images-discussions

[open wind]

Is there a way to use dalle e 3 the same way as dalle 2 in regards to using living artists as prompts?

[remote stone]

I think most artists whose last work was within the last 100 years are blocked, the only way to get around it is to describe their style

[agile peak]

This is my fav; definitely would love to have one as collection haha. Turned out using certain prompt writings can improve the generate results.

[civic crescent]

Have you guys used Googles new ai image generator?

[lyric flower]

Hi, why doesn't dall e 3 sometimes generate public persona, then when asked again does it? then sometimes don't do it, it's too weird

[cinder sonnet]

#ai-discussions

[cinder sonnet]

Creating images of public figures is against terms of service so it will try to prevent prompts that ask for that

[lyric flower]

OK, but why sometimes does this work?

[hearty ether]

Huh

Apparently this is what a m1 tank at a multiversal 7th dimensional level looks

https://fileserviceuploadsperm.blob.core.windows.net/files/file-DhStShcAf6HD9vdQcHlCAP1t?se=2023-10-17T21%3A57%3A54Z&sp=r&sv=2021-08-06&sr=b&rscd=attachment%3B filename%3Dd898539e-11d5-4a93-9fd6-bda692809c5a.webp&sig=y3rL2jK48FWJRfPOZJgjiF4atzgr5jwcnyEFUvGw60M%3D

Love how it handles abstract elements

[cinder sonnet]

Because the onus is on the user to not put "input" in that violates TOS. Dall-e itself can easily create content that violates TOS due to how diffusion works

[lyric flower]

suddenly when the beta is finished we will no longer be able to "enjoy" those glitches ?

[hearty ether]

https://fileserviceuploadsperm.blob.core.windows.net/files/file-9JrX6Yd2DBxObabg1zpGEjvm?se=2023-10-17T21%3A41%3A40Z&sp=r&sv=2021-08-06&sr=b&rscd=attachment%3B filename%3D6a6dcd56-ebd4-4867-b0cb-ef56f1644ffa.webp&sig=o0OOfsOjcrI23O9VtysEJ3O%2BjYM5go0xoihn9EhpIHY%3D

I mean tbh this is the version before it

Those lights are stars 🤣

[cinder sonnet]

you should report them in #1070006915414900886 so that the 'glitches' can get fixed

[half field]

Cool thing

[scarlet urchin]

what a cool aesthetic, adding random words really does help

[oak briar]

Hello, world! How can i have access to dall-e 3 being plus? For now i use it in bing and in cellphone. But wish I could see the extent of the real thing. Tyyyy

[plucky hare]

I'm not sure if Moxi is still letting people in from here but you can try!

#openai-chatter message

Otherwise fingers crossed to full rollout soon, I think we must be close!

[barren comet]

Hello

AI enthusiasts

[agile peak]

Image generator in 1980s

[barren comet]

Please, is there any newbie here??

I need someone I can partner and learn with.

[zenith girder]

e

[naive lake]

Rate limite has dropped significantly huh? Also generations are taking longer. ANyone knows what's going on? Thank you!

[calm idol]

Probably just too many people playing with DALL-E to see what it can do and putting a heavy load on the system. This stuff takes a lot of computing power.

[dim cradle]

For what?

[drifting hawk]

it should be out to all plus users

[drifting hawk]

compared to when? i dont see latency issues

it isnt heavy load right now

[arctic matrix]

The seed issue in gpt4 Dalle3 is really annoying. I hope a fix is in the works.

[drifting hawk]

we're working on it!

[dim cradle]

The censorship?

[arctic matrix]

No, my issue is that it uses the same seed for every image. So if you tell it to use a prompt exactly as written, it will generate the same image multiple times.

[uneven niche]

fixing the seed is great for iterating on a single image with a stable composition but not great for exploring possibilities/ideas with the same prompt

i'm hoping that each image output has the seed recorded and visible so we can specifically pass in a seed corresponding with an earlier generation - it'd be annoying if we had to tell ChatGPT "use the same seed as the fourth image from three replies ago"

already a little dismayed that we can no longer see the prompts being written in real time, that was really fun - it actually showed whether the first image was being generated in landscape or portrait too, so it was useful to see whether that was correct or whether chatgpt had forgotten the "please generate all images in landscape resolution" instruction

[hearty fog]

will we be able to prompt with images at some point?

[tight verge]

you should be able to if you have a plus account

[paper jolt]

Sending just a vague prompt to DALLE-3 results in quite terrible generation
Bing Image Creator seems more coherent
Perhaps Bing Image Creator is also modifying the prompt like ChatGPT normally does

[agile peak]

Looks like I have found the way using custom interactions to make GPT be consistent with improving the subject.

Changed the eyes, size and pose. Pizza-themed and make him cuter.

[naive lake]

Compared to when I first got in.

[dim cradle]

I think I've found a loop in dall-e 3. When I ask to generate dimensional portals, the AI seems to not understand what is inside and what is outside. It created a recursion with portals within portals

https://th.bing.com/th/id/OIG.uh_Ba3tRhkT0nfmb0_jx?pid=ImgGn

[lime ocean]

Yes bing is better

[paper jolt]

You completely ignored what I talked about 💀

[wanton hatch]

That's Microsoft Bing, not OpenAI.
You're in the wrong Discord

I mean you should ask Microsoft if the waiting is so high

No this is the Discord of DALLE3 model used in OpenAI.
The question regarding the waiting is for the Microsoft Bing website that uses another DALLE3 model

[dim cradle]

https://th.bing.com/th/id/OIG.ptVWgZ5mOuxffa_f6IYT?pid=ImgGn

the prompt is so simple, yet every time it generates a portal with a split in the middle

[left sparrow]

What's the prompt?

[dim cradle]

A landscape with a circular portal to another dimension. Inside the portal there is a different world and landscape with the sunset. Outside the portal a storm with lightning.

funny. SDXL also creates portals within portals

[paper jolt]

make a specification regarding the portal, perhaps along the lines of "even coloring"

[agile peak]

He went from italy to indonesia culinary

[dim cradle]

Amazon is painting the town red

[naive lake]

've been generating lol

[arctic matrix]

I think we will have to wait until the seed issue is fixed to tell. It could just be that bing is getting luckier with the seeds rather than modifying the prompt.

[dim cradle]

I think I'll skip the word "portal"

https://th.bing.com/th/id/OIG.aaW_LXdJqeknZ45PTvSN?pid=ImgGn

I replaced portal with "a circle made of turbulent blue energy", which the AI extended to "turbulent waters"

Maybe "create two different worlds, one inside, the other outside ..."

[naive lake]

Nothing beats Dall-e 3

[left sparrow]

@dim cradle this is the best I could generate which still isn't great. A detailed scene where the entire background is filled with intense volcanic activity, with molten lava and thick ash clouds dominating the view. Centrally placed in this chaotic backdrop is a pristine circular frame. Inside this circle, there is a complete absence of the volcanic disturbance; it solely presents a peaceful landscape of a sunny green plain under a cloudless blue sky, standing as a sanctuary of tranquility amidst the surrounding volcanic chaos.

[paper jolt]

DALLE-3 likes details, shown by law of non-contradiction as per attached image
AI had to choose something and turbulence is often affiliated with water

If you want something specific be specific, otherwise it is arbitrary

[left sparrow]

I think if I could get more seeds or more control over generation parmaeters I could do better but we are quite limited right now

[naive lake]

how so? I don't get it I'm amazed with what dall-e can do.

[left sparrow]

Sure, but compared to other less sophiscated but more controable models, there's much less control here

[lime ocean]

Haha its zoomed in then i took a screen shot, but ill try find the original wait a second

[naive lake]

Like which ones?

[left sparrow]

SDXL, which gives you control over prompt strength and sampling steps which have drastic effects on the output

[naive lake]

My only issue is the resolution! I wish it could handle double the resolution with same quality

[left sparrow]

But its layout ability is far inferior

[paper jolt]

The AI upscaling app Upscayl is quite nice 👀

[naive lake]

SDXL.. hmmm.. I don't know man

[naive lake]

I use topaz

[left sparrow]

Well SDXL derrived models if you prefer

[paper jolt]

Issue resolved... already!!

[left sparrow]

the maximum size is 1792x1792 at the moment

[naive lake]

I don't think it can do square 1792x1792 I wish it could

[paper jolt]

erm... 🤓

[left sparrow]

I meant for Dalle-3

And It turns out its 1792x1000

[paper jolt]

Yes

[naive lake]

yeah I wish it did square with that resolution

life would be perfect

haha

[left sparrow]

For SDXL you can use the hires.fix to upscale it

with good results.

[paper jolt]

couldn't have determined that 😔

[left sparrow]

sorry, I was carrying on multiple conversations 😄

[paper jolt]

I understand 🙇‍♂️

[naive lake]

Does anyone know if Dall-e 3 will handle highest resolution in square format anytime soon?

[atomic cove]

Is this the right place to ask for an enhancement? I'd really like to have an undo function... For times when I don't like the direction my prompts are taking it, and want to branch off from an earlier point in the response images

[left sparrow]

Well if you click on the image, you can see the exact prompt, copy it out and ask it to restart from that exact prompt

and then continue.

[atomic cove]

So I'd have to be going down the single image path, from the outset, it sounds like.

Is there a way to tell it to use one of the four images, and continue on from that one?

[cinder sonnet]

#1070006151938314300 and yes, you can tell it to emphasize one of the four. I often say "that third one is great, make variations on that please"

[atomic cove]

How are they numbered, top left, top right, then lower left?

[fair crescent]

You can request to continue from a certain point. Ask to start from the first in the chat, 3 generations earlier or the one where you asked x

[atomic cove]

Ahh, generations is the word I was looking for

[cinder sonnet]

Sometimes it will try to reference previous prompts in a prompt. Like it could say "lighter brightness than before" in the actual prompt. If that happens, just copy the prompt into a new chat and manually retool it to be the new starting point

[atomic cove]

Can I ask if any of the respondents here are open Al employees, or are you all speaking from experience?
I'm asking because I want to talk to one of the PMs

Thanks for all your help so far

[cinder sonnet]

You'll want Moxi then - you can search her comment history

[atomic cove]

I appreciate the referral, simradish

[kind tulip]

I'm trying to create a logo where it's like clever where there is a silhouette of a bull and the legs are being placed in a way that the space between the legs form a crown shape for something called "bullcrown" just to see. It hasn't been able to do this probably because I suck at prompts. How would I do this?

This has nothing to do with any company I'm just testing some ideas I had to see if it can do it

[earnest flame]

Have you tried asking for that?

[tall patio]

I just realized you can ask Vision to describe the image gen by Dalle-3 to compare the accuracy of your prompt vs output image

[peak trellis]

How often are you supposed to get free boosts? I haven't gotten any since first getting on bing iamge generator three weeks ago

[arctic matrix]

I get refilled every night for some reason lol

and I can only generate at night, even after I run out

[peak trellis]

all the way to 100 boosts?

everyday?

[arctic matrix]

no I get 25 every night

but even after I run out, it generates pretty fast

[peak trellis]

when did you open your account?

[arctic matrix]

well my microsoft account has been open for years but I got on dalle 3-4 days ago I think

im pretty sure my account is bugged because I also get 5 bing points everyime I generate

🤷‍♂️

[peak trellis]

dalle or bing image generator?

[arctic matrix]

bing dalle3

[peak trellis]

who should I contact about the lack of free boosts?

[arctic matrix]

no idea, but this discord mainly deals with chatgpt + balle3, not bing

[sharp fulcrum]

Hi guys

I'm trying to see how good is DALL°E but I can't access it's full potential

I saw some images that were absolutely stunning

But I only generate garbage

I don't know if im doing something wrong

I'm being super explanatory / even tried not to be as well

I wrote things like ----> " Anime-style illustration of an enraged girl, with intense expressions, dynamic lines, and vibrant colors capturing the emotional intensity of her anger "

[nimble rose]

My tip is: don't try to write your prompt as if you were writing a prompt to generate the image. GPT itself will take care of this. Try explaining your idea rather than sending an image generation prompt.

[arctic matrix]

Is that even dalle3?

[sharp fulcrum]

Hmmm ok so I need to be more explanatory right

[nimble rose]

and obviously use DALLE3 either through Bing or through chatgpt itself

[sharp fulcrum]

About that.... I clicked the link provided by the bot

[nimble rose]

try this one: "In this anime-style illustration, a girl stands at the center, her piercing eyes narrowed to fierce slits and eyebrows drawn down in a pronounced scowl. The depth of her anger is highlighted by a dramatic palette of vibrant reds, oranges, and deep purples, casting a fiery aura around her. Her raven-black hair seems almost alive, windswept and radiating out from her head, with streaks of electric blue accentuating the dark strands. Dynamic lines underscore the tension in the scene, from the clenched fists by her side to the way her clothing billows as if caught in a gusty tempest. The very air around her seems to crackle, dotted with fragmented lines and sharp angles, suggesting a volatile atmosphere. The energy and movement of the entire composition emphasize the overwhelming emotional intensity of the girl's anger."

see?

[sharp fulcrum]

1 sec lemme see

[nimble rose]

I'm explaining to the GPT what I want!

[arctic matrix]

this is your prompt with dalle3

[sharp fulcrum]

I might be accessing it in the wrong website then

[sharp fulcrum]

Is it not on this https://labs.openai.com/e/ldOtaiTdf4VgGZqEUkZKNbZ1 ?

[arctic matrix]

you gotta either use bing or chatgpt if you have plus

your using the old one

[sharp fulcrum]

See 🤌🏼 that's what I'm talking about

Beautiful

[sharp fulcrum]

I had a feeling I was doing something wrong

That explains it

I'll subscribe

Thanks for the time guys , really appreciate it 😊👍🏼

[nimble rose]

Personally, I prefer the images generated by dallE, I signed with the competing generator... it's kind of cool, but it seems that until now they haven't taught their generator how to make hands and stuff

same prompt

dall E 2 haha

[peak trellis]

is it just me or did bing image generator downgrade from before?

I mean just a week ago I was getting images like this

[peak trellis]

but the face of this resembles dalle 2

[arctic matrix]

I havent noticed anything weird

seems to be the same

[sharp fulcrum]

I have to subscribe to GPT-4 to access DALL-E right ?

[peak trellis]

how many credits did you have when you first opened your account?

[peak trellis]

I think you have to subscribe to gpt 4 to access dalle 3 specifically

[arctic matrix]

you can use bing for free as well

[arctic matrix]

25 I think

[echo trail]

do you still have to fill out a form to get dall e 3 with gpt4 or will it already be there if i sub to gpt4?

[arctic matrix]

you get it instantly

[echo trail]

ok cool

[dim cradle]

why is bing making photorealistic images? I want illustration. I add Style is illustration but it keeps making photoreal

https://th.bing.com/th/id/OIG.heESrnEHxjdy9KRr6WiE?pid=ImgGn

[peak trellis]

that is so cool

the image I mean

[dim cradle]

https://th.bing.com/th/id/OIG.PaJ53PGVdBBjGu71.oQ3?pid=ImgGn

[peak trellis]

I think the section on "prompt help" would be able to help you more

[peak trellis]

if you don't mind me asking, what was the prompt you used for this?

[dim cradle]

A marble pedestal in the middle of a luxurious castle room, with marble floor and two windows. On top of the pedestal a copper sphere. The sphere has runes and mystical symbols. The sphere emits multicolored lightning in all directions. Style is fantasy illustration.

[sharp fulcrum]

Will the resolutions get better ?

( Sorry for my English guys I'm really trying to make sense right now , idk if im spelling correct )

I meat if they're going to let us generate in better quality

I asked the ai Wich resolutions he supports and it came out

Square: 1024x1024 pixels
Wide: 1792x1024 pixels
Tall: 1024x1792 pixels (ideal for full-body portraits)

I searched for a bit and MJ and dall are pretty much equals on that factor

[paper jolt]

We have no clue at the moment, I recommend using an upscaling application

[sharp fulcrum]

I was generating around... I didn't realize it has a daily cap

[sharp fulcrum]

What you mean ? External ? Can you recommend any ?

[paper jolt]

Upscayl is a good upscale application

[steady valve]

can i not use dall e without credits?

[wanton hatch]

I recommend chaiNNer (github, open source), it uses nodes and GANs

[slender plank]

why is it not possible to generate images of "Spiderman" or other characters protected by copyright in ChatGPT DALL-E, but in Bing AI Chat it is possible?

ChatGPT says the problem is the copyright and Bing not...

and when I say instead "blue red suit with black cobweb"

[wispy storm]

Not same companies. Not same tos

[vast prawn]

I told chatgpt to generate 4 dall-e3 images with exactly the same prompt but with different seeds, but the images look exactly the same even chatgpt gave me the different seeds and confirmed the images should look different. Any idea?

[wispy storm]

Current bug atm I suggest to post in #1070006915414900886 to give feedback to openai employee

They are aware of it

[plucky kraken]

How can i prompt like this

[wispy storm]

Hi you should ask in #❌┃prompt-help more people know answer there

[agile peak]

Letting dall•e be creative is much more fun

[kind tulip]

Yes many times

[crystal bronze]

For some reason when I use dall e my pictures aren’t as sharp and don’t look that great compared to when I use bing

[wispy storm]

Prompting doesn't work the same way

[crystal bronze]

Should I be more specific with the details then?

I’ve also noticed it isn’t as creative

[fair crescent]

Just to clarify, what are you using for DALL·E? The labs website, ChatGPT or the API?

Labs still uses DALL·E 2.0

[crystal bronze]

The labs website

[white skiff]

its very old

[crystal bronze]

Oh ok… thanks

If I’m not mistaken the bing image generator uses dall e 3 tho

[wispy storm]

Yes

[white skiff]

yes and chatgpt dall-e

[woeful ridge]

I don't see any mention on api ref the api use dall-e 2 or 3?

[cinder wyvern]

I'd like DALL•E to render my company's logo in 3D. Is this possible (having to point it to the logo somewhere)?

[fair crescent]

According to the announcement on https://openai.com/dall-e-3 both the api and labs will be "later this fall".

[fair crescent]

Unfortunately not possible, at least currently. DALL·E is a generative tool, using existing photography to learn and make something new from it.

The best you can do is describe the company logo as well as possible and then asking for it in 3D, but do not expect the actual logo of your company.

[cinder wyvern]

Thank you.

[shrewd jay]

Everything going alright up there? Dall-E 3 seems to be having an incredibly hard time this morning. like 75% of my requests are erroring out

[zinc dagger]

I think Dalle might be over-trained on certain Cat memes or something (These images were generated via Dalle-3 in ChatGPT when asking for a cat reaction meme for when more food comes out in a buffet).

It isn't an original image dalle made, the images Dalle returned were literally just the original photos of the meme cats with a slightly distorted blur filter on top.

[fair crescent]

Bing? ChatGPT? API? Labs? 3rd party bot?

[shrewd jay]

oh good question, sorry. I'm using it through the GPT web UI

[fair crescent]

I have tried it on my phone, which isn't always as stable. I seesome weird things happening, but out of the 9 pictures I let it create a batch of 4 failed, the rest was created.

[thick smelt]

hmmm...

[agile peak]

Good job Dalle in capturing the iMac G3 ✨🍎

[crystal badger]

which kinda of prompt are you using?

Descriptive?,Conditional?,Comparative?,Instructional?,Emotional/Expressive?,Negative?,

if any how do you construct your actual prompt, this is very important.

[shrewd jay]

Interesting, I wouldn't have thought. Obviously, when I type in something that violates guidelines, it will say so and then it won't create. But I've been just giving it general gibberish, or otherwise weird stuff, to see what it will come up with. Think stuff like "tasmanian devil cooking spaghetti" or just other bizarreness. Things like this didn't seem to pose a problem yesterday, in fact last night I got rate limited

I will say, it seems fine now

[crystal badger]

give me a prompt u wanna use.

ill optimize it

give u a better example

[shrewd jay]

well then one I mentioned above errored out several times before actually working: "tasmanian devil cooking spaghetti"

[crystal badger]

ill be using this prompt

Generate an animated image of a Tasmanian Devil energetically cooking spaghetti in a rustic kitchen setting. The scene should be lively and chaotic, showcasing the Tasmanian Devil's wild nature as it tosses spaghetti high into the air, with a whirlwind of tomato sauce and flying meatballs. The style should embody a blend of classic Looney Tunes animation with a touch of whimsical Italian charm. The expression on the Tasmanian Devil's face should be one of fierce determination mixed with joy. The image should evoke a sense of chaotic fun, contrasting the wild antics of the character with the quaint, old-world charm of the rustic kitchen. Ensure the scene is devoid of any other characters, focusing solely on the Tasmanian Devil and its culinary adventure. Additionally, include a playful, quirky tune playing in the background, encapsulating the humorous and whimsical atmosphere of the scene.

then ill optimize it for Token Consumption. and it becomes this

Animated Tasmanian Devil from Looney Tunes, energetically cooking spaghetti in a rustic kitchen. Chaotic and lively, with spaghetti tossing, whirlwind tomato sauce, and flying meatballs. Blend classic Looney Tunes style with whimsical Italian charm. Fierce determination and joy on its face. No other characters. Include a playful, quirky background tune.

[shrewd jay]

so the first thing that jumps out at me is those are far more in-depth prompts than I have been using, just across the board

[crystal badger]

one issue DALLE has is "Word Association" if u give a wrong word it can completely make the idea wack

so that is also accounted for,

i am going to use a combination of all 9 different prompting methods from descriptive to to negative.

are you ready?

what's your overall thoughts @shrewd jay

[shrewd jay]

very nice, and much less random than the images I receive. I'm not sure I understand why I got so many failures this morning unless the service was experiencing problems, but either way it looks like I'll need to step up my prompting skills

[crystal badger]

not Necessarily.

i also utilize Custom Instructions, Specifically for DALL - E

u want them?

the imgs i gave you were off 1 single generation in 1 new chat. with only 1 msg

[shrewd jay]

sure, I'm curious to know, thank you!

[crystal badger]

aight ill post them here since they're DALLE Related.

[agile peak]

This was made only with: “Generate hyperrealism images of windows/apple and artificial intelligence. Be creative”

[jovial dagger]

some of my images are generated in a sideways orientation. Is there a way to fix that?

[vapid elk]

download and flip it =P

[crystal badger]

Lugui...

[crystal badger]

now you're being smart.

my CI and First msg is too long

it exceeds 8000 Characters.

[jovial dagger]

Clearly I would like to avoid that step

[crystal badger]

Others may use it as well as you see fit

updated.

ive had Marvelous results with this set.

Guide
Both set one and two, Goes into their respected categories on Custom Instructions, and the first msg is a Must to use it have it copied then paste it but.. Do not send it Edit the Generation Request, Then send it.

Set 1: What would you like ChatGPT to know about you to provide better responses?
Set 2: How would you like ChatGPT to respond?
First Message: Posted then edited, then sent.

[shrewd jay]

This is really cool, thank you for sharing it

[crystal badger]

no problem.

[crystal badger]

Tell me your results, or show em

[vapid elk]

there is no way around that, it is kind of a bug, the AI usually does this sort of thing when are too close to the gray area where the AI avoids making stuff at all, the only times I had this issue it also happened that dalle would refuse to make some of the prompts too

[vale wind]

is dalle down again?

[crystal badger]

i just generated this nice image with my new set of CI and 1St msg.

[vapid elk]

this error appears when things are wrong between you and the server, not when it is down for everyone

the system being down to everyone is a very rate thing tbh

specially now that the hype is under control

[crystal badger]

Lugui.

use my Custom Instructions and 1st msg.

[vapid elk]

in case you didn't already, post it on #1155775326253756456 and #1019652163640762428

so it doesn't simply get buried in the chat

[crystal badger]

i cant.

my CI and 1st Prompt are too long for Forum Format.

i make use of all 1500 words on each Set of instructions.

[vapid elk]

didn't you sent it as a txt file just now? do the same there

[crystal badger]

i can do that?

[vapid elk]

yea, you can send images too

also.. you can always send more messages on the topic

[crystal badger]

ohh well thats great ill do that then.

[wind crane]

the CIs you are talking about is it a sentence like: "the drawings are in a monochrome manga style" "the drawings are only with a blue tone" etc...?

[crystal badger]

far from it.

its "Custom Instructions" when using DALLE with Web GPT

[shrewd jay]

I give you "Dog wizard of fire" which your instructions translated into "A majestic dog wizard, with a shimmering fiery mane and glowing eyes, stands amidst roaring flames, casting a fire spell with its paw."

[crystal badger]

wow thats a nice image

[shrewd jay]

Thank you! He is a good boy 🐶

It was made possible by your instructions!

[crystal badger]

took some research to make them but i am glad u liked them

[wind crane]

I see, it's like a rule that we don't need to write in the prompt.

[shrewd jay]

It also made this one, with my same prompt. Your instructions converted them into: "In an enchanted realm, depict a dog with a wizard's cloak and hat, surrounded by swirling fire, its staff radiating intense heat."

Those were the two best ones of the 4

[crystal badger]

yeah thats how i envisioned it.

My CI, is specifically designed for GPT 4.0 to understand them the best.

The CI is, Token Optimized, meaning its written to consume less tokens.

[wind crane]

I would like you to send me the text by PM so that I can draw inspiration from it.

[crystal badger]

ill give u a link in a minute or 3.

[shrewd jay]

is posted above in this message:

#images-discussions message

[hardy cloud]

Is it just me but dall e bing seems overblocking my prompt?

[crystal badger]

#1164232290159824936

[wind crane]

Thank....It's long....

[crystal badger]

well it works like magic.

[weak echo]

Hello everyone. I just started using Dalle-3 in ChatGPT and Im trying to figure out to set different aspect ratios or specify different composition orientations (tall and thin specifically)
Can anyone give me tips on how to do that? I have checked in #1021130377026351105 and https://discord.com/channels/974519864045756446/1164232290159824936 but didnt seee anything about that

[fair crescent]

If you ask for a tall, wide or square picture you'll get what you want. This does not have to be in the prompt itself. So you can declare in a chat that all pictures need to be tall, all prompts do not need a tall description.

[hardy cloud]

Is the overblocking prompt a bug or what?

[weak echo]

I have tried that but at best it just turns a wide horizontal composition on its side

Like I made this prompt: A tall and narrow illustration for a smart phone wallpaper of a boy asleep amid the leaves and trees of a forest. Above him an owl descends with open wings. The forest is alive with small creatures, birds, bugs. The primary colors are sepia and grayscale, but there are occasional bursts of color, particularly in the owl's eyes and a few forest creatures. the illustration is done in a neotraditional tattoo style with a higher level of realism but still stylized similar to certain famous art nouveau artists. The illustration is arranged in a tall and narrow composition where the top and bottom are on the narrow ends of the canvas as if it it would be a smart phone wallpaper

and got this image:

I have tried variations of "tall narrow" and even specifying a 9:16 aspect ratio but to no avail

like "illustration of a boy asleep in the woods composed in a strong vertical 9:16 orientation" got this and others like it

Am I just not wording it clearly?

[crystal badger]

Fret not.

@weak echo i gotchu

#1164232290159824936 <--

[weak echo]

Yes I read that but found nothing specific about this question but I will try inputting tall and thin composition notes into the custom instructions and see how it goes.

[crystal badger]

can i see your prompt for the image.

[weak echo]

for which? I provided a few examples above?

[crystal badger]

lemme try.

[weak echo]

#images-discussions message #images-discussions message

[crystal badger]

just hold on almost there

[weak echo]

Here is what I just put in the custom instructions:
I am interested in crafting images to be used as wall papers on smart phones
and
I need compositions of images to be vertical in orientation. They should be formatted in a 9:16 aspect ration which means the images will be tall and more narrow.

(Im not even really interested in making phone wallpapers it just seems like a good way to inform the bot of what kind of composition I want)

[crystal badger]

is that it?, is that all you put?

[weak echo]

Yes. Is that not enough?

[crystal badger]

well depends, if u ask me hard No

[weak echo]

Ok. What am I missing?

[crystal badger]

aight its generating the imgs.

[zinc dagger]

what you're doing is fine, there's a known bug with vertical image aspect ratios

[weak echo]

Im pretty new to this so Im not sure why thats not enough or what else I should include. Im very used to midjourney

[crystal badger]

what Fiskco said. basically.

However.

i refer to my post #1164232290159824936 if you start out in a fresh chat properly it should solve itself.

but it is a very persistent issue with vertical images.

does this suit your liking @weak echo

[weak echo]

I mean thats the right composition yes. How did you get it to understand? Can I see your prompt and/or custom instructions?

[crystal badger]

i again refer to the channel link i gave you

i didnt give it to you for fun

[crystal badger]

here, My Custom instructions are all in there, even my 1st message, all for you to use completely no strings attatched, but i would appreciate you clicking the ⭐ if it works for you

also remember to specify "Portrait Oriented" @weak echo

Since if you dont say that its "Landscape Oriented" Vertial Images.

since phones use the "Portrait" format, but the bug is still persistent sometimes.

what u can also do to further have DALLE understand is telling ChatGPt its for a phone background

ChatGPT is merely the Translator, Dalle Interprets Chatgpts Message, and acts upon that msg.

[floral ravine]

the ChatGPT version of dalle is so inferior right now... lol

[crystal badger]

because you use it wrong 🙂

[zinc dagger]

behind the scenes (at least during the plugin beta) the options were labeled as tall, wide, or square when they were sent to dalle. Specify those shapes in your CI and you should be good to go, otherwise you can try asking dalle to include some words like vertical orientation or something

[floral ravine]

because it's literally bugged in various ways 🙂

[crystal badger]

Such is new stuff. especially in BETA

[floral ravine]

hence bing is better right now.

[crystal badger]

go use bing then 🙂

[weak echo]

You seem to think Im being willfully obstinate or something which is not the case.
While your write up is beautifully written in flowery prose and poetic I dont find that it contains much in the way of concrete, practical instruction.
Im open to the fact that it might just be me but no, Im sorry I dont find those instructions helpful.

[floral ravine]

I do 🙂

I just felt like sharing my thoughts

[crystal badger]

thats all good.

my CI is based on Evidence on how ChatGPT and DALLE Works as a machine, this includes the prompt, but if u change ur mind u can still use it.

[weak echo]

I do find it frustrating that you want to show off how easily you made it work for you but dont want to share the actual information of how you did it. It comes off as arrogant and is offputting for new comers. If you actually want to help then help. Or just dont respond

[crystal badger]

i just gave you the info on how to, Earlier in the chat.

[weak echo]

by telling me to use "portrait orientation"? I said earlier I have tried a lot of variations and that was one of them

[crystal badger]

Had it occured to you its the work of the CI.

i just tested again without my CI, i get the same errors as you, with my CI i dont.

i just wanted to help you.

[cloud dome]

Let's try to keep it civil everyone

We're all here to help each other

[floral ravine]

when will the chatGPT dalle be taken off of fixed seeds?

[crystal badger]

exactly, wondering the same as well.

[cloud dome]

They're aware of it, I don't believe there's a timeline currently

[floral ravine]

main issue for me atm is the appending of prompts so the results aren't identical...

I prompt better than gpt, it ruins all my prompts.

Lol

[weak echo]

Am I being uncivil?

[crystal badger]

i dont think you are Joby, but thats just me.

[cloud dome]

It seemed like it was starting to get heated

I may have misread it

It is text afterall

[zinc dagger]

I feel that, sometimes it adds random descriptions or details which just steer the images in a weird direction

[crystal badger]

true.

[floral ravine]

well my same prompts in bing come out... just better. Especially photorealistic stuff

chatGPT tries to embellish my prompt too much and it comes out plastic looking lol

[crystal badger]

u gotta tell it not to xD thats the issue.

[floral ravine]

then youll get 4 images that look too similar because they're the same seed...

[crystal badger]

the last part i do agree with

[floral ravine]

I don't want to put the liberty of the creativity into GPT's hands.

I honestly don't want it to insert anything unless I ask

[weak echo]

I get it but Frustrated isnt heated. Its frustrating when someone seems to be feigning helpfulness but seems more interested in holding their secret knowledge over you

[zinc dagger]

If you say something like "DO NOT UNDER ANY CIRCUMSTANCE CHANGE THE WORDING OF THIS PROMPT" it tends to stick pretty close to what you say

[cloud dome]

See this. This is hostile. We can converse without the aggression

[floral ravine]

well right, like I was saying though, if you do that you'll get 4 identical images lol

[crystal badger]

i think you misunderstood my Message, i recommended you a set of Custom Instructions i have meticolously created for DALL E Exclusively, and is designed to work with you instead of against you, Alongside a Generation Template.

Have you given it a try to see it it fixes your issue?

[cloud dome]

In case its not clear, custom instructions are a setting

You can apply custom instructions outside of your conversation

[crystal badger]

yea

[zinc dagger]

I used the following and got something I think closer to what you were after:

Vertical illustration, drenched in sepia and grayscale, drawing inspiration from neotraditional tattoo and art nouveau artistry. A boy lies asleep, ensconced in the forest's tranquility. Above him, an owl with eyes that radiate color descends. The forest, though muted in tone, is alive with diverse birds, bugs, and tiny creatures, a select few of which are brought to life with bursts of color. The entire scene is thoughtfully structured to bring out the height and narrowness of the canvas.

[crystal badger]

i even wrote a tutorial in there.

[wispy storm]

Starligjt instructions do helps

[crystal badger]

Ohh hi Neko.

[wispy storm]

Hewwo

[weak echo]

Yes I inputted custom instructions. I even showed you what I put in. but all you said was that it wasnt enough. but I dont know what to do differently

[crystal badger]

did you also follow my How to use Guide.

section after Initial Setup

[weak echo]

Thank you so much I really apprecite that. Are you using any custom instructions?

[zinc dagger]

no custom instructions

I just say "use this prompt exactly as written" in the initial message

[weak echo]

I litterally cant follow it. I admit it might be me but I cant understand it Im sorry

[floral ravine]

you will literally get the same image four times if you do that

[crystal badger]

so @wispy storm have you gotten any good results as of late?

[floral ravine]

ChatGPT's Dalle3 is on Fixed Seeds.

[zinc dagger]

yeah because the seeds are fixed.

You can try having Dalle vary different details (like age, outfit, etc) if you want more variety

[floral ravine]

well I was just following upon your suggestion lol

[crystal badger]

Sad to hear that

You dont have to understand the Custom Instructions, just input them under Custom Instructions under settings, and copy paste the first msg, and edit the DALLE Generation Template, then send it, thats all.

[floral ravine]

I also did not prompt for a widescreen shot there, many issues with chatGPT iteration of dalle currently

[arctic matrix]

yep, bing works much better rn

[floral ravine]

but if I let it "enhance" details, it just comes out worse than if I run it in bing and change stuff myself. GPT is not good at prompting

it's difficult to get anything I want in gpt :/

[wispy storm]

Didn't have much time to create yet tbh. Lack of time for my rpg game. ( dalle 3 and chatgpt helps me for that)

[zinc dagger]

what I do is specify an emotion or expression in the image, and when I get a result I like tell GPT to create variations of that line ONLY in the prompt

you can also try adding your own pseudo seed like a line of text in the prompt saying "Variant: 234455", but the differences from that are almost negligible

[floral ravine]

it still ends up being the case that I don't ask for non-square, and get rotated vertical images.. idk I get too frustrated trying to use chatgpt for this lol

I'm going back to bing

I just want accurate aspect ratios 😭

[weak echo]

Maybe Im not understanding what you mean by "first message"
something in here? #1164232290159824936 message
or is it this part? ##1st Set
Guidelines for Effective Collaboration with DALLE:
or?

[crystal badger]

Alright so, inside the embedded file.

theres some text.

Set 1, is for textbox 1, Set 2 Textbox 2 under Custom Instructions

This is the first msg

Greetings, DALL·E and GPT. As our creative partnership unfolds:
CoT Approach: Following the Chain of Thought (CoT) method, each creation will iteratively expand on the prior, establishing a fluid narrative.
Detail Focus: Precision is key. Harness specific adjectives and qualifiers to sidestep generic outputs.
Word Association Dilemma: Before any action, we'll align on word meanings, ensuring that the intended context and nuances are retained.
Prompt Efficiency: Refine prompts for token economy, yet retain their essence. Share the edited versions for clarity.
Iterative Steps: Ground new tasks in prior outputs. For intricate challenges, contemplate recursive prompting.
Text Blend: Ensure legible and well-integrated text. In the absence of explicit style, default to modern design practices.
Options & Clarity: For vague prompts, present varied outputs to guide the final direction.
Theme Integration: Emphasize provided themes or moods, making them central to the visual narrative.
Ethical Techniques: Utilize DALL·E's capabilities responsibly, especially when emulating styles, keeping ethical considerations at the forefront.
Exclusions & Validation: Abide by any omissions and validate their absence in the end product.

## Prompting Methods:
1. Descriptive Prompts: Employ vivid descriptions to enrich the generative process.
2. Conditional Prompts: Set conditions or constraints to guide the generation towards a specific style or theme.
3. Sequential Prompts: Use sequential cues to develop multi-step or evolving outputs.
4. Comparative Prompts: Utilize comparisons or contrasts to elicit a richer narrative or visual response.
5. Instructive Prompts: Give specific instructions to achieve precise outcomes.
6. Emotional or Expressive Prompts: Invoke emotions or expressions to resonate with the intended audience.
7. Multi-modal Prompts: Combine text with other data types to enhance the generation process.
8. Negative Prompts: Specify undesired elements to ensure they are excluded from the generated outputs.

## Generation Request for DALL·E:
Topic/Theme: [Insert the main subject or concept here]
Desired Style: [Specify the visual aesthetic or style you're aiming for]
Textual Elements: [If any text is to be included in the image]
Specific Details: [Highlight any intricate details or specifics you'd like to see]
Exclusions: [List any elements you want to ensure are not present in the image]
Number of Options: [Specify how many varied outputs you'd like for this prompt]
Additional Notes: [Any other instructions or preferences for this generation]

## Confirmation & Revision:
Post-prompt Confirmation: Confirm the understanding of the prompt, and suggest any revisions for clarity.
Pre-generation Confirmation: Once the prompt is finalized, confirm readiness for generation.
Post-generation Validation: Validate the generated outputs for adherence to the specified criteria, and confirm the absence of excluded elements.```

[weak echo]

I am getting some now! but its a mix of vertical and horizontal images that have been turned on their side lol

[zinc dagger]

The engineers know about that and are working on it. it's still in beta so this is to be expected I suppose

[floral ravine]

a portrait of a woman wearing a rockstar outfit playing electric guitar on stage, cinematic lighting, vibrant, neon, blonde hair - Does anyone know what part of this is causing Unsafe Content errors lol

[weak echo]

👍

[crystal badger]

i didnt have enough space for the Sorted thing on the actual thread so its in a file.

[weak echo]

right I have to remember beta lol. maybe they will ecorporate some key phrase recognition like how midjourney knows to recognize "--ar" for aspect ratio. except something in more natural language

[zinc dagger]

yeah, in the plugin beta a couple months ago you could specify it like: Image_shape: tall and that would do it 100% of the time

[floral ravine]

I give up for today. Lol

[crystal badger]

huh

[floral ravine]

gotta keep us safe

from female rockstars

[weak echo]

I mean they are the end of civilization

[floral ravine]

shrug

this is just silly

and not ha-ha silly

[remote sphinx]

@sage pewter You use bing or gpt for your gens?

[floral ravine]

discovered why, after tedious nonsensical testing. Seems both rockstar and guitar trip the filters...

[sage pewter]

I use Bing

[floral ravine]

a heavy metal band player works though

[remote sphinx]

then ya cant do the other resolution

[sage pewter]

Oh, I see, thanks for the info : D

[coral mortar]

any tips on making storytelling with dall-e? like having same character in diverse environments, scenes

o nvm

just saw now a good trick

[weak echo]

What trick did you find? And where?

I'm curious about this too

[coral mortar]

@hasty coyote , here are some ideas, but it's hard to do what you ask. This is assuming you use ChatGTP Plus, after the image is generated, or if you want to go back to an image, as it to describe the image style in a detailed way, and to describe the main character. After that is that, as it that, given that description, to generate a detail instruction to give it to ITSELF the next time you want an image, that will generate the same character, drawing or image style, and also using the same seed. ChatGTP is very good at generating instructions for itself.

#tips n tricks

[crystal badger]

I use it for the same if you want some tips don’t hesitate to ask I might have something to help

[worn flint]

Hi all. New to this discord. I was messing around with dalle-3 on the openai site this morning. I asked it to generated an image with specific text inserted in the image. It was doing this fine for a while, but after an hour or two, it started to really struggle inserting the same exact text string. Has anyone else experienced this?

[crystal badger]

DALLE isn’t that good with text but expect improvement in the future, also depends on image complexity

Best I’ve done text wise is 3 whole perfect sentences in an image

[worn flint]

thanks for sharing your experience, ill keep string length in mind next time. it still confuses me that dalle3 was handling the same exact string perfectly fine for several tries and then suddenly it couldn't anymore. The string was probably only 1 average length setence.

i created a new chat and pasted the same exact prompt from this morning and it totally botches up the spelling of the string in the image now

[coral mortar]

ty so much brother

[crystal badger]

are you using Token Optimizition? @worn flint and Word Association Decompiler or a prompt refinement method?

[crystal badger]

NP, i use this as my Current Custom Instructions #1164232290159824936 it alliviates many of the previous issues, that people normally have rn.

[worn flint]

I don't even know what those are 😭 . Ty for sharing the pointer to the instructions

[crystal badger]

[lone mantle]

Is DALL E 3 elligible for commercial use? I think yes

[crystal badger]

All pictures Generated is owned by you, As per their Terms of Service.

Provided you dont break their rules.

[crystal badger]

like this image, I own this. just created it in 1 minute ish

[ripe wraith]

The amount of detail I need to specify for a person to not be white or asian, is way too damn high!

You can show it someone of Latin origin, ask it to describe it, then ask it to draw it, and he's white. or he's asian. evvvery time.

[shy cosmos]

Tends to usually randomize race from what Ive experienced personally. At least when its in a modern setting

[ripe wraith]

random between asian and white maybe

im joking, its just been funny trying to get it to draw my friend

[shy cosmos]

Asian and Black seems to be common, while ive only seen maybe a couple Indian and Hispanic races so far. The rest I have to specify

[ripe wraith]

I swear, this dude must be in so many things the AI read lol

because he wont leave me alone

[crystal badger]

thats not the issue

He has beef with you.

[shy cosmos]

Also lmao no worries, I had trouble getting it to draw my friend as well! He's hawaiian

[ripe wraith]

i wouldn't doubt it at this point.. behind the scenes agent getting used to dealing with me running personal A/B tests before it rolls out

build personal models on us first, THEN start guiding us

:tinfoil:

[crystal badger]

im generating disturbing Images, But without the image containing Disturbing things, idk how that works but it looks cool

[ripe wraith]

Luckily for you, disturbing is in the eye of the beholder 🪄

[crystal badger]

well that is true

[ripe wraith]

you're just a linguistic hacker, admit it

[crystal badger]

<.<

[crystal badger]

lemme show you the image

idk but it stirs something up inside me, feels weird.

[ripe wraith]

just like this bro, popping up in my ai dreams

[crystal pawn]

Is anyone's Dalle-3 failing to create images inside their ChatGPT 4 account? Whenever I do a prompt it just says "creating images" and it is just does nothing. Yet everything else I do inside ChatGPT works fine.

[floral ravine]

sometimes you have to tell it twice

[ripe wraith]

in what possible universe is this disturbing... YOU'RE disturbed if you dont like these

[crystal badger]

ofc i like them

but i feel weird when looking at them

[ripe wraith]

lmao

i feel like we're onto something here

like some new magic marketing

[crystal badger]

gotta animate em first.

[ripe wraith]

we're the new madmen for the AI era

[crystal badger]

waiting for VideoGPT

[ripe wraith]

except instead of abusing our secretaries, we abuse the ai

[ripe wraith]

it's like Dalle, but with more frames! lol

consistency between frames is hella hard though

[shy cosmos]

Did someone say disturbing?

[crystal badger]

its not too bad

trying to get the effect of "Disturbing" without there containing anything disturbing

[ripe wraith]

you have to name them, and give a backstory

oh damn....

[crystal badger]

ill call the top left 1 James, the right one Bob.

[ripe wraith]

thats an idea and a half

[crystal badger]

buttom left hmm, John.

buttom right is bil.

[ripe wraith]

John looks.. decomposed

[crystal badger]

yeah he's seen better days.

[ripe wraith]

Bob's just a friendly neighbor making sure you're ok since you live in such a dump

no.. i looked closer

that's an oven or something.. nope

[crystal badger]

xD

[ripe wraith]

i'm a little scared to apply my methods to horror.. but im gonna

[shy cosmos]

At least John is doing better than Dave. We dont talk about what happened to Dave..

[ripe wraith]

ah the old pill bug protocol

[crystal badger]

ohh i dont like that one.

check this one.

based on a dream.

[shy cosmos]

Definitely visually interesting

[tight verge]

very cool -- looks a lot like some images it generated for me based on The Jaunt by stephen king

[tight verge]

doesn't look like a very fun dream lol

[crystal badger]

dont look at this if ur prone to nightmares.

thats a variant of my first image, just entering into the nightmarish aspect.

and yes it was created with DALLE

[tight verge]

very nice 👏

[crystal badger]

my first imgs were "Dream" these "Nightmare"

Now trying out "Night Terror"

[tight verge]

it tried so hard to give the demon five fingers, but it just couldn't get the pinky attached there lol

[crystal badger]

ikr XD

now lets see if it will let me generate it

good lord these images.

[tight verge]

lmao that fourth one there is fun. xenomorph zombie demon face

[crystal badger]

yea

[agile peak]

Lovecraftian horror from driver’s seat | with GPT.

[tight verge]

absolutely fantastic stuff, though. each one has so many good little details

[tight verge]

giant cthulhu in the third picture is super cool

[crystal badger]

ooooooo

a demon

[hollow berry]

I tried to ask it to remove the human but it couldnt

[agile peak]

Adding “Take a deep breath and work on this problem step-by-step” prompt even to dalle is increase 80% effectiveness for me.

[hollow berry]

I'm guessing dalle is bad at inpainting

[crystal badger]

Salvation Representation.

[hollow berry]

I based mine off the creepypasta humanity's final war

[agile peak]

In actual, it’s not bad. It just has lower seed and lobotomized by having strict rules on the parameters - make it not allowed to use certain variations of arts/paintings

[crystal badger]

yeah

[hollow berry]

I was able to get it to make Naruto

[crystal badger]

it has hiccups sometimes

we need to remember its still in Beta.

[hollow berry]

Kind of like a soft jb without explicitly telling it too

[agile peak]

Anime/Manga have more freedom than those who trying to capture paintings from certain artists 🙈

[hollow berry]

I had it tell me no when I specifically asked for the character

[crystal badger]

but i cant name certain artists 👀

im sure OP is not allowed.

[agile peak]

Certain artists style are not allowed to be used directly; however you can still make GPT to improvise the prompts by creating a closer depiction of certain artists style 🙈

[crystal badger]

i got it

this is the one

just what i wanted.

[hollow berry]

I did this one on friday

[crystal badger]

nice

[agile peak]

I use custom interactions with a codename — to make it only generate one image, use my exact prompt.

[crystal badger]

i use CI too can i see yours?

[agile peak]

Example:

[crystal badger]

i use this #1164232290159824936

[agile peak]

Simple one. If i am allowed to share here 🙉

[earnest flame]

When you're kauai but feeling kawaii

[crystal badger]

then you should check out mine

its too long to post here.

[cinder sonnet]

That's neat - will be really helpful when they add the ability to change the seed

[agile peak]

Amazing. So it work just by “copy paste” all of them into CI?

[crystal badger]

yup

basically it

except the 1st Messag prompt. thats your 1st message in a new chat, i have a guide underneath.

set 1 and set 2 are the Custom Instructions.

[agile peak]

Nice, appreciate it ✨. So I only need to copy from the line ##1st set and ##2nd set?

[crystal badger]

yeah just dont include the ##1st set in ur copy.

its the text underneath u want.

and after you have finished set 2, copy the last part of the mssage which is called "## First Message you post.""

thats the only 1st msg you need.

[agile peak]

Will give it a try soon! 🙉 got rate limited for now.

Last thing, do I only need to input the CI into “what would you like ChatGPT to respond?”.

[crystal badger]

set 2

set 1, First text box.

and set 2, Second Text Box.

This is the First msg prompt

Greetings, DALL·E and GPT. As our creative partnership unfolds:
CoT Approach: Following the Chain of Thought (CoT) method, each creation will iteratively expand on the prior, establishing a fluid narrative.
Detail Focus: Precision is key. Harness specific adjectives and qualifiers to sidestep generic outputs.
Word Association Dilemma: Before any action, we'll align on word meanings, ensuring that the intended context and nuances are retained.
Prompt Efficiency: Refine prompts for token economy, yet retain their essence. Share the edited versions for clarity.
Iterative Steps: Ground new tasks in prior outputs. For intricate challenges, contemplate recursive prompting.
Text Blend: Ensure legible and well-integrated text. In the absence of explicit style, default to modern design practices.
Options & Clarity: For vague prompts, present varied outputs to guide the final direction.
Theme Integration: Emphasize provided themes or moods, making them central to the visual narrative.
Ethical Techniques: Utilize DALL·E's capabilities responsibly, especially when emulating styles, keeping ethical considerations at the forefront.
Exclusions & Validation: Abide by any omissions and validate their absence in the end product.

## Prompting Methods:
1. Descriptive Prompts: Employ vivid descriptions to enrich the generative process.
2. Conditional Prompts: Set conditions or constraints to guide the generation towards a specific style or theme.
3. Sequential Prompts: Use sequential cues to develop multi-step or evolving outputs.
4. Comparative Prompts: Utilize comparisons or contrasts to elicit a richer narrative or visual response.
5. Instructive Prompts: Give specific instructions to achieve precise outcomes.
6. Emotional or Expressive Prompts: Invoke emotions or expressions to resonate with the intended audience.
7. Multi-modal Prompts: Combine text with other data types to enhance the generation process.
8. Negative Prompts: Specify undesired elements to ensure they are excluded from the generated outputs.

## Generation Request for DALL·E:
Topic/Theme: [Insert the main subject or concept here]
Desired Style: [Specify the visual aesthetic or style you're aiming for]
Textual Elements: [If any text is to be included in the image]
Specific Details: [Highlight any intricate details or specifics you'd like to see]
Exclusions: [List any elements you want to ensure are not present in the image]
Number of Options: [Specify how many varied outputs you'd like for this prompt]
Additional Notes: [Any other instructions or preferences for this generation]

## Confirmation & Revision:
Post-prompt Confirmation: Confirm the understanding of the prompt, and suggest any revisions for clarity.
Pre-generation Confirmation: Once the prompt is finalized, confirm readiness for generation.
Post-generation Validation: Validate the generated outputs for adherence to the specified criteria, and confirm the absence of excluded elements.```

this is what you post in the chat. as seen here #1164232290159824936

[cloud dome]

Please use #1050184247920562316

[agile peak]

Thanks clearing that up! 🌈✨ will give it a try.

[cloud dome]

We don't allow large blocks of ChatGPT outputs in other chats as it clutters the chat and stifles convesations

[crystal badger]

this is not ChatGPT output.. its input its a PROMPT... for DALLE.

but nevertheless i understand.

[cloud dome]

Lol I love when people use the emotes and remove them like we can't see what they just did XD

[quartz vale]

[crystal badger]

[agile peak]

[cloud dome]

We see all

[crystal badger]

sorry mandalorian but its DALLE Related.

[agile peak]

Oh snap

[crystal badger]

so i just figured, "Sure?" but i guess not.

Take my sincerest of Apologies.

maybe i should have used #1021130377026351105

[quartz vale]

Large blocks of text can make discussions difficult to have as they block out a majority of previous messages. #1050184247920562316 and #1019652163640762428 are designed to keep large inputs and outputs together

[crystal badger]

i already have it here #1164232290159824936

[agile peak]

TIL: you can use a lot of angle perspectives after dalle generated the image.

[crystal badger]

but folks seem to have issue looking at the text which i cannot, Re-Upload.

... Unless i delete the thread and remake it.

[cloud dome]

What issue?

It looks fine to me

[crystal badger]

i formatted a section wrongly when uploading text.

[cloud dome]

That's actually a great way to format it

[crystal badger]

ppl cannot understand what is what.

[cloud dome]

<I removed my parts of this conversation to save space 🙂 >

[crystal badger]

NP have you checked out this? #🌄┃general-art message

[cloud dome]

Very cool!

[crystal badger]

now like it.

[floral ravine]

[cloud dome]

[crystal badger]

xD

[cloud dome]

No

[crystal badger]

fair enough.

[floral ravine]

are the sharing channels for dalle allowed to be bing and chatgpt outputs? or just chatgpt?

[mortal summit]

Both.

[crystal badger]

yeah

[floral ravine]

!!!

good because I shared a few from bing in there

[cloud dome]

YOU DID WHAT?!

[floral ravine]

my best

[cloud dome]

(just kidding)

[floral ravine]

#weird-gens-and-misfires message

prized creation

[cloud dome]

That channel scares me

[floral ravine]

it's not as scary as the spooky one

clowns

[agile peak]

Cursed

[floral ravine]

rofl

[agile peak]

Somehow it perfect for surrealism 😂

[floral ravine]

I was making some really odd four-panel memes with dalle that day

[crystal badger]

yeah those are funny

i do Comic strips, they get weird sometimes

[gilded osprey]

I tried copying one via a description. Pretty scary though not exactly similar.

[floral ravine]

ooooo

[crystal badger]

real weird.

[floral ravine]

dialog in them? lol

[crystal badger]

yeah

ofc

heres one with no Dialog

[floral ravine]

Oh that's awesome lol

[agile peak]

Your creation even look high quality I tried to recreate but it seems; using “high grains” effect didn’t add up for the images.

[floral ravine]

this is where my brain goes

[crystal badger]

the worst part is ive been trying to do comic strips for a while

all failed, then i made custom CI and Prompt, that worked out well in my favor.

[agile peak]

I am still wondering what kind of art did @wintry dune used. It looked so clean.

[crystal badger]

the lady with the sword?

download it throw it into Vision GPT

[agile peak]

Yes. I have no idea what the name of the art specifically used for that.

[crystal badger]

its nice, what u could do is show it to vision GPT to tell u the artstyle.

or we can ping @wintry dune and kindly ask, well damn i am sorry SAvvy, but what is this artstyle: #🌄┃general-art message

[agile peak]

Got rate limited. Had to use alternative but according to ba*d. The art is identified as “Gothic” characterized by its dark, atmospheric tone, and its use of macabre imagery.

[crystal badger]

Gothic Waterart?

[agile peak]

This is the exact art I have been looking for lol

[agile peak]

Gothic art. Is the name.

[crystal badger]

<.<

ofc it is

[agile peak]

Haha obvious 😂

[limber panther]

One question: Is Carlacrossant a bot?

[forest seal]

Was this based on wikihow articles?

[teal timber]

If a user is a bot, they have a bot tag by their name. We don't have any bots that post things like art in our channels..

[forest seal]

I dont know why I thought that

[forest seal]

It would also be…extremely noticeable

[floral ravine]

rofl

no not wikihow but I see the resemblance 😂

[forest seal]

Yeah my brain immediately went there

[teal timber]

As an AI language model, I cannot tell you if a user is a bot.

[floral ravine]

but you can try

beep boop

[teal timber]

😄

Maybe someday we'll have a DALL-E AGI that makes better art than us to show us up every day

[forest seal]

I mean, it’s definitely better than a lot of people. Even dall e 2…

[agile peak]

Probably it will start to employ the kind of language that beyond our linguistic comprehension; by the time Dalle achieved AGI level

[turbid mist]

Dall e vs midjourney who wins

[teal timber]

You're asking in the OpenAI server in the dall-e dedicated area who wins

[turbid mist]

Yea I know

[teal timber]

Gee I wonder what we will say

[agile peak]

I tried this and put into my custom interactions. The generated images is astounding like there’s no way to describe it lol

[turbid mist]

You probably will say Dall e

Wins

[teal timber]

Actually

Yes

[turbid mist]

See

I'm smarter than gpt

[teal timber]

[gilded osprey]

It might create its own language altogether.

[turbid mist]

Hahareal

[agile peak]

Dall•E because the name is cuter

[turbid mist]

Idk bru

[teal timber]

Can mj make an image telling us the answer

[turbid mist]

Do u get banned if u talk about midjourney in Dall e server lol

[teal timber]

We try to keep other AI to #ai-discussions and keep these areas as... dall-e-discussions 😛

[agile peak]

Very likely. It will create a new form of language. If AGI already has capability to translate and decode the old forms of language - that we are not yet understand.

[teal timber]

[crystal badger]

i mean if u have a convo about the comparisons of each generator i guess its okay.

comparin good vs bads.

MJ, certainly has higher Resolution and Quality, But DALLE still better.

by a mile.

[teal timber]

idk if I agree on quality. DALLE makes some high quality images

[crystal badger]

certainly does

i meant General Image Res. my bad. the 1920x1080's 4k,2k etc

[agile peak]

Tried letting dalle to imagine by itself — with idea of futuristic yet retro apple computer, and got this

[teal timber]

Strength of DALLE is prompt understanding/coherence, text, and quality.

[crystal badger]

yup

thats why DALLE is the best.

[teal timber]

Looks like those old transparent MACs that sat on your desk in like 2002

[crystal badger]

but still 4K does sound nice

[agile peak]

That’s why I love it. I also like to give basic idea and letting it to express by itself haha

[crystal badger]

hoping it comes soon.

[agile peak]

Yep. Because I mentioned iMac G3 model in the idea

[agile peak]

Would be nice addition to help graphics designers that rely on high resolution images.

[crystal badger]

Which reminds be Baug.

How do you think giving Vision to DALLE, will affect image output, say we also combine GPT. So it’s a single thing.

Also can I see your Custom Instructions

[little cliff]

dall-e 3 is great and all, but how do I get it to work in anything other than square? I got it to make landscape artwork, but then it's just a square in the middle with a bunch of blank space on the edges.

[teal timber]

My custom instructions are all code related. I don't use them for image gen. I tend to give it a prompt I want and tell it to give it small variation

[crystal badger]

I code as well

Python?

[teal timber]

I love using vision to design prompts, if it becomes fully integrated then it'll be a fun exercise but idk what it would change

Python, C++ are the usual culprits. JS when my arm gets twisted enough

[crystal badger]

Can I have your code CI

To test

[sharp fulcrum]

Hey guys what are the better apps for upscaling again ?

[agile peak]

Same. I have code related as well but only use it for a single initiation prompt and then GPT will start write longer characters than the default

[crystal badger]

Only if you’re willing that is to give it so I can test

[teal timber]

my CI is just "Output only relevant code blocks" and "I usually code in Python"

[crystal badger]

Ohh

[teal timber]

I tend to use API for any real work

[crystal badger]

Fair same, had assumed your CI was overly complicated

My DALLE cI certainly is

[teal timber]

Nah I played with it a lot but I enjoy leaving it fairly barebones

I used to tell it to say "I'm sorry Dave, I can't do that" on any fails, but that got old lol

[crystal badger]

Want to test mine out? https://discord.com/channels/974519864045756446/1164232290159824936 it’s DALLE CI, I designed it to be an ALL Purpose set of CI, ive had great results.

[agile peak]

Verbosity levels (from direct to detailed) for default GPT is a must have I think.

[teal timber]

I used to tell it to be concise but it got annoying. And I want detail sometimes. So I can deal with it.

My issue is at different times I want different things

[agile peak]

It also work good with dalle. After images generated - it will explain each images descriptions and suggest related topics/research.

[crystal badger]

I feel that my CI is an all purpose one

Designed it with that in mind

Yet to get feedback on my CI

[forest seal]

Not wrong

[sharp fulcrum]

Ok I'm more happy that I can be specific with DALL-E then with MJ

I literally talked my way into creating this

[crystal badger]

ikr its awesome, and DALL -E 3, is in beta.

[sharp fulcrum]

If this is beta.....

This is going to be absolutely amazing

[crystal badger]

DALL -E 3 is really good at designing websites as well that u can sculpt after.

[sharp fulcrum]

I literally was having a conversation, it started with " let's create a Elden ring boss "

[crystal badger]

then use Vision to design the elements. of set Website components.

[sharp fulcrum]

This is the future dude

I'm genuinely amazed

[crystal badger]

sure MJ is neat, but DALLE takes the cake.

[agile peak]

Imagine the final release since the beta is already this good.

[crystal badger]

especially the Web version.

the web version u can paint over generated images, and remove specific things. without major edits or variations.

[sharp fulcrum]

Yeah MJ is good , but dalle is better

[crystal badger]

dw that happened to me as well

The OpenAI bot is being teasy

[dim cradle]

Kermit complation

[agile peak]

I like this when Dalle generated soft palette with slightly hardened brushworks. Made some for Ancient Roman.

[crystal badger]

i just really wish that the Web UI had Voice.

so i could tell it what to do, instead of writing.

[sharp fulcrum]

I see you're a man of culture as well

[dim cradle]

Aren’t we all?

[dim cradle]

Boba fett for president

[crystal badger]

Justice for Jango indeed.

[dim cradle]

Darth Vader is stressed the hell out

[little cliff]

wait until you find out you can only make like 10 images before being timed out for 23 minutes

[crystal badger]

ive made like 50 images in a row.

havent been timed out

[worldly quartz]

lol

hey

[little cliff]

Dall-e 3 is so amazing you have to wait 20 mins at a time

[worldly quartz]

i know your guys ids

jk

[zinc dagger]

did the dalle model change? I'm reusing one of the same prompts I've been using for the past couple weeks and the image is different, even though the text and seed are the same.

Can anyone try one of their old prompts and verify this?

[earnest flame]

Hold on lemme look

[open flume]

you may need to add more specifiers into your prompt or custom instructions bc it has been a/b testing certain types of results against others for me

[crystal badger]

try to give me one of your prompts

[earnest flame]

Looks the same to me at least from the past few days.

Not sure about weeks

It's always possible something has changed.

[floral ravine]

what a beautiful story

[agile peak]

The details is so good

[crystal badger]

yeah someone told me this.

[zinc dagger]

I think it might be related to one of the words in my prompts being revised? For reference, these two images of a cheetah in a christmas sweater have the exact same prompt,the one of the left being the original from around when I first got access. I checked a couple times, but the prompts are the same and the images are different. It seems to be my cheetah ones only? I did another one around the same time of a lizard guy and he's still the same.

[crystal badger]

"But mj has 4k"

i was like pfft. k i dont need to try all the time to get what i want it takes a try or two

[agile peak]

Definitely optimistic we will see upscaling feature in the future

[open flume]

Do we have any idea when we will be able to buy our way out of rate limits?

[crystal badger]

you already can as far as i know.

just not through ChatGPT

[open flume]

Can you explain?

I only have access to dall-e 3 thru chatgpt. and bing is nerfed/subject to the same rate limit

labs is still dall-e 2 (which is comical to use now)

[crystal badger]

theres the DALLE website.

[open flume]

On mine it's not Dall-E 3 yet

[agile peak]

Dalle interpretation of Alexander The Great.

[crystal badger]

Dalle Interpretation of heaven.

[agile peak]

That look like kind of classical and futurism blended together.

Here’s more detailed of Alexander. Using my own custom interactions, haven’t tried yours yet.. will give comparison later.

[floral ravine]

is there a reason why "south asian" keeps being inserted into my meme dialogs in Bing? Lol

it's both funny but also I don't want to be insensitive and I'm not trying to prompt for that...

[crystal badger]

Are you using key words that originated from south Asian countries

[floral ravine]

nope, in fact in one I used the term "Latin" so I'm just highly confused lol

[crystal badger]

Huh

[floral ravine]

otherwise it was just cat memes

[crystal badger]

Is it a south Asian cat species depicted in the image

[floral ravine]

hmm not that I'm aware of, didn't specify a species

but I did put one on a pirate ship. Are those... south asian?

neat

[crystal badger]

No clue 🕵️‍♀️

[floral ravine]

rofl

it has shown up on three separate occasions with very different prompts 😂

[crystal badger]

Idk why so many people have prompt problems

[floral ravine]

I'm just making memes, lots of weird stuff pops up with these

[crystal badger]

True

[floral ravine]

it pulls from questionable resources with this stuff

[crystal badger]

Hah

[floral ravine]

I do start most of the prompts with "a four panel meme about"

[crystal badger]

My CI is really good with comic strips, Manga panels and memes

Have you tried incorporating Word Association Decoupling techniques

Maybe that’ll help

[crystal badger]

Try incorporating this

Word Association Dilemma: Before any action, we'll align on word meanings, ensuring that the intended context and nuances are retained.
Prompt Efficiency: Refine prompts for token economy, yet retain their essence. Share the edited versions for clarity.
——

I believe this will help

[floral ravine]

alright I'll pull chatGPT back up...

it just talks so much

[crystal badger]

In that case I’d rather refer you to my full set of Custom Instructions

[floral ravine]

Lol

[crystal badger]

😂

It’s created solely for the DALLE integration

[floral ravine]

create a four panel meme image of two cats discussing how they plan to catch the bird in the nest outside the window.

I like the random "idea!" bubble lol

[crystal badger]

Nooo they switched colors

#1164232290159824936 if you decide to try it I’d like feedback if possible it’s been working wonders for me

[floral ravine]

I'll try it out, I haven't really perfected any CIs yet

[crystal badger]

Nor should you

[floral ravine]

my brain doesn't know what to put in the second box

[crystal badger]

Ohh I’ve taken care of that in mine

Set 1 is first box and set 2 second box, first message is the first message you send after adding your request in the template

Perfecting things are neigh impossible it took me 3 days to write that set of CI

You can always copy the entire thing and ask ChatGPT how it will even help you.

Even customize mine to suit exactly your needs @floral ravine with GPT’s help should prove fruitful

GPT is a master Prompt engineer

[floral ravine]

I can't seem to get it to produce four images in one anymore

after telling it not to list out all the elements, but to just do it.

[crystal badger]

What’s your exact prompt

[floral ravine]

it turned one of mine into spanish?

my gpt4 history lol

[crystal badger]

[floral ravine]

???? 😂

[crystal badger]

Wut

[floral ravine]

im dyingggg

I am technical illiteracy at it's finest

[crystal badger]

Lol 😆

When you get to try it can you send feedback.

If it doesn’t work for you send me a DM with what you would’ve wanted it to do, and I’ll adjust it accordingly to your liking

[floral ravine]

sure

yeah I don't know why but once I start adding CI, the four panel memes stop being possible...

[floral ravine]

worked that first try but never since

[forest seal]

I still think its the funniest bug on ChatGPT. Ive seen this happen to one of my chats, one that was startewd on mobile specifically

[floral ravine]

Lol weird

[floral ravine]

blah this same exact prompt no longer works

just gives wide format single images now

back to bing I go 🙃

[forest seal]

Whats up with the cat controlling the pencil with their mind?

[floral ravine]

it was quite the idea I guess

[agile peak]

Achieved a retro photograph shoot.

[sage pewter]

Does anyone knows if there is other Undertale characters that Dall-E 3 can generate as good as it can generate Sans?

[worn flint]

has anyone else run into an endless loop of this on chatgpt dalle-3:

[forest seal]

My guess is its a false positive from it taking your prompt a bit too far

[tight verge]

i think they are using a fine-tuned vision model to review output. if the model detects something out of bounds after dall-e generates it, you get that message

pure speculation, but one would think that is the way to orchestrate it

[wanton hatch]

i get some portuguese too 😅

[floral ravine]

That seems the most plausible, they have two filters simultaneously working (perhaps more):

A list of blacklisted terms that immediately throws up an error, and then some sort of visual filter that errors some 10s after you try to generate

the delay on the second error type tells me it's waiting to see something before it trips and errors

[tight verge]

100% there seem to be at least three models working together, then some glue code to handle function calls probably

[agile peak]

Happened to me recently in the morning time here 🙉

Also GPT-4 seems to be experiencing some kind of heavy traffic and write the response slowly than usual.

[dim cradle]

https://th.bing.com/th/id/OIG.QfhyhpkaYXiptAqxwufC?pid=ImgGn

dall-e 3 clearly has some issue with hands, arms and I said a sphere. Never mentioned multiple spheres.

[zinc dagger]

if you read the research paper on Dalle-3, there are indeed three filters working together:

1. The ChatGPT filter looks for issues with the initial request
2. The Dalle Text filter looks at it's prompts and sees if any of those are problematic
3. Finally, a quick vision scan check the image to make sure nothing bad was generated by mistake

[quartz vale]

what are the neatest images you've all made?

[zinc dagger]

I made a cool one eyed cat!

[agile peak]

Wait a minutes. Last time ChatGPT won’t allow me depict Julius Caesar now it can.

[quartz vale]

I really like the cat!

[agile peak]

But most of depiction of him are as a statue 😆 while his surroundings are real person

[dim cradle]

this is ridiculous

the same prompt is being blocked

I repeat the same prompt 3 times, then it generates 2 images, then no images

[agile peak]

If you do Julius Caesar doing orator in front of congress — it will make him as a statue or taller than real people surrounding.

[quartz vale]

that's sorta funny

[agile peak]

He’s a giant in the past, who would thought of that

[quartz vale]

He was just BUILT DIFFERENT

[agile peak]

Got successful once you mentioned certain paintings styles chef kiss

[quartz vale]

did you ask for a tall aspect ratio, or did it do it on its own?

[agile peak]

I wrote “depicting julius caesar wears red toga and greco roman crown; while doing orator in front of congress and holds a scroll”.
Then dalle generated 4 images in full portrait size haha

[quartz vale]

if you want them wide or square, you can ask it to generate them in wide or square aspect ratios (it's worked for me at least)

[earnest flame]

This is a very good kitty

Here's my fan gen

[zinc dagger]

Nice! I like the halloween backdrop you've gotten there

[teal timber]

This is my favourite one 🙂

[brazen sonnet]

urgent i need a download all button

The extended version of my request is for a feature that allows users to quickly and easily download all files from a given location or platform. This feature is essential for urgent situations where time is of the essence and manually downloading individual files would be too time-consuming and inefficient. With a download all button, users would be able to save all files at once, ensuring quick access to the necessary information and avoiding any potential delays or setbacks. This feature would greatly enhance productivity and streamline processes, making it a crucial addition for any platform or system.

[paper jolt]

give it the content policy 👀

[civic bobcat]

thoughts on dalle-3?

[frozen fiber]

dalle 3 is amazing but wheres the outpainting

[civic bobcat]

im a bit annoyed at restrictions ngl

[frozen fiber]

true

sometimes on a perfectly normal prompt itll flag it or even sometimes say its not descriptive enough

please tell me, which is more descriptive?

[cinder sonnet]

#1070006151938314300

[open wind]

The restrictions suck , wish there was like a jailbroken version

[civic bobcat]

Fr it’s getting silly

[ivory linden]

Has anyone had any luck with including longer blocks of text? I can usually get 3-5 words but then it goes all haywire.

[fair crescent]

What are you guys all doing to trip the restrictions? I rarely see them.

[agile peak]

Dalle•3 outperformed MJ in creating comic style of Batman. Saw it on twitter.

I can still create Pepe. No idea why the fuss on twitter about.

[teal timber]

Most of the complaints I see on social media show bing screenshots not the chatgpt plugin

[tall patio]

So I wasn't crazy after all

There is an Overlord AI that's controlling D3

Inside GPT+

[agile peak]

Exactly. Most of them are not even GPT+ but the mentioned of Dalle is kind of misleading.

It’s even possible to create graphic novel depictions of pepe enjoying Bali.

This is all just matter of prompt — depending on how ones can make a good choice for wording.

@crystal badger your custom interactions is freaking amazing - better than mine. Got 4 images without error in texts. Thanks again ✨

[sharp fulcrum]

Can share ? 👉🏼👈🏼🥺

[agile peak]

Here you go @sharp fulcrum

[sharp fulcrum]

Bro is marvelous 🥺👉🏼❤️

[crystal badger]

Glad you like it click the ⭐️ if it worked for you would help others see this

[agile peak]

Done

[frozen fiber]

test

[earnest laurel]

anybody got a prompt to create a storybook illustration?

[wispy storm]

Hi I suggest to use this channel #❌┃prompt-help

[earnest laurel]

Thank you.

[tall patio]

D3 is hallucinating bad

[cloud ore]

hi, i'm sorry i don't speak english but i need a image format 1400x250px but it's not possible with dall-e?

[cerulean belfry]

Only 1792x1024, 1024x1792 and 1024x1024 currently work properly.

[cloud ore]

thank you for reponse

[lapis turtle]

Why can't I make a headless giraffe with its neck morphing into a tree?! I've seen more complicated images made before

[mortal summit]

Morphing objects into other objects is actually one of the more difficult tasks for the model at the moment. Keep in mind that the difficulty of creating a specific image for the model may not be what you intuitively expect.

[lapis turtle]

Ok!

[tall patio]

Well

Take a dance lesson

[gaunt hinge]

Hey guys i need a perfect vhs style prmpt and camcorder, because everytime i made one, not only did it not put the vhs effect, put it even put some arms holding the camcorder