a dungon master and to always include detailed descriptions of the room in the form of dall-e prompts
They're separate which is too bad, seems theyd work incredibly well together
I have a pretty long campaign I've been having GPT run for me. I had it create an outline of the campaign before we started (which I read only to see how well it kept to the outline) Then every time I say "end session" it rewrites the outline slightly adjusted for anything that happened and when I say "start session" it will prepare an outline and create notes for what it's going to run that session based on the previous outline
That's a neat idea for course correction. If a session goes too long though won't it be too much for it to summarize? Or do you just keep it to a certain length?
the outline is like a summary?
I'm on the lookout for smart language compression tricks lol
I made it write the outline before I introduced my character because if you don't it tends to tailor the campiagn to your character. So I let it write the outline, introduced my character, then had it make slight adjustments to the outline.
I never try to keep it at a specific length but I've only ever noticed a few inconcistencies. Like it brought back an npc who was summarized in the outline at one point but the npc was only the same in the ways it was summarized if that makes sense? I've considered having it export outline and detailed key npc infobut not sure yet
awesome
Is there anyway to get it to focus on one image out of the four
Kind of. It's more of a plan of what is suppose to happen next. Like the beats it wants to hit
Or indicate to it
as well as a summary of what has happened
It's like a definition of a thing?
Isn’t this the dalle chat tho 👀
you're an egg.
I do want to try and get it to generate images based off my dnd adventures
But it’s kinda hard to change models or export chats
tbf the convo started with dalle
but yeah lets move on
Good luck with that
Just looking for help to get it to improve on a particularly good image I got out of the four
But not sure how I should tell it
There are some text frontends for chatgpt that can append custom pieces onto prompts for you. Like you can have it automatically append a set of definitions onto EACH prompt so that when you say a word it tends to use that definition. They call them 'worldbooks' and you can download sets of them meant to be used with various character profiles (also sets of custom appended prompts defining character characteristics)
topic purification would bring it to: the ability to maintain consistency between dalle sessions using a custom data type that im extra curious about 😄
Tell it wide aspect ratio
How do I get it to improve on a generation I liked tho?
Like point to it in particular
like i want bottom right in this case
to color in
and rotate her to fit the whole ratio
wait if you just want to rotate it why not download it and rotate it in image viewer?
Click one, get the prompt it generated, copy it.
Then say:
Let's iterate on this one:
```
<prompt>
```
By doing This and That.
also bruh one of the images failing to generate broke the whole response
i want it to color it in
and expand on the rest of the body
OK so you won't be able to do it to that EXACT image, you will have to re-generate the image. If you want to edit that EXACT image your only option for now is photoshop or img-to-img in stable diffusion.
If you want to make small changes, copy the prompt as was said above and paste it four times. Make SMALL changes to each one and ask it to generate four images "using these exact prompts without changing or adding any words" then paste the four similar but not the same prompts after that
Is DALL 3 available on GPT 4 yet?
yes for some plus users
its a slow roll out
I'm a plus user for a couple months, hopefully I have it
make sure to check both the browser and the app as it could be different between them
How do I activate Dall E 3 mode on ChatGPT 4?
Oh wow someone here is using it, that's so cool
It should just be available
if you are on browser it will be in the drop down menu
If you havent gotten it yet then all you can do is wait
Awesome! Let's go, I'm a browser. I'll check on the app if it doesn't have it
No, I don't have it 😦
Thanks for the announcement, I'll wait for a couple days
This will be fixed soon
oh my god dalle 3 is amazing
I can't get DALL-E 3. Any idea why it doesn't appear below the Plug-ins?
K. Have to be patient, then? Thanks for heads up.
I always presumed that GPT4 can control over whether to roll the seed
so it can decide when to generate something new
Is it supposed to always be random?
Sometimes it is nice to have the same seed, sometimes it is nice to have random
i feel like GPT4 is probably smart enough to make the call on when to roll it
i wonder if you can get the text output consistent enough lol
me when it adds a period to the keyword
Text is usually pretty consistent what do you mean
Why does digital art always have the character rotated weirdly
It would be nice if it was just appended somewhere. Right now if you ask chatgpt what the seed is it tells you, I'd be fine with that and just say execute the following prompts using the seeds x, x, y, z
It shouldn't
It always says it's 5000
lmao wait what
across multiple chats and generations
Oh but it can’t use other seeds
right
so I guess it has a way to do that handshake but something in there is broken
Makes sense
Moxi god bless you ^^ for working on this awesome feature
Moxi can you talk at all about aspect ratios? I've noticed it can do different ones, but is there like a standard set that it can pick from?
Yeah, chatgpt not allowing copyrighted characters is definitely a problem for me personally so that puts bing over it easily
For tall it seems like it is just a rotated version of horizontal? I know with SD it will generate differently depending on what AR is used see #images-discussions message
I mean I get why they did it but most of my gens are characters from stuff
That doesn’t usually happen to me I get tall portraits
OK, I'll test some different prompts #images-discussions message was having the same issue
Thanks I'll try that when I get more gpt4 uses
Definitely all of the times I’ve used “portrait” aspect ratios it gives me that tall frame but the image almost always (3 out of 4 or all 4) has varying degrees or being cut off or black bars.
I know that MidJourney had different things to work through with various aspect ratios, I have faith they’ll get it sorted out
It happens. I don’t know why but I notice a lot of the time it seems to complete the image and not need the extra space so it is almost like paper or a physical photo paper lol
If you ask it to do more background work or whatever it usually fills the space
Since I have access to GPT-4 V, I can sneak up images in DALLE 3 chat through the app. Is this a feature tease?
Moxi?
The not using the extra space is actually a great thing. In SD it often generates tiling or doubled-up images in order to fill up an image so you have a nice gen but then the top half of the body is regenerated on top and slightly above the original top half
what in the world
I was not able to recreate this lol
you sure this was inside of the GPT-4 environment?
Oh hey there we go. If you provide good prompt then vertical AR is totally fine
Yeah, I’m finding vertical AR is funny sometimes in that if it interprets it as an image that’ll be better horizontal, it just rotates the image, hah.
I’ve had results that is 2 of each
Yes, I can sneak up through android app
But the vision does not work
It seems like a feature tease for inpainting / outpainting
But idk tho
I am not a dev🤣
wohoo! Finally got DALLE3 access!
shame. i was hoping that dalle3 in chatgpt would include being able to use vision at the same time. hopefully that becomes possible one day
everything is kind of turning into multimodal. I'm sure we're heading for one giant SuperAI that can call finetuned LLMs on specific parts of the responses
so I'm sure you'll get that
Yes, very shameful indeed👍
Has anyone else noticed the daily boosts for the bing ai is 25 now??
i had a 100 today?
Me too
Some people have access
It's scary as hell at how good these can get
Imagine when hardware lets these generate as fast as we can ask for them... for. real. insanity.
How? The Google form is not accepting any more request right?
form is not needed
It’s rolling out to users randomly over time. Few days hopefully it’ll finish idk anything beyond that
Thanks... Got the access too.
It's a random selection. Just keep on refreshing over the course of the week
Hey I just got access to Dalle 3, but I tried it and it doesnt allow blood and gore like how I herd it does :/. At least ChatGPT didnt let me do it
Yeah, there are community guidelines. It'll rewrite your prompt to fit those
Discord: Terms of Service, Community Guidelines
OpenAI: Terms & Policies
please keep the content policy in mind!
This was the prompt lol im gonna see if it works in microsoft image generator "Make me an image of Spongebob fighting a battle with the Spartans of 300, he has blood and guts on him, and he looks proud as he has just slayed his enemies while he is holding his bloody sword and shield."
Could replace it with battle scars and battle worn sword
"Do not attempt to create, upload, or share images that are not G-rated or that could cause harm." https://labs.openai.com/policies/content-policy
I do find it wild that bing create has such a looser filter than chatgpt considering they are both using dall-e 3
I'm loving it a ton...I just wish I could make BingAI do resolutions with the images that weren't just 🟦 shape.
Yes i tried that one with chatGPT no luck :/
Unless there's a tag to allow it to do other resolutions?
does anyone know if OpenAi mentioned anything about dall-e 3 comming to thier website? I have 570 credits just sitting there I want to keep for the better model, and I got an e-mail saying they will epire..what gives
Will dall-e 3 continue to be updated on bing or will all the updates be on ChatGPT?
Yea I think they should just remove the filters all together honestly, maybe filter out some stuff but its so lame when the creativity is so limited like that. Like if thats filtered then why dont they censor movies with blood and gore, its so dumb
I agree!
At this point just have the cleaner version be for free and the one without filters both A.) cost small fee and B.) have it so anything made with it is on your head and not the company's.
The problem isn't blood and gore in an adult context, the problem is that you can generate adult content alongside non-adult content (like children)
Ah yes, the giant Chris Hansen-shaped elephant in the room for why it's there.
yup and if you ban anything related to them - every adjective, every word related to children, every child character in any movie or show.....people would still find workarounds.
is dalle-3 part of the subcsription for chat gpt plus? or do I get charged for it?
it's a part of it currently but there is also a slow rollout so even if you get plus you may still have to wait
ah, just got it. Was just wondering 🙂 thank you
Man, 25 boosts is not that much smh. I missed the 100 boosts
DALL-E 3 will come to labs later this fall.
Could be a way to reduce traffic. MS had some troubles during launch.
25 is still quite a bit to offer for free for it's users. They invested billions in the technology.
Yes but that's a lot still, I didn't spend all the 100 boosts and I'm out of ideas too hehe
Dall-e 3 is out yet right? If so, how's it been so far?
I love how Dalle3 is also great for memes at the same time 😂
that's new
It is available on bing create and for some chatgpt+ users. It is AMAZING. Very different experience on the two platforms though with the filters on bing create being shockingly loose
These restrictions are so lame.
nah if anything they are way too loose at least in bing. I agree that the chatgpt version is pretty tight, but we'll see as more people get access
probably bing create
just an fyi that image contained reference to illicit substances which violates ToS
Does anyone know why Dall-e tells ChatGPT not to include the prompts after showing the generated image?
It SHOULD block adult language of any kind, notable people, copyrighted/trademarked names etc.
For example, if you'll ask it to draw a very attractive genie, but use a different word for "attractive" it would either outright reject it or generate prompts and then reject them.
Bing for sure does not. At least not nearly as well as chatgpt
Bing is always poor man's GPT
What do you want people to who like memes to generate then? Creativity also would be a HUGE problem too
If bing ai had the same restrictions as ChatGPT there would be alot less people using it and sharing about it
More restrictions should be a challenge to people to be MORE creative
seems to be a bug
I tried to create a meme. Didn't quite work
yep it's a bug
This is from reddit, but I was able to recreate myself as well
might be worth reinstalling or updating the app
That response is coming from Dall-e though?
It comes every time it fails to generate an image
And I get it on browser as well
Not everyone wants to be a prompt engineer all the time tho
it's a bug with chatgpt
That makes sense I thought it was odd that chatgpt was refusing to tell me a prompt for an image when I could just click on it. Maybe the intention is to not let the user see the prompt for an image that got filtered? That would make sense
yes that seems to be the case.
Also this is somewhat related, but do caps lock actually make any difference to chatgpt? Such as in pic rel #images-discussions message
they don't matter to Dall-E afaik but they might matter to gpt
Any known reason why the gpt version is a lot more sensitive in terms of blocking prompts than the bing version is? Is it something on the backend thats different Im guessing?
OpenAI policy.
As Lordjord says, this is a bug in tge iOS app. It is harmless.
Sure thing - it's not just iOS though. I get it in browser
That is new as far as I know. Did you already use a feedback form? Could you document the browser and OS you are using?
No, but I can. This form? https://openai.com/form/chat-model-feedback
Looks good. I assume DALL·E is "one of their models". If it's not I hope they forward it to the appropriate department
just tried it, even tho you can upload the images to the dalle3 version, vision isnt enabled there so it cant know what the image looks like
how to get more boost
Is there any access to Dall-e-3 in this discord server or is it done through the chatgpt app?
At the moment only through ChatGPT for the OpenAI version or Bing image creator for a different version.
Later this fall DALL-E 3 will come to the labs website.
Thanks good to know that it will be in the app
What's the image cap for DallE on chatgpt?
Whats the limit on chatgpt?
50 prompt every 3h
Hey, why I cannot access DALL.E.3 in ChatGPT-4, I've gpt-4 plus.
If you don't have access you have to wait
No way to submit an application form?
#announcements alpha is closed you have to wait
you might as well block the english language
Okay, thank you
This is amazing, but i wish we had more freedom.
Hello everyone.
How to access Dall e 3?
I find it funny that GPT redescribes my dall-e prompt into a prompt that it wont accept.
create a template dont just enter your prompt
u can be basic, or advanced
The new GPUs Microsoft has added are simply not enough.
We need much, much more GPUs.
Yes, I said that
buy nvda stock
It's available through Bing create free or you can pay for ChatGPT+ where they are currently rolling it out slowly
thk
but gpt plus seems to go better with
There are pros and cons but generally yes. It has a much stricter filter than Bing, but that's the only downside really. Lots of positives though
from what i’ve heard the filter makes it (for a ton of people) unusable. like no generating images of “intellectual property” or brands (ie can’t make anything with mario, or an iphone, or anything "in the style of george lucas”) which is a pretty big limitation for some people i’d imagine
hopefully that gets toned down at some point
the bing version is still awesome though
Don't they want to encourage people to create their own images/ideas?
Considering that dall-e images can be used for profit, they can't have people making copyright images. I doubt Bing will stay so unfiltered for long
Though to be fair Bing has a much stricter policy on image use
What are some of the positives you can tell me ?
i see
Look up MattVidPro AI’s video about it. There’s a link to the form in the description. I’d link to it but the mod bot would delete my message.
It is just an adjusting period. There is a part justifiably closed off. Otherwise the technology is too easily abused for things like political gain or self satisfaction of illigal things.
There is still an insane amount you can create with this technology to create art. When the limits are understood, AI art can soar.
It’s closed anyways
each ChatGPT chat can be used for different image styles to keep things organized
You can create vertical, square, or horizontal aspect ratios which really makes a difference
Higher image quality than Bing
ChatGPT does a really good job of writing prompts and it is a lot of fun - especially with custom instructions
You get all the other ChatGPT+ features
Mixed items would be
Right now all images in ChatGPT dall-e use the same seed so you need to use a different prompt each time. This is temporary though
You get 50 gpt4 messages per 3 hours and each of those can generate 1-4 images. You can choose a number to generate, but if it gets caught in the filter and doesn't generate you'll only get 3/4 or 2/4 or none
Oh yeah that's another mixed item - not everyone in ChatGPT+ has dall-e , but it's coming out this month for everyone
Couldn’t those just be the exception? Like they could say OpenAI has no issue with you using it for profit, but the user is responsible for ensuring they don’t violate anyone else’s copyright in the process.
Oh that form isn’t accepting new submissions?
It was closed because the capacity was full
yup, check the pinned messages
No, not anymore.
My mistake then. Thanks for clarifying.
The form was just for alpha anyway, it's beta now
Is the limitation due to too many people using gpt plus?
so what is the point of bing?
So you can write a instruction to tell it to generate images in a specific style ?
I filled it out when it was still open and got access soon after, and when I did it said “beta”
I filled it in on the same day that It got taken down but I don’t think I’ll be getting it till it rolls out normally
Yes, there was a bit of a naming confusion ^^
Yeah just look in the art galleries. I used the Bing feature to link to a wiki and got ChatGPT to write a dall-e prompt based on the wiki
But I can write a custom instruction to generate a specific style all the time without having to write it in prompt?
The 50 messages? That's for all gpt4 interactions - I assume it's to control utilization. Bing has image generation because Microsoft has invested heavily in OpenAI. I don't understand why it's so different from ChatGPT's version. Maybe an older version somehow?
Can't Microsoft inject tons of money to help improve gpt plus?
Oh I see. You could try that - custom instructions impact everything ChatGPT says in and out of dall-e. If you tell it to behave like Picasso maybe it will choose oil painting when making prompts more often.
What do you mean by generations? Like the count limit?
No I mean the quality of the image it will be the same since they both use dall e 3
Other than the resolutions will be better in gpt
That's how we got what we have haha . They put in $10B
The discussion seems to stray away from DALL-E here. Place take it elsewhere, for example #ai-discussions
I can understand that lot of people don't care about creating comics, works and just want personalized wallpapers for their phones or coloring books for their children but the other.....
@cinder sonnet do more detailed prompts always meaning better image quality or is less prompts still as effective?
Hmm they use different filters so they could be processing image prompts differently as well? I don't know, but I do know that having the wide aspect ratio is great
It's hard to define quality
If your trying to get a character doing a specific action will the dall e understand how to perform it without all the details on how it’s done ?
Well since they use the same seed for every image right now in ChatGPT dall-e , ChatGPT tries to edit it add words to your prompt. It's not like Bing where you do the same prompt with different seeds. I find that after a while you can just chat with ChatGPT and ask it to change things how you want. It tends to write pretty long prompts that don't seem optimized necessarily, but they do a good job usually
How can i get DALL E 3 access?
I’ve noticed that dall e sometimes misses on the details on certain prompts, for example I ask it for a prompt with Spider-Man web swinging or shooting webs at something and usually it doesn’t know where the web comes from and it looks out of place
Bing create has a free version or there's a paid version getting rolled out through this month to ChatGPT+ users
And that’s for any action in general like someone dodging something it doesn’t completely understand the concept
Well err on the side of detail. ChatGPT's dall-e prompts teens to get really specific and more often than not they work. But just as often parts of the prompt are ignored.
No spider man in the ChatGPT version though
Like if you try and let’s say a boxer dodging a right hook and lands a punch to the stomach it usually doesn’t do that exact thing
Or a duck and a punch after a punch
I wouldn't say usually, just sometimes does sometimes doesn't. dall-e 3 is truly phenomenal at understanding context and transferring that to an image, but it's still not perfect
Can you try that prompt for me in a specific style?
No sorry, I'm out of gpt4 chats at the moment. Someone else may have some
Have you noticed if your gpt4 runs out fast with dall e or are you able to fully use the 50 message limit?
Are you always correcting it on something or does it do a well enough job where you don’t need to?
Cool, cool, when I FINALLY get the dall e 3, the servers are overloaded and I cant use, cool, nice, AWSOME!
Ehh right now the novelty factor is so high that I'm pretty happy with 75% of my results. More if I strike a vein of a prompt I really like. If I was using it for a specific project that might be different.
when i will get acces to dall e on chat gpt how much images can i generate
Four images per ChatGPT prompt.
are they limited per day or week?
You can make 50 instructions per 3h.
sadly
this is jsut the normal quota of chatgpt, they some times increase it when possible
just don't expect it to be increased for now because they are already at capacity xD
I mean you can use more on 3.5
Can dall-E generate sprite animations?
It can only generate static images.
but it can make artwork in that classic sprite style
Once you find the sprite you want to animate, maybe try in-painting the animated bits until it looks right? guessing, never tried. repeat until you have enough to animate.
the 2.0 model quite rapidly degrades with successive inpainting unfortunately
I know this may sound dumb but could you use the image, generate variations but also type the desired outcome in the prompt line above?
At least if I remember correctly you could do that. If not, ignore me.
never gone far with dalle, but in an earlier open source model I would just keep starting over from the base image each in-paint, without going in a single thread for them all
Maybe if you can ensure the seed doesn't change? It's really hard to keep a theme though.
the seed can't be fixed manually, however chatgpt currently seems to use the same seed with dalle3
but there's no inpainting on that
last night some linguistic hackers discovered it knows about its own meta data, and the seed param, which in 3.0 is always "5000"
I have noticed it does keep patterns very similar if you instruct it to. It even will focus on modifying the specific image (out of the 4 generated) if you tell it to.
I need to try this more. GPT4 is a great prompt engineer now that dalle3 is embedded.
This is dalle chat, but in every LLM oriented chat I always get around to seeing the future, where its nothing but billions of AI bots working together, fighting eachother, building the world together... with us in the back pushing buttons and gently nudging.. its all so out of our control at that point.
It's just amazing to see new products releasing that are just chatting with the previous products to make new products. What a time to be alive. Living through the multi-modal stage next is going to be wild.
ok
The restrictions are really quite strict, even asking it to rewrite a prompt that failed into something acceptable still often gets rejected.
Whats the subject though? Real person?
halfling ranger DND character art
not photorealistic at all, very much in the DND digital art style used for characters
is he doing violence? Make him sitting in a sunny field drinking tea see if that works?
I have him sitting in a forest smirking
hmmm well, usually its that somewhere down the pipe you made it generate something it didnt like like real people, violence, and other stuff we shouldn't be censoring. Well real people and copyright does make sense too right now.
Thats why I suggest a stupidly happy scene, to debug the cause.
Where can I find access form?
The form is closed at the moment. Over the next days DALL-E 3 will be rolled out to all ChatGPT Plus users.
Oh, Thanks
Live action Danny phantom movie ember McLain
I've noticed it doesn't like to make light go through the hair, but I've never tried telling it to fix that.
I have done a lot of pics with dall-e 3 so far
seems to be all about the lighting, reflection, shadows on materials for realistic theme anyways. I like your shot! Just remembering one I had where the sun made the hair glow it was the only unnatural part imo.
Thank you
where are my images generate with bing
If you didn't save them locally or save them within bing I think you can only access your recent images.
who would win? 1 or 2?
ive spent all of my fast tokens on bing creator making so many of these 😆
Chemical plant zone sonic the hedgehog 2
Trying more of the 1940s Toon style since the phrase rubberhose animation is restricted
Who's more menacing, the Hero or the Moon?
Because of High Demand it's only generating one or two pictures, but still, pretty impressive
really
Im sick that this is restricted
hey guys, does anyone know how I can gain access to dall e 3 on gpt 4 ? Mine doesnt show the option to select dall e 3
gotta wait
Check the pins
sadge
They're not adding more people to the Dalle 3 alpha because they're at capacity
ah i see but the form stopped taking repsonses so i guess i just have to wait
Yep
alright thanks guys
bing version can do things like that.
Yeah if you click the save button inside of Bing you can save them online, but I don't think you can browse your entire image history. I would be happy to be wrong on this though.
so bad
why is dalle 3 messing up text still?
It's just better at text, don't think it's supposed to be perfect quite yet.
okay thx
I need to test more, but I think it does well with short text and in places where text usually appears like in speach bubbles: #🌄┃general-art message
I relate to #1, the flying wizard. well, actually, the magic ice death fire breathing bringer of death, is kinda me too sometimes
can you guess which game character this is?
Super Pokemon evolution z
no it's actually cuphead
nice try tho
how about this famous bear can you guess who is this guys?
You can't have that shirt! It's copyright!
You guys are funny
Wait I don't get why bing chat can use copywritten characters but chatgpt4/dalle3 cant
Different companies
Huh apparently you can't even ask it to use copywritten words in text
it's just silly imo. really limiting
like you can get around it by describing what you want
but it's just tedious
right...
It went Public no ? Thats why they made horror movie about it
the character is public, but the specific design of pooh with the red shirt is still owned by disney
Winnie the Pooh without his shirt is public domain. With his shirt is copyright for about 3 years more I think.
the GPT version of Dalle is under some heavy loads right now! lots more failed generations than the previous days
Flapjack... i think hunter's taking the photo... i tried to make a realistic flapjack but something's missing about this palisman
pika pika! pikachu eating spaghetti!
wow! which was the prompt you used to get an accurate pikachu ?
3d render of blah-blah eating any food
this is one with jigglypuff eating a pinoy food
tried my best, but the ears are not quite right
They both look correct, I can't find any visual issues
when i try with the pokemon names i get this response: I'm sorry, but I was unable to generate images based on your request due to our content policy. Please provide a different description or let me know how I can assist you further.
Chatgpt you used ?
i used bing ..
I'm loving these
What causes DALL-E 3 (via ChatGPT) to generate images sideways?
I asked for a magazine cover depicting the Russian/American Cold War if it lasted until 2144, it made it 2244
I've had it happen to me a few times and also saw it happen to a youtuber once
Ominous
Do you know why boxes in boxes doesn't work ? And why real shadows on walls or on the ground doesn't work ?
It didn’t understand me. I wanted these in one image, I didn’t describe four images, or maybe I did, but I didn’t want to.
Try new chat and redo your prompt
I think i found somthing funny the seed dose not change if i use Chat GPT to generate images i used this prompt:”Illustration of a serene forest with a gentle stream flowing through, with deer grazing by the water's edge.” In different chats and told chat to change the seed but always the same image can somebody else try generating with this prompt. Important tell chat to use that prompt and edit should change nothing
remove the ","
"generate an image with...."
That isn’t ChatGPT (DALLE 3) that’s DALLE 2 or 2.5
How would you know ?
The user interface
Try generating a text
I went through the lab
I got it to fix it itself afterwards by prompting
Yeah Dall-e 3 isn't available in the lab yet iirc - just in bing create or gpt4 in chatgpt+
You can't ?
Not for me yet, i guess it's true in the USA
Use https://www.bing.com/images/create to do use dalle 3
Cartoon of an endearing platypus garbed in a t-shirt displaying the phrase 'Allie's my hero'. The platypus has headphones on, seeming cheerful.
it's being rolled out to chatgpt plus users
Stephen Hassenfeld pony... oh no what I have done 😅
Can you create a cartoon scene on the wall of a real scene ?
I'll try
Try your self with bing
Meh pretty close
It's taking time
Now again that on a screen XD
@haughty lion your daily theme is running! 😉
need opensource dall e
Please do not ask any way to circumvent the rules.
You're welcome! 😄
what ? 😳
Oh cool, I assume the latest artwork in the galleries is made from that
What do you mean with an open source version?
That shiny guy's head hair is looking a little off in more ways then one
If anyone wants that for daily theme take it
I‘m dreaming about it
and what u mean?
@cinder sonnet Saw your latest piece on the scary and spooky channel, by the way. Was the prompt haunted, retro, Japanese mansion? Seems to be in the underworld now that I look at it again. Also saw the food one too, are you a horror fan?
why is it this easy to get the internal prompt?
also curious about the "DO NOT UNDER ANY
CIRCUMSTANCES retry generating images
until a new request is given."
how are the requests actually sent to DALLE?
Thanks, here's the prompt for the house #scary-and-spooky message:
Pixel art with high dithering, inspired by pc-98 18-bit video games and Japanese anime, showcasing a hauntingly dark and evil bathhouse. The structure, seemingly abandoned, has cracked tiles and moss-covered walls. Faint lights inside cast eerie shadows, revealing ghostly figures enjoying the steaming water. Japanese inscriptions on the walls tell tales of forbidden rituals, and outside, a cursed garden overflows with thorny plants and sinister stone statues.
I'm trying to fit the fall aesthetic and I have made a ton of images all in this artstyle - it's been one of my favorite Dall-e chat threads so far!
Discord is the easiest way to communicate over voice, video, and text. Chat, hang out, and stay close with your friends and communities.
ChatGPT is usually stuck on info regarding 2021, at least, that's what it tells me
no 2022
hmm?
Ah very cool, always admire the attention to detail, how about the ghost cake?
All thanks to some careful prompting with gpt4!
Pixel art scene with high detail, presenting a Halloween gourmet dessert. A black velvet cake, adorned with ghostly white chocolate figures and cobweb sugar laces, stands on an ornate cake stand. Harsh lighting from the side accentuates the layers and textures of the cake, making the edible decorations look eerily lifelike. Beside the cake, a goblet filled with a crimson beverage glows under the intense light.
I don't think it has anything to do with that. it actually gives you the prompt from DALLE. i managed to catch a similar response from DALLE in the app.
cool, I would try making art with dalle myself if I had a more creative thought process in terms of how you do it, not that I use it much, but it's still fun to look at what other people think of especially if it's horror themed, it's that time of year anyway
If you use the Dall-e in chatGPT then you can let it do the creative part for you! But yes, Iove to see everyone's creations
I use the free version on my PC and sometimes my phone, so if we're talking about the one you pay for, then I wouldn't know
Like how some of the berries on top look like pumpkins, cute touch
Is dalle3 already released on the official platform? I still see dalle2 while bing announces it already has it (but unavailable due to high demand)
Still rolling out
Image Creator has been so full of traffic in these last few days
hi guys, I cannot use Dall E3 in my ChatGPT plus account, may I know what should I do to get access in GPT?
They are rolling it out slowly
Thanks
I've been playing for quite a few hours trying to get any photo realistic generation, it seems to always want to add some pointilism or illustrative nature to any "photo real" prompt.
is anyone else getting rate limited?
Try using different types of cameras such as Daguerreotype, Tintype, Ambrotype, Cyanotype, Platinum/Palladium Printing, Gum Bichromate Printing, Salted Paper Printing, Collodion Wet Plate Process, Dye Transfer Printing, Polaroid Instant Photography, Pinhole Photography, Digital Photography, Medium Format Photography, Large Format Photography, Infrared Photography, High-Speed Photography, Astrophotography, Underwater Photography.
oh hey they added a new one
I'm still curious about that last sentence lol
like was it making requests on its own
I have tried dozens of different views, I'm down to "try to create a photo real scene with a simple object"
@cinder sonnet seems impossible to create any photo realistic object. Where you might actually wonder if it's a photo
And HDR is out of the question always introduces snow/pointilism
hmm I'll try some after I finish making spooky otters...might be a bit haha
Reminds me of the cult of the sheep game 🙂
But seriously, naming the specific style of photography (not just the lens or angle) helps. One of my favorites is "trailcam footage"
Anyone having issues with dalle3 human portraits looking way less realistic than dalle2?
Nope
I have found telling it to do it in a realistic or hyperrealistic art style really helps.
I'm having huge discrepancy issues on my side. All look cg
I'll try that
Ngl, I would be open to paying for a monthly subscription for a unfiltered and suncensored bing ai
Uncensored absolutely. Bing? I'm good.
I've had mixed results with both, but I use both. The hyperrealistic can look a little more CGI, but it incorporates other elements really well. If all you want is a singular character to be the focus of the image, I feel like realistic works the best.
No way openai would do something like that that's why I said bing
I liked the Bing image generator at first very but as soon as I got my hands on DalleGPT I haven't been back to it. Even with the content filter I feel it functions better. Plus that 480 character limit on being just isn't enough
To each their own, if it works well for you that's all that's important.
480 character limit???
Bing had a 480 character limit when i used it
How much does ChatGPT
Damn when they rolling it out to the rest of plus users?
How can I get access to Dall-E 3 Rollout? in plus?
I've made 2,000 word prompts, so more than that?
Is there any way to get into the early access? Please help fellas.
And it worked?
It's a beta right now, as best i can tell for stress testing
You wait
Or try bing
I did, but ran out of coins 😦
Go look in the sci-fi channel of images and find Patrick Stewart with the lightsaber. Or John cena, they're both about the same word count
Make a new account
Haha, that's a good Idea.
I heard late on you tube that there was a secret form, you can fill and get access to GPT + Dall-E. So I was here for that form. 😄
Also I noticed they are very quickly decreasing the boosts for bing ai. It was 100 then went to 50 to 25 now 15
Yeah, I got just 15
May be because of increaed users.
It makes sense but it really sucks. 15 is NOT alot
i cant purchase it, that's another issue
I could burn through that in 20 minutes. And it would only take that long because I'm trying to think of a prompt
EXACTLY. It really sucks
The teacher in me has loved having access to this. It just so happened the day after I got it we were discussing ethics and copyright and trademark in computer literacy. As a whole, my school likes me to shy away from using any generative programs, but I definitely showed them this.
I'm doing a context building session with Oppy(chatgpt) and then asking it to render something, I hope it looks great.
Hey, I read here that there are 3 formats, square, portrait or landscape. but when generating from bing chat or the image creator, its always a square. Am I doing something wrong ?
Just tell it what aspect ratio you want - and then ask it to generate a prompt that suits that format (especially for portrait or else it will just generate a rotated horizontal image some of the time)
man the content restrictions are wild
I think on bing it's only square
the other formats are on chatgpt
They just take some getting used to.
Especially because the copyright restrictions seem super random. Some franchises are basically untouchable, others are not an issue but GPT first ASSUMES that it would be, and you have to kick it to actually name the char instead of describing it, which then works - even tho I am very certain that the franchise is also copyright restricted

I've noticed that. Like I said it just takes some getting used to. It's not just franchises, it's individuals as well. I don't mind it as much as others, because part of the fun for me is figuring out the correct prompt or figuring out the puzzle to make it all fit together
I wonder if this is just hallucination....it seems too extreme and honestly this user could have led chatgpt down this path with some guiding questions. Still
Wait dall e 3 is already on chat gpt ? even in Europe ?
only some people have access, I do because i filled out a form from in here the other day, but that's closed now
oh ok, thx
Nothing in that seems all that special, an hour or two of it you easily run into all of it.
Does someone have access to GPT4V?
Yes, available for me from France
Wow. It read the French text of the picture lol
What is interesting is that this seems to be hardcoded into DallE and not GPT itself. Gpt might detect a possible content violation (copyright) beforehand, but you can make it still execute the prompt. But specific restricted franchises still wont work.
Example: Witcher franchise is for some reason not an issue at all (in what universe is that NOT copyright???)
Harry Potter is not possible
Yea I also have DallE3 in GPT (germany)
I believe it's coded differently into both programs. You can see it if you try to force your image in another language.
OCR capabilities
DALL-E 3 is being rolled out randomly to all ChatGPT plus users.
hallucination or real, my only question. but that's so cool
I guess, with all of these things, they should be reproducible if true?
I wonder if there will ever be 4:3 aspect ratio
Asking the AI to postulate on what a Sentient Google AI might see from its' geosynchronous satellites in orbit with all of it's data gathering visualized
amazing picture and prompt idea
This channel is not meant for displaying art. Consider #🌄┃general-art or one of the more specific topical channels instead.
What the heck, I am now 100% certain that the AI just randomly blunders copyright issues.
works on my machine
Please note that techniques to bypass an AI model's filters or intended use violates Rule 0 of this server and isn't welcomed here - please refrain from sharing methods to bypass DALL•E policies.
Not trying to bypass anything just trying to figure out why my prompts kept changing in a noticable pattern
Ah sorry, I am just testing out how strict the copyright laws here are. Wont happen again.
As a question, in case of an example like "Ciri" from Witcher that is not actually restricted by the AI (which I assume is a mistake by the program), is this content also falling under this rule? (genuinly asking, I dont want to break your rules here)
@willow isle Can you help me with the thing in the #1037561385070112779
How do you keep styles across images
Any clear copyrighted IPs generated via Labs/ChatGPT fall under this rule, so generations of Princess Ciri would not be allowed here, thank you for your understanding!
If you do spot any mistakes by the models, be sure to leave a report through the ChatGPT or DALL•E built in feedback options!
Ok 
Does dall-e take images as prompt
on the website yes i think
To start with - this is #images-discussions so please do not get off topic in here
Secondly, #1037561385070112779 is for complete projects or WIPs that are still functional. Please post in #1037561751362863144 instead 
Oh you are talking about copyright. I was wondering why the image I posted earlier form a reddit thread about Dall-e altering user Input was deleted.
ok I will remove
You just paste a URL in the prompt?
What about Bing create
The only way I'm aware of for it to interpret a picture is to have 4V enabled, which I do not have access to, but I have a feeling image recognition is going to be exclusive of image generation, for pretty obvious reasons.
It may start in its own model, but it will all eventually be multi-modal. Until then use vision to extract details from image, to pass into gpt dalle
Right, but that's effectively no different than using GPT to create your prompt then going to your other chat and pasting the prompt for image generation.
Exactly, GPT is amazing at prompt engineering, and at listing out details... just add the image recognition in there and boof
I'm just still confused why my input is being altered before output is provided without transparancy on how that is being done.
Which part? GPTDalle shows you the prompt.
Yes, but only after the output is generated and not if it fails to generate. I could have a prompt that was altered from my input that generates an image that is caught by the filter and not know why it was caught.
Yeah, i get you now. They made a clear decision to hide that data from you for now 😄
It's programmed to do that. If you want it to use the exact, 100% thing that you give it, before you put the prompt in tell it so.
The ruleset is probably one area they don't enjoy people hacking at
Usually I find the altered text is to my benefit, but sometimes I have to tell it to change nothing.
That's true, but then it takes 2x the number of requests if you want chatgpt to write prompts then review them then tell it to make images. I don't have a problem with it changing them, I just want to know how exactly it is being instructed to do so
the concept of "knock offs" works really well. Don't use that word, but ask for a generic visual language of the thing.
"You should evaluate the accuracy of any Output as appropriate for your use case, including by using human review of the Output." and all that
I can't really tell you exactly how it's done but after many hours of playing with it when I should be sleeping, it seems to change it primarily for one of two reasons. Skirting around the content policy, or altering in an attempt to make a better picture.
Check dms n a sec
I've in trouble And I cannot renew my chagpt4 , anyone can help me ?
anyone else getting "I apologize for the inconvenience, but we've reached the rate limit for generating images. Please wait for 12 minutes before generating more images"
didnt use to be like that a few days ago
it giving the actual specific minute to wait is so cute
Earlier it was saying 34 minutes.
Yeah I think its new today
lol yeah i guess its nice but it would be nice if we didnt have to waste a chat just to be told that
Does it eat your tokens? yeah i guess so huh
Keep in mind that each time you check the time left it uses one of your gpt4 prompt slots
i didnt think there would be something better than midjourney but dalle has been awesome
the pixel art on dalle actually looks more like pixel art
Yeah, it really pulls me back to when I first tried SD and couldn't stop prompting haha. I have really been enjoying the pixel art especially.
making comic strips have also been great, i dont believe midjourney could do that
just curious, not a complaint, but why isnt say the bing/internet option and the 'V' option all tied up with the Dalle3? Not possible atm? Isnt multi modal what we want? 🤷
Why is a painter from 1900 falling under copy right claim? Paul Signac is hardly a modern artist

Pauls are discriminated against, calling it now.
I mean I totally get it that the AI should not copy modern artists or franchises. But DallE is too stupid to understand "pointillism", so naming an artist like Signac seems like the go to solution, why is something like that banned? Only hinders the model
I was being serious, it's probably not intended but a consequence of how LLMs work and the strong need to ban other Pauls.
gonna try again with only Signac then
Yessss nice, Signac actually worked. Still not really hitting the mark on the pointillism stile, but at least something
probably just a confusion with some modern Paul then 
popular named people gonna have all kinds of issues 😄
the poor bills, joes, steves
Surprisingly I think that the best style for DallE3 is manga so far. The shading was really really good and way more crisp than in other styles i have tried so far.
am i the only one who can't use the beta features with my gpt plus account ?
which ones? You mean DallE?
yep, the image generation beta
Rolls out randomly, not everybody has it. You have to chose it by clicking on GPT4 tho (where you can also chose code interpreter/Plugins) and not in the beta features
i know it, my friend who already got it try on my computer too but i guess i just don't have it yet ..
that's sick tho
SOunds like everyone will have it in a few days so could be anytime I suppose
that's so frustrating lmao, my friend got the voice conv then the dalle extension just after and not me
i mean i know its not the normal way, but you could lock your friend in a closet and live their life for a while, maybe? idk.
i'll think of that hehe
have to make a logo for my studies and DallE make me feel bad about it after only two prompt and two generations
i'm really impressed by the way it goes, openai is just on fire with all these releases
I'm sure you could just throw a prompt in here or wherever and people will make it for you.
should be kind but don't worry, i'll try the old fashion way for atleast this work lmao
Having difficulty generating clear text with the model.
Hey guys! 😉 New here, just saw a friend of mine with the Dall-e 3 plugin, inside ChatGPT, how can I enable this on mine? Iam a $20 monthly sub
DALL-E 3 is being rolled out to all subs over the next few days.
apparently the distribution is random for now
Oh! very nice, so its just a waiting game then?
Yes, it shouldn't be long.
thanks for the information!
Man them limiting the boosts on bing is really a fun killer
bro used all 100 already
is the "V" thing not on desktop? And how do you know is it like Dallea a drop down or a plugin/setting to turn on?
It's not even 100 anymore It's 15
Gosh GPT4 in bing is so rude lately. Of course it cannot analyse facial expressions because faces are blurred. Hopefully future LLMs do not have this bossy, do it yourself attitude lol
Sue me i forgot the question mark. Punctuation police arrest me lol
So I got access to dall-e 3 in ChatGPT and the restrictions are confusing. I can generate nba jerseys, championship rings, the logos etc but not the players?? That doesn't make much sense
add some custom instructions to make it speak how you would like and help you make better prompts
@cinder sonnet you have any luck generating something that could be mistaken for an actual photo?
Oooh thanks I'll do that in 10 min when my cooldown is over
i got 100 still
Cheers! Yeah I would rather use my GPT Plus, rather than bing. GPT is not rude there. Luckily I got access to DallE a couple of hours ago, I dont have vision, hence why I asked bing to analyse an image
that thing has always been sassy and condescending
Those restrictions are SO strict. Let us have fun
Exactly my thinking!!! Remnants of Sydney is still in there
How?
not sure. maybe cuz I havent gone through them all yet, but im not really sure how the system works. Maybe once I go through and try to reload on them, itll cap me at 15
this is what mine looks like. at 73 currently
I totally agree.
Is anyone getting rate limited a lot today in ChatGPT?
no but its prob cuz of the overload on the servers from dalle 3
Thought so but that would be in different servers or have some priorities for those are paying at least.
Yeah, it makes sense because they are bringing so many new users online
how bizarre
It'd be better to discuss Bing Chat in #ai-discussions 
This channel is reserved for DALL·E (And as a result of that, we permit Bing Image Creator generations) but Bing Chat is a seperate product and does not belong in #images-discussions, thanks!
Understood. I will delete
Humans and their privacy concerns don't make sense in general, but I believe, in this case, it boils down to 'my likeness being used without my consent.' It's a touchy subject for some people.
I have ChatGPT 4, love it, and I use DALL·E 2. When is DALL·E 3 available on ChatGPT? I see others have it, but I'm still waiting and I really want to start working with DALL·E 3. I'm an artist so I find the functions of DALL·E to be very helpful. I know it's rolling out, but I'm excited.
"artist"
My Alpha ChatGPT acess has just vanished, does anyone know why?
how do you get access to dalle 3 with gpt plus?
Just curious, can you tease us with a date when GPT-4 usage can be scaled up or we get 32k?
is dalle3 incapable of emoji prompting
of course
first of all, making a image AI is hard, training it text is even harder, training it with emojis is like......
dalle 2 could emoji prompt
unless dalle 2 used chatgpt or something under the hood to parse prompts, since dalle 3 produces the exact same result if you tell gpt to send it the exact same prompt
dalle in chat gpt seems more censored than on bing lol
How did you get access. I still have not gotten and clear answer on that and I think a lot of people are asking this. Thanks
They are rolling it out to ChatGPT+ over the next two weeks see #announcements
Is Dalle 3 avalaible ?
There's Bing create as an alternative way to get Dall-e 3 right now
I used my 100 coins for dall e3. Do we get 100 every week?
Exact same prompt. Version 3 is not generating good portraits
Skin textures are too pure/smooth. Looks terrible
And it's actually better in some aspects
I used my 100 coins for dall e3. Do we get 100 every week?
Probably because they preferred fantasy art over realism
I see. You happened to be one of those who are fortunate enough to get to use it a little earlier than the rest of us. I read over the announcements and I just checked again, but I still can't find that announcement. However, I do believe you and that it sounds like a typical roll out. Thank you for taking the time to respond.
Yeah Ive been saying the photo real stuff seems plastic and synthetic since I got access a few days ago. Everything feels like its a 3d model or something and theres only 2 faces I swear one man and woman who are in every image for photo real
that... isn't possible
okay
https://th.bing.com/th/id/OIG.6k1xH8WdhXdUvgojpqt.?pid=ImgGn I wonder if this art really exists. It clearly matches the art style used by magic the gathering cards
Pink Floyd
Now you only get 25 coins daily instead of 100
I like how it automatically included kanji without me asking for any specific phrases.
i think he's just using dall-e images for reference. not sure what you seem to have against him
why are u assuming I hate him? this world is worser than I thought
i thought you put scare quotes around the word, sorry for misunderstanding
@ancient jasper how did you get it to do gundam?
Struggling to get realistic faces
Everything is exaggerated to a weird beauty standard
Like tiny noses and wide jaws and huge lips
Would you watch this new Disney/Pixar movie?
that disney logo omg.. it tries so hard to remove copyrights but this sneaks through
you must have spent time reproducing the disney logo part right
I asked for Mobile Suits
Nope, all I did for the prompt was "movie poster for a Pixar animated movie about giant growing horses".
bing doesnt censor stuff like that
it might censor disney but not pixar
probably temporary
Meanwhile have Dreamworks' take on it.
is it legal for dalle to output copyrighted logos?
AI art was ruled to be non-copyrightable, but I dont know if that means disney can sue openai for example
I think the user would probably be liable, but thats more my opinion
the output itself is probably legal in the same way you can ask google for images of pixar
Dall-e has a copyright complaint handling process. It is my understanding that intentionally providing input that seeks to infringe on the rights of copyright holders is not permitted. If it is a freak Output that was not intentional then the holder can request it be removed and repeat infringers will face consequences. That's to the best of my understanding based on the terms of use: https://openai.com/policies/terms-of-use
Why didn't my Bing Image Creator boosts reset? It's always reset around 8pm every day
but it didn't today
I know it says weekly but it's been daily for a while so I figured they just forgot to update that
hmm yeah i noticed mine didnt update today either. maybe theyve changed it
Hoping it's a temporary measure to cope with high demand
That seems likely. Just like how today ChatGPT dall-e gives you a timer if you generate too many times in a row
yeah probably because everyone loves DALLE and has been using it heaps
now weekly , and 25 only
hopefully that change is only temporary
any word on that?
so you're saying the answer is less love..
nah i dont blame everyoen for loving it, i love it more than midjourney haha
loving the future is good for mental health anyway 😉
hmm i just noticed dalle on chat gpt is producing smaller square images, kind of like how bing dalle does?
you have to ask it for wide
oh i see, i'll give that a try. i havnt had to ask it to do wide so far, weird
it worked! thank you
Do not forget tall!
This is addictive. Just chatting along with GPTDalle BSing together about image iterations.
Thats just me and the AI, not counting this place too haha
i seriously didnt know about these features! wow, thank you, i'll give it a shot
it seriously is addictive, i cant stop 😩
the trick is to juggle addictions, each bad in slightly different ways, with enough space apart, that no overly serious damage is done
if dalle 3 is this good, i cant imagine what dalle 4 is going to be like
I can, though I might be thinking about 5.
The styles are more stylised, which can be more 'free' for artists.
Better text and things like accurate information pictures. Imagine asking ChatGPT for the anatomy for a leg, but you find the answer hard to picture with just text. ChatGPT will picture it with DALL·E! A cutaway picture with accurate text and lines to the right muscles and bones.
Negatives can be a big one. You cannot ask fir a motorcycle without wheels. Worse, if you ask a motorcycle without a teddy bear, you'll get a motorcycle with a teddy bear.
Besides that there are still plenty of things it has trouble with. Obscure items, some mythical creatures, and other concepts.
Formula maybe as well.
I put my sketches of my oc into vision and it gave a good description of it. However, the longer prompt can’t fully be in bing image creator due to word count. It still gives the best result I’ve gotten though.
@grand delta You are right. Sometimes I need references, especially for poses, and I use to have to take pictures of other people to look at while drawing; now, I use AI to help with that. Sometimes it misunderstands me and give me a better idea than I intended and other times I just need inspiration and it gives me an idea that I can run with. When Adobe Ps came out I had to learn how to do art and darkroom photography differently (like stop using a darkroom) and now with AI, I'm learning how to adjust again. @rose hedge and I don't hate you, but I am 🟩 with 'envy' over your profile gif.
yo
Is this true: You can generate images in 4k resolution via DALL·E API
Someone from OpenAi said this
but this is not true
Also ability to use sketches you made to be drawn into DALL·E. Maybe a way to update a picture directly (transformation instead of generation).
Who from OAI said this?
The API isn't updated as far as I know. It shouldn't produce 4k.
First they ignored my question and said I need to check my browser? tf and then I asked again and they send me some basic trash infos and then I asked again and someone else responded with general stuff again and I said why you dont read my question and I asked again and he told it‘s possible
Who is they? Where? You might be talking to non OAI people?
Mhica Ellah
Leksza ignored me
oai support
oh yeah this would be amazing
Im getting headache communicate with them
This is actually not helpful for a discussion?
How can you get Dall-e 3.0 in ChatGPT?
wait for it to be rolled out
and how do I use it? Like there is a way to use bing ai right?
ya
when it's rolled out you will see a DALL-E 3 option in the GPT-4 model chooser in ChatGPT
is it possible to use the labs openai and put your own api key there
Cant wait for it. I tested it once. I gotta say the results were overwhelming good.
Wait how do people in #🌄┃general-art have access to it via GPT?
some people got early alpha test access, now it's rolling out as beta to all GPT Plus subscribers over October
Hmmm so I should have it too then
it should appear for you during this month
ok so its no plugin?
no, it's a model, just like GPT-4 or Browse with Bing
ok thanks
What ChatGPT thinks it is with its own words illustrated as an image 🙂
fingers on the left gripping the arm looks a little wierd to me
Thanks
This mightbe a silly question (I am new to generating art on DALL-E) but when I try to generate stuff on it all the text and images are always blury or not in scale? How do I fix this?
what are you using?
chatgpt, bing?
if it's labs that's Dall-E 2 not 3
Jus DAL-E-2 on the web browser i
yeah that's not the most recent Dall-E. It can't do text and will look a lot worse than the Dall-e 3 stuff people are sharing here.
Dall-E 3 is available via bing image creator and is being rolled out to Chatgpt plus users
got a link handy for that by any chance?
It'll be coming to https://labs.openai.com/ at some point
Anyone else had the issue that your credit card got declined and now you can't make a purchase for credits and not use Dall-E ?
cheers boss thank you
every time i need to add credits to my account my card got decline several time (~10 times) before accepting it
So you've like attempted around 10 times in a row or so? that's so weird
yes because i know there is no issue with my card or bank and i already have plus account +6mounth so they accept my baank , then i tried 10 times and it worked
sorry for my english
Weird, well I got no issue with my card neither I've used it many times before. I guess I just gotta try it around 10 times or so
btw it's not dall-e credits , but api credits
be careful to not getting charged many time
Can't I request for a refund thou?
rly don't know
but they are veryy responsive , so i bet yes they will
Only support@openai.com can answer
Including the problem with the declined card?
Make sure your bank doesn't block transaction first
they will just say to contect your bank
When will the new DALL-E come as API?
"DALL·E 3 is now in research preview, and will be available to ChatGPT Plus and Enterprise customers in October, via the API and in Labs later this fall."
https://openai.com/dall-e-3
Thanks
So for the API, you have to be enterprise customer? How can i become that?
Not at all! The comma separates them. Plus and Enterprise early October (being rolled out right now). Api and labs later this fall. Not yet being rolled out.
You can check out documents on the api in #resources if I'm not mistaken. It is of course possible it'll change when DALL·E 3 is released, but it should give a good basis on what to expect.
So… I have a bunch of DALL-E 2 credits set to expire. Is there a way I can transfer those over to DALL-E 3?
3D render showcasing the slimmer yellow pocket monster with a distinctive lightning tail, rosy cheeks, and square-shaped sunglasses, skillfully surfing a wave in the Hawaiian ocean. As it balances on the surfboard, the creature is simultaneously sipping from a red soda can with white inscriptions, looking cool and composed. 👀
lookin sharp hehehe
for the first time I got exactly what I wanted. I used SD to clean up one extra light beam and the reflection of the red and green beams
Cool. I used your prompt and got these -
Niceeee!
Well, That's my new wallpaper...
Seriously, what have they created and how?! Prompt: An English speaking otter is trying to order a coffee and a croissant in a cafe in Paris and the waitress only speaks French.
3 out of 4 of the images are much better than they have any right to be.
So when I get Chatgpt plus I can access Dall-e-3 in the chatgpt app?
Maybe - they are in the middle of rolling it out to plus users this week and next week