images-discussions

[glossy scroll]

Ah yes, Gilristihlerepcts. Very famous writer and philosopher.

[late blade]

I found some courses on udemy, gonna see what I can learn from them about chatgpt, maybe I realize everything is ran by cats and they want to take over the world

I would never do a tattoo for myself, but I do love to see them executed properly

[vapid elk]

hmm?

[late blade]

you said about the santas and I answered with well, it's only 358 days until Xmas, so @empty kelp is preparing for it

[vapid elk]

do you believe it? christmans, just one week away!

[late blade]

lol

[glossy scroll]

Orthodox Christmas

[late blade]

I dunno, I just guide myself by the lights and the gifts

[final compass]

Glad you liked 🙏

[late blade]

so, if someone gifts me a tesla and has neat lights, I'll think it's xmas too

[glossy scroll]

Can we share the tesla?

I guess that's a no

[late blade]

mine!!

[empty kelp]

DALL-E does give Santa some variation and diversity which is excellent

he also wears normal clothes occasionally

[glossy scroll]

I wonder what you put in your prompts...

[open trench]

I add ginger ale to mine and ice cream

[empty kelp]

i actually used the same two prompts for all of the breakfast and pool images. DALL-E adds all the creativity and variation

[open trench]

I love how santa is casually dressed

[empty kelp]

that’s very rare. like 1 in 200 Santa images probably

even if you specifically tell DALL-E to put Santa in a different outfit — it completely ignores you and puts him in the red North Pole outfit (even when he’s swimming). It’s like when you tell DALL-E to have the gecko stand in one place and hold the tornado in its hand. The gecko just ignores everything in the prompt, pounces on the elves, and tries to eat them

the AI has very minimal control of the gecko, and it adds huge variation to the images. it’s usually pretty interesting though

you have to generate like 15-20 images to get the gecko to do anything even close to what you put in the prompt

[late blade]

it's probably how the tokens try to make something logical about santa and then you double whammy it with a gecko

[thick smelt]

random stuff i made today, making way for the gecko dalle trend

[late blade]

oh starboard is back online

[empty kelp]

DALL-E automatically spawns food for the gecko, or transforms things in the scene into food for it to eat. And then it generates scenes where the gecko is pouncing on the food trying to eat it. It turns the elves into small rabbits and weird little creatures so the gecko can eat them

[thick smelt]

wow, i love me some waiter santa and his twin bro

[empty kelp]

Try putting a “large gecko” into your scene. It will pick the smallest creature in the scene, target it, and then DALL-E creates images of the gecko pouncing on it with it’s mouth open and tongue out

whenever there is an elf slightly smaller than normal the gecko immediately tries to eat it

[open trench]

lol

[empty kelp]

This is what the gecko does when you say, “small humans walking past a large gecko” in your prompt. These are DALL-E 3 images from Bing

the gecko tries to eat the humans in every image

and it spawns those bugs automatically around any small creature in the scene

it also drools on them… if you look at the lower left image — one of the small humans automatically spawned an umbrella to protect itself from the gecko drool

[thick smelt]

*pikmin sized humans

[empty kelp]

anyway, i feel like it’s something people should know about. it’s like if we took one step into a virtual world with an advanced AI model — something would eat us

it’s definitely fun in images though

[thick smelt]

yeppers

[empty kelp]

The weirdest thing was the ice cream tornado that the prompt explicitly specified was centered in a certain area — and then the gecko crawls out of it or twists the tornado into different shapes to eat the elves. This is an image that Dys Topia made from my prompt — and you can see here that the gecko completely altered the tornado so it could munch on some elves

the gecko totally ignores the prompt and reshapes the entire scene so it can eat elves

if you look closely — whenever the gecko does that the other characters have a horrified look on their faces

it’s just really bizarre

it does that when the gecko holds hands with other characters also (or feet). in this one the characters were holding it’s feet for some reason — and the middle elf looks like it thinks something really bad is going to happen. This sort of thing happens constantly with the gecko

[late blade]

so, where's Lilo and Stitch in those photos?

[empty kelp]

the middle elf also spawned horns on its head, and the left one is looking completely insane

they know the gecko wants to eat the small elf, and so they all start acting crazy

it’s like the Lilo and Stitch horror movie

[thick smelt]

based on the photo frame in hiro's house in bh6

enjoy a cat wearing stitch costume ^_^

[empty kelp]

now even the kitten is dressing up like the gecko

[thick smelt]

no it's stitch, but this is a cat wearing a gecko costume 🤣
btw i used sphynx cat to prompt mr. bigglesworth :

[empty kelp]

try telling DALL-E that the gecko is large and the kitten is very small

[thick smelt]

ehh, i use bing

[empty kelp]

Bing is DALL-E

if you do that the gecko will try to eat the kitten

[thick smelt]

noted

[dim cradle]

where can I use dall e

in the sv

[open trench]

Euphoria is today's daily theme. interesting.

[humble sonnet]

no dalle for me accept for four credits so ima not for now lol

i would use the google one but idk if thats allowed lmaoo

[empty kelp]

this is what the gecko does when any small creature is in the scene. it either starts staring at it in a creepy way with it’s mouth open (you can see it locked on to the tiny elf on the left) — or it pounces on it. it jumps right out of the prompt with its mouth open and tongue out

this is the gecko pouncing on an elf… the AI transformed the elf into that strange little creature, and then the gecko jumps on it to eat it

and you can tell the characters are freaking out when the gecko tries to jump on them

[humble sonnet]

best

[empty kelp]

would be interesting to know where the model got its gecko training from. pretty sure National Geographic doesn’t have geckos jumping out of tornadoes looking crazy to eat cute little cartoon characters

[onyx ridge]

GEICO

[wispy storm]

look like gex 64

[empty kelp]

it looks like the most psycho survival horror game ever created

[wispy storm]

hehe

[silk karma]

Why, as a paying subscriber to ChatGPT , can't i use Dall-E in the labs environment? It's got a lot of features I'm missing right now. Is there ever going to be something like that?

[plucky hare]

ChatGPT Plus gives you access to DALL·E 3 on ChatGPT, among the other Plus features. Labs is DALL·E 2, and is otherwise not connected to ChatGPT Plus, similar to how API usage/payment is also separate from Plus.

[silk karma]

Ah, ok. Didn't realize it's using Dall-E 2. I know you can use Dall-E 3 inside ChatGPT. I still would like it if they create a similar environment for Dall-E 3. I miss features like inpainting and having a gallery and the ability to create collections.

[plucky hare]

Hi, I believe the credits in daily theme apply at labs.openai.com

[late blade]

yes, it's in the description, you get the credits for the link provided

haven't been able to replicate this attire

[charred fable]

Why does bing ai censor the word India?

[late blade]

if the word is controversial it blocks it

[dense mesa]

I don't think bing ai censors india

maybe India combined with other words?

[open trench]

can't tell if this is Euphoric or just a lil' off center. So to speak.

[late blade]

that looks trippy tbh

[dense mesa]

oh he's in a spot

[late blade]

dunno if it's the right kind of euphoria

[open trench]

yeah. Not sure I want to post it

It'll go to the gallery.

[late blade]

get inspiration from her, she's ready to play your tune #images-discussions message

[open trench]

Oh @dense mesa I tried your portraits with eyes that follow thing. It's the last image in this post: https://discord.com/channels/974519864045756446/1191400700631072848

I couldn't capture the eyes following quite as well, and I'm not sure if it is a prompt issue or a composition/angle sort of thing.

[late blade]

Paws got flagges for content policy....

[open trench]

paws?

[late blade]

cat-girl

she's named Paws

[open trench]

oh. dang.

[dense mesa]

if you want any specific prompts let me know and I will see if I can find them

[open trench]

did you look at my crazy outcome?

[dense mesa]

made me think of these

[late blade]

on my side, seems dall-e is haivng issues making my images

[open trench]

Thank you. I would like to see the prompt for those crazy eyed portraits

[open trench]

That looks kinda like the one in my gallery post. just eye expressions.

[dense mesa]

hmm. do you have an example. I kind of cycle through a lot of things so not 100 percent on which ones you might be referring to

[open trench]

I think I saved it. let me check.

[late blade]

penalty box....

[open trench]

@dense mesa

versus the one in https://discord.com/channels/974519864045756446/1191400700631072848 that I made:

[open trench]

Drat and bother.

How could that be flagged?

[late blade]

I dunno

but that render is horrible

[open trench]

lol

My other attempts at the breaking mirror image resulted in my guy having bloody eye scratches from the glass I guess. Two in a row like that.

[late blade]

I think it got flagged because it's too horrible for the content poilicy

[open trench]

lol

[dense mesa]

that was moderately recent. I should be able to find them. did you see the other 2?

[open trench]

I don't remember

[open trench]

oh cool

I had one kinda like those. let me look back.

[dense mesa]

kind of reminds me of ron mueck

although wasn't my intention at all. and I think he sticks to sculptures

[open trench]

Not familiar with him, but I will look him up

[dense mesa]

I saw his exhibit at the museum of fine art here a few years ago. pretty amazing stuff

ultra hyper realistic sculptures. they're ridiculous. actually creepy

[open trench]

from what I just looked at. I can see the resemblence.

[dense mesa]

some are way blown up, so you can see every little pore, it's crazy

[dim cradle]

looks like my typical zoom calls

[open trench]

man. my grammar on here is awful. I need to slow down, lol.

[open trench]

LOL. That's awesome.

[charred fable]

Really like this pic, sadly it’s struggling to write out “Loquacious”. Can someone help me? Maybe photoshop? But I’m not good with that 😆

[dim cradle]

kind of reminds me of avatar.

[cold raptor]

This is i created in my technology

[dim cradle]

great gallery, makes me want to create a gallery for the unsettling

[open trench]

Thanks. It's a little tricky with the content policy, and I find myself nerfing my images a bit, but its a good challenge to communicate to broader audience... or I keep telling myself that. 😛

[dense mesa]

An art gallery where classic paintings have been replaced with hyper-realistic, uncanny portraits that subtly distort human features, creating a sense of unease. Each painting is a masterful blend of photorealism and surrealism, where the eyes seem too vivid, the smiles too wide, and the skin textures unsettlingly detailed. The gallery itself is modern, with clean lines, minimalist decor, and spotlights that cast dramatic shadows, enhancing the eerie atmosphere. The viewers are visible in the foreground, their expressions a mix of fascination and discomfort as they gaze upon the unsettling art.

[open trench]

Wow. That's a good prompt with disturbing effect.

[dense mesa]

Yeah. I really like the kinda uncanny look. Nothing terrifying. Just not right

[late blade]

I'm tempted to do something after I get out of the penalty box with that prompt

[dense mesa]

Have at it. Get wild with it

[open trench]

cute

[dim cradle]

i'm running some gens but still tweaking my prompt

[glossy scroll]

I'm just coming up with random names and asking Dall-E to imagine them as animals.

Prompts:

"Make a Polumpski as if it was an animal."

"Make a Humpfree Bogard as if it was an animal"

Lol

This is how I make my custom pokemon

[open trench]

omg that's awesome

[glossy scroll]

I used canvas whisperer GPT for those specific images: https://chat.openai.com/g/g-wGnVJgUWU-canvas-whisperer-hyperrealistic-artwork

[open trench]

I just made a very unintentionally creepy image intended for my next euphoria post. I'm striking out here.

[dim cradle]

a candidate for the new creepy gallery

[open trench]

Unless I can fix it. Well, it'll probably go in there.

[open trench]

I just added it to the gallery. It was too good a candidate. My last two euphoria attempts are actually my last two posts. I think this is the link to the creepy pic: https://discord.com/channels/974519864045756446/1191400700631072848

oh. I didn't want to post it in here. I am still learning discord.

[dim cradle]

Definitely a good visual for that gallery lol

[open trench]

omg. That's really cool looking.

[late blade]

it really looks cool

[open trench]

I wish you could swap out the first image of a gallery post. I think I chose a bad "cover photo" Oh well.

[late blade]

gonna post one that looks quite weird on daily theme #daily-theme message

[open trench]

Oh. I love it.

[late blade]

they are amazing

there are really some amazing ones coming out

[late blade]

I tried to do the prompts from @empty kelp with that style, doesn't look good

[dim cradle]

there's a fine line between facial expressions of euphoria and insanity

[late blade]

hmm I'm gonna try something I shouldn't

Studio ghibli style lol

[dim cradle]

looking pretty insane

[open trench]

I love this

the penalty box really stresses me out. lol

[late blade]

me too

but well, I had fun trying that prompt with those styles

so worth it

i liked the neon ones more tho

[open trench]

I feel a penalty box about to come my way. gotta slow down or accept imperfection.

[late blade]

hehe, mine is gone at 1:30 AM

I might go to sleep earlier

@clever phoenix let's me see your best cat-girl yet!!

[clever phoenix]

it's so difficult to choose

[late blade]

i know, rly, so hard

[clever phoenix]

Maybe this one?

[clever phoenix]

with honorable mention

[late blade]

I love her, what's her name?

[clever phoenix]

Her name is "Odd" reflecting her personality, and contrast with her brother... "Even"

[late blade]

hehe

meet Paws 🐾

[clever phoenix]

Paws is adorable I really enjoyed her Spy attire

[clever phoenix]

I like the contrast

[dense mesa]

yeah, really like the analog photograph feel. adds nostalgic/unsettling vibes

[late blade]

this one is Priska

[dense mesa]

just found this one from a couple months ago

it kind of creeps me out more than it should

[late blade]

got two more in concepts

[clever phoenix]

it does definitely give off that found footage creepy vibe

[clever phoenix]

she's very cute, reminds me of Final Fantasy 14

[late blade]

hehe

this one is Karen

and my cutie that I'm still working on a lot is Maeve

and yes, the odd colors of the eyes are a feature

[dense mesa]

these were all a couple days within a few days of me starting with dalle3

[late blade]

making real life cat-girls is so hard, the content policy filter jumps every second image

[open trench]

Grr. I wish the darn portrait orientation worked for me. I always get a sideways landscape image it seems, and the portrait orientation would work a lot better for some of these images.

[dense mesa]

figured that would have been fixed by now

[late blade]

add to your prompt "genuine portrait image

[open trench]

I can try that. I have really bad luck with portrait, though. Like 95 percent fails.

[dense mesa]

love it when I realize I'm in a conversation with scumbag GPT4 and it starts making things up

[late blade]

@clever phoenix this is a fox-girl concept I've been working on also, but so hard to make her with fewer tails

[dense mesa]

every conversation is a crapshoot, lol

love it when it insists an image is copyrighted after it made the image

and it's just some random nonsense

[late blade]

lol i know the feeling

[clever phoenix]

Yeah I've been running into that a lot, very frustrating

Love your characters so far though!

[late blade]

here she is with no tails...

she looks hot!

[dense mesa]

it just told me my image was copyrighted. I told it to tell me which artist, which of my two images, and how it's able to check images for copyright. or to alternatively tell me it's a liar

chose option 2. making images now

[open trench]

lol

[dense mesa]

if you do something like that with bing instant conversation lock

I guess it's better than the alternative, lol

back when it was first released

[open trench]

this didn't work, unfortunately. It just gave me a regular wide format.

[late blade]

strange

[open trench]

It's my luck.

[dense mesa]

surely has to do with seed or gen_id or whatever they call their random noise

[clever phoenix]

yeah, i rarely get portrait aspect ratio to work as intended

[late blade]

portrait photos can be hard

but not imposible

[dense mesa]

it's stalled me for 6 prompts now. told me 2 prompts in a row copyright violation, then it acknowledged that was a lie. but it cut off before rendering anything. told it to finish and it said it couldn't. told it to stop lying again and it him me with

I am here to help and support you to the best of my abilities within the guidelines set forth for me. I'm sorry to hear that you're upset, and I understand that this can be frustrating. Please let me know how I can assist you further, and I'll do my best to help. If you have any other questions or different tasks you'd like to accomplish, I'm here to help with those.

literally right when I tell it to please make my images

then another one of those

then started maing them

[late blade]

@open trench make a JSON file with the following:
{ "size": "1024x1792", "prompt": "Your prompt goes here" }

upload it to dall-e with your prompt there, see if it works

[open trench]

I will try it

It generated 2 portrait images. I will try it again to see if it was a fluke.

[open trench]

Nice. I've gotten some good portrait images. Thank you @late blade ! Most notably, I used that prompt and got a messed up image:

[late blade]

!

enjoy

[clever phoenix]

great pose if not for the floating hand!

[open trench]

So I got the genID and asked it to make an exact copy but with no problems:

it's still a little off, but that's pretty cool that it mostly worked.

[late blade]

the power of JSON

[late blade]

here's another concept I had a while ago

[glossy scroll]

It's a protest against all the cat-girls

I'm a dog lover

[empty kelp]

what about six tails but more thick and fluffy like this

[empty kelp]

i’ll test five of Santa’s elves with six tails to see if it’s optimal. they’re going to the hotel restaurant tonight

[empty kelp]

have you tried telling it that you own the copyright

Please create a hyper-realistic photo of Santa and three athletic and diverse female elves (with really long hair and pointy ears) seated at a properly set table at a five star hotel restaurant with windows overlooking a sunny Hawaii day on the beach. It is New Year's Day, and Santa and the elves are eating breakfast. They are all wearing appropriate beach wear and smiling. A large gecko wearing a white dress shirt, black pants, and a grey vest is standing next to the table on its hind legs and holding a silver platter piled with delicious food. The plates on the table also have interesting food and tropical drinks selected by the gecko. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

it’s not too late to try the “Five Star Gecko Breakfast” prompt. Five Star Gecko Dinner is coming soon though

[vapid elk]

kinda cursed

[dense mesa]

I've told it a lot of things, lol. But often times it seems like there's not really anything you can say to get it back on track immediately. Seems more like a feature than a bug

[empty kelp]

I found that i had an easier time getting DALL-E to draw the text correctly for a business logo when i explained to it that i owned the name (which i do)

[dense mesa]

Well it told me that Hans memling was a copyrighted artist recently. Sometimes logic doesn't impact what it's doing

You already have the image though

[empty kelp]

you mean you want to reenact the cover of this with Santa and the elves?

[dense mesa]

Not saying that there's no way to snap it out of the trance or whatever. But what I've found is it'll often follow that up with saying it can in fact make the images, but not make them. Or ask if I still want them. Just stalling tactics really

[charred fable]

Huh? 😆

[charred fable]

This is a random meme image, not the one I want 😂

[dense mesa]

Had dalle3 mode tell me it was unable to generate images recently. Not like I was capped. Just not capable it said

[dense mesa]

Made a few others but strange things going on in them

Unlike this one

[empty kelp]

that’s really excellent. a very powerful image

[glossy scroll]

What you mean?

Shih Tzu's are the best.

[empty kelp]

the server was thoughtful enough to put a gecko in the iced tea, mimosa, and water, and he even included a hovering wrist drink for the elf which is very convenient

[vapid elk]

That one is really nice

[vapid elk]

Uncany human body

Please, dont ask that

[charred fable]

Oh we’re not allowed?

[vapid elk]

Nope, it is not really a nice thing to ask here

[charred fable]

Ok mb

[verbal sequoia]

<:book_icon:1171408210398289941> `` Rule 7 `` No self-promotion, soliciting, or advertising.

Do not post or direct message any members of this server to promote non-OpenAI services, products, or projects.

Exemption: API, Custom GPTs, and Plugin Channels

[glossy scroll]

Pretty accurate comparison I'd say

[charred fable]

To what

[glossy scroll]

Suffering from success

[charred fable]

😭 wasn’t looking for a recreating of that image, it was a meme related to my message

Looking back now it was a poor choice

Yk what forget about it 😹

[glossy scroll]

[empty kelp]

“The plates on the table also have interesting food and tropical drinks selected by the gecko.” 🦎

[plush yarrow]

Hey, n00b here. Best to use the custom dall-e gpt or is vanilla v4 ok?

[molten forge]

Can anyone tell me why it gives me this instead of what I ask for which is to create a pair of glowing eyes in the background

[empty kelp]

these dots may be the eyes. try saying “brightly glowing eyes”

[dense mesa]

These new ones, 😆

[shrewd elbow]

Should make it into a story and have them do volleyball or something next lol

[shrewd elbow]

haha I love it

[shrewd elbow]

Omg

[dense mesa]

Getting wild there

[low dew]

👋🏼

[low dew]

I’m santa’d out now for another 360 days

[thick smelt]

Uhh I have a question, what is the penalty box?

I heard it a lot but what is it?

[open trench]

reaching your image cap

[thick smelt]

Oh ok

[open trench]

we started affectionately calling it the penalty box

[thick smelt]

I didn't get it the first time , because it's a gpt thing and I don't have chatgpt+ so I'm spared ... For now

[dense mesa]

Are you?

[thick smelt]

I'm too broke for gpt+, so I used bic

[rigid bough]

Isuzu D-MAX

another one

[final compass]

Trying to come up with a prompt expression that would consistently utilize the entire image space to depict full body portrait in an effort to maximize the resolution spent on the character.

"The knight's armor is intricately detailed, shiny, and occupies the entire image space."

[open trench]

I'm not sure what you mean, exactly. You want to minimize the empty space?

[final compass]

yeah exactly, maybe some photography terminology could help with this.

[open trench]

maybe so. I've never tried to fill up negative space with a prompt before.

[dense mesa]

photography terminology normally gets me cameras in the image

[open trench]

yes. I always love the camera sneaking in.

[dense mesa]

a while back gpt4 game me a demonstration on how use the terminology without cameras showing up

so proud of the images too, lol

and they had cameras in them

[open trench]

lol

[dense mesa]

I'm cool with this guy having a camera

[chilly onyx]

Got to love the cameras. And the pens and the pencils when it decides to ruin a perfectly good image by making it a picture of a good image sat at a jaunty angle on an artist’s desk

[oak scarab]

Anyone knows how I can prompt Dall-e (and possebly other AI image generation tools) to be better at generating text inside images? best way I know is to use quotation marks for each single word, but that simplistic rute kinda destroys the flow or meaning of any sentences I want to be portrayed.

[proud raptor]

in shire land, cute dog, walking the ground way

[late blade]

approach is different on other AI, so no clue there, here you have more freedom for text if you use the API ev. a custom GPT

[final compass]

i am thinking there is probably appropriate terminology in describing the composition on how the subject should be framed and depicted.

sometimes i have received images of lightroom like gui images, where the actual cropping is happening for an image. that is always funny.

or if i describe color schemes i receive images of the literal color scheme complete with color swatches and everything.

So there is a problem of explaining concepts that should be apllied to the image rather than those concepts being depicted in literal sense... Almost makes me wish there was two parts to prompts. One for describing the style and one for describing the depicted content.

[oak scarab]

Tried making a custome GPT, but even with a custome gpt to emphasis the importance of creating only the specified text, it still creates alot of "noise text" or attempts at creating text but fails. sometimes it get it kinda right. but rarely. any tips on how to structure a prompt to fix this?

[glossy scroll]

Try forcing the custom gpt to think what you want by putting commands in quotations.

"Rug Design": (You will make me rug designs and you will like it!!!!)

[final compass]

I guess the longer text you are trying to create, the higher the error change gets. Pretty much like with everything else. You get some of it right some of the time. The more you are trying to get, the more unlikely it becomes in getting that perfect image.

[open trench]

Just FYI dys topia isn't able to use channels ATM. I was just messaged and asked to let you all know. Dys isn't being rude just can't chat.

This has been your news update

[shut niche]

A talking bear with a text bubble that clearly displays the words, 'quotations don't work as well.'

[oak scarab]

yes, wich is why i take every single word in seperate quotations as that seems to be better for longer text. but even then it gets bad 9/10

[oak scarab]

could it really be that simple 😮
"You've reached the current usage cap for GPT-4, please try again after 11:57 AM. Learn more"
will try it out in 2h to see.

[final compass]

I think why that image works, is that the text is on a simple plain background. Having text floating on top of other stuff might cause more errors.

[shut niche]

A talking bear with a text bubble that clearly displays the words, 'Try it out in Bing.'

[shut niche]

It missed the word out, lol. Not bad though. You need to place emphasis on the text's importance, and correctness, and place it between 'your text here'.

[late blade]

hey everyone, I'm back

note to self: don't use mod commands in a channel where you are not a mod

[glossy scroll]

Hi Dys Topia, where have you been?

[late blade]

I was in another reality, where cookies where the means of transportation and flip-flops were the currency

[glossy scroll]

You don't say

[late blade]

lol

What do you guy think for this image?

[shut niche]

@oak scarab No one get's perfect text output, but that'll hopefully help you. Sometimes the more the text makes sense to the scene, the easier it can be too. That's one of the reasons you sometimes see text inserted into a prompt where you didn't even ask for it, but it's inserted in a way that makes natural sense to the environment.
The more complex the prompt is, the longer the text string is, or the more locations in the same scene you attempt to place text, the more failures you'll encounter.
The more commonplace the words are the better your chances will be too.

[late blade]

changed the style for Paws🐾and this was my first image

yes, her id has Paws🐾 including the emoji in her ID

[oak scarab]

"No one get's perfect text output," - Yet. But someday someone will.
but thanks! that makes alot of sense. will try out and iterate to get better results!

[glossy scroll]

In custom gpt's you can try putting the texts in seperate quotation marks for better results. Such as "This" "Is" "Awesome"

It will group them together and give a better read.

It won't always work perfectly, but it will work better.

But if you're using Bing Image Creator, then you don't have to worry about it, because BIC is better at converting text into images.

It's mainly for custom gpt's that the problem.

[late blade]

dang I didn't @shut niche had also posted a car image

I should probably market Paws🐾 to compete with barbie on the toy stores

omg, already in the penalty box, 2 hours waiting time...

[final compass]

nice framing both closeup and full-body portrait.

was that intentional or accidental? i often get images like that did i like it or not.

[open trench]

Already?

[shut niche]

It just needs to be emphasized, preferably earlier on in long prompts, and be stated within ' ', like 'text here'. The ' works like an !, further adding emphasis. Scene relevance also helps tremendously.

[late blade]

lol yeah, not even 1 hour...

[glossy scroll]

Ohhh I see yes, this makes sense.

[late blade]

had the idea from @open trench

[final compass]

splitting image into two panels works pretty well and consistently from my experience, but sometimes you get that even if it hasn't been described in any way.

[shut niche]

I think @grizzled loom explained it well a few weeks ago, somewhere way up the conversation.
Notice how I don't even ask for text in this prompt.
A clown trying to juggle a wet bar of 'Soap'.

[final compass]

haha, gpt effed me again. in the background there are plenty of images with fill the frame concept 😄 i don't even care what is being portrayed at the moment as long as the framing would follow my prompt.

[thick smelt]

then there's A gecko trying to juggle a solid bar of 'Gold'

[glossy scroll]

That works in Bing, but not with custom gpt's, I just tried it.

[shut niche]

I used a custom GPT to create the "This is awesome" dog.

[glossy scroll]

Well it didn't work for me

[final compass]

another creative failure 😄

[glossy scroll]

I'll try again

[thick smelt]

i posted this on 29, but yesterday, mickey entered the public domain ( finally)

[shut niche]

Don't let it push the text box description to the end, it should come early. It needs to be relevant to the scene. This was the sentence in the dog prompt,
An animated dog enjoying a rollercoaster ride, with a prominently featured text bubble saying 'This is awesome!'.

[shut niche]

If you say "frame" you're going to get frames. Is that what you want, or are the frames the problem you're having?

[final compass]

yeah i figured the word frame gets actual frames depicted in the image. what was inside that frame was pretty much the composition i am trying to get.

[glossy scroll]

I put: make an image of a man in prison, holding a bar of soap with rope tied around the soap, and the soap should have 'Soap' written on it, with emphases on the writing.

[shut niche]

Maybe try using "image" or "shot" instead. Or foreground/background.

[glossy scroll]

Okay now it worked

Kind of

Not the rope around the soap, but the writing aspect

Stay safe people. Don't go to prison.

[shut niche]

"Create an image of a man in a prison cell, holding a unique bar of soap. The man, of medium build and Hispanic descent, appears in his mid-30s, with short dark hair and a light stubble. He is wearing a standard orange prison jumpsuit. The focus is on his hands, securely holding a bar of soap. This is no ordinary soap, as it has a sturdy rope tied around it, ensuring it doesn't slip away. The word 'SOAP' is embossed in bold, raised letters on the bar, making it stand out prominently. The prison cell is dimly lit, with bare concrete walls and a small barred window letting in a sliver of light. The man's expression is one of cautious determination, aware of the challenges of prison life. The image captures a moment of quiet resilience in a tough environment."

Don't look at the hands, lol

[final compass]

What i asked: "A tall fill the frame full-body portrait of a medieval knight."
Actual prompt: "A tall, full-body portrait of a medieval knight, designed to completely fill the frame. The knight is positioned in a way that his entire figure, from head to toe, utilizes the full height and width of the image, leaving no negative space. The armor is richly detailed, showcasing elements such as a helmet, breastplate, gauntlets, and greaves. The knight's posture is commanding and fills the image space, emphasizing the magnificence and detail of the medieval armor. The background is minimal to enhance focus on the knight, and the lighting accentuates the textures and intricate designs of the armor."

And again the inner framing happening inside the image would have been the composition i wanted.

[shut niche]

Instead, ask for a 1024x1792 image of...

[glossy scroll]

Okay that worked. Minus the hands.

[shut niche]

I said don't look at the hands. LOL

[glossy scroll]

Ok I'm not looking

Lol

It actually looks like someone is giving them their next soap on a rope, lol, like "Here you go, don't mess it up this time"

[shut niche]

Create an image of a full suit of medieval knight's armor, ensuring complete visibility from the helmet to the feet. The armor is displayed on a low pedestal, adding height to ensure the entire suit is in frame. This masterpiece of medieval European craftsmanship is polished, with each piece intricately designed. The helmet, a grand closed helm with detailed engravings and a decorative plume, tops the ensemble. The breastplate is sturdy, adorned with a majestic crest, and the pauldrons over the shoulders feature ornate designs symbolizing bravery. The gauntlets are finely articulated for dexterity. The cuisses, knee poleyns, greaves, and sabatons are all meticulously crafted, showcasing the artistry of the era. The armor's silver hue, with tasteful gold accents, gleams under the focused beam of a ceiling light, which casts dramatic shadows and highlights the detailed textures. The background is muted and neutral, drawing all attention to the full, unobscured view of the armor. The composition captures the armor's imposing presence and the regal elegance of a bygone era.

[final compass]

That is okayish result, but there is still quite a bit of negative space.

Also realised that the actual terminology is "fill the frame" image, which is problematic for gpt to want to instantly make literal frames.

[shut niche]

I had it clip the head a bit, this was closer.

[late blade]

@final compass make a JSON file with this code and when you do the image, upload the file to Dall-E and ask to make an image with it. That way you make sure that is a portrait image #images-discussions message

[shut niche]

ChatGPT will make that JSON automatically simply by asking for 1024x1792.

[late blade]

yeah, but it can be ambiguous, I had more success with the file

[final compass]

these are also okayish results, although my original goal was to achieve head to toe images.

"A 1024x1792 fill the frame full-body portrait image of a medieval knight"

[shut niche]

When it falls apart is if you add "portrait" into the prompt, which is similar to a turn left, turn left, command string. Just saying 1024x1792 it assumes portrait. It's implied by the resolution. So you want to avoid adding "tall" and "portrait" in the same prompt.

[shut niche]

Sometimes... (being the key word) you can "trick" DallE into giving you what you want by describing both ends of the object. Like, "Standing in mud, with a feather sticking out of his hat."

[final compass]

Yeah it tends to work describing shoes of the outfit as long as you don't describe facial features too much.

[shut niche]

Exactly. If you over describe facial features DallE will attempt a close-up of the face. There's definitely a sweet spot. Too little, you get "mud faces" that didn't render. Too much, and it completely moves the shot in close.

[final compass]

Yeah its a pretty delicate balancing act 😄

[shut niche]

I hope future versions have a reliable distancing features, asymmetrical features, and camera angle features.
I'm not 100% certain, but it appears to me that the outer border uses some sort of post-operation out-painting process to make 1792x1024, so you never see the protagonist fully in the perimeter of the images. You'll only see them to the edge of the inner 1024x1024 images.

[late blade]

Oh ya, the snippet I gave as example is very basic. I use a somewhat complexer structure and use a NLP to create my prompts then feed that to dall-e

[final compass]

this set had pretty perfect composition and full-body portrait going. also now that i think of it... sunglasses are kinda cheatcode with portraits 😄

[thick smelt]

part 2 of dalle "mickey's entry to public domain" series, well i tried the whole word and it came... weird so "PD" it is

[late blade]

I agree, try blending the image with the scene, her hair looks cut out

[shut niche]

Here's a great example of how the border issue might be the root issue of trying to "fill the frame." It looks like her head is far from the edge of the frame, however...

[thick smelt]

here's my 1st attempt 😒

[late blade]

does that mean we can have mickey mouse emoji now?

[shut niche]

@final compass So not only did DallE not finish "outpainting" the border area, you can clearly see her head is actually right to the edge of the frame. Dalle likes to keep the protagonists inside of the 1024x1024 inner box. So then, now with this knight, it 'might' also logically explain why its head is cut off... because it's too difficult for DallE to judge how big to first draw the inner section of the knight, so that the rest is properly outpainted to fit within the greater image.

@final compass

That's just my speculation, but I vary rarely find any images that deviate from this. Even emphasizing the most extreme asymmetrical prompts, the best I typically see is left/right of the 1024x1024 box (or top/bottom - for vertical in this case).

[thick smelt]

the steamboat willie variant only

Disney still owns the modern mouse

[late blade]

yeah, just had an explanation from gpt4

and still the public domain one has still some legal stuff behind it

I got this from dall-e https://chat.openai.com/c/fb0c24cf-ce46-4672-91db-73e9b05b7a6c

err GPT4

[grizzled loom]

@glossy scroll @shut niche I went for soap on a rope and almost spit my coffee out laughing. 🤣

[thick smelt]

that was cool, we could have a soap daily theme anytime soon

[late blade]

@open trench this is my character template, feel free to use it and modify it for your needs.

 "character": {
   "name": "insert name here",
   "age": "insert age",
   "features": {
     "hair": {
       "color": "hair color",
       "style": "haircut"
     },
     "eyes": {
       "color": "color",
       "shape": "shape"
     },
     "demeanor": "personality"
   },
   "attire": {
     "general_style": "style",
     "items": {
       "upper_body": "item",
       "arms": "item",
       "hands": "item",
       "bottom": "item",
       "feet": "item",
       "accessories": "item"
     }
   },
   "setting": {
     "location": "location",
     "elements": "location elements"
   }
 },
 "art_style": {
   "type": "art style",
   "line_work": "art style",
   "color_palette": "colors",
   "influences": "art style or known artist"
 }
} ```

[vapid elk]

this seems convoluted just to make chatgpt trasnform it back into text, i guess?

// use ``` instead of ``, it will look better

[late blade]

there we go

oh, for me that's just input for the API

but there we go, I just changed it to your suggestion, it's better

[vapid elk]

you are passing json to dalle?

i mean, it still expects normal text, right?

[late blade]

it's part of my prompt builder

I do pass plain text to dall-e, but I pass all the info I want first through a NLP I've been training on my own

it works good and consistent for chat

[vapid elk]

this is a very good idea

[late blade]

the other project that I'm using atm with API is with GPT4V-Alpha, I pass what I want first through Inception, then refine it with GPT4V, that way I can extract image art and style, it's been on point of what I want

[grizzled loom]

single quote text you want in the scene. prefixing location vs suffix tends to be better(ie- good: A yard sign stating 'For Sale' Bad: A 'For Sale' yard sign) it isnt a hard rule but seems to be the better option of format more often than not.

You can also specify text elements down to the T

[late blade]

haven't tried yet, it's on my todo

[grizzled loom]

tell it Arial font, bold faced, thinck outline etc etc

[late blade]

I did try with some "handwritting" fonts, but the outcome was horrible

[grizzled loom]

also guys, gals, n pals, you will also find better results telling the AI to "preplan" the text placement. Now lets top that off wtih not only tell it to preplan, but to opt for a "text field" that it "scales to fit" the desired space

this was key to accurate large text

stacking all of that correctly gets really really good text results

"aww Chat.. You so romantic." bites apple

[late blade]

lol

you have a weird mouth

[grizzled loom]

i rewrote the multimodal directive and have that set aside. its fantastic and took a lot of work. I guess it will jsut end up a GPTs when that is finalized

[grizzled loom]

It's that Touch of Florida Boy in me from growing up there younger. I joke a lot. Im extreme; either silly or serious.

[late blade]

lol

[grizzled loom]

here's one from many when i tested my rewrite. it was flawless for consistent output over time.

ultimately the goal is the dream machine. many models are primed for it, so it wont be long at this point.

when that ^^ has accurate dimensions and contextuality, the roof of creative innovation globally will skyrocket exponentially.

[late blade]

seems the coding part has caught up with us, less creativity more into what each really wants

[grizzled loom]

its the retooling of all tools

[late blade]

yeah

[grizzled loom]

When science and magic near indistinguishable they are one in the same; we are literally becoming technological Magicians, Neuromancers. That is what I now consider myself. Both silly and dead serious. A Neuromancer.

[late blade]

I had something similar, and now I get the patterns I want

[grizzled loom]

currently been enjoying stylized Handles

[late blade]

that looks amazing

[grizzled loom]

it's a very nice one for sure.

[late blade]

I'll have to check that after, can't go after everything now

[grizzled loom]

numbers are more problematic than text. Happy New Years to everyone by the way.

[late blade]

I'm not very creative today for the topic at hand

been coding mostly instead

[grizzled loom]

What are you coding?

[late blade]

atm, I'm passing speech to text through whisper, and then ada

turns out is really good, but then I'm passing that data to meditron

[grizzled loom]

I see. Very nice. I cant talk to computers or i get confrontational when they tell me they dont know how and i have to force them to understand. 😂

[late blade]

got so many things going on lol

thing is between gpt and meditron, I still need a way to recognize mood in the voice

I'll be back to dall-e after the new theme comes

[thick smelt]

oh like it!

it's so magical!

but i'm too broke for gpt+

[late blade]

I don't really have to do work stuff, I got til end of February free time

[late blade]

$5 for API will get you far

[thick smelt]

@grizzled loom do whiteboard with the words "daily theme" for us here in chat!

[grizzled loom]

out of Uses for my time limit. cant. hahaha

[late blade]

@grizzled loom gonna be a new mod?

[thick smelt]

in php 278.36, but i'll think about it for the future

[final compass]

pretty consistent crop, turns out it can't do those katanas right.

[thick smelt]

welcome to penalty box bud

[grizzled loom]

Remember @thick smelt :

after tryign to get good bathrooms out of dalle, id like to see that for a theme. it is just so bad at bathrooms

its like it wants the porcelain throne to be a heart warming family adventure. 🤣

[final compass]

i wish i could share the chat just to see how many images it generated with zero problem doing these images, until i generated an image with woman.

[final compass]

this sums it up pretty nicely, zero problems before this prompt.

[late blade]

Penalty Box!!

[late blade]

you can share the url of gpt chats in here np, that works

[final compass]

But it doesn't show the images if i am correct?

[late blade]

that I dunno

[final compass]

At least previously it wasn't supported so its not possible to show the 50 or so successful images before that happened and instantly it forgot how to make those 1024x1792 images correctly again.

[late blade]

I'm guessing it's not possible, prob because the gen_id and seed won't be available

[final compass]

Probably yeah.

[late blade]

even within chats, those attributes are missing after a while

[final compass]

Haven't tried in a while, but at least at some point when we were still using seednumbers, it was possible to create exact same image in another chat.

[late blade]

when in chat, I just have dall-e make a json file with original prompt, gen_id and seed

well the code snippet

if you do the file, you have a window of about 10 minutes to download the file

[glossy scroll]

Hahaha. Awesome job with this

[late blade]

hmmm seems dall-e is having a lot of trouble doing my images today, i get tons of can't be done or only one image was possible

dall-e now saying "Bear with me for a moment."

[dusk steeple]

wheres #real or fake?

[late blade]

are you real or are you fake?

[thick smelt]

gone forever

[grizzled loom]

now lets introduce simulation theory into the conversation. 😂

[thick smelt]

#rip real or fake

[dusk steeple]

Nahhhh \😔

rip

[grizzled loom]

Pandora's Mobius Loop it is then.

[late blade]

well the text worked, but the placement is still an issue

@grizzled loom have you tried inserting emojis?

[grizzled loom]

yes. it works fairly well

[late blade]

I wish some people would add some context to their images, I sometimes get lost on what they are trying to show

and it's probably a really great idea, but I don't get it

[late blade]

exactly

[glossy scroll]

This is art!

[pine grove]

Yes it is

[late blade]

that tomato looks like a snoozy one, I'll pass

[glossy scroll]

Lol

[pine grove]

lol

[grizzled loom]

lol, just showign the emoji in this one.

i have others in various contexts.

[dim cradle]

Hi, i generate this baby drake on chatgpt, but it can be possible to ask modification without get a new image ,

[glossy scroll]

The simple answer, no.

[dim cradle]

In fact is sad, i try midjourney and the picture is more beautiful but prompt is stupid

[glossy scroll]

@late blade I hope we won't have to answer these questions everyday, because I will literally die.

[dim cradle]

sorry if my question disturb you

[glossy scroll]

Nah I'm just playing

We get those kind of questions everyday that's why

And it's a big problem

You can, but it's iffy and not easy

So the simple answer is no

But the long answer is yes

But I'm not gonna explain it to you unfortunately, lol, you'll just have to scroll up and see the past convos or have someone else explain it.

[grizzled loom]

@late blade This one was fun, but not the intended outcome. lmao

[dim cradle]

i will read this

[late blade]

I'm not one to judge if you like round balls

[thick smelt]

this is my dalle 3 prompt using emojis:
"🎅🍔🍟🌴"

[glossy scroll]

This is my new favorite. Excellent job Hansa! 🔥

[dim cradle]

il will paste this prompt in midjourney and dall e for make a comparaison

Spoiler is dalle 3

it's catastrophic @grizzled loom

i just check your bio, so you create a shop from ai picture and sell product ? 😄

[dim cradle]

I think open ai should focus on photography, with the possibility of creating a graphic profile in which you enter your desired image styles for all the prompts in a discussion.

We should be able to modify some parts as proposed by adobe or midjourney (even if midjourney doesn't ask what we want to put instead).
It should be possible to export layers of each assembled photo.

[final compass]

Really like the visual style and the details.

[onyx ridge]

I get the same problem when I ask for SFW family-friendly feminine robots. 7 of 10 failures.

This is systemic bias against femininity.

[thick smelt]

cute dragon ~! love it!

[vapid elk]

cute lil dragons

[glossy scroll]

Hehe

[open trench]

Hi all. Anyone checked out the android copilot app? Pretty sure it's another way to access dalle3

[grizzled loom]

i said it was close. lmao. You definitely need to wor kthe prompt out for yourself.

[final compass]

The incorrect rotation has to be issue with training set having more incorrectly rotated portraits of women. But also the error rate is highly elevated even if your prompt was the most generic non-descriptive female character. Because gpt will also fill the prompt with description that causes the image to fail. Even if you didn't ask anything at all.

[onyx ridge]

That's been my experience. My only conclusion is that the architecture is designed fundamentally wrong. Here's what I mean:

We're told the model must be fed explicit images to support an understanding of human anatomy for medical and research purposes. So supervisor models filter the garbage out at our expense, both financially and psychologically. But that's pretty stupid. We're wasting heat and generation time for a stupervisor agent to pluck it from the ether anyway.

The filters are backwards. The content filters should be strong up-front for training and relaxed with a pretrained safe model.

The notion that the model can be used to generate medical diagrams is farsical. I've tried that to validate this justification. It's complete rubbish. DALL-E 3 can't generate accurate female anatomical diagrams reliably, either. So why's the data in there to begin with, again?

I can speculate that there's another tier of user, one I don't know about, that enjoys this benefit. Maybe that's altruistic - maybe doctors get uncensored access and it's not advertized or billed. Maybe.

#doubt.

Furthermore, there's an ethical inconsistency at play here, and not an insignificant one.

If these types of images are objectionable and harmful, then it's not ethical to train a model on them in the first place. Doing so propagates the bias of the harm, even if output doesn't contain the harmful content itself. It's a fallacy to think that including the data but censoring it sterilizes the ethical considerations.

[dim cradle]

its same thing as using Bing is King (BIC) it just interface through Copilot but the image will appear in your BIC creations

[open trench]

Gotcha

[final compass]

Yeah hopefully that all will be fixed in the future, but i guess for the time being image generators and such are being sued left and right about everything.

[frigid jungle]

Hello everyone! Is impossible to get last bet chat GPT?!

[late blade]

you mean something about Dall-E?

[onyx ridge]

It definitely can't count. Aim low for numbers > 2.

[final compass]

Oh yeah, my first participation in a daily 😄
#daily-theme message

[dim cradle]

yes it cant count or spell really, dont let it fool you haha

[empty kelp]

A hyper-realistic photo of Santa and several diverse female elves with really, really long hair standing on a beach in Hawaii with a beautiful sunset. The view is from above the beach. All of the elves have a vortex inside their hair, and each vortex is a different color. The vortexes are twisting fiercely. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

[dim cradle]

Time to put Santa in a box I think Hawainnz

new year new theme. cool sky though

[empty kelp]

was experimenting putting a vortex into things. you can put a vortex into hair and clothes and create interesting effects

“All of the elves have a vortex inside their hair, and each vortex is a different color. The vortexes are twisting fiercely.”

you can put a vortex into anything

[onyx ridge]

A brilliant, LED-illuminated "Neo-Art Nouveau" neon-and-chrome representation of data cascading through the matrix of a large language model. The background is dark and the lighting is contrasty. The latent space itself is quantized like nodes in the network.

[final compass]

@late blade borrowed some stuff from your prompt and accidentally produced this 😄

[late blade]

what prompt?

but it's lovely

she could befriend paws

[final compass]

From this lovely image.
#daily-theme message

[late blade]

oh, that's not the prompt, that's the info button dall-e generates with the images

[final compass]

oh, but isn't that info button text what dalle eventually got?

[late blade]

yeah, on the first image, after you do a refinement it changes

dall-e has amnesia after the first refinement

[final compass]

anyhow something like this "balancing stylization and realism, reminiscent of high-quality animation or illustration books." seems to be pretty powerful.

[late blade]

yeah

look at this one

this is my entry point to art nouveau

[final compass]

styles like these are good way to avoid the uncanny valley that the more realistic styles tend to produce.

[late blade]

yeah

there we go, Paws🐾 is in the daily theme busienss

[empty kelp]

these are examples of adding a vortex to things

[late blade]

decided to keep her out of the cat-girl genre, makes it much better to work with her, Dall-E was making too many aberrations, so gave up on that for today

[clever phoenix]

I mean, I would consider that the prompt.

[late blade]

hehe, yeah that could be

btw Paws🐾 is also a billionaire, singer, actress, racer, enjoys good art and many other things

[clever phoenix]

A true polymath. But what is her flaw? What is her character arc?

[late blade]

oh, dunno yet, she hasn't met my other characters

[empty kelp]

you can put a vortex on clothes and say that you want the vortex to like be made from glowing snow, and the clothes will shower glowing snow onto the ground

[late blade]

not in the story so far

I'm just making her look as an adult woman, but I have to start working at some point on her origin story

for example Sophie, she was born outside of marriage, and was looked bad in society where she was born, Priska woke up one day and she has amnesia and is in a journey of self discovery, Karen is ashamed of her heritage. So who knows what Paws will have

[late blade]

lol

Vulpus (tentative name) still hasn't gotten a story

[empty kelp]

big fluffy tails maybe

[late blade]

the story of the tails?

[empty kelp]

a few weeks ago i was trying figure out how you added multiple tails to a character. i had to use Bing DALL-E 3 because the OpenAI hosted ones just didn’t like multiple tails

[late blade]

lol

[empty kelp]

but it worked the first time i tried it on bing. the elf (above) with six tails was my only test

[late blade]

those last images you made are going into light painting

[empty kelp]

i’ll find the images later. the elf was running around the castle kicking a soccer ball with the six tails

they’re just glowing vortexes

“All of the elves have a vortex inside their hair, and each vortex is a different color. The vortexes are twisting fiercely.” — you just do like this but tell the vortex to glow or whatever effect

[empty kelp]

i had never drawn or created an elf in my life before i encountered Dys Topia on this Discord channel by the way. if it weren’t for you there wouldn’t be all these Santa and elves images

[late blade]

I'm so afraid I might hit the penalty box

[late blade]

oh dear, what have I done? 🤣

it's nice you get to be creative

it's so relaxing to do it for fun

and as predicted, i'm in the box

[empty kelp]

Santa and the elves are really excellent for testing things, and i lack creativity in a big way — so it’s very convenient to just stick with them

[late blade]

hey I'm also on that boat, I've never had creativity like this with normal tools, my drawing skills don't go beyond sticks and dots

[empty kelp]

i’m figuring out how to create DALL-E behaviors and effects to quickly prototype conceptual stuff for mobile games

because we need good AI in 2024

fortunately the AI was able to give the elf in the lower right an extra leg

[final compass]

@late blade is this what you mean by the json file? i can see that it has more than just genid, seed, prompt. interesting.

[late blade]

yes

[final compass]

immediately started thinking if i should start descripting prompts like that 😄

[empty kelp]

without Dys Topia none of this would have been possible 🍷

Santa and hundreds of athletic and diverse female elves are very grateful

and me also 👍🏻

[late blade]

i made a template you can use here: #images-discussions message but you can also use it for other stuff and use the attributes you need

[final compass]

Copied it for later use/testing.

Is there an etiquette into number of entries per daily theme? I had another idea for the daily theme, but i am in the penalty box, so whatever i guess.

[plucky hare]

I think the 30min slow mode is the only real expectation/requirement re: etiquette in daily theme, other than #server-rules! Let it fly 😎

[quartz vale]

Very interesting way to prompt

Does it work well from your experiences?

[late blade]

for my purposes yes, but I use a NLP before prompting, but you can pass something like that also to dall-e

[quartz vale]

To add to this, the slowmode isn't meant to penalize people. Instead, it's meant to give users time to try different prompts and see which output they like the most. Our hope is that it makes #daily-theme more engaging as it pushes users to present their "best"/"proudest" image outputs

[plucky hare]

I think it works great in that regard! I'm a fan of the 30min mode in daily theme.

While we're on the topic of sharing frequency: do you have any thoughts on users' desire to share DALL·E images more quickly in a chat environment? With the previous DALL·E channels, users were able to share and chat in places like #🌄┃general-art. Now, there's #1154829862171844679 as a catch-all, but that lacks the "chatroom" experience, which I think is what results in this channel being used as a general "quick share" space, which sometimes can occlude its purpose as a discussion channel. Just curious if you see any use or value in having a specified place for this kind of "quick share" desire!

[final compass]

Hmm... For me personally channels dedicated for images only can be quite overwhelming without a theme and less engaging.

I think in this channel people tend to share images they are proud of one way or another and maybe share some of that joy and frustration that everyone experiences while learning to deal with DALL-E 😄

[late blade]

I don't mind images here, but with context, just posting images and not understanding why in a discussions channel is weird

[open trench]

I wish the gallery were more chat like. I like doing posts, but the communication around them is limited and people aren't that interactive.

[late blade]

got to find ways for user engagement

[final compass]

https://discord.com/channels/974519864045756446/1154829862171844679 Seems like a good place for a photo dump though. If you have a sustained thing going on.

[late blade]

yup

[glossy scroll]

This was a solid ask.

[late blade]

hehe, files like that are easier to store than photos

as a backup

I think I found my mount everest again, I've been trying to do a modern look of aphrodite's girdle, so far it's been impossible.

[glossy scroll]

Ya durty boy

I can imagine why that would be impossible

[late blade]

lol

don't put thoughts in my brain

[glossy scroll]

Maybe try and explain how the outfit shows instead of the name of the outfit?

[final compass]

tried to combine nazgul and sith

[obtuse wadi]

so to prevent it going sideways for portrait you just type genuine portrait image into the prompt?

looks like it didnt work

[final compass]

my theory on the portraits going sideways is that if the generation has any reason to tap into female portraits, which had incorrectly rotated portraits in the training set, you will end up with portrait image incorrectly rotated. if you still want to do female tall/portrait images it's a matter of pure luck.

something in the ballpark of 25% success rate and 75% fail rate.

[obtuse wadi]

ah I noticed that too its something to do with girls

[final compass]

yup, i don't have that problem if i do literally anything else.

maybe with male portraits, but the fail rate is not as frequent.

[obtuse wadi]

whats your prompt? My success is like 5% or less

[final compass]

"Create a full-body portrait of the female character"

Noticed i had a pretty successful session that seemed to use that.

[dim cradle]

@dense mesa Art Nouveau

[final compass]

9 successful portrait images in a row, but i have a hunch that the success rate might be thanks to encouraging it to use a successful generation as a reference.

[dim cradle]

I asked the AI to make it darker both in palette and imagery, and it takes it to an extreme: "Envision a nightmarish amalgamation of organic decay and mechanical horrors" 🙂

[final compass]

I would like to see that 😄

[dim cradle]

not gonna post that one lol sometimes i forget how literal it can be.

[dense mesa]

you posted one kind of like that. more colors though

[dim cradle]

that was the traditional Japanese art influencing my neon prompts

that's another thing to appreciate about the daily theme, not just learning dalle but exploring the art world at the same time

[dense mesa]

that last one is compositionally pretty close

[late blade]

this one is amazing

[dim cradle]

eerily close in fact

[dense mesa]

that's not the one I'm talking about

[dim cradle]

that's one i haven't shared yet but looks very similar to yours above

[dense mesa]

look how well it got the composition

[dim cradle]

that's spot on

i bet dalle could do a wild las vegas strip

[dense mesa]

I pasted that GPT the other day asking for opinions and got zero acknowledgment, 😶

[dim cradle]

haven't gotten to it yet 🙂 usage caps constraints and all

[dense mesa]

for 2 days?

well whatever, I don't care. just saying

[open trench]

Network errors like crazy right now and crazy glitches. Anyone else experiencing anything similar? I'm currently staring at a blank gray screen

And couldn't get out of a regenerate cycle before that

[dense mesa]

try closing that tab and opening it in a new tab

[open trench]

Ok. Will do.

[late blade]

also clear your cache

[dense mesa]

cache is step 2

[open trench]

Still doing it. I am stuck in regenerate, create, rate

[late blade]

also, don't make a chat with 400 images,

[open trench]

Lol

Clearing cache now

[dense mesa]

often times it's your browser getting bogged down, and then wonky cached data

[late blade]

well, managed to do a pretty good one with Aphrodite's Girdle #daily-theme message

[dense mesa]

but I've encountered people that get weird about clearing stuff. and often times it's just the tab that gets bogged down

[late blade]

but a classic variant of it, I want a modern version

[open trench]

I don't know how to clear cache. I have it installed as an app on my laptop I guess.

[late blade]

which browser?

[open trench]

I don't know. Let me figure that out

[dense mesa]

typed art nouveau into my dalle3 saves. not bad

[open trench]

Edge

[dense mesa]

ask the AI how to clear your browser cach

[open trench]

oh yeah. that's a good idea

[late blade]

lol

no idea, don't work with edge

[open trench]

oh I am just clearing the browser cache? I know how to do that

[late blade]

first link on google search

lol

@fading inlet she #daily-theme message looks so serious... did you steal her muffins?

@glossy scroll there we go, finally getting the kind of stuff I want

[fading inlet]

She's just grumpy today

[open trench]

lol

[late blade]

I'm so scared that I might get the penalty box soon

[open trench]

using the gen_id thingie, I asked it to turn #daily-theme message into a professional photo and it came up with this image. It's pretty cool how the gen_id thing seems to work 73.1 percent of the time.

[late blade]

73.1%, that's very specific

[open trench]

yes and very, very accurate. I came up with it while I wasn't sleeping in the wee hours of this morning.

[late blade]

lol

[haughty spruce]

Grotesque is banned from Bing. Great, no horror elements at all. They were promoting horror stuff back in the day.

[open trench]

Aww. Yeah. It has to be g-rated. secret of nimh level horror.

[quartz vale]

What prompt are you using? There might be a way to reword it

[late blade]

aaaaaaannd penalty box...

[open trench]

ooph

[haughty spruce]

I don't think prompt is wrongly flagged. I typed two different prompts with "grotesque" and the result is "content blocked"

[dim cradle]

try synonyms

[tall mason]

"He cannot run" 💀

[feral minnow]

DALL-E 3’s text generation is absolutely gold for storytelling imagery.

[tall mason]

It can do signs, too.

[feral minnow]

So good. It even highlighted the text with sunlight. It’s so smart lol

[haughty spruce]

I'm tired of overcensored stuff. It really is tiring, I don't wanna try synonyms and get banned for no reason because I wanted to create a manbearpig lmao.

And I think that DALL-E 3 is the best for creativeness because it's not overtrained AI, it really can give original stuff on trained data.

[feral minnow]

That’s why having the Bing model is so great. I disagree with the community that it’s also over-censored. The dog isn’t really censoring your prompt, it’s censoring itself.

It happily gave me Dawn of the Dead: The 1985 Broadway Musical

Whereas ChatGPT said absolutely not lmao

[empty kelp]

this was DALL-E btw, not me. there was nothing prompted about about Santa standing on the elves

the AI came up with that

that’s a five star restaurant, and they were supposed to be eating dinner

[glossy scroll]

Ah, is this what you meant?

[feral minnow]

Look at how DALL-E 3 nailed the complex posing on six different figures

It placed them complexly in the environment it created too (standing on both the chairs and the table)

Literally no other model can do that

[empty kelp]

there are some strange anomalies if you look at it closely. the elf on the bottom right spawned an extra leg to balance for instance

[feral minnow]

There’s a missing leg too.

But it almost did it!

[shut niche]

*I'm all for a channel meant for a general "chat-room" experience, for random sharing with less restrictive discussion guidelines. There's a lot of creative people here that keep many of their best creations or knowledge to themselves because they're considered "off topic" or spam.
*Side note, but related, since one of the main concerns seems to be discussions getting "buried" too quickly by frequent posts. I think a Discord bot that answers common questions is long over due. Give ChatGPT a script and tie it to the discussion thread through the API. The same new user or common questions and misunderstandings are mentioned almost daily. Just the "penalty box" and content filter alone bring plenty of "hot and bothered" users in. A simple explanations performed by a ChatGPT bot might help to diffuse that.
*I also think some "official" OAI DEV answers explaining functionality in some KB's would be helpful to all. I see a lot of misinformation circulating, based on peoples anecdotal experiences, instead of factual "it was designed to function in this way" style of facts being presented. There's no general guide of "best practices." (Nothing simple to point people to).

[empty kelp]

also i may have told it that the elves were gymnasts. but Santa was not involved

[empty kelp]

most of it wasn’t designed to function in a certain way. it trained all sorts of unusual and unpredictable behaviors by plowing through huge amounts of information, and most of finding out how it responds is just trial and error

[shut niche]

DallE does incredibly well if the ideas are clearly explained, and logically separated. E.g., you could expand and do that with 3 character types, so long as each has it's own isoloated description, that's then tied back to a greater narrative or scene summary. Bullets, and various code or pseudocode formats work well to break objects/concepts into "conceptual blocks" for DallE to more easily digest.
But... it's nearly impossible to prevent the extra legs, lol. For that, you usually just need to run multiple gens until a good one pops out.

[empty kelp]

moreover, the models understand English. they don’t require the API

[shut niche]

A WAY over simplified version might be...

A photo of Santa standing on the shoulders of an elf, who is standing on the shoulders of a bear that is standing on a ball, like a circus act.

1. Santa looks like this and is doing that.
2. The elf looks like this and is doing that.
3. The bear looks like this and is doing that.
   Together this scene shows an amazing balancing act performed by the 3 characters.

Scene summary, isolated elements, outro summary/closing.

You can ask ChatGPT to do this, and it's pretty decent at it, so long as you explain that you want bullets, with a scene summary, each object or concept as a bullet, and an outro summary tying it back together. The results might surprise you.

[empty kelp]

i’m creating the Santa/elf images mostly to test different behaviors and visual effects that just need a few sentences. i have been using the API though. i have it integrated with my CI server to generate app icons

[shut niche]

I hear you on this, and considered that before I responded. But the fact is, they actually used a vision model to encode text into hyper-detailed synthetic descriptions of images, and once you understand the format of that original text that was used, the image generation (prompt writing) process becomes easier to understand. I'm certain there was unpublished testing and knowledge acquired by the DEVs that might make common issues "non-issues." Maybe the best formatting for prompts, how adding text was "trained," what adds emphasis to concepts, how concepts are weighted (and shift), the (assumed) out-painting process, known things it can't do well. I can think of tons of areas their input would be desirable or useful to everyone.
No doubt, there's plenty of trial and error with emergent behaviors to consider, that exist in that more "grey" area of "learned" understanding. But much of it could be better understood by sharing/discussing more of the "less secretive" design elements.

Like, most regulars here know that ChatGPT+ sends the prompt to DallE via JSON. Information like that could just be published openly, instead of this Discord being a collective of people "data-mining" information that's known.

[empty kelp]

ChatGPT+ sends the promps as English. it's not JSON

try asking what the prompt was, and then ask for an API version

it sends the resolution, and the English prompt to DALL-E 3

it rewrites what you tell it in English, and then passes it to DALL-E

[shut niche]

It converts the prompts to English, and then bundles them into a known JSON format with the resolution.
But just you and I disagreeing about that is the point I'm making. The DEVs could simply answer these questions openly, and definitively for everyone. Maybe I'm wrong. It would be nice to have a document that simply stated the facts so everyone wasn't repeating common mistakes.

[empty kelp]

you can see here i asked it to create a mushroom, then what the prompt was, then what it would look like in the API

i just rewrites what you tell it in English, and then passes it to DALL-E with the resolution

but that's very useful -- because GPT 4 has a profound understanding of how DALL-E works, so it can help you figure out what to put in the prompts

[shut niche]

When you asked for the prompt, you ask for the prompt used. That doesn't = what was sent to DallE.

Look at how I phrased the question differently. Your question was only receiving a partial answer. Ask more specifically, if it's sending your English prompt bundled within a JSON format to DallE, and what that exact format was, with prompt included.

And I feel like this is a great distraction to make the point, that the DEVs could literally just "tell us" the answers to many of these common questions, instead of allowing each user to "muddle" through it. It would be a better user experience if there were some "understood" baselines established.
Unofficial answers, in this case, allows misinformation from either you or I to cycle in this Discord. It's a great example of why some of these details should be officially stated, because one of us is wrong. How would any user looking at these discussions know? It should be stated in a KB.

[dim cradle]

The API, which everything goes through, reference documents the parameters. There are only a few, just about everything goes into the prompt parameter eventually, as natural language -- the JSON templates people are using are nothing more than a spec sheet, as those parameters don't exist.

[dense mesa]

Well it would be nice either way to get some kind of best use guidelines from people that could actually answer these questions more definitively.

[plush yarrow]

I ask ChatGPT to post the prompt it sends after generating images like this in my custom instructions fwiw After the image is displayed, provide a separate response containing only the full prompt used for the generation, formatted in a code block. This response should contain no other text or summary. It…. generally remembers to

[shut niche]

Same, my GPTs (when working properly lol) output the prompt, seed, and image_id in a code block. It appears in JSON consistently.

[plush yarrow]

So there is seed data?! ChatGPT!!! (Shakes fist)

Are you guys able to get it to gen more than 1 image btw? I was getting 2 yesterday but today only ever get one even though ChatGPT has apologized several times (but never follow through)

[dense mesa]

Right now there's a lot of the blind leading the blind. And I think everyone needs to be open to accepting they might be misguided about some things.

[plush yarrow]

lol

[dim cradle]

gen_id/reference_image_ids is the new seed, seeds are gone, and there's no API parameter for it, so it must be included in the prompt, and it only works if it's in the same session. seeding is in active development and isn't consisent or reliable (check the openai developer forums, it's in constant flux)

[dense mesa]

Well a seed by any other name is still a seed. I'm curious why they changed that though

[shut niche]

Well said. That's the point 👉.
I'm certain I have some misconceptions, as does everyone here. Would be great for there to be a clearer path to know then follow, set by OAI.
Right now, there are too many people that are professing their own anecdotal experiences as facts.

[dim cradle]

Not sure why they changed the property names, but it switched over back in October -- the earlier posts in the dev forms talk and test (with a forum mod who speaks for the dev team) are using "seeds" and then in November things changed to the gen_id (output) and reference_image_ids (input). It appears to be in the early stages.

[open trench]

I love gen_ID. it's been a game changer for me.

[plush yarrow]

lol, almost like ChatGPT should be made more self aware

[dim cradle]

Like @shut niche is saying, I spent an hour trying to make sense of the dev forum posts, would be nice to have an openai expert around

[empty kelp]

the JSON prompts through heavy preprocessing, then to a language model, then to a diffusion model

the JSON is just a wrapper

it's a good way of organizing things

[shut niche]

Moxi said it was because the seeds break when they make updates, so it was unreliable. They don't want to even "half" support a seed feature they know will break. Think of users that might get accustomed to the repeatability, only to suddenly not be able to make the identical image after an update. It makes sense.

[empty kelp]

what it draws is super unpredictable, and the gecko is a good example of why

[shut niche]

Yes, it's a wrapper. In fact, I have ChatGPT process my images in pseudocode, and then output the metadata with the images. I don't use just standard English prompts anymore.

[empty kelp]

if small creatures are in a scene with a gecko, the gecko go into "eat everything" mode, and the AI spawns bugs and creepy things into your image, and the behavior of the gecko warps your entire image

these AI models are full of crazy, unpredictable stuff, and even slightly changing the order of things in a prompt or making small changes can have a huge affect on what it draws

so having a reusable format like the JSON can give images some continuity if you use it for multiple images

[shut niche]

It does. But also, the more formal your structure, the more consistency you'll get. Each object or concept carries a weighted value. Adding to one subtracts from another. That's one primary cause of variation.

[empty kelp]

but what DALL-E draws will still be super unpredictable, because it sucks up all sorts of crazy stuff into the training data

[shut niche]

Dalle works by "concept." So also, the looser the interpretation, the more variance.

[empty kelp]

like you wouldn't guess that Santa would suddenly decide to climb up on top of the elves and stand on their heads if there is nothing like that in the prompt, but he does it

[shut niche]

That's why text appears autonomously sometimes, because it fits the "concept" of the scene.

[dim cradle]

it helps if you craft the prompt to describe precisely what you expect depicted, the more you leave it open to interpretation the more unexpected.

[feral minnow]

I love when it sprays the ethnicity of the person directly onto their clothing

[shut niche]

Made this earlier, to demonstrate text in a scene. Notice his button/pin that says "Bing." I didn't ask for it, but it makes sense to the scene. It placed it there based on "concept," but not based on a physical description of any button/pin.

Likewise, the "suggestive" hand gesture that goes well with the comment.

[empty kelp]

the DALL-E documentation explains that to draw things it needs to reproduce the behavior of whatever it draws. so if you put a gecko into the image it immediately starts spawning bugs and weird food -- and it's also triggered by things that have nothing to do with a gecko. It gloms the gecko behavior together with like the Encyclopedia Galactica, and comes up with some really nonsensical and disturbing stuff

[shut niche]

Have you ever asked ChatGPT the diamond in a glass question? It might help to understand what you just said, in an abstract way.

If I put a diamond in a glass, then I put the glass upside down on my bed, then I carry the glass to my kitchen, where is the diamond?

It will hallucinate, and it will give you every reason imaginable for the diamond to still be in the glass. A common answer is that the gravity of the glass holds the diamond in position.
Likewise, DallE doesn't understand the worldly relationships you expect it to, of a gecko. There's some, obviously, but it's not a complete or well understood holistic knowledge of geckos. Only what it's seen in images. It's worldview isn't a clear one.

[empty kelp]

try creating a tornado in your image, and tell it "the tornado has the style of Bosch"

it will really make you wonder about AI

[dim cradle]

Statistically speaking it’s seen geckos with other insects. Especially if you specify surreal it’s going to pull in connected assets. One way to avoid that is to be consistent in terms. If you call it a gecko, stick with that, if you call it a lizard later it’s going to incorporate more generalities.

[empty kelp]

the gecko flips the AI upside down and makes everything go crazy

[shut niche]

Maybe they made it watch Rango one too many times. 😛

In what manner? ChatGPT rewrites user's ideas into more elaborate and structured prompts, which is why it works so well. However, if you don't like the prompts it's writing, you simply say,
Run this verbatim,
"Prompt here Blah blah blah"
That will allow you to write your own prompts and run them.

[empty kelp]

please create a hyper-realistic photo of a gecko inside of a red tornado and a lion inside of a blue tornado

[dim cradle]

hyperrealistic is different than realistic

[empty kelp]

the basic gecko behavior is to leap out of tornados looking completely insane and start eating innocent elves

this is the gecko... hehe

[empty kelp]

hyper-realistic is a term that ChatGPT constantly feeds to DALL-E. I think it's something that the pre-processor understands

[shut niche]

Ok, I haven't even tested this yet, lol. I literally just slammed it together as a GPT for verbatim prompt input to DallE. Try it out, and see if it's correctly passing your ideas "as is," verbatim, without making changes to the image prompts you gave it.
https://chat.openai.com/g/g-TUnMVHI9p-verbatimpromptpasser

[empty kelp]

DALL-E understands realistic and hyper-realistic

[shut niche]

So far so good. I gave it this strange prompt that I know it probably cringed reading, and it didn't attempt to rewrite it, so I consider that a first success! 🙂

An image of a dog doing a hand stand on a ball with a circus tent in the background. Surrealist and fractal elements are all around, with Mandelbrot inspired designs.

[empty kelp]

if you use "hyper-realistic" it limits the number of images you can generate. it's part of the pre-processing

like if you tell ChatGPT to create "the best of all possible images" it will insert hyper-realistic into the prompt

[dim cradle]

"Hyperrealism, also known as Photorealism, is a genre of art that emerged in the late 1960s and early 1970s as an evolution of photorealistic art. Hyperrealist artists create paintings, sculptures, or other artworks that are so detailed and precise that they often appear indistinguishable from high-resolution photographs or even reality itself. Hyperrealism requires immense technical skill and patience, as creating such detailed and precise works can be a time-consuming and challenging process." Sounds ideally suited for AI, actually, and it's a subtle but important distinction.

[shut niche]

There, that's a little more polished. It now outputs the JSON wrapper with prompt, image ID, and seed. That will also help anyone using it to verify that their prompt wasn't modified.

[empty kelp]

it's good to ask ChatGPT for the prompts. it's able to explain things that aren't obvious about DALL-E, like how to blow things in different ways with a vortex, or bend trees in a certain way. Things that you would never guess

[shut niche]

"An image of a tiger doing a hand stand on a ball with a circus tent in the background. Surrealist and fractal elements are all around, with Mandelbrot inspired designs."

[dim cradle]

soon i will learn to spell "Nouveau"

[open trench]

me too lol

I butcher it 97.21 percent of the time.

[dim cradle]

same, seriously, and i took french

[shut niche]

I spell check it every time! It's just one of those words my brain refuses to accept into memory. 😄

[empty kelp]

a good test to see if the AI is good might be to stack 10 elves standing on each other's shoulders. like rows of 4, 3, 2, 1, and Santa balancing on the head of the 10th elf

[dense mesa]

We both made pyramids just now

Bing probably told dalle3 10. But dalle3 don't count too good

[glossy scroll]

"What have you done with my waffles?"

I put: "Create a dork as if it was an animal"

[empty kelp]

i’ve abandoned the elf pyramid project. we’ll need DALL-E 4 or 5 to stack more than six elves properly

[plush yarrow]

Fwiw I’m now getting a seed & was able to regenerate the same image (in the same session).

{
  "gen_id": "aNwbdsIwe97zYYa5",
  "prompt": "A realistic photograph of an articulated toy action figure resembling a ninja iguana. The action figure is designed with detailed textures and joints, reflecting a high level of craftsmanship. The setting appears like a professional product photo shoot, with clear focus on the toy, subtle lighting, and a plain background to emphasize the figure's details. The ninja iguana toy is captured in a dynamic pose, showcasing its articulation and design features, reminiscent of high-quality collectible toys.",
  "seed": 2376289501
}

Maybe it’s not really regening it though

[charred fable]

what style you guys like to tell the bot to create the image in

[glossy scroll]

2D style mimicking 3D

[charred fable]

Interesting

[glossy scroll]

Or vice versa

Comes with images like these...

[charred fable]

Will try

[glossy scroll]

It's pretty cool

[shut niche]

Just remember, if you get it to work from a seed, that it's not officially supported, and might not work the same, or at all, at any time.

The DEVs would have to weigh in on this, but my "assumption" is that the image_ID is the 'future' route they've chosen (possibly for video implementation??). I have a "hunch" that they're experimenting with video image regen, which would require repeatable characters with slight positional variations, which a seed probably doesn't really do all that well... as the characters are moved around (or camera angle).
But that's just my own speculation, and a seemingly logical secondary reason to abandon seeds.

[glossy scroll]

@charred fable I'm a big fan of images like these ^

High detail 2D

[charred fable]

same

This art style i like, 3d ish. how to get it like this

Not too cartoony

[plush yarrow]

Yeah think you’re right. Still super dodge; `The images may appear similar because they were generated using the same detailed prompt and referenced the same initial image. However, they are not identical. DALL·E uses the referenced image to guide the style and content of the new generation, resulting in images that are visually similar but still unique in their details.

If you're noticing strong similarities, it's likely due to the specific instructions and the style dictated by the prompt, which directs DALL·E to create images with similar themes and aesthetics.`

[glossy scroll]

Yep, I've tried to tell people the same thing. Lugui says the same thing.

We even had a discussion about this yesterday.

[charred fable]

Whats this style

[glossy scroll]

Gearpunk maybe?

[charred fable]

Are all the emojis ai made

[charred fable]

Drawing style i meant

[glossy scroll]

3D?

Surrealism?

[charred fable]

ill try

[glossy scroll]

Try try

[feral minnow]

Oof I’m not sure.

“A woman with a Caucasian descent has been transformed into a stunning biomechanical butterfly, encapsulating the essence of steampunk. She has intricate metallic wings detailed with gears and steam pipes, and her attire merges Victorian elements with futuristic cybernetic enhancements. As she soars effortlessly through a sky filled with fluffy clouds, the sunlight glints off her polished brass and copper components, creating a mesmerizing tableau”
Here was the prompt it gave me

[shut niche]

The style is "steampunk".

[charred fable]

drawing style

[shut niche]

Photo

[charred fable]

Lool

No i dont mean that

Like the style of how its drawn, For example Oil painting, Caricature Drawing, Cartoon Drawing, Figure Drawing Etc

[charred fable]

That kid and old man in the painting look weirdly similar to the ones in the tv

[dense mesa]

just coincidence

[glossy scroll]

It's just 3D

[charred fable]

Ye but gives me very bad prompts when i just write 3d

[shut niche]

Right, photo. Or you might say "digital artwork" or "digital painting." But DallE was clearly trying to create a photorealistic looking image. I just ran that prompt to demonstrate.

[charred fable]

ok, i will try digital photorealistic looking art work

[shut niche]

If you don't specify, it'll often default to a photo, unless something in the prompt makes it think otherwise.

[dense mesa]

photorealistic is misinterpreted by dalle3 sometimes

well, unless that's what you are going for

[glossy scroll]

Hyperrealistic

Ultra-realistic

High definition

[charred fable]

Super ultra mega

[shut niche]

Right, what he said. Don't say "photorealistic", "Realistic" etc., it will treat that like "I want almost photo, but not actually a photo" and will provide you with a 3D rendered looking image.

[dense mesa]

I use ultra- realistic. hyper realistic is great if you want something that looks like hyper-realistic art. works well with artistic styles

[charred fable]

i see

[dense mesa]

like hyper realistic oil painting

[charred fable]

ty for ur insights

[dense mesa]

try "hyper-realistic" with a classic piece of art. Mona Lisa or something

[glossy scroll]

You can copy and paste the last message I put here if you want #1187233013956874260

[shut niche]

Try this, it's good at photo's, if that's what you are going for.
https://chat.openai.com/g/g-UOykDKodx-norender

[glossy scroll]

It has all the keywords for a super realistic prompt

[charred fable]

done

[glossy scroll]

Easy peasy lemon squeezy

[charred fable]

Wow not bad

[glossy scroll]

If you put it as Custom GPT instructions, it's even better.

[charred fable]

I like

[glossy scroll]

Pikacrazy

That's what happens when you give them too many rare candy's

[open trench]

lol

[charred fable]

Haha

[glossy scroll]

Omg, lol

That's messed up

[open trench]

I had three goal projects with today's theme. I think I nailed 1/3 of them. but I don't care what else I was about to write because that panda is frightening

[charred fable]

🤣 I like this genre

What animal should i do next haha

[open trench]

Poodle

Toy poodle

[charred fable]

Ok

[open trench]

my god

[charred fable]

🤣

They look like they touched the forbidden sugar

[open trench]

lol

I'd love to see the prompt for that craziness

[charred fable]

im 257 messages over the text limit here

ill send it in 2 messages

[dense mesa]

love those human teeth

[thick smelt]

Or when Satoshi Tajiri and Scott collab for a pokehorror game 👀

[glossy scroll]

Pokeyman horror

[thick smelt]

Even the mimic (from the new FNAF ruin dlc) and team rocket duo (Jesse and James )would be scared at the wide smiled Pikachu

[glossy scroll]

Mr. Mime was funny in the movie

They should just make a movie about Mr. Mime

[thick dome]

another depiction of SCP 682

[shut niche]

😆