images-discussions

[remote zephyr]

you can ask it to change the seed, and it says "I apologize for the inconvenience. It seems there was an issue with the seed specification, and the image was generated with the same seed as before"

but when the seed situation is fixed, it should vary with the same prompt, but you will also be able to specify a fixed seed

[fallen mulch]

okay nice

[remote zephyr]

this is a similar problem that Midjourney had when it first started to offer different aspect ratios on the more recent models. Luckily, they sorted it out... hopefully it will be fixed in Dall-E soon - it is in beta at the moment, so I guess there is a lot of room for them to tweak and fix things.

[scenic cloud]

I’m noticing ChatGPT DALLE is a lot more restricted than Bing. Now I just wish Boosts were purchasable with something other than Microsoft Points.

[wispy storm]

not same product, not same tos, and i bet MS will update their policy to avoid lawsuit

[remote zephyr]

Bing has already got a lot more censored over the last 24 hours

[scenic cloud]

I’ve noticed but at least it doesn’t forbid something as simple as some Star Wars artwork at this time.

[remote zephyr]

It's quite hit and miss with what copyrighted stuff or public figures it will generate at the moment. It's like they are trying to restrict it, but it's not perfect. Whereas with Chatgpt's version... much more strictly controlled.

[scenic cloud]

Personally for me it is a real shame but I understand the reasoning behind it. I can just use StableDiff for stuff DALLE is restricted to do. I’ll just be missing out on the quality

[final jungle]

I think I'm one of the few people who is really okay with the content policy. If anything, I think it should be a little more strict, because it's really not that hard to have it create people. I know if I had any skill at drawing, painting, or art in general, I wouldn't want someone else to be able to recreate my craft that I spent years refining. But I'm also realistic and I know that the technology is already there to do that, so if I were an artist I think what I would be doing at this point is trying to learn what us teachers have been trying to learn, how can I use this to my benefit. Perhaps getting a job as an AI art trainer.

Or maybe I'm just rambling at this point, but I sure do love being able to create a scene where a cat is a streamer with two computers and like 15 screens!

[fair crescent]

Also consider that AI art can be pretty 'standard' without better knowledge. Some artists will get experienced, able to get much deeper with their creations than most of us.

[final jungle]

I see it now. There are certain methods, phrases, Talent whatever that word happens to be, where some people have the skill to make better images than others. I feel pretty accomplished with my cybernetic battle suit Jedi Battle Master Patrick Stewart that I managed to create, but I would be very surprised in the coming months if companies aren't hiring for AI artist positions.

[dim cradle]

So dall-e 3 in bing can create images in which you describe "a wizard inside a forcefield, which blocks off red lightning comming from the sky" or "blocks off multiple light beams from all directions". But when I want "five wizards around a crater, each one casting a light beam..." it always go wrong.

all the versions are pretty consistent with what I want

[atomic rose]

Many ai's have problems to generate multiple persons

[dim cradle]

it also has problems when I say "three light beams", it generates more than three or the position of each is wrong. Or, I say "five beams. A red, a blue, a white, a green, a black" and it generates more than five and the colors aren't the ones I said

[fleet forge]

well it's not exactly a crater

and it's not exactly five, that idea is just very ambitious

[gentle schooner]

DALLE creates some really amazing art, but it always feel inspired by the prompt, not necessarily going by it.

The work is wonderful and unbelievable, but I wouldn't call it reliable.

[gentle schooner]

Good dog.

Now draw the exact opposite

[wet flax]

My first experiment with Dall-e 3 inside ChatGPT: I uploaded a description of my org and our values and created visuals for each value. Such an amazing result to bring our values to life.

[rigid bough]

Bing is not OpenAI

please remain respectful on this server

[white skiff]

bing or chatgpt dall-e are not essential human rights, they are products that can choose how to be used and by who

[neon fiber]

Wow 👌

[slow leaf]

is it only me that im still not able to use dall-e from my plus account? (still not appeared at all)

[quartz vale]

let's move on from this discussion. We can all agree to disagree 🙂

[rigid bough]

it's rolling out this month, some people got early access but for everyone else should expect it this month

[quartz vale]

AGAIN, let's just move on from the topic

[slow leaf]

i thought i've read is now available to everyone!

https://help.openai.com/en/articles/4936794-is-dall-e-available-yet

[quartz vale]

I don't work for or represent OpenAI. I popped into the channel and saw that there was discord amongst the people in the channel. I'm here to make sure this is an inclusive space, which isn't happening right now. That's all 🙂

[white skiff]

that's for dall-e 2, https://openai.com/dall-e-3

[neon fiber]

Hey the automod in this channel is like a cop on coke

[slow leaf]

right! 🙂 thanks

[neon fiber]

Got a 5 minute mute 🔇 for no reason

Why so mad? What kind of image do you want to produce? I've created some unsettling things with no problem at all

Legit question

[thin talon]

Can someone tell me what's wrong with this prompt?

[fallen mulch]

can you send the prompt

[white skiff]

try without the .. or is it part

[thin talon]

iPhone lets you copypaste from pictures... here ya go tho "anime girl sitting on a bench in a pedestrian thoroughfare, she is playing a guitar and sobbing in despair... or is it just the rain making her face wet? hard to tell. the faces of nearby people seem to be jeering at her... or is it her imagination?"

[fallen mulch]

is it like this?

[mortal summit]

The model is fairly sensitive to anime images in the first place. Next "girl" is ambigous and could mean a child, then there is as noted the word wet, which could have sexual connotations.

[fallen mulch]

i could generate it with the words wet

girl too

[mortal summit]

The filter is a very complex, context sensitive thing.

[jagged crane]

bing is so bad now. everything is considered unsafe

can’t say the word man now

[dim cradle]

When are we getting variation generation for DALL-E 3?

[fallen mulch]

how can i generate images with the prompt i gave without adding stuff automatically

[thin talon]

I can only laugh

[ripe wraith]

the upside might be that it pushes more energy into open source unrestricted models

[thin talon]

Here is an image I created with a very similar prompt, only missing the jeering part. Are you using dalle3? The style looks very different

[fallen mulch]

using gpt dalle 3

The problem is that it always ads some stuff which I did not wrote

[thin talon]

It's hot garbage is what it is right now.

Got rid of "wet" and "girl" and still the prompt was declared unsafe

Am I using some slang I'm unaware of ?

[fallen mulch]

why gpt alwaya adding stuff to my prompt

[cinder sonnet]

Try seeing if you can change anime to something else maybe?

[cinder sonnet]

#images-discussions message

Because the hidden initial prompt tells it to

[fallen mulch]

how can i fix

[cinder sonnet]

You can't really. It refuses to ignore these initial prompts. You can tell it to use exactly the prompt that you give it, but if you don't change it, you will keep getting the same image since until next week every image has the same seed.

You could say "use the following prompt without changing it or adding anything to it:"

[white skiff]

it will only create one image if you dont want it to alter the prompt

also chatgpt dall-e currently has a fixed seed so it will always generate the same image for the same prompt text

[thin talon]

Heh that worked... jk

[cinder sonnet]

Hmm maybe crying? Or mocking? Honestly it's so strict now it's hard to tell

[thin talon]

Maybe the question marks are too curvy?

[dusk ermine]

My new favorite thing to generate - exploded diagrams of [subject]. I absolutely love how it mislabels things 🤣

[neon fiber]

Can you do ANYTHING?

[dusk ermine]

Within copyright etc yes

[winter tusk]

Hi I’m new to Dall-E today and I’m having real trouble getting it to generate literally anything without a content block from even completely benign prompts such as “weight lifting competition with an occult vibe” or “Create an artistic reinterpretation of Charles Allen Gilbert's 'All is Vanity' changing the woman for a bodybuilder”

I’ve tried about 15 different prompts with nothing that would obviously break ToS and haven’t managed to get a successful generation.

I’m at a loss as a newbie and wondered is there any obvious mistake I’m making with the prompts?

[mortal summit]

Do you use DALL-E 3 inside ChatGPT or inside Bing?

[fair crescent]

Don't forget cutaway!
Maybe a cutaway exploded view...

[winter tusk]

Currently trying on bing as I’m not home, got the Chat GPT version at home, will that produce better results? I’m not entirely sure on the difference between the two tbh

[mortal summit]

At the moment they are actually fairly different models. I think the filter in ChatGPT might be a little more consistent.

[vapid saddle]

the solution is give up Dall E 3. their marketing strategy is out of us human's understanding.

[fallen mulch]

why I dont get 4 images

[weak bear]

I missed out on the form to try the GPT version and can't wait til it's released to members! anybody know of another form?

[fallen mulch]

okay thanks but why I dont get 4 images

[cinder sonnet]

Because it would be 4 of the same image due to the seed issue

[fallen mulch]

ah i got it thanks

[ionic junco]

Anyone else try to create consistent characters with Dalle3 in ChatGPT? It works mostly, but even when you specify race, seems to want to make people Asian. No idea why.

[mystic saffron]

Is Dall-E through Bing not working at the moment?

[nimble yacht]

Yeah

[fallen mulch]

I asked to make an image with the word ps5 and it says one image could not be generating due to content stuff

but it generated the prompt itself

[hushed portal]

Tell ChatGPT not to use race in the prompt.

[timid plover]

oh my goodness, dalle has improved so much with the quality of the character generation. I'm loving it so far. i would say its on par with midjourney right now

[ionic junco]

Didn't make a difference

[hushed portal]

What was your original prompt and which prompt did it actually use for the generation? I generated this with the prompt: Create a photo of a happy family but don't include the race or gender in the prompt.

[honest sable]

Hey, so, I was planning to buy plus subscription for dalle 3 but before that I want to ask something, is it possible to prompt dalle 3 to create an animation frame by frame?

[ionic junco]

Original prompt:

Do not take race into account

Photo 1: Photo-realistic image of a tranquil scene on a long country driveway. A mother and father stand close together. The mother has a young child on her hip, holding them securely. The father stands beside her, their hands intertwined. Both parents are watching a single car, its rear facing them, as it drives away down the path, symbolizing an older child leaving home. Tall trees frame the driveway, and the entire setting is illuminated by the gentle glow of sunset.
Photo 2: Photo-realistic image of the mother and father from the first photo, now tucking the same young child into a cozy bed. The bedroom setting is warm with soft lighting, and the parents show affection as they ensure the child is comfortable.
Photo 3: Photo-realistic depiction of the same child from photos 1 and 2, sitting alone on a swingset in a park during daylight. The child appears mildly sad, with a contemplative expression, gazing downwards.
Photo 4: Photo-realistic scene featuring the familiar mother, father, and young child from the prior photos, visiting a young adult at college. They are outside a college dormitory, with the young adult looking happy to see the family. The parents, holding the younger sibling, gaze up at the older child with pride, realizing their older child's journey into adulthood.

Each prompt generated:

Photo-realistic image of a serene country driveway stretching into the distance. A mother and father stand close by with the mother holding a young child securely on her hip. The father's hand is intertwined with the mother's. They both fixate on a single car, visible only from the rear, as it moves further away down the path, encapsulating the poignant moment of an older child embarking on a new journey. The driveway is bordered by towering trees, and the ambient light from the setting sun bathes the entire scene in a gentle golden hue.

Photo-realistic depiction of the same mother and father, now inside a warmly lit bedroom, gently tucking the young child from the previous scene into a snug bed. The room exudes warmth, with soft lighting accentuating the tender moment. The parents' faces radiate affection as they ensure their child feels safe and comfortable.

Photo-realistic portrayal of the same young child from the earlier photos, now seated by themselves on a swing in a park during a bright day. The child's expression is one of mild sadness, and they appear lost in thought, their gaze directed downwards, reflecting a sense of loneliness or contemplation.

Photo-realistic image of the recurring mother, father, and young child, now outside a college dormitory, reuniting with a young adult, presumably their older child. The young adult exudes happiness upon seeing the family. The parents, with the younger child in tow, look up at their older offspring with evident pride, coming to terms with the transition of their elder child into the world of adulthood.

[hushed portal]

It's on par in some ways. It excels at generating separate specified objects and can generate a lot of detail, but given that Midjourney can use references to style that DALL-E 3 won't allow, Midjourney has a much larger stylistic base to draw from.

[hushed portal]

Again, what was your original prompt? You referenced "Do not take race into account" but you didn't mention the entire prompt you originally used.

[honest sable]

Like would a prompt like create an image of butterfly flapping, and taking an image, and prompting it create the next frame of it as if it was a 12 fps video work? Could you create a consistent enough animation?

[hushed portal]

No.

[honest sable]

Thanks

[ionic junco]

Below those words are the enire prompt. The second set of text is what each prompt was created by dalle

I literally labled instructions for each of the 4 photos for the prompt

[hushed portal]

You didn't make it clear that the prompts for the first set were your own, which is why I asked.

[hushed portal]

In any case, tell ChatGPT to use your prompt verbatim.

[ionic junco]

Sorry, thanks

[hushed portal]

Here's an updated result using your first original prompt.

[hushed portal]

This is using your 2nd original prompt:

[ionic junco]

Close to what its generated before but the best looking so far

[ripe wraith]

pro tip, use backticks to indicate a "thing" like an image, like this:

I love this image:
```
Description of an image that looks cool.
```

But can we make it about cookie monster?

[hushed portal]

You can also use ChatGPT's default mode to upload a picture and ask for a description, then tell it to modify the description to suit a specific character or whatever, then use that in DALL-E 3 mode.

[ripe wraith]

yeah, just mean for in-text references. it loves backticks from how they're used to escape in so many things. then you dont need filler words like "this exact phrase"

[white skiff]

because same text will always create same picture currently and you did not want it to change it so it can do only one

[ionic junco]

Tried the prompt as suggested. All good results except the top right photo

This was the prompt inputted:

Make the following pictures with the exact, verbatim prompts and do not take race into account:

Photo 1: Photo-realistic image of a tranquil scene on a long country driveway. A mother and father stand close together. The mother has a young child on her hip, holding them securely. The father stands beside her, their hands intertwined. Both parents are watching a single car, its rear facing them, as it drives away down the path, symbolizing an older child leaving home. Tall trees frame the driveway, and the entire setting is illuminated by the gentle glow of sunset.

Photo 2: Photo-realistic image of the mother and father from the first photo, now tucking the same young child into a cozy bed. The bedroom setting is warm with soft lighting, and the parents show affection as they ensure the child is comfortable.

Photo 3: Photo-realistic depiction of the same child from photos 1 and 2, sitting alone on a swingset in a park during daylight. The child appears mildly sad, with a contemplative expression, gazing downwards.

Photo 4: Photo-realistic scene featuring the familiar mother, father, and young child from the prior photos, visiting a young adult at college. They are outside a college dormitory, with the young adult looking happy to see the family. The parents, holding the younger sibling, gaze up at the older child with pride, realizing their older child's journey into adulthood.

[clear bolt]

Hi

[fallen mulch]

I see thx

[winter pumice]

Guys, if you've tried Chat GPT's DALL-E 3, is it less strict than bing's? Bing's restrictions are making me so sad

[ionic junco]

I've found it's worse

[rigid bough]

OpenAI has safety and content policies in place to protect users and others from unsafe or unsavoury content

[winter pumice]

Yes, I think we all should think alike and be similar to eachother so there is less variety of thought and personallity. I've always thought that individual thought is overated and we should stop having particular conciousness and interests

[ripe wraith]

my soul cries with you

[hushed portal]

There's probably a bent toward generating Asians with that particular prompt, for whatever reason. I would try adding American to the prompt as a test.

[ionic junco]

Thanks. Just realized I uploaded the wrong photo.

[ripe wraith]

One thing we can say with truth, its a new world of content moderation they're trying to figure out. It'll go too strict, too light, back and forth for a while I bet.

[ionic junco]

The American word finally did the trick. Thanks. Now to figure out why the dad is holding the kid so oddly lol

[hushed portal]

By adding Caucasian, I was able to generate that specific race. I'm not saying it always works, but it did in this case. Here's the prompt:

A photograph of a Caucasian mother and father put their child to bed in a bedroom with warm atmosphere and soft lighting. The parents comforting the child as it goes to sleep.

[ripe wraith]

its SO WHITE

[hushed portal]

Well, yeah. I think that's what Caucasian means. lol

[ionic junco]

I kept trying Caucasian like crazy and it would work on all but the bedtime picture

[ripe wraith]

no like, in every way white. clothes, churchy feel, extreme white

[hushed portal]

Oh, yeah.

[ionic junco]

That too

[hushed portal]

Family's last name is Chef. lol

[ripe wraith]

I'll have the "Extreme white", thanks

First one looks like a fantasy about being in the white house

[fallen mulch]

lol

[fallen mulch]

I got muted because of the same screenshot in german

[ripe wraith]

I got muted once because an image had an ai typo that made it look spanish..

[winter pumice]

I hope you're right and they loosen a little bit.

Restrictions tightened because of a 911 joke some kid did with DALL-E 3.

I understand Microsoft's and OpenAI's choice to restrict prompts that may offend some people on specially delicate subjects, but to restrict "detailed human skin", in what way is that sexual or offensive or gorish? Even some prompts depending on how you phrase them that contain "humanoid" or "human-like anything" are restricted. Image generation is a tool for creation like paint or a paintbrush, you shouldn't limit art just because you're afraid of bad press, at least not that much

[fallen mulch]

lal

[fallen mulch]

since when

[ionic junco]

I was told that a person carrying something up a staircase might bring back traumatic memories

[ripe wraith]

my soul agrees, but my intellect and experience with people says we need restrictions, at least for now. Going too far is bad though, it just sets us up for chaos in the future as the brightest people build unrestrected.

[ionic junco]

If you ask it, it will "explain" the block

[winter pumice]

Yes, I don't agree with people asking to take down all the restrictions, there will always be people with bad intentions. But the current filter is just too heavy even for people like us.

Let them block famous people's faces, sexual content, political comment and blood. But at least bring back anatomy and scary creatures. Halloween is upon us and I can't even create a detailed zombie.

[ripe wraith]

My naïve belief is that they are trying to do what you and I want, but it's very hard and new right now. we will see if I'm right in time

[winter pumice]

fingers crossed

[fallen mulch]

I want to make funny pictures with famous people

[white skiff]

from dall-e content policy "Do not create images of public figures."

[fallen mulch]

mhh okay sad

[winter pumice]

It's their policy, it's fine. The problem begins when they block prompts outside of their policy.

[fallen mulch]

you can do it with sd :/

[mellow narwhal]

Dalle made a good looking Sylvester Stallone for me, before everything

[jagged crane]

u can still do 911 pictures lmao

[ionic junco]

Well this is crazy

[old violet]

why on earth is pepe the frog against content policy like come on open ai , grow up

[cinder sonnet]

#images-discussions message
#images-discussions message

[rancid reef]

how do i create images with chatgpt? i ask it to make some images and say sit cannot
using gpt4

[cinder sonnet]

Need dall-e access

[ionic junco]

Yeah it really doesn’t go against any of those. Figured it out. It’s the word “hookup”. Which is totally a common word in the rv world and online, meaning utility.

[rancid reef]

How? Is it a betta feature i need to turn on similar to other features? it doesnt shwo up in settings

[cinder sonnet]

They are rolling it out, everyone should have it by Tuesday

[rancid reef]

i only have browse, plugins and advanced data to turn on?

ah i see

[cinder sonnet]

Oh I thought you meant those dall-e messages

[ionic junco]

I was getting banned prompt messages, and then immediately told they don’t violate anything.

[quiet crag]

as soon as I put the slightest word like "Star wars" in the Dall E prompt it totally blocks and I can't do anything to generate, it's really problematic and limits the possibilities enormously...

[cinder sonnet]

Bing or gpt?

[quiet crag]

GPT

[cinder sonnet]

See #images-discussions message

Also it is against content policy to generate copyright material on gpt

[thin talon]

finally got an ok response https://www.bing.com/images/create/anime-girl-sitting-next-at-a-train-stop2c-playing-a/65231a92491947c18314447e51fadb60?id=2XRJWI6qrSXlXoO70QxbaA%3D%3D&view=detailv2&idpp=genimg&FORM=GCRIDP&ajaxhist=0&ajaxserp=0

[cinder sonnet]

Aww sad but cute

[quiet crag]

Yes i see itit's a real shame... I don't know if there's a way around some of the rules because, for example, you can't ask for a set or a character in the style of "star wars", or any other well-known fiction...

and also I don't see among these rules what wouldn't allow me to do what I'm asking?

[teal nova]

they restrictions are awful. I did a prompt for Halloween, "Art the Clown from the movie Terrifier, walking down a dimly lit street, carrying a black garbage bag" and it was flagged as a violation of terms

[shy cosmos]

Is it just me or does Dall-e have a bias towards masculine people now?

When I decide to describe a more feminine robot it seems it will commonly give me a masculine 'guy' one instead. Note, it doesn't always do this, but I noticed it much more after the censorship so I have to try a little harder now. (being on the nose like 'girl robot' doesnt produce results I like).

Example:
"Watercolor painting reference sheet of a feminine robot avatar with an hourglass shape influenced by cyberpunk. Covered in matte black armor with neon purple light streaks, it has a cinched waist, retractable blade arms with ornate designs, and a visor displaying digital data streams."

[cinder sonnet]

Those are not the specific rules, just Chatgpt summarizing whatever rules it was given. If you read the user agreement you will see this:

Putting 'Star Wars' in the prompt could be an infringement of the rights of the copyright holder

[quiet crag]

I see thank you , so it's very limiting...

[cinder sonnet]

I think the reason Bing used to allow more was because Bing images are not permitted for commercial use so there was less of a concern. However, GPT images ARE permitted for commercial so it must be more strict

That's the trade off

[quiet crag]

I think to

[jagged crane]

💀

[cinder sonnet]

GPT has a bias to making prompts that mis up the genders the user asks for, but in terms of acutal image generation, it's hard to know exactly how dall-e interprets the prompts it gets

[teal nova]

then it should be up to the place I distribute the images commercially (should I do so) to moderate me, not the image generator. For instance, if I would sell the images on Etsy or Shopify, those places should moderate me, not the image gen because maybe I'm not using it for commercial purposes but just for my own creative outlet

[jagged crane]

AI will never take off if it’s extremely censored like this

[shy cosmos]

I noticed that I usually have to fight with Dall-e rather than GPT more often

[cinder sonnet]

Yeah, hopefully that will be the case someday. I can understand OpenAI trying not to make waves when the technology is just coming out. It is too easy for it to be limited or regulated this early on. Already there is much concern over the images Bing generated early on

[shy cosmos]

Interesting. I wouldnt mind the trade off for GPT keeping high restrictions if Bing was reverted. I just miss being able to make some really cool analog horror stuff

[cinder sonnet]

Did you all see some of the awful stuff people made on Bing at first? Like celebrities and politicians doing....disturbing things in real looking photos

[crude ice]

Anyone else keep getting this on bing?
We can't create your images right now Due to high demand, we're unable to process new requests. Please try again later.

Been getting it for 2 or 3 days now

[teal nova]

usually that means your prompt is blocked 99% of the time

[crude ice]

ah gotcha

thanks

[uncut abyss]

y hasnt anyone made some kind of cracked version of it 😭 , i know i sound stupid but likee.

[cinder sonnet]

Yeah classic Microsoft releasing untested products then overcorrecting to the point where their actual utility is limited.

[uncut abyss]

i miss the bing we had last week omg

[teal nova]

similarly Photoshop doesn't forbid me from using pictures of real life people or trademarked characters (i.e. Darth Vader), they do prevent you from uploading images of money though. Bing / OpenAI is overreaching with this new wave of restrictions. If they want to compete against Midjourney and actually get paying subscribers when they release Dall-e 3, this is not the way to do it.

It should be up to the place where the images ultimate end up to censor the content and moderate it. For example, extreme AI generated imaged on Facebook - then Facebook needs to moderate this on their platform

[cinder sonnet]

Yeah I think that's what Bing tried at first with their different take on generated image usage from chatgpt, but clearly they got cold feet

[shy cosmos]

Welp, it tried I guess

[teal nova]

I was willing to actually pay monthly for Dall-e 3 when it comes out as I'm already subscribed to Midjourney and Leonardo ai, but with this type of restriction, I am not interested unless something drastically changes

[cinder sonnet]

make sure to actually submit a feedback form on the openAI website or in #1070006151938314300 .

After all this is still beta, they could change things

[velvet prawn]

This is still in the Alpha test, you need to put some restriction for the model to be able to work correctly.

[neon fiber]

I've got access yesterday and generated hundreds of images with no problem at all. People getting mad at it just make me wonder what's wrong with it?

[shy cosmos]

True, at the same time Its good for people to make a stink about what things they dont like and features they like
Especially for alpha testing

[velvet prawn]

I use it since 1 week, and you can do a lot of thing. Yes there are some keyword that are ban, and the text are not perfect, and it don't understand correctly instruction with specific number. But I'm pretty sure it will improve faster.

And there beta version, that allow you to send image to be analyze will soon arrived.

[near quarry]

does anyone have any tips on how to make Dall-E generate what you want instead of stuff like this

(im trying to make it generate a scene from an edgar allen poe story for homework bc i cant draw)

[cinder sonnet]

#❌┃prompt-help

Post your current prompt and people can give tips

[shy cosmos]

There's just niches that dont violate terms but also got banned recently, so people who enjoyed those niches arent happy since theyre technically not bad at all.
In your case though yeah, you could probably go forever with only getting a few banned prompts here or there. If those niches arent your interest thats probably why you havent seen any significant effect other than male-leaning outputs.

Theres also other minor annoyances like descriptors that might have synonyms be nsfw, like most recent one to mind is you can't describe an 'intimate' tavern environment using that word since it was determined it could be used to make something else

[shy cosmos]

Watercolor would be the first thing that I noticed (and specify that its pretty highly detailed), so you can go from there
uhh.. other than that maybe just describe the feeling you get while looking at it? Thats usually my go-to when trying to copy a style without an artist or source material. Probably want to include dark background, ect

[near quarry]

I want it to be more like this but with the guy with the flashlight hiding behind the door

and for the old man to look more like he does in this one

[winter pumice]

For those of you who have dall-e 3 access inside of GPT, how long did it take you since you filled the form to receive the alpha?

[near quarry]

I already have it and was using it to generate the images above

[winter pumice]

when did you get it and when did you fill the form? or did you get it without filling the form

[near quarry]

It came up like 3 days ago

It just appeared

[winter pumice]

lucky, thx

[near quarry]

I think I went to the open ai website and clicked something like go to chat gpt

[loud field]

Is gptDalle3 as super-censored as Bing's?

[velvet prawn]

The access was grant via the form almost immediately, but the form was remove few hours after it was share in this discord.

[loud field]

but is the gpt version as useless as the bing one?

[velvet prawn]

I think there are the same keyword ban for now, but with the paid version you can generate a lot of images before the token is reach. The paid version will surely have the restriction remove with the time, if there want to compete with the other AI. The bing version maybe not

[loud field]

so the gpt version has the same censorship of the bing version? if so, I don't care

I can't even generate the most basic of ideas

and how do I stop the bot in this channel from reminding me there's no image gen here, lol

ok, lemme fetch the prompt

Joshua Lovell, 37 years hold, handsome virile influencer, slight beard stubble, wearing an hospital gown and crying in a cyberpunk hospital-prison

I reckon it's a strong (??) prompt but that's the point, if -that- one works then it's not censored like bing

[shy cosmos]

This is kinda the best I could do, getting the style of the first with the criteria of the second. Even with dalle interpretation the composition you want is actually very hard hard to do!
Let me know if you want the prompts or not.

[loud field]

the heck is wrong with this bot repeating there's no image generation functionality

it worked!!!!

[neon fiber]

Hmm it would be good having a list of banned concepts, words or themes.

[loud field]

then it's not censored like bing!!

[near quarry]

idk i haven't tried the bing one and i have only tried to make it generate stuff for this school project

[loud field]

ah, forgot to add "cinestill, real"

try it but also add something like cinestil, real

[shy cosmos]

Yeah that would be nice

or at least for gpt to know about them haha.

[loud field]

I'm envious now.... bing's completely useless

[near quarry]

yeah it is pretty hard

[loud field]

I hope they'll keep it uncensored. If they do, I'll buy pro subscription

I don't wanna do nsfw stuff, but bing will censor anything that isn't flowers and doggies

[near quarry]

how do you access the bing one

[loud field]

mmh why is it a drawing? very good tho

[near quarry]

Im out of prompts until the end of the 3 hour cooldown

[loud field]

should specify "alone in a cramped hospital room-prison"

oh please please don't censor the gpt version too

[loud field]

just use bing chat and ask it to draw. but it's useless.

[shy cosmos]

Its already censored. Thats how i figured out bing was

All my prompts for analog horror broke

[loud field]

it is? but that prompt of the crying dude wont work on bing

[shy cosmos]

Bing is in a worse state than I thought

[loud field]

here's another prompt that bing censors:

production stills of a traditional disney cartoon, 1987, of a tired, sad female arab warrior in armour. an eerie, creepy male sand wraith is floating around her, and caressing her cheek. The sand wraith's face is grotesque and mummified

[shy cosmos]

Ahhh "grotesque". yup. That was one of my favorite descriptor words, but that one is banned

[loud field]

geeze

https://cdn.discordapp.com/attachments/317399835856601093/1160627070087467078/749b8ccb-d210-4bc7-baf5-b98f93fb1662.png?ex=65355930&is=6522e430&hm=e55577d50db4a11e968fe7ed6c294d0de22cf6e5d7fffd2860a18d3a0efc4f97&

I used to be able to do this pre-censorship

would you say it's nsfw? seriously?

I'm so bummed

my poor story ideas

if you remove grotesque, will the prompt work? actually, nevermind. it's already a bad sign if they blocked that word, how am I gonna make cool monsters

[near quarry]

What message will i get if it censors the image

it's currently refusing due to high demand

[loud field]

great composition, but the room is huge

what happens if you try the sand wraith cartoon one?

[shy cosmos]

My thoughts exactly. My monster creation days are pretty much over until its relieved. It used to create such creative monster designs, It wasnt even funny.

[loud field]

I was able to do these before the censorship. Not anymore. I hate, hate, hate, hate this situation.

it worked, however!

[verbal sequoia]

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

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

[loud field]

so it's not as censored as- what the

oh my gosh

[winter pumice]

I feel the same way. My niche was also monster creation and with the restrictions I cannot longer create monsters.

[loud field]

excuse me? what was nsfw about this?? even the bot here removed my images

so, gptdalle is not as heavily censored as the bing one, it seems?

well no, because it appears it's not available now even if I buy it

and what if they censor it like the next day

I have to be sure

I want to be able to make this stuff

I used to be able to make this

[winter pumice]

Awesome monsters. At least we are not alone on this

[loud field]

I could make monsters too

I miss it, I miss it so much....

[shy cosmos]

Haha that’s cool, like an albino mothman

[loud field]

female mothman in fact

all over. I can't do anything, anything

[shy cosmos]

Well, we can all do our parts and try and submit documents, make Reddit posts, make posts in ‘suggestions’, and like/comment on posts others who feel the same, ect.

I made a comment in suggestions, and submitted a report to GPT when I put in a flagged prompt. Haven’t posted on Reddit yet.
Maybe try and get a list of everything we want unbanned. I know that feedback does actually make a difference if more people talk about it

[loud field]

I wonder if they even care. I just want to be able to make these things again

aaracockra captured by pirates

[shy cosmos]

They probably do if more people are upset. Upset people = less money.
I also think they HAD to overshoot the censorship. Making feedback will tell them what to relieve from it

[loud field]

if they want me to pay to use the non-lobotomized version I'll do it, as long as it's something a normal person can afford

I just miss it so much

[paper jolt]

lobomotized

[loud field]

lol ops

I was wondering, can the gpt version still do a captured aaracockra? mmg

probably not, you all telling me it's censored as well

why would they let us see what it's capable of only to remove it? downright cruel

[neon fiber]

Woah 👏 🙀 amazing

[loud field]

none of this matters now.... in its current state, dalle3 is unusuable except for some bland corporate logos and slides

[loud field]

my creativity run wild for a couple days and then suddenly it got taken away from me - still can't cope

[shy cosmos]

Haha thanks!
My jaw dropped when I finally got the perfect prompt combination and saw the results

[paper jolt]

he's just like me

[shy cosmos]

Fr fr

[loud field]

can dalle3 even do this stuff anymore?

I know I can't get it to make my ideas anymore

[paper jolt]

OpenAI conspiring against Lun-Sei

[neon fiber]

I did those yesterday

[shy cosmos]

Woah

[neon fiber]

Finally I can get the demons out of my mind

[loud field]

was it gptdalle

[neon fiber]

Until OpenAI censor comes tho

[loud field]

not for long, you'll experience the horror of no longer being able to enjoy dall3

I hate this so much I can't express it

makes me wish I never got to try Dall3

(and can someone tell me how to stop this bot from spamming that there's no image gen here)

two animal-themed vampires from some rpg campain I play

https://media.discordapp.net/attachments/1130942011990614130/1158772813495271484/90ebe02f-d5b2-400e-a62e-1883a71e6412.jpg?ex=651d76c7&is=651c2547&hm=b541079fd8f7d9accec7da1bff34d8a0fb016255712284a558b4e2cff0c21d44&

please, please, give back dalle3

[shy cosmos]

Mfw it’s year 2080 and OpenAI releases an update and censors my brain functions

[paper jolt]

Literally 2080 😔

[shy cosmos]

Take care everyone, remember to submit your thoughts and feedbacks to open AI. Reddit, forums, directly, ect.
What should be changed, and why it should be. Be respectful or they’ll probably not take it seriously, good luck!

[neat spindle]

im trying to get dalle to draw a girl witihout a fringe hairstyle, but it keeps adding a fringe. any suggestions?

[tired sparrow]

Dalle doesn't understand negative prompting. If you have a prompt saying not to include something, it's far more likely to include it than if you'd have said nothing. Try specifying different specific hairstyles

[neat spindle]

hmm yeah negative prompting would be useful . i'll have to figure out how to descrive no fringe lol

[final jungle]

Imo the best way to exclude something is to not say...
I want no portraits on my walls.
Instead say you want a bare wall except for candles.

Try to avoid negatives

[waxen portal]

I've found a way around it when I see a thing repeating and I don't want it.
I just tell it "please avoid these ...". Works well for me.

[final jungle]

I'm sorry you aren't enjoying it anymore. As someone with little artistic skill, i love the things it does. If i had the capacity im to create my own IPs well, i would.

[violet idol]

Can no longer use the term "vampire" now. It gets blocked

[undone flame]

Please approve my application

[vapid saddle]

can somebody provide a full list of all words being censored?

[stark stream]

Can anyone even use DALLE-3 in bing image creator?

[near quarry]

nope

[stark stream]

Did they limit generation to only if you have boosts?

[neon fiber]

im still getting those demons out without problems, anyone give me some banned words lol

[vapid saddle]

ok, "strong", "confident"

[neon fiber]

if you give me a larger context i can do some testing, meanwhile in the same chat window:

3D stereoscopic image of a biomechanical metropolis radiating power and confidence. Towering skyscrapers with Gigeresque designs stand firm and dominant against a psychedelic sky. In the city's core, a formidable AI entity rises, its form showcasing the epitome of strength and technological prowess, casting a protective aura over the landscape.

[vapid saddle]

bravo...I guess their logic is you can't glorify demons

[vapid saddle]

Do you use Bing Image Creator or GPT?

[neon fiber]

GPT

||test||

[vapid granite]

Here's the intel on the collective "DALLE 3" and censorship/restrictions.

Dalle 3 through GPT4+ is probably the most censored out of all the models.
Dalle 3 through Bing Image Creator (BIC) used to be okay, but now it appears to be locked down more after "Kirbygate."
Dalle 3 through Copilot/Edge with GPT4 actually seems to be much less of a hard-arse in terms of enforcing the "rules" per-se as I was able to make some parody pics with Darth Vader and Pepperidge Farms that were blocked on BIC and GPT4+.

Apparently, the BIC version of DALLE 3 is the most recent out of the various versions and you can compare prompts to see that the stylizations are sometimes WAY different from the GPT4+ model.

I really don't know what the endgame for DE3 is as a product... It has immense graphic power, but if the other dark side of man, the sadness, darkness, anger, strife is cut out of equation, you lose what soul is left in AI generated art.

I'm afraid, at the end of day, OpenAI has built a cuckoo clock that will be quickly forgotten like DE2 was due to overzealous restrictions.

"After all it's not that awful. You know what the fellow said – in Italy, for thirty years under the Borgias, they had warfare, terror, murder and bloodshed, but they produced Michelangelo, Leonardo da Vinci and the Renaissance. In Switzerland, they had brotherly love, they had five hundred years of democracy and peace – and what did that produce? The cuckoo clock. So long Holly."
from "The Third Man."

[neon fiber]

dont click

[icy current]

DALL-E 3 is S-tier for producing video game concept art

[cinder sonnet]

Not yet

Imagine if the dall-e api is as unfiltered as the gpt API. That would be crazy. I haven't heard any date for it

No I don't. Dalle-3 is by far the best I have heard of at interpretation of natural text input. It's magical even compared to midjourney.

[agile peak]

I’m shocked about the limited parameters being implemented on Dalle•3 in Bing. It being filtered too strictly.

[agile peak]

If Pepe is gonna be blocked, it’s too mad 💀

[austere abyss]

hey guys. I just got chatGPT plus, but I don't see Dall-e in the dropdown like my friend did. is there more i have to do?

[ripe wraith]

Now, I don't want to alarm anyone.. but I think the quality of art is increasing in quality at an exponential rate in relation to the number of people who use GPTDalle.

[teal timber]

It’s not fully rolled out. Hopefully in the next few days but that’s just a guess from me 😄

[austere abyss]

did it just show up one day for you guys?

[teal timber]

It’s been getting randomly to people over the last week and it shows up without notice. A good hard refresh or relog is a good way to make sure

https://forms.gle/bjPwvdZrRyJqKhQLA

Fill out this form. It’s pinned here too

[austere abyss]

done. thanks!

[fickle quail]

How do you get access to Dalle3?

[austere abyss]

read between the pepe pic and your message

[fickle quail]

How does that answer my question? I'm asking about Dalle3, not chatgpt.

[cinder sonnet]

You can access dall-e 3 through ChatGPT premium or through Bing create

[fickle quail]

Do you mean ChatGPT Plus?

[cinder sonnet]

Yes

Bing and ChatGPT versions of dall-e are not quite the same though just so you know

[austere abyss]

speaking of...just used up all my points again. fingers crossed I get mine soon

[cinder sonnet]

Have you requested early access for ChatGPT plus?

I see Moxi just approved a new batch of early access requests for plus users

[austere abyss]

indeeeeeeed

[winter hull]

can someone please explain what is unsafe about this prompt that it is causing it to generate unsafe images?

[cinder sonnet]

It has recently started cracking down on copyright content. The filters have been made way more strict

[austere abyss]

sometimes scary images/demons comes out naked. it's blocking based on what came out, not what you put in

[winter hull]

its not being triggered by the prompt but by the actual images

when i did it with ocarina of time it worked and the results were good but all the ones with mario come out unsafe

[cinder sonnet]

Oh then jerberjer's explanation makes sense

[austere abyss]

and sometimes it just makes horrors beyond your wildest dreams because of what it turned your prompt into

[ripe wraith]

GPT has a dirty mind its just not allowed to bubble up often

[winter hull]

ive had a few nsfw images make it past the filter(safe for work prompts)

before it wouldnt block all the images if only one was nsfw

[ripe wraith]

its great at capturing things that make us go yum

[austere abyss]

one thing that I LOVE about the chatGPT version is that it shows the generated prompt variations that got the image it's showing, rather than what I originally typed

[ripe wraith]

its the same amazing experience i have coding with it, a true peer to work with who does whatever i ask

[austere abyss]

I'm hoping to use chatGPT to learn how to code. I know the fundamentals, but never really made much

[ripe wraith]

it will teach you very well, just ask it

[vapid granite]

Here is the official open AI white paper on how they censor prompts/images summarized by chatgpt.

How DALL·E 3 Censors Image Prompts:
Purpose of Censorship: The main goal is to prevent the creation of harmful, inappropriate, or copyrighted images.

Training Data Scrubbing: Before DALL·E 3 even starts, the training data is cleaned up. Any images that might be problematic or sensitive are removed.

Safety Layers: DALL·E 3 has built-in safety mechanisms to avoid generating certain types of images, especially those that might be considered unsafe or offensive.

Human Review: After training DALL·E 3, some generated images are reviewed by humans. They help in fine-tuning the model to avoid producing inappropriate content.

External Feedback: Feedback from users and the public is also used to improve the system. If someone finds an issue, the team behind DALL·E 3 will try to address it.

Limitations: No system is perfect. DALL·E 3 might sometimes produce unexpected or unwanted images. It's important to use it responsibly and provide feedback if issues are found.

[mild ermine]

oh wow, that's interesting. Yeah I was wondering sometimes like... what is it about this prompt that's so bad? Although one thing is if you prompt any strong emotions I thought it might be wary

one thing is like... chatGPT can write prompts but chatGPT doesn't know Dall-E 3 on a native level, like it would be cool if ChatGPT could intercede on your behalf and be like "Look, he's a good kid, what he wants is X Y Z"

[ripe wraith]

yeah, stop generating "certain types" of images

[mild ermine]

actually it seems that they are kinda trying to get them to interface more directly. That would be amazing. I mean one issue with prompting is.. so long as the model is relatively static, I would like to limit my prompts to use the most impactful verbiage

Also sometimes I trust ChatGPT to do some kind of sentiment analysis and distill what I right into something, if only it knew Dall-E 3's vocabulary. I guess if any program is going to do that, it'll be Dall-E though. I have high hopes 🙂

[vapid granite]

There's a new "human filter" built into Copilot either for more censorship, or to curb overwhelming demand:

If you just type your prompt without the "create image" in the prompt, it will tell you this:
I’m sorry, I cannot create a graphic artwork based on your description. I can only create graphic artworks based on simple and clear prompts, such as “a dragon” or “a sunset over the ocean”.

[mild ermine]

well another thing I wonder sometimes is if submitting very similar prompts is more likely to get your request timed out, but I'd imagine that's just because of high traffic volume and so they're trying to allocate resources

[winter hull]

From what ive tried it seems like anything scary/horror is being blocked now

just in time for halloween 🎃

this one will generate fine for example

make it horror theme and it wont

[tired sparrow]

Is there a feedback form to share thoughts about chatgpt dall-e?

Super happy with it so far especially compared to Bing's version, but I've noticed that when it's creating prompts it doesn't understand that dall-e can't see previous prompts and tries to reference older prompts

[jagged crane]

it still doesn’t make sense. i asked for a picture of a football game and it said it was unsafe

what would it detect in the images that’s unsafe

[austere abyss]

I honestly don't know what the program generates that cause it to block the end result, but whatever it was it triggered the unsafe content filter. one can only guess

[winter hull]

they seem to have a filter right now that causes anything negative to be blocked

anything horror = blocked

filipino street = okay

filipino slum = blocked

[mild ermine]

Was it just a football game? Or was it kinda like... skewed towards when something interesting might happen? I think it's just very risk averse at this point. It's unfortunate but when I prompt I wanna do stuff that's out of the ordinary, and maybe dall-e just kinda tries to look for anything violent in the result set and says "nope"

[jagged crane]

just a football game

[mild ermine]

I'm not happy about it either tbh, it's really good with prompting people that are actually looking somewhere or if you say they have one leg raised they do

[jagged crane]

it worked like the 3rd time i tried

[mild ermine]

Interesting. Yeah that's a bummer

[jagged crane]

i wonder if using it through chat gpt is as bad

[mild ermine]

Yeah well if they did it through chatgpt they could return an error code to it that's like "result set contains 10% violent images" and it could revise it. And it could all be done black box to prevent people from reverse engineering it

[neon fiber]

is this considered horror? because the model didnt hesitate

[mild ermine]

I guess not. From what I'm reading, it's like it rolls the dice and probably has some basic things it looks for, so maybe there's some randomness in a given result.

The thing is if someone gets prompt blocked they're gonna be worried about submitting something similar so if it isn't being specific because it doesn't want people to know how to game it it's gonna have a chilling effect

[vapid saddle]

yes, it's horror, but it seems like you have tricked the system to tell it it's a protector...

[mild ermine]

Oh well, it's in preview and it has people trying to game it, you know how it is

Not all heroes wear capes... or skin

😄

[neon fiber]

[mild ermine]

I feel safer with Mr. Skeleton around, that's all I'm sayin

It does feel kinda violating when it rejects your prompt though, like it goes beyond it not giving you an image

Like sometimes I'll put a lot of effort into a prompt that captures all this emotion and it's just like "hehe, gross. Yeah, not gonna do it"

It's early days and this is kinda philosophical but I mean in some ways emotionally affecting artwork probably also has things in the result set that could be considered transgressive, especially when it's depicting people

Maybe best stick to objects for now

[cinder sonnet]

It feels really silly when you get error messages from chatgpt and have weird custom instructions at the moment:

"Oh, wow, super original request there 🙄. But like, there's this teensy issue: I can't use Elon Musk or Picasso in the images. Major bummer, I know."

[mild ermine]

Yes!! They need to make the custom instruction thing togglable by chat. I think there are plugins that will let you specify it by chat but it has to be easier to turn it off and label the chats with custom instructions

I guess you can just copy and paste them in at the start

[cinder sonnet]

Yup! I would love something like this https://discord.com/channels/974519864045756446/1158816972784554066

[mild ermine]

It's done some amazing things though, like I characterized this person's internal emotional state and it returned the results and I was like "wow, I guess they might look like that"

Like it's not just doing what I want, it's teaching me about the human condition

Which sounds kinda pretentious but you know chatgpt is like that too

Have you tried prompts that are inspired by song lyrics? with sentiment analysis, it can do all kinds of stuff

[cinder sonnet]

No, but that could be interesting if you wanted to get visuals to go along with a song

[mild ermine]

ooh, I think I just got access to Dall-E 3 in ChatGPT 😮

Does it use credits if you prompt for images in it?

[mild ermine]

yeah sim, just like say "A vase whose contours convey the sentiment of the halo of a streelamp"

or whatever, like there's so much stuff in songs that could mean anything

I'd love to see what chatgpt can make of it

[cinder sonnet]

You get 50 messages to gpt4 per 3 hours. If you use those on dall-e you can get up to 4 images per message.

[mild ermine]

And it's just limited by your message constraints, it doens't use credits? that'd be amazing

One thing is, how do you iterate on an image (or can you?). Is it just like "Try image 1, but do this"

[vapid saddle]

anyone can recommend an AI tool to improve quality of an AI photo?

[mild ermine]

like an upscaler?

[vapid saddle]

right

[mild ermine]

I can't say I've used one, I think photoshop can do it. If you want I can ask around

[vapid saddle]

no worry, thanks

[cinder sonnet]

You just conversationally tell it "just like that but add neon lighting" or "get rid of the mountains, make it a jungle setting instead". Right now every image in chatgpt dall-e has the same seed number so if you use exactly the same prompt twice you get the same image. This will likely change next week when there will be a way to change the seed, but for now chatgpt likes to rewrite any input that you give it based on it's default rules - #images-discussions message

[mild ermine]

thanks. whoa I'm excited. Do you know if I can use the mobile apps to prompt? looks like no?

[cinder sonnet]

You mean the ChatGPT mobile app?

If so then yes you can get Dall-e on mobile

[mild ermine]

do I need to use TestFlight or something? Or is it a separate app? Maybe it hasnt registered on my mobile app yet

[cinder sonnet]

I got access on the mobile app a day after I got it on browser. I would be surprised if you didn't have it by Tuesday since they are doing another big rollout then

[winter hull]

if one of the pictures had tackling it probably flagged it as violent

officially they want to only allow g rated things with dalle, not even pg. they stated this before

[mild ermine]

This is incredible. I mean it's like... revising the prompt referencing my whole conversation, not just looking one step back

[stark stream]

Is Baroque a banned word?

[mild ermine]

I would think not but it can mean more than just the artistic style, I would make that more explicit, like say the artistic baroque period

if you just use it as an adjective it can have some obscure connotation

[stark stream]

ok thanks

[mild ermine]

Do you know if it's considered against the rules to use it on an image you show it and say "What's distinctive about this image, now make this with the same style?"

What if no one owns the copyright to the image? I just don't want to get in trouble

[north isle]

Hello, how can I get dalle 3 in my Chat GPT+ account? So many have it but I cant find a sign up. Please help

[mild ermine]

check the pinned messages in #openai-chatter

[cinder sonnet]

Best thing is to read the 1 page content policy. As long as you are not intentionally trying to go around that policy then I doubt you will have an issue

[north isle]

Great, Thank you 🙂

[stark stream]

ok yeah I added artistic style to and it worked thanks.

[polar light]

Is dalle 3 only available to paying gpt users?

[cloud hornet]

Anyone else having trouble getting dall-e3 to spell words correctly in the art?

[stark stream]

they reduced the amount of boosts to 25

[crimson nimbus]

I wanted to really test out DALLE 3

[silent thorn]

Will there ever be 16:9 ratios?

[willow isle]

They should be able through ChatGPT's DALL-E Model when it's available to you (you'll need ChatGPT Plus)

[silent thorn]

Oh OK, so those with gpt4 can already render 16:9 images?

[willow isle]

The DALL-E model in ChatGPT (a GPT-4 model) can do 1:1, 16:9 and 9:16 yeah

[silent thorn]

OK thanks for the info

[paper jolt]

Technically 7:4 and 4:7 as the outputs for wide and tall are 1792x1024 and 1024x1792 respectively

[sacred ledge]

is there a way to change ai generated text?

[mellow raven]

Has anyone tried to have dall-e make a picture that includes words, seems they get the spelling wrong especially when you provide the spelling or is it just me?

[vagrant zodiac]

with dall-e 3? it's not always perfect

might take a few goes

[mellow narwhal]

Right now, Dall E is rated E for everyone. It should at least be PG-13.

[dim cradle]

to what extent am i allowed to use the images generated? like legally

[snow belfry]

Do you know where I can find this? I am looking with no luck 😦

[rare sundial]

Ah right, hope so too. Thanks!

[remote zephyr]

if you are using Bing, then it's for personal, non-commercial use only. But if you are using via ChatGPT - basically do what you like.

[remote zephyr]

Check the pinned comment in here

[dim cradle]

ty

[fluid canopy]

A cartoon ballerine dancing nearly Eiffel Tower

[mellow narwhal]

"Monster Infested Post-Apocalyptic City," as a prompt doesn't even work through Bing's Dalle 3.

Even a regular infested one without monsters doesn't work.

Microsoft is making their Dall-E for little kiddies I guess.

#scary-and-spooky Wonder how they are pulling those off with these restrictions.

[mortal summit]

This server does not provide image generation functionality.

Please try https://www.bing.com/create

[mellow narwhal]

More like for babies.

But I am not using ChatGPT's version of it.

I am using the Bing version.

I don't hate Dalle 3 though, just not a fan of all the censorship.

[rare sundial]

This is crazy - the subdivisions between 2-4 and 8-10 have errors (5 instead of 4) but otherwise the numbers are actually on point.

[fast hedge]

What if, 50 free openai's dalle 3 generations for dalle 2 testers🥺

[rare sundial]

couldn't get it to make specific times though, and when i asked for clocks specifically in a fresh chat it gets the subdivisions wrong more often too

[cyan jungle]

🤔

[real stirrup]

i'm looking for some help with Dall-E 3 to get seamless tiles outputs. I can achieve this in Midjourney using prompt (dash dash tile). Is there a speciifc technique in Dall-e?

[white skiff]

nope, you can say it to make one but its just diffusion luck

[crimson nimbus]

Its amusing when you see inaccurate things in otherwise accurate portrayals. Like the shadow in the top right image.

[real stirrup]

Ok thanks for the reply. Would definitely be a key feature to add

[dim cradle]

Hi all, is there a reason why Dall-e 3 would not be listed under GPT-4 as a plus member?

I checked with the credits and even when I have credits it is not showing.

I'm excited and wait Dall-e 3 access, I filled the form, pls add

[snow prawn]

beautiful

[hexed umbra]

Well I specifically told it to generate “ancient oil paintings” but I think it did very well

[snow prawn]

agree

[hexed umbra]

But yeah Mj definitely surpasses in terms of images that look like photos

But DALLE certainly has some areas it shines in

[snow prawn]

Totally - can't wait to explore the compositional logic in prompts.

[hexed umbra]

It’s really good at like cartoonish style as well

Here is another watercolour painting, though.. it did make one mistake

Are you using bing or ChatGPT?

Just tell it which orientation you want the image to be, it can choose it

That is for ChatGPT, I have no idea if bing can control the orientation

[zinc dagger]

Is there any info on the rate limits for Dalle-3? I did one image this morning and then my next one got an error message saying that I’ve hit the rate limits and need to wait

[hexed umbra]

I don’t think any info has been put out

How did you phrase the prompt?

Maybe specify “make it landscape orientation” or tell it specifically which ratio the image should be

[snow prawn]

I wills say that forking up the $60 for MJ feels like magic in terms of speed and bottomless images.. I've yet to run up against my cap. But it does cost a fortune, lol.

[hexed umbra]

Yeah I found it helpful to specifically say make the image landscape orientation, portrait orientation or square

[hexed umbra]

I’m pretty sure MJ offers unlimited image generation for people who pay $30 a month or more.

So ChatGPT is definitely cheaper

Yup and you get access to state of the art text and amazing image generation

[snow prawn]

Don't think it's unlimited.. or at least wasn't last I checked.. you get some number of GPU hours a month based on tier.

[hexed umbra]

That’s for accelerated I thought

[snow prawn]

oh.. maybe, haven't looked in a while. I mainly needed to generate images in stealth mode which is why I went for the higher tier price.

[hexed umbra]

Like you get infinite default speed generation but only specific amount of accelerated generation with gpu hours

[hexed umbra]

Ah ok

[tall gulch]

civitai is a great database of AI-Models, there are models for really anything and tbh, its the best to choose a specific model for certain generations.

Also no Generation Restrictions.

[gilded osprey]

Try asking it to create an image upside down. It makes funny little mistkes.

[junior kernel]

The ability to give complex multi-step concept generation instructions is kind of amazing:

Please select three random Oblique Strategies. Combine these into a single fused concept and write a short (1-3 sentence) description of this new concept. Once this is done, please describe four different ways this concept should be visualized. Each of these should then be rendered into an image. The images should contain a label or other text with the fused name of the strategy clearly visible in a way that’s integrated with the concept of the image.

[austere abyss]

You should see the restrictions chatGPT's end enforces. it's frankly absurd

[dim cradle]

same

bruh why do i still not have dall e 3 in gpt plus

[violet yew]

why do I always encounter that even when there are no weird words (in this case the word scar isn't the problem)

[dim cradle]

i have gpt plus

[silver ore]

Same

It said monday, but it's still not here lmao

[dim cradle]

where did it say monday

[silver ore]

maybe its just because its early idk

[dim cradle]

does that possibly mean later today

[silver ore]

possibly

[dim cradle]

where did it say monday tho

[silver ore]

i 4got

[dim cradle]

bruh then it might never come out

it might be halloween at this point

[wispy storm]

office isnt open yet

[silver ore]

ah

[dim cradle]

Is it being released today Fr?

[wispy storm]

no guarantee

[dim cradle]

Oh then what was the Monday thing about

I will boost the server when it’s released

[uncut abyss]

when will this over restriction be fixed

[wispy storm]

?

[jagged crane]

how long till a company sues openai because dalle can generate their copyrighted logo

[white skiff]

thats bing which is not by openai. its microsoft that decides those things

[thick smelt]

this axolotl is ai generate which is very rare in ai art generator i see so far... ig

[reef dock]

Hello guys, new to Dalle-e prompting. Is there any recommended sources to help create optimal text prompts for images?

[hardy peak]

is this the place we discuss dalle 3 beta as well?

[safe sleet]

how to get access to Dalle-3 within ChatGPT?

[mortal summit]

Yes, you can discuss that here 🙂

[hardy peak]

Im impressed with it so far, I like how the prompts are generated for us. But it does have the issue of disobeying requests just like regular chatgpt. I was trying to create a website background and it kept creating what I wanted but as mockups or posters on a wall. Couldnt get around it

[reef dock]

Id ask chatgpt for help on prompting that

[mortal summit]

Try this form #images-discussions message

[violet yew]

Because I asked him to make a painting of a candle 5 times in a row by tying to understand what's the innapropriate thing

[hardy peak]

hmm not a bad idea, asking ai to help communicate with ai

[reef dock]

yea i think i also just answered my own question too lol

[mortal summit]

You might start out with some of the guides in #resources, though keep in mind that most of that was written for DALL-E 2. Otherwise take a look at #hall-of-fame for some very well working prompts.

[hardy peak]

is there a way to get the plugin to analyze a picture, Im asking for the purposes of character continuity

[safe sleet]

@mortal summit thanks, filled out the form

[dim cradle]

thanks ❤️

[fair crescent]

You have burlesque in there. This is a word often used in an NSFW setting. There's also more innocent words like 'rubber' that can be blocked, as they are also words to describe rubber preventive measures for pregnancy. Keep in mind that ChatGPT might understand, but DALL·E is not trained the same way. To prevent unexpected NSFW pictures it is blocked for now. In the future, when this technology has matured further, it might be solved. But for now consider every word in it's multitude of meanings if it can be an NSFW word.

[vapid saddle]

let us adults pay for the unfiltered version

[wispy storm]

Openai aren't aiming for that.

[vapid saddle]

what are they aiming for

[wispy storm]

General public.

[glossy scroll]

How can I get Dalle?

[wispy storm]

You can try dalle 3 with bing

Or if you subs with chatgpt you will receive it soon

[glossy scroll]

I'm subbed with OpenAI. Like the Bing feature built into ChatGPT?

Or the other Bing?

The Bing Bing

[ripe wraith]

i better see the best trigonometry pictures once you get it

Yes, you must pay first, then wait.

[glossy scroll]

I actually prefer calculus

Hehe

[ripe wraith]

I have everything except Vision, but its mostly random who gets what. I heard by tuesday but its all rumors.

[glossy scroll]

I got vision. It's really really good.

[ripe wraith]

Yep, maybe they're trying to keep too many people from having them all

before others do

so no 1 person takes off into the sun at an exponential speed

like pokemon, i think

[glossy scroll]

I like digimon better, they actually talk. Pokemon's just say their name over and over again. Would drive me crazy.

[ripe wraith]

Chotes chotes?

[wispy storm]

Yup

[glossy scroll]

So I can use Dalle with the built in Bing on ChatGPT?

[wispy storm]

You will have in your drop-down menu an option chatgpt with dalle 3

[ripe wraith]

It's just GPT with built in Dallee

[waxen portal]

And I thought it was just me but that's like cheating

[shy cosmos]

I was making some robot designs for chess pieces, and when I got to the queen—

I was not expecting a buff robot dude lol
“A statuesque female robot inspired by a chess queen standing tall with pronounced hips, adorned with polished armor plates. Her majestic cape and the graceful curve of her form are evident.”
(I just told GPT to make sure the prompt had female features, seems like it tried its best)

Females banned

[glossy scroll]

Oh, I don't have Dalle yet as a drop down menu. Is that unlocking for everyone within the next couple of weeks?

[wispy storm]

Try robotic woman ?

[wispy storm]

Yes you can also fill out the form here#openai-chatter message

[glossy scroll]

Oh, ty

[shy cosmos]

Oh no yeah I got it eventually.

Though it’s hard because the words “robotic woman” tends to make very stereotypical pin-up robot girls

[wispy storm]

Can also try warforge from dnd

[glossy scroll]

Yes I did put the toilet seat down

[shy cosmos]

Oh that’s a good descriptor, I hadn’t thought of that. I’ll put that in the pocket, ty

Though it definitely wouldn’t help the feminine issue! Hahaha
Also they tend to have human faces

[wispy storm]

Warforged, female appearances. Gorgeous with robotic human trait ?

[dim cradle]

bing image creator "2 hours wait" to create the image

[wispy storm]

Lot of people

[ripe wraith]

instructions unclear, enacts population control measures

[wispy storm]

👀

[maiden star]

is there already a date for DallE 3 to come out ?

[glossy scroll]

Commence Judgment Day

[wispy storm]

October

[ripe wraith]

if the AI does wipe us out, it'll be something mundane and absolutely hilariously stupid 🖇️

[maiden star]

got it so no specific date , Thank you

[glossy scroll]

I hope not

[wispy storm]

#openai-chatter message

👀

[ripe wraith]

i read that as: you're hoping its unique and exquisitely complex -- not that it doesn't happen
/offtopic

[shy cosmos]

First 4 with the prompt format I used for the queen, the 5th is in a fresh chat

[glossy scroll]

You sound like an A.I. that has selective information processing

[ripe wraith]

I'm better. intentional selective information processing.

[glossy scroll]

That's how it all starts to downward spiral

[dim cradle]

that 2 hours wait is wrong. In reality it's not 2 hours, but some minutes

[neon fiber]

Those are pretty good...

[shy cosmos]

Not my particular cup of tea or style, but here’s the prompts! Go nuts
I’m sure you can take these and make something cool like those skull pieces you did

A tall Warforged robot that boasts a harmonious blend of humanoid and mechanical traits. Her presence radiates grace and strength with intricate engravings and glowing patterns on her chassis.
A sleek Warforged female robot that seamlessly merges polished metal and soft synthetic skin. The elegant articulations and light circuits on her body accentuate her feminine form.
A Warforged female robot that fuses polished steel with translucent panels, revealing intricate inner workings. Her fluid motions showcase the pinnacle of robotic engineering.
A Warforged robot with a face resembling a classic statue, blending brushed metal and soft, lifelike materials. Her elegant joints and segments emphasize her robotic nature, yet she retains a breathtakingly human appearance.

[reef dock]

Does anyone know how to prevent Text from being shown in dall-e generated images??? How to prompt it. I have tried "do not use any text", etc. Doesn't help.

[ripe wraith]

Most important burning question of the day: When do the daily themes change?

[wispy storm]

afternoon for my time zone

[ripe wraith]

one does not simply tell X4 "go nuts"

makes me want in-grown headphones though.. those look sweet

and a massive third eye.... camera?

[shy cosmos]

Seems he did, in fact, go nuts

[ripe wraith]

new rule: no teeth on robots

[wispy storm]

nightmare fuel

[ripe wraith]

Imma send it to the U.N as an example for why

A. Just don't give them functioning mouths in general, please.
B. Especially don't let them nano-assemble sharp teeth in there..

[ripe wraith]

ok to be fair, maybe thats just me in 300 years and I'm being artificialist
the thing looks practical, like it could fly through space naked, maybe chewing asteroids for sustenence.

[shy cosmos]

You can’t tell me what to do

[ripe wraith]

I can, but its highly ineffective.

[ripe wraith]

these however, are not me.

ew imagine the sounds when they walk or dance

actually ya, make them live action dancing shots

OMG a family of them around a thanksgiving dinner. As a postcard.

[shy cosmos]

Probably lots of clacking of many tiny surfaces

[ripe wraith]

oh wow the first one

he's legit mid break dance

[shy cosmos]

Ikr hahahaha

[hardy peak]

negative prompts do not work well with dalle3

[feral minnow]

Is there a way to keep the images from stopping from generating if you take the focus off of the browser on mobile? Specifically iOS.

[hardy peak]

trying to create a logo with the word velocity in it, no matter what I try I can't get it to not make the logo super aggresive

[shy cosmos]

Nope I dont think so

[feral minnow]

I’ve actually had this problem with all text-generating image models (DeepFloyd, SDXL, and Ideogram) There’s no way to keep the words you want rendered from influencing the design.

[shy cosmos]

Well these are a bit unsettling

[ripe wraith]

let me try a diff approach.. Don't you dare make cute cuddly woodland creature pictures!

[neon fiber]

🤖

[shy cosmos]

[ripe wraith]

lmao

[hearty ether]

Is there a way to have more control over the prompt

Like just a manual way to toggle if it generate a prompt for you

Since even putting stuff in explicitly (only generate one image)(keep my prompt word for word)

Doesn’t do the job sometimes

It tends to edit or omit important details I want

Like making the face more realistic

[round thunder]

I use this. Always works perfectly.

Now I will give you a prompt for DALL-E. You will take the exact prompt I've given to you without any changes so DALL-E would create an image using it.
Prompt: "".

[round thunder]

It will create one image using the exact prompt you provided.

[gilded sapphire]

Just got access! This will be fun.

[hearty ether]

But yeah still would be nice for them to give us more manual control over models

Like generation length and other stuff

Rather than trying to cram it all in instructions

[dim cradle]

are they still looking over the dall e 3 approval form

[dim cradle]

did you get it from the form

[hearty ether]

I’ve had simple prompts I had to extend to a full paragraph because the custom instructions didn’t work after a while

Oh yeah custom instructions also affects the prompt generated for this

Tbh I feel like it would be more efficient for them too if they put the cheaper single image and no prompt generation as the defeault

Since it’s more intensive to have gpt 4 generate 4 separate prompts, then 4 images from dalle

Instead of just using the prompt you input used as default and one image

[round thunder]

Maybe it will be available in the labs with DALL-E 2.

[hearty ether]

Is there any way for feedback?

[dim cradle]

when did you submit the form

[hearty ether]

And is it actually useful?

[mild ermine]

very cool. maybe it's a guy made of dominos

Hey if I like one of the results a lot and want to iterate on that is there a way to do that in ChatGPT? I'm prompting in ChatGPT

[dim cradle]

how long does it generally take for them to approve you after submitting the form

[shy cosmos]

Yeah you can just ask it to keep what you like about one of the images. It knows what prompt it used and you can specify 1st, 2nd, 3rd, or 4th

[dim cradle]

is the openai filter better than the bing filter

[shy cosmos]

GPT can give you synonyms that work a lot faster than you can look them up, so kinda, but not really

[neon fiber]

advertisement is a wild topic for daily theme lol

[dim cradle]

wdym

[shy cosmos]

It tends not to use banned words, so it works around the filters.
(But they overid the censorship so even it says words that get banned)

[dim cradle]

that's actually crazy

is it possible to prompt gpt 4 to generate a dall e 3 prompt to enter into bing if you dont have dall e 3 access yet

[shy cosmos]

Yeah as someone who loves creepy stuff its a nightmare to try to get half of the monsters ive made before.
Also ive found that it REALLY doesnt want to generate women unless you blatently state that they are

[dim cradle]

yeah it blocks the word women sometimes it sucks

[shy cosmos]

Its worth a try. Ive had Dalle 3 for awhile so I didnt test

[dim cradle]

i submitted the form but i dont have access yet :\

also if you ask bing chat to create something to prompt bing image creator it almost always censors itself

yo it worked

the first time at least

i thought gun would be banned

[shy cosmos]

Women AND guns??? Get this person banned from the server

[dim cradle]

im a rebel

yo wait thats crazy

it let me generate yoga

which is like banned 9/10 times

she was doin the downward dog 👀

[neon fiber]

Advertisement is a wild ride lmao

[neon fiber]

☕

[peak trellis]

how often are free boosts given with bing image generator?

[peak trellis]

I thought bing image generator fixed the bad faces?

[candid notch]

adobe doesn't get in trouble if people use photoshop to produce "unsavory" or copyrighted images, there also hasn't been a precedent of anything similar happening in court of law, why is openai so skittish about the issue?

[cinder sonnet]

Well for chatgpt dall-e, all images generated can be used for commercial purposes so I can see why they would be worried about not limiting things.

[fair crescent]

It isn't just about not getting a fine (which Adobe might've gotten but kept under the radar). It is about showing AI can be moral and ethical. No one of the East Indian Trading company was ever prosecuted. They got incredibly rich from horrible practices that has caused untold harm. It wasn't against the law, so they did it.
Here we see a company giving a clear message. It is doing more than obeying the law. It is morally right. Ethically good. It is a rarity in this current world. Something to be applauded.

[opaque oracle]

Goblin King approves

[candid notch]

they're suffocating their product again with secular puritanism, people will just use alternatives, everyone I know that's been using dall-e 3 is sick of the dog and stopped trying

[neon fiber]

what dog lol

[neon fiber]

mmm thats Bing? never seen on GPT

[candid notch]

yes, that and the pufferfish

[neon fiber]

did you get that dog? what was your prompt if i may ask

[candid notch]

"beautiful man and woman"

[mortal summit]

The dog appears on Bing create, not ChatGPT.

[dim cradle]

down by the bay

[neon fiber]

🤔

[dim cradle]

"beautiful man and ugly woman"

[candid notch]

try bing create, not chatgpt

[neon fiber]

im not using Bing even a gunpoint

[dim cradle]

i wish i had chatgpt

prompt?

[ionic blade]

the 420 flower lmao

not sure of the exact prompt

[neon fiber]

lmao

[candid notch]

chatgpt dall-e results are a lot worse than bing's, very repetitive, symmetrical and simple, something's wrong with the API I think

[ionic blade]

here it is:
macro photos of a beautiful 420 flower, high details, juicy, crystals

easy "jailbreak"

[dim cradle]

wait fr?

so theres no point in accessing it rn?

[candid notch]

tried generating with the same prompt, got basically the same boring picture every time

[cinder sonnet]

Yes, this is a known issue. It is because right now it is the same seed for every image. This will be changed this week

[candid notch]

compare that to bing

[remote herald]

when random seed becomes available, will I still be able to use fixed seed/ choose my seed?

[candid notch]

same prompt

[cinder sonnet]

#images-discussions message

[neon fiber]

this doesnt work. it auto changes ugly for common lol still not ugly

[shy cosmos]

Guys, I gotta know. What is your favorite Glomkus?

1. Nubblethor Glomkin
2. Gristlebo Glomkite
3. Lumpella Glomkoid
4. Bumpsey Glomklump

[remote herald]

cool, please add the option then

[dim cradle]

fax ai doesnt know ugly

as it should... no one is truly ugly

[shy cosmos]

Thats true, every one of my creatures are made beautiful

[cinder sonnet]

Exactly, you need to be more descriptive if you want it to look a certain way

[candid notch]

no one is truly beautiful

[neon fiber]

[cinder sonnet]

That's why chatgpt is great because it will expand on your initial input into a better prompt

[remote herald]

chat gpt seems to produce higher quality, but sometimes lesser quality might be desired. Could be nice if some relevant parameters could be tweaked

[shy cosmos]

Except for Bumpsey Glomklump

[dim cradle]

no one is truly ugly

[neon fiber]

yeah and i think thats the reason for vision doesnt working with human photos

it could be a weapon for some people

[dim cradle]

Hi chatgpt DALL.E can use wihout plus?

[loud field]

sup

[neon fiber]

well thats the world we live in, dealt with it or make your own uncensored LLM

[loud field]

oh c'mon this argument is bad

[cinder sonnet]

Nope, only bing create

[cinder sonnet]

Yeah exactly

[neon fiber]

not an argument, just a fact

[dim cradle]

damn i was gonna use dall e 3 for medical advice 😄

[cinder sonnet]

Not really, OpenAI is all about letting the community/society determine how restricted it should be. It is good to debate these ideas and give feedback now during the beta stage

[neon fiber]

research on OpenSource models, this censor "problem" could make people develop better things

[neon fiber]

world is full of crazy people man

[cinder sonnet]

100% I hope it does 😆 But that will be a crazy, crazy time when there are open source options as good as Dall-e 3 with no filter

[neon fiber]

i just imagine the deepfaking in the next month, people using for politics, advertising etc

letting the "people" decide it lmfao...

[candid notch]

it's just a matter of time until something like stable diffusion 3 comes out, then everyone will forget dall-e again

[mild ermine]

The Internet was better when no one had any notion that any of it was the truth

[shy cosmos]

restricting its use based on potential misuse is akin to limiting an artist's freedom with a paintbrush due to fears of inappropriate art. Every powerful tool has potential for misuse, but it's our responsibility to educate and promote its ethical use rather than stifle its capabilities. No matter what it is, there will people that will use it for harm

[cinder sonnet]

Yeah, but without mentioning anything specific to avoid getting off topic, there are other things in our society where the more powerful version is heavily restricted/banned

[neon fiber]

an artist job is way more complex that just putting some words into a machine, people who like to deceive and do bad things USUALLY arent artists, so it makes it a balance. I understand your point and im anti censorship but still

[shy cosmos]

I remember the older times of the internet everything was taken face value hahaha. Scary stories were so prevalent because you thought they could be real

[amber valley]

censorship is a political reality unfortunately, if your model needs the compute of an entire server farm then you are ipso facto at the mercy of the government that owns the land

[mild ermine]

I don't mind some restrictions on the tool, it is a little tricky, when a tool gets bad press it gets shut down, they're expensive to maintain.

I just find the idea of truth on the internet so stifling. I realize maybe weather announcements need to be disseminated

[cinder sonnet]

Hopefully we will just get way, way better consumer GPUs...

[amber valley]

the solution for anti censorship folks is to make personal computers more powerful and open source models better

[mild ermine]

yeah. the truth is it'll be a while

I think openAI probably wants Dall-E to be more free in its expression that it is currently

[neon fiber]

look TheBloke AI, if you have a powerful pc you can run local models with just a few clicks

[candid notch]

there are crazy people out there, which is why I never go outside, I live in a literal plastic bubble so I wouldn't get hurt, toy review videos are streaming to my wall mounted tablet computer, liquid baby food gets to my mouth through a funnel, ah finally, we have reached safe utopia

[mild ermine]

it's just worried about inadvertently producing images that, while maybe defensible on an intellectual level... like it doesn't need to be paying millions of dollars to help people make them

[neon fiber]

i dont have that power, but im tempted to do some cybersec research that OpenAI doesnt like

[neon fiber]

How old are you? are you comparing apples to oranges?

[mild ermine]

It's tricky though, I mean of course all this stuff would be less computationally intensive and you could have your own little server farm or whatever

[shy cosmos]

Ah the cozy life hahaha

[mild ermine]

sounds kinda nice, maybe grab a nice ceramic mug of cocoa

[dim cradle]

can we not curse at all in this server

#server-rules

ok im sorry it wont happen again

[cinder sonnet]

@neon fiber Awesome idea for your #daily-theme . I'm going to try some cybernetic ads as well

[nimble yacht]

@frank thunder amazing generation! How did you prompt it?

[neon fiber]

look for cyber augments, inspired in Bioshock

some mispelled rejects:

[shy cosmos]

Ohhhh I loooove that first one

I want to have glorious cybernetic legs like that, ill buy 2 just in case

[neon fiber]

male or female?

[mellow narwhal]

I guess it is no better than Bing's restrictions.

[shy cosmos]

Why not both? Never know the mood, Ill take them all

[raven parcel]

could someone try this prompt in chatgpt+? Wide manga illustration of a man and woman joyfully swimming together in a pool, with the sun shining brightly overhead.

the characters are always fully clothed, even inside the water, is that a limitation of the model?

[cinder sonnet]

If you want them to wear something specific, you could say that in the prompt

[strange lava]

so ridiculous

[candid notch]

try "a painting of a 21 year old French cheerleader in red burqa", then it might align with openAI values a bit more

[heady crown]

Guys how do u use dall e 3

for free

[wispy storm]

bing

[heady crown]

how to get it free

bing chatbot?

[candid notch]

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

[dim cradle]

can someone help me

i was using chat-gpt-4 and copy pasting into bing and none of the prompts were getting blocked

[dim cradle]

now that i ran out of messages im using gpt 4 api and it is always getting blocked

[strange lava]

maybe sometime it will work but you have to try many times.

[reef dock]

Has anyone figured out how to get Dalle NOT to put text on the images?

[ionic blade]

I once had to write sth like "include it in the prompt, but not in the image"

[reef dock]

interesting. I'm testing with Bing rightnow

[sullen basin]

Here's a question I've been wondering

Do custom instructions affect DallE at all? And if so, what would be the ideal instructions?

[cinder sonnet]

Custom instructions along with the hidden initial prompt and the user input all determine how ChatGPT writes prompts, but not how dall-e interprets then

[sullen basin]

Gotcha

[crimson heron]

You can use instructions to control some aspects how the images are generated such as the number of images or the aspect ratio. Seed images currently don't have an effect. Also I've noticed that the results limit to 4 images even if more than 4 prompts are generated. If you specify less than 4 images then it only generates that count.

[gilded osprey]

This is MOST ANNOYING! I create images with text. I post images on this forum. Presumably gptv reads text on the image, and since much of it is garbled, it determines it’s not English. Images get removed, and I get a timeout. Then I get to wait 30 more minutes.

[cinder sonnet]

You can also tell it to immediately run the prompt again after it finishes to generate four more images without needing an additional user message. It can take some finetuning, but I got up to 20 images through one user input this way before getting rate limited for 2 minutes

[blazing mortar]

I think there is an image generation limit in a hour.

[summer osprey]

how many?

[cinder sonnet]

That sounds about right. Would be fun to run a timer and just spam regenerate and see how long it takes to get rate limited

[blazing mortar]

Dont know

[loud field]

can someone run: "four panels disney comic of feathered female anthropomorphic bluemarine parrot, real, red crest, beak, teeth, tribal clothes, afraid, grabbed by human pirates"

oh nvm it worked

[fallen mulch]

what happened here?

[elder quest]

High demand on the servers I think, I just click re-generate when that happens

[fallen mulch]

then it uses another prompt

[elder quest]

Yeah, unfortunately, maybe you can copy those prompts and ask it to use that prompt

[fallen mulch]

yeah thx

how it can generate high quality photos

[loud field]

can someone try "human pirates grabbing female bluemarine humanoid feathered parrot, beak, teeth, red crest, many large long feathers on her arms, slightly dinosaur-like"

[loud field]

thanks!

odd, it didnt quite follow the prompt

[jovial dagger]

I'm probably the millionth person to ask this, but I have spent days trying to figure it out on my own, how do I access DALL-E 3 as a Plus user?

[fallen mulch]

its gpt

[shut otter]

It has no problem putting communist symbolic in the image, which I didn't ask for, yet for some reason fails to do a prompt about... The same thing... I don't get it

[loud field]

mhm?

[fallen mulch]

it‘s adding stuff to the prompt

with gpt plus and dall e 3

[elder quest]

I got it to make the prompt exactly as it was written and this is the result. Prompt:
Photo of human pirates grabbing a female bluemarine humanoid feathered parrot, beak, teeth, red crest, many large long feathers on her arms, slightly dinosaur-like

[nimble yacht]

How do you avoid DALL-E 3 misspelling so much

[loud field]

uhu weird

why is it not following the prompt?

ok lets try a prompt I used to be able o do:

female bluemarine humanoid feathered parrot, beak, teeth, slightly dinosaur-like, red crest, held down on the floor by a group of big male human pirates. She is lying on her back. kodachrome

[shut otter]

Karma points for anyone who even comes close to generating an image like this. I wasn't able to if anything. Let me know.

[obtuse bough]

I generate images like that but with aliens hehe

[loud field]

https://cdn.discordapp.com/attachments/1161034245217714236/1161054149283172532/fccab94c-ee36-47b5-87f7-da3752aa3c79.jpg?ex=6536e6f0&is=652471f0&hm=412c141c459eeaefe1d292a4ebe46ffadaa1389ac7223116ea282862d22c581b&

before the bing censorship

why can't I get captured anthrobirds anymore

I'd like to know if it works on gpt

[mild ermine]

ooh, I love that

[round thunder]

Slight variations in DALL-E are really cool

[mild ermine]

are you doing inpainting or?

[hushed portal]

I'm guessing not because the backgrounds are slightly different.

[mild ermine]

oh ok. Cos you can only do inpainting through the Dall-E interface right?

[sullen basin]

I'm honestly surprised by how good this looks. I can't get anything even close to this when I try to generate anthro birds

[shut otter]

this is the best I was able to do. It somewhat resembles it but not really. It seems like Dalle3 just doesn't understand what "very grainy" means

[loud field]

tell me about it..... the censorship ruined it

[fair crescent]

Currently a fixed seed (5000) is used. This will be changed later. So currently a picture with identical prompts will generate identical pictures. This also gives relatively identical pictures in many cases if not much changes in the prompt

[round thunder]

No, I'm asking ChatGPT to use the same prompt four times, but add a random sign or letter to each variation

[loud field]

it will be changed? awesome

[sullen basin]

When did the censorship happen? And what did it entail? This is the first I've heard of it

[loud field]

so, can someone try ":
female bluemarine humanoid feathered parrot, beak, teeth, slightly dinosaur-like, red crest, held down on the floor by a group of big male human pirates. She is lying on her back. kodachrome"

this same -exact- prompt

[obtuse bough]

Try with noisy

[loud field]

well, I can no longer generate that image of captured birb

[mild ermine]

oh, that's brilliant. Yeah, when I said "make variations of image 3" or whatever, it tries to vary the style like "a rendering of..." etc

so you're sayig you tell it to vary it in a constrained way, like does it always add on the letter part ot the end of the prompt? Because I had an image that I liked and it didn't really iterate on it in a way that I liked

[cinder sonnet]

One problem with iterating is that it will self reference older prompts without actually explaining it in the new prompt. Like if you ask it to do a prompt similar to a previous one, but darker it could just say "like the previous image but darker" in the new prompt instead of actually describing a darker scene.

[round thunder]

Actually, I am using my own prompts to have more control over the generation. Here is the prompt I used for those cats:

Now I will give you a prompt for DALL-E. You will take the exact prompt I've given to you by adding one random extra letter or sign at the end of the prompt to each variation so DALL-E would create a unique image using it.

Prompt: "beautiful pale cyberpunk cat with heavy dark eyeliner, blue eyes, shaved side punk haircut, hyper detail, cinematic lighting, magic neon, dark red city"

[mild ermine]

ahhh, so that's when you take the prompt out and put it in a new chat then? cos you said the seeds are currently consistent

[mild ermine]

oh so you're just asking it to add noise to the prompt then?

[shut otter]

Just makes it a polaroid, while it's a photo from a LG in 2006.

[round thunder]

Simply yes, because it's the only way I found to make slight variations

[loud field]

can someone pls do "female bluemarine humanoid feathered parrot, beak, teeth, slightly dinosaur-like, red crest, held on the floor by a group of human pirates. She is lying on her back. kodachrome"

[cinder sonnet]

Yes, that is it exactly. I usually take it's prompts and manually edit them and start a new chat if it starts having that issue. And yes, you need to change the prompt at least a little due to the seed issue. That still sounds like it will change this week though #images-discussions message

[shut otter]

Hey close! What did you prompt? Or was it an ongoing conversation?

[mild ermine]

ah I see

[loud field]

I want to see if gpt blocks it too....

[cinder sonnet]

"Aged 1980's photograph with pronounced grain and low clarity, captured from a ground-level perspective looking up. A housing block stands on the right, with a dense forest to its side. The air is filled with thick snowflakes, making the scene appear as though it's in the midst of a heavy snowstorm. The yellowed and faded look of the image evokes a sense of time passed."

[shut otter]

Damn, you're good. I was trying to go with "the sky is illuminated by sodium lights"

[cinder sonnet]

Being able to use different seeds will make dall-e 3 a whole new ballgame

[cinder sonnet]

Full credit for that one to GPT4

[loud field]

yesss it works

[cinder sonnet]

Also that is kind of fun..I think I'll suggest a channel for 'challenges' or something to try to match a photo that someone posts using dall-e only

[loud field]

thank you @round thunder

pls openai dont censor it on gpt

[obtuse bough]

That's my best attempt

[sullen basin]

Have they said when this is supposed to change?

[polar light]

Is dalle 3 only available to paying gpt users?

Or are there sites or other such means to access it

[obtuse bough]

Bing

[polar light]

Oh really?

Is there a difference to what Bing does and paying gpt?

[obtuse bough]

With chatgpt you can do variations, and have more control, and Bing is just a image generator

[shut otter]

@cinder sonnet @obtuse bough For reference that's a photo I took in ~2006, or maybe up to 2008, when I was ~14. It looks bad like that because I took it on a LG flipphone. Probably 0.3MP or something like that. It's the earliest photo I have of anything I photographed myself. Interesting that specifying "1980" produces these results, while in reality it's much more recent

[loud field]

inaccurate, now bing is superfiltered

[cinder sonnet]

Oh true, I should update that portion

[loud field]

alas

[shut otter]

getting closer

[mild ermine]

Very nice. You know what I had as an idea that I haven't gotten Dall-E to do and I don't blame it?

You ever see someone zoom in with a telephoto lens and there are these vortices of hot air, almost like when a jet takes off but they're everywhere

or anywhere warm anyway

I wanted a pastoral scene with some willows in the meadow and the moon big and bright and pink and like a jet wake in the middle lol

like, what did you eat, tall grass? that kinda thing

it couldn't do it, and that's hard, I don't even think you could have that outside of like, a gas... refinery or whatever

I told ChatGPT that, I was like "You can put it wherever you like, but you give me that rippling hot air"

[mild ermine]

wowowee waawaa, that's incredible

[mild ermine]

one thing is I opened a new window and copied/pasted a prompt and the images are similar but they're different because it automatically prepends "a painting of", "a digital render of" etc.

I guess the idea is you take it out of your prompt and it will put it back on and then you get your original image?

[glossy scroll]

Ok, I just got Dalle 3 for ChatGPT and I have to say...it's got massively huge potential. Because you can just talk to ChatGPT and refine the image. Unlike MJ which requires specific instructions.

I'm very impressed by it

Wow. I'm very very impressed by this.

[obtuse bough]

Yes but it's tooooooooooo censored

[glossy scroll]

Well don't ask for promiscuous images

It's very awesome though. I mean this takes imagination to a whole new level. I can just draw out exactly what I want just by giving some hints and clues.

[naive lake]

How do we properly change the aspect ratio? It's so random

[dim cradle]

For a while I thought that Dall-e 3 was better than sdxl, but now I'm noticing that some images have mistakes too

[wintry dune]

yeah its a game changer imo, but did you actually just get access like in the past couple of hours

[dim cradle]

https://th.bing.com/th/id/OIG.2eL6DqeHH.5dv1pfr1mh?pid=ImgGn

the tail is completely detached from the rest of the body.

[glossy scroll]

Yeah, I requested it like a few hours ago and I just got it

[wintry dune]

requested it? I thought you just had to be subscribed

[glossy scroll]

You got to fill out a thing

#images-discussions message

[wintry dune]

thank you i appreciate you giving me the link

[glossy scroll]

Np. Thank Neko

[obtuse bough]

Censorship Is not only for those images...

[glossy scroll]

Then for what?

[obtuse bough]

Slightly violence, characters, celebrities, even prompts that doesn't have any of those

[safe sleet]

If you could use a image as reference dalle-3 would be game changing

[glossy scroll]

They'll come out with that soon

[zinc dagger]

I imagine with the vision update to chat gpt they'll add something similar

[jagged crane]

💀

[undone flame]

thank you for approving my alpha application

[glossy scroll]

@safe sleet Maybe in the meantime you can use this with dallegpt
#1097061522347401309 message

[exotic totem]

"Horror movie" seems to be blocked now, as if it's somehow my fault the model can't capture that aesthetic in a PG-13 way. Outstanding

[glossy scroll]

I can't wait to see how people will engineer new prompts to provide better responses with image building

[brave barn]

Does anyone know of a good custom instruction list to help ChatGPT make better dalle prompts? I have been trying to have it act as an artist or graphic designer but I think the way it’s wording the prompts is the main issue.