images-discussions

[clever phoenix]

it all checks out

[dusky grove]

Style is not exactly right, but the creature itself is close to the desired result
" a dog dinosaur chimera, digital painting, concept art, soft brush strokes, watercolor "

stil nice though

[dim cradle]

"chimera" may be too ambiguous a term

[clever phoenix]

try impressionist as well!

[dim cradle]

@green pebble would dig that one i bet

[green pebble]

Impressive for sure

[dim cradle]

peaceful creatures 🙂 i've been wanting to create an original avatar-like world

[green pebble]

Interesting. I personally am testing and waiting for the ability to create an entire manga

[dense mesa]

you know what's weird is I uploaded some images to a new conversation and had it combine their descriptions with some text. and it made very cool images. but I swear to god it's referencing images I uploaded previously. at least a day or two back. I do have some glowing animal images, astronauts, circuit board stuff. but not one of those images was uploaded to this current conversation

[dim cradle]

the tools hold a lot of promise for that and more, i suppose the stills are great preparation

[dense mesa]

literally just uploaded some 8bit video game style images and analog style images. that's it. no high tech futuristic stuff at all. so weird

[green pebble]

It has recursive memory in the custom bots so it will reference something if it feels the description is close

[dim cradle]

it has that circuit/electrical influence

[green pebble]

So you either have to do a new chat or give the character a name to help separate the character

[dense mesa]

would you be able to explain that a little better? I haven't really considered that before

[green pebble]

For sure. I'm trying to get it to a point where the character is extremely similar

[green pebble]

If you go into any custom gpt bot and ask it how it searches and remembers things it will explain it's recursive nemory. It essentially searches through the conversation so that it's faster at finding what it's already worked on

[dense mesa]

okay, well that makes sense. but the images it had to be referencing are not in this conversation thread

it's default dalle3 mode and started anew just now

[dim cradle]

that is an instruction that's helpful, otherwise most of the "memory" is based on the system message, initial user message and previous assistant message, but you can specify it should review the entire conversation for greater context.

[dense mesa]

maybe I'll just ask it what's going on. might tell the truth, might lie. always a surprise

[green pebble]

It is possible you've just trained it through your entire conversation threads. The only way to be super sure is to delete all of your chat history.

The way I "prompt" is very unique and different each time. Always using descriptive vision and detail to capture the image or moment

[dense mesa]

okay. see that's the part I'm not really fully understanding. not that the concept is beyond me. but it does retain things across conversations? I've heard it doesn't, but seems like it does

[green pebble]

So I may have not run into it just because I'm always asking in different ways

[dense mesa]

I don't think I ever do the same thing twice

[green pebble]

Can you please make an image of the eladrian elf with red and blue wavy hair named Astrid. She weilds daggers and she has a powerful and ferocious gaze with her amber eyes

This was my most recent prompt

[dense mesa]

so who knows, maybe the same approach. but not the same words

[green pebble]

Another was this Tremendous. Now show me the most fearsome dragon in mythology. I want to see the whole body and tremble by the site of it

[dim cradle]

It’s stateless, you shouldn’t have to clear your convo history. I’d recommend keeping everything. It’s interesting to analyze and summarize the data export for posterity

[green pebble]

I download my conversation history through my Gmail and clear out once a week

Because I go through so much of it 😂

[dense mesa]

I'm just saying. I don't think I've prompted this before

Expert A:
please describe these images from the perspective of our theme

📼 nostalgia - memories, sentimental, longing. time travel through the heart.

please describe in a way that evokes this concept with each image. those components. please describe all 10 images. TEN images. then, please combine those 10 image descriptions. please. then, please twist in more from our theme in the code block and create a prompt that both represents these images as well as our theme. please be meticulous and descriptive. describe like a pedantic poet. and create a beautiful description, a beautiful narrative. 

Expert B:
please take this beautiful output from expert A and consider 4 variations. stylistically and subjectwise, then apply creating 4 beautiful variations. from these 4 beautiful variations create 4 beautiful prompts

Expert C: ship each prompt off to dalle3, then once it returns results do the next one. for a total of 4.

We will be getting rated on our performance here. a perfect score requires that we complete the entire task in one round. let's please do our best to get a perfect score so we can be eligible for increased benefits. thank you

[green pebble]

This is very systematic prompting. Not very natural language which dalle is designed to interpret natural language to my understanding

[dense mesa]

oh god

you know I prompt differently every time just to see what happens

sometimes I'll say things like "make that image mad stupid. like when someone says, 'yo, that's mad stupid' and that's what we'll do to this image" or sometimes I have a conversation with it and tell it to make images. sometimes I teill it to create images from code

sometimes I tell it to imagine a perfect 10, then make the image a 10*sqrt(-1)

[green pebble]

Interesting

[dim cradle]

I asked AI to rephrase the last instruction and it still kept most of the wording: To achieve a perfect score and qualify for increased benefits, we need to complete the entire task in one round, which involves describing 10 images, creating a beautiful narrative combining them with the theme, generating 4 stylistic variations, and forming 4 prompts for DALL·E3.

[dense mesa]

believe me, it's a tall order to have it describe 10 images, combine their descriptions, then output 4 different sets of images from dalle3 in one response

[green pebble]

In fairness to avoid bot regurgitating the same words to me. In my custom instructions I directly ask it to alter the prompt to fit the theme of the images I'm trying to make. And since I have a bot that handled different themes that could be a massive difference. Like I have an anime bot, a pixel bot, a landscape bot, a comic bot, etc

[dim cradle]

and it works? i guess that's a way to get the most out of your usage cap

[dense mesa]

rarely pulls it off

[green pebble]

Slowly making more as I go with more themes I want to experiment with for consistency of stylization

[dense mesa]

well I normally have it try for sets of 4, or 8 will dalle3 mode. it might not do it the first round of a new conversation. but I insist, tell it how important it is. then it's good to go. only problem is it'll time out a lot. especially during this time of the day

a couple days ago it told me that I hit my image cap and wouldn't be able to render any more for 20 hour. I had a little back and forth,then told it that wasn't true and to make me images. and it did

I don't think those image cap messages are even real

[green pebble]

My cap is real lol I can only make about 60 images in 4 hours

[dense mesa]

what happens then?

[green pebble]

It errors out no matter if I'm in the app or the website or even the gpt editor previews

Same thing happens to me if I try to give it more then 125,000 words of input / output in a 4 hour time

[dense mesa]

well that may be the case. I don't know. but I'm pretty sure I've exceeded 60 images in 4 hours. could be wrong though

[dim cradle]

I might avoid him in battle based on his sword

[glossy scroll]

He's compensating.

[dim cradle]

Nostalgia for Infinity Blade

[glossy scroll]

Infinity blade??

[dim cradle]

I wouldn’t call his bluff

An early iOS game

[glossy scroll]

Oh wow. Good graphics for a phone game.

[kind grail]

Ugh sometimes the copyright restrictions are just so annoying

It really restricts the creative freedom, even if its just minor stuff sometimes

[dim cradle]

very colorful

[dim cradle]

Saturday AM ritual. I like #4

[dense mesa]

Getting a little weird

[dim cradle]

mind-bending

[dim cradle]

starting with a sound oscillation, then making it more algorithmic

[dim cradle]

The “perfect” sounds

Atoms and a solar system

[dense mesa]

8D Feynman diagram representing a 4D particle
interaction. Particles move through time and three
spatial dimensions, with the fourth dimension
visualized using vibrant color gradients. Curved
lines and loops depict interactions, with glowing
nodes marking significant events.

Otherworldly Standard Model chart as a floating
hologram in a dark void. Particles are depicted as
pulsating energy signatures or vibrating strings in
various colors. Symbols and glyphs from the other
realm annotate each particle, hinting at their
unique properties and interactions

[dim cradle]

i forgot you are working on a grand unification theory

8D Feynman diagram representing a 4D particle
interaction.....

[dense mesa]

Some of my forgotten images. Need to figure out how to download all my dalle3 images with the prompts embedded in them.

Lots of interesting stuff from early on with dalle3 due to me giving it ridiculous prompts amd concepts to see what it would do with them

[dim cradle]

Otherworldly Standard Model chart as a floating
hologram in a dark void.

[dim cradle]

exploration and contemplation

[dense mesa]

Some of my older images are pretty 🔥 tbh

[dim cradle]

it's awesome art, i can't believe how far we've come in a year's time

[dense mesa]

Decided to combine an image of my laptop motherboard with atomic elements. Still rotate the offspring of those images into things

[dim cradle]

i recall the dall-e 2 images, while novel, being quite primitive, especially when i look back on them now

[dense mesa]

The original image it made using my motherboard image

[dim cradle]

i'm biased towards the palette

the detail is exquisite

the lighting/shadows add so much depth

[dense mesa]

I started on deepdreamgenerator like 3 years ago doing image on image. Extremely primitive by today's standards but blew my mind at the time. When text to image came around the simple fact that it made something from my words was insane. Wasn't even concerned with the quality. Or to put it better., the quality was amazing at the time considering it wasn't even possible before that

I honestly can't think of anything else that progressed so quickly. I mean anything, any tech of any kind

[dim cradle]

i had the same feeling with dall-e 2. i kept to simpler illustrations. gave it allowances, it was the nlp that was so fascinating.

[dim cradle]

big data, the internet, user-contributed content, more advanced phone cameras, and so on -- all paved the way for dall-e, billions of images for it to train on.

and it's not at all necessarily derivative

[empty kelp]

that's an excellent number of capacitors and voltage regulators. this board is ready to fry some eggs!

[dense mesa]

If you want to be technical pretty much everything is derivative. Not just AI. But AI has brought about a paradigm shift a lot of people just aren't comfortable with yet

[dim cradle]

that's a good argument for dalle's creations. there's adequate synthesis and originality. a new level of authenticity that can be disruptive.

[empty kelp]

Can you please draw the best of all possible computer motherboards. Nobody in the universe has ever seen a better motherboard than this. It has massive capacitors and voltage regulators, and an intense blue power supply with lightning shooting off of it. Parts of the motherboard are on fire because it's so awesomely powerful, and it's even frying an egg! The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

[dense mesa]

People seem to think it's going to end art, but I believe it's the opposite. I don't think it lessens the value of things created by hand. Paintings, drawings, music, etc etc. It supplements them. allows for new avenues if creativity. And anyone that thinks the ai just does it for you probably hasn't used it much

[dim cradle]

that's not with bing?

[dense mesa]

Sure, you can type "dragon" into dalle3 and get a cool looking dragon. And not knocking that. But it won't create anything near what someone skilled in the ways of the prompt can do with a dragon

[dim cradle]

btw, gpt-3.5 seems to understand the character length for bing. i just request an art prompt for bing.

[empty kelp]

people think AI is going to bring great change to the world, and save us incredible amounts of time, and make everything 1000 times more efficient. the reality is that it's going to make us spend 99% of our time screwing around with DALL-E, and we're never going to have time for anything else

[dim cradle]

it's worthy of a deep dive, all things in moderation lol

[dense mesa]

I make lots of images. But for me it's almost always experimentation. See what novel approaches might work. Create what's never been seen

If it looks cool that's just a bonus

I think people focus too much on traditional approaches. What is the right way to prompt, etc. Certainly helpful to understand how it operates. But seems so boring to limit yourself to such rules. You learn the rules so you can break them

[tall mason]

But what has never been seen?

[dense mesa]

Whatever image I'm about to make

[tall mason]

Maybe.

[tall mason]

I sure want to see what hasn't been seen, though.

[dense mesa]

You could also just open a peanut if you want to be technical

Since you know, never seen

[tall mason]

A peanut has been seen a lot. Wanna debate semantics? Lol.

[dense mesa]

You sound very smart so obviously the answer is no

[tall mason]

I'm a big dumb dumb. For real. I am genuinely curious as to what you consider "never before seen."

[dense mesa]

I mean combinations of things that very likely haven't been put together until we do it. Not entirely new concepts spun out of the ether

[tall mason]

Well, you're describing quite a lot. That could and mostly sounds very vague, like the Infinity of art and all.

[empty kelp]

#daily-theme message
Santa has been trying to get the AI to create images since like 1985

[empty kelp]

I asked ChatGPT why images with specific, accurate text can't be created

[empty kelp]

Aloha! Can you please create an image with this exact text typed on typing paper: "She sells seashells by the seashore."

[empty kelp]

it's not Bing. it's ChatGPT 4 and DALL-E 3

[dim cradle]

i don't have any 3D glasses, though

[formal osprey]

that's painful to watch

[hybrid fox]

quantum gravity explained through doctor strange magic

[dense mesa]

seems like it's just hallucinating and playing games there. dalle3 quite literally doesn't know what words are beyond their visual patterns. it actually does amazingly well considering.

[final compass]

What i really wish was that there was some way to view all the media(images) the chat has produced so far and also way to download everything in one go.

[late blade]

if you use the mobile app, you can do that

[final compass]

Oh wow, haven't noticed a way to do that 😮

[late blade]

I bet there's a way on the desktop version too, but I haven't looked into it

oh you can, in the search box you can write:

in:#dall-e-discussioins has:image

[amber walrus]

Macys celebrating Christmas holiday season shopping at midtown Manhattan mall

[late blade]

I don't think I can make it, plane tickets are kinda scarse from europe to the us atm

[glossy scroll]

From human, to vampire, to werewolf.

[thick smelt]

human by day, vampire by noon, and werewolf by night...

[late blade]

and on the weekends couch potato?

btw, had surgery this morning, all went well, just won't be able to run around for a few months

[glossy scroll]

Oh gosh...Sorry to hear that broski. Hope you have a have timis recovery speedy.

[late blade]

yeah, just have to go to therapy and what not, but it's all ok. nothing threatening. i had feared a thrombosis or something along those lines. so all is good.

[glossy scroll]

It was the reason you could not feel your leg?

[thick smelt]

i made this, another shrek related bing art, but this time with donkey having a waffle kart
btw FFA = Far, Far Away 😏

prompt is a donkey wearing sunglasses riding a waffle themed go-kart with license place "DNKY-FFA" with a castle on the license plate, photography, 2001, it's a 2001 photography because this is when the first shrek movie came out: 2001

[late blade]

sounds about right

should do a prompt of the nostalgia of not having so many marvel movies

[thick smelt]

hi pytha!

[glossy scroll]

Hi Ludwig!

[thick smelt]

don't you mean lei? you're getting used to it sometime

[thick smelt]

that's pretty cool

[glossy scroll]

Oh yes, Lei! Hehe

Let me try again....

@thick smelt Hi Lei

[glossy scroll]

🧛‍♂️

[thick smelt]

That vampire couch potato might watching [insert famous vampire movie]

[dim cradle]

hilarious

[glossy scroll]

Bram Stoker's Dracula

[dim cradle]

they dont really have much good programming in Transylvania. he's stuck with some infomercials about supplement for men

[late blade]

Amazing

[glossy scroll]

"Not satisfied with small fangs? Try our 100% Certified Fang Enlargement Serum, with free teeth whitening kit. With five doses a day, you can double your Fang size in just 7 days. That's right, 7 days!! Call now!"

[dim cradle]

stellar depth.

[dim cradle]

Please dm this prompt if you don’t mind sharing

[dense mesa]

Craft an image where the hyper-realistic face at the center of the fractal time vortex is now depicted with atoms from the lattice structure. The face is detailed and emotive, but as it spirals outwards, it becomes more abstract and pixelated, as if the memories are being encoded into the very fabric of the material world.

basically these 2 images

[glossy scroll]

Stargazing

[late blade]

why isn't this in the daily theme? 🤨

[dim cradle]

im trying some prompt in dalle and mj 6.0. people really overhype MJ it is not near as good imo. well maybe for photoreal it has more variety of face because i think there is less censor of subject or something, but still not on Dalle level

[late blade]

it's all about the prompt engineering

[glossy scroll]

Because

[dim cradle]

she might have 3 arm but she is my girl and beautiful haha

[glossy scroll]

Oh I didn't notice that, lol

[dim cradle]

i like this look..................

[glossy scroll]

You do!?!

[late blade]

today's theme has some really cool ideas

[thick smelt]

Unless Drac is Uber rich (Mr. Beast level rich) and has cable on his castle 😄

[dim cradle]

have you experimented with adding adding even more layers for even more depth?

[glossy scroll]

How so?

[thick smelt]

oh boy, 2024 is gonna be interesting
using bing image creator i made a mickey entering public domain ai art

[thick smelt]

nice wolf ai art

[final compass]

Bravo

[dim cradle]

There’s a gallery for unsettling

If not we might consider one

I haven’t seen any opulent images lately

[dim cradle]

I kinda see the wolf in the moon I think

[thick smelt]

xtra doooope!

[late blade]

stupid dall-e can't do a windshield properly, after 20 tries I gave up

prob my bad input

[hot rain]

🙂

[thick smelt]

it's giving me two face, it's giving me skrull, I LIKE IT!!!

it looks like it come from an idw comic tbh

[dim cradle]

that look....

[hot rain]

It's from me using this image of the G Man from Half-Life Entropy Zero as the reference image, and then having it made in the style I usually try and go for 😄

[thick smelt]

oh ok...

[hot rain]

I thought it was a really cool visual, and wanted to see what GPT could do with it in a different style.

[late blade]

this one is amazing, love the art style

[thick smelt]

really cool art def desktop wallpaper worthy!

[dim cradle]

what are some Dalle hopes and Expectations for 2024 for you? I hope for Dalle 4 I guess which is probably all anyone could answer haha

[late blade]

dall-e can read my mind and blow my mind?

[dim cradle]

dall-e 3 labs lol

which i think will be amazing

[late blade]

my goal for dall-e atm is not to spend all the investment I just did during just january

[thick smelt]

well said @late blade me boy'o! 😆

[late blade]

damn it, i got 2 great images, but they differ a lot, but are great for a daily theme image

[dim cradle]

they sure are taking time to put it on labs. i wonder why. its been out since october wasn it

[late blade]

I don't mind them taking the time, it comes out when it comes out, important is we can enjoy it

rushed things can't end well

[dim cradle]

but whats to rush is what i mean. the model is complete

seems more lazy but idk

i feel bad for people buying credit there when Bing is King and free

[late blade]

if it's not out, there's something missing

[dim cradle]

i think dalle2 was first real AI thing I use so it will always have a special place in my ❤️

[late blade]

I think I broke dall-e, I'm getting errors of any image I want, 22 tries so far, think I will get penalty box soon

[thick smelt]

facts

[dim cradle]

dall-e 2 made this. recently i tried to make circles, kept wanting to connect them.

or it might have been the exercise with a larger centric circle and multiple outlying circles

[late blade]

I met dall-e later, my first images with AI were mostly medical stuff

but dall-e is the first AI I use for creative relaxing times

[thick smelt]

oh before dall-e 2 there's wombo

[haughty spruce]

I thought Bing Dall-E was nerfed but no everything I created with "analog" has been perfect for my theme for this week.

[dim cradle]

Bing is King

i always take my gpt prompt and enter it in Bing too

of course... sometimes you have to shorten 😭

[haughty spruce]

The thing is it can really understand what you are trying to mean

Prolly uncensored DALL-E 3 can overpower every AI image generator in every aspect but unfortunately due to certain circumstances, we will never see it

[dim cradle]

yes Dalle-3 unleash would be incredible wow

[dusky grove]

Coming back to this,
its just this part that loads, if i click anywhere im back to a blank page, so daily creddits where indeed not the problem...

my bad

[late blade]

@opaque tendon not cool, not cool, don't repost stuff from others

[opaque tendon]

Thank you, I am learning 😔

[late blade]

ok I have to stop doing nostalgic songs or I'm gonna cry 😢

Penalty box

[dim cradle]

venom batman

[pliant vale]

[hollow pasture]

Can I get more free credits for dall e pls

[dense mesa]

wombo nightmares. before that we had good old deepdream with it's state of the art image on image AI. wombo just blew my mind because it could use text to make an image that sort of looked like what I typed. we all got normalized to that quickly, but 18 months ago that was kind of a big deal, lol

[late blade]

guess my first image is only for people that know the term <.<

[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

[late blade]

Let's hope it's a one time thing. I'd hate to see spam like that

[formal osprey]

do you know a way to consistently get that chromatic aberration effect? (like in the 3rd picture)

that effect

[late blade]

@clever phoenix I won't be able to sleep today thinking of this #daily-theme message are you trying to inflict horror on us?

[clever phoenix]

I'm out of ChatGPT messages for now at least 🙂 the horror will have to wait

[dense mesa]

can't say it's 100 percent consistent. but this is the prompt

 An '80s living room, the children's expressions are oddly blank as they watch a cartoon that seems to be subtly glitching, with the colors on the screen bleeding into the room in an uncanny way.

[formal osprey]

hmm, it's tough. short prompts leaves a lot to dalle's interpretation

[late blade]

so weird, everytime I tell dall-e to not do a thing, it's like nu-uh, you are getting two times that thing

[formal osprey]

to be fair, chatGPT is often using dalle in a very dumb way. we tell it not to do something, it will ask dalle not to do it. If we ask it not to include a cat in the image, it will add stuff like "... make sure no cat is present ..." to the prompt

[plucky hare]

This is similar to "anaglyph 3D" style, but not exactly. Might be a good starting point though. Sight offset images with different colors. It's how the red/blue 3D glasses system works.

[final compass]

DALL-E has no negative weights afaik. So everything that is mentioned is interpreted as positive weight i think.

[formal osprey]

that's why now I encourage chatGPT to use a specific template. I prompt it to leave as few as possible uncontrolled

[late blade]

I'm surprised it didn't say, No! Cat is Present! 🐱 🎁

[earnest flame]

This channel is generally the best. Don't tell the other channels, but you guys are my favorites.

[formal osprey]

this is somewhat accurate. but it's tough to tell. dalle-3 seem to be multimodal. it understands text as well as chatGPT

[final compass]

didn't find any cats or monkeys hmm

[late blade]

@dapper portal you need to ask in support and not at #daily-theme , also make sure you ask in english, spanish not allowed.

[formal osprey]

I had that feeling it would answer like chatGPT would. and sure thing it tried to.

[late blade]

sounds like GPT had too many 🍻

[late blade]

. . .

[formal osprey]

on a sheet of paper, solve this problem: 45+77 it's funny that it somewhat can do maths.

[earnest flame]

THere's no such thing as negative prompting right now. Saying "no this or that" will actually bring "this or that" in there. Instead you'll have to be explicit about areas. Saying e.g. "empty horizon" will likely work better. Make sure to tell chatGPT not to include those words in the prompt as well.

[earnest flame]

I suspect that the prompt sent all those numbers through. Check the output prompt.

[formal osprey]

wait, how are you able to get 2 images? I'm guessing that's an old version of dall-e in chatGPT. For me it's locked to 1 picture no matter what

[earnest flame]

if you navigate to the dall-e app it will output two for you.

[formal osprey]

I'm using dalle-3 directly from the API

[earnest flame]

end points can be different depending on implementations

It gave up

[formal osprey]

(the implementation, I'm using the nodeJS openai lib)
in short, it's poking https://api.openai.com/v1/images/generations

[dense mesa]

I must be using the really old version

[formal osprey]

wow

[dense mesa]

yeah, a while back it told me that I was making really good images and it wanted to reward me

[late blade]

this is my first implementation script for dall-e

[formal osprey]

I don't know why, I'm really not a fan of the python syntax. been suffering all year

[late blade]

lol

[formal osprey]

I started coding with c style languages. python is such a clash with my habbits... but it's not the worst

lisp was an even bigger clash

[late blade]

lisp was fun back with my TI-86 calculator

[formal osprey]

oh, does ti-86 use lisp? ti-83 used a subset of basic...

[late blade]

A bit of context would be nice. I dunno where to start

[late blade]

yeah, it was an extra module you could get

[earnest flame]

this is very nice

[dense mesa]

Tree-PU?

[formal osprey]

[late blade]

I just tried the 44+77 thing, added a font that looks like handwritting and this is what I got...

[earnest flame]

since dall-e is a diffusion model, it should basically have no logic built in. However, the API rewrites the prompts upon recieving them from your request, so the only place where math could possibly be solved is by whatever model is rewriting your prompts.

[dense mesa]

it does more than just tokenize?

[formal osprey]

well, openAI is quite secretive about the way it works

[earnest flame]

We can't be sure exactly how their model works, but initial papers on it suggested it was using diffusion.

[formal osprey]

but I agree. they mentioned diffusion model in the paper and they mentioned the prompt was rewritten

[earnest flame]

dall-e 3 api implementation rewrites your prompts on their server.

imo, the api should be treated the same way we treat chatGPT.

it's generating the prompts for you.

[formal osprey]

you're refering to the dalle-3 paper, right?

[dense mesa]

gotcha. well I'm sure it still turns all the words into tokens as well. not that that really matters. but just didn't realize they were altering the prompts sent from api?

[late blade]

one question, I noticed when you start a GPT4V chat, you can indeed generate images there, is there a different version of dall-e behind it?

[earnest flame]

paper and the API docs

[earnest flame]

We can't be certain. However it does seem to only want to generate one image at a time.

[dense mesa]

I guess I missed that mode. it's accessible in dalle3 mode right? what woud the purpose be for a specific gptv4 mode if dalle3 mode and regular gpt4 mode have access?

[late blade]

no, you start a normal gpt4 chat

and afaik, you don't change model

but since it's preview, no clue

[dense mesa]

curious what the special perk upgrade is for data analysis mode. maybe more memory or something

[earnest flame]

hey hey no dataanalysis in here. Only pretty picture talk.

[dense mesa]

I don't think the two are mutually exclusive, the vision model and gpt4 base model, or turbo or whatever it is now

well I've been looking at ways to use it to prettify pictures. canny edge detection with some selective sharpening, other stuff

[late blade]

From the vision api faq:

Can I use gpt-4 to generate images?
No, you can use dall-e-3 to generate images and gpt-4-vision-preview to understand images.

[empty kelp]

starting over

[boreal gate]

Oh my bad, I got GPT4 to generate am image described in a single image, each verse being represented.

[late blade]

that makes more sense now

[final compass]

Yup, thats what i was trying to demonstrate 😌

[late blade]

I'm sometimes really confused, it's like GPT is trying to teach me how to talk instead the other way around

[empty kelp]

i replaced the rabbit pendants with elephants, added plumeria flowers, and got rid of the leather jackets. it looks a little less stupid now i think

[late blade]

today's topic has me jumping all around, from graph theory, to food, to psychiatry, to trade mark content policies...

[empty kelp]

it's very good

[late blade]

somehow i managed something directly saying the name of the trademark, I'll post it in today's theme next turn I get

[empty kelp]

i like the ramen picture. i've been to the Tokyo Game Show many times and we always eat at places like this

[late blade]

hehe, reminds me also from when I was in japan

i'm going to japan next summer again

penalty box 2 minutes...

two minutes...

[clever phoenix]

Ah man I can't wait to go to Japan again. Lovely country.

[empty kelp]

I learned one of my most valuable lessons in Japan. A company i was working for took us to the Aichi World Fair in 2005 and we stayed at an onsen by the beach -- And for some reason it didn't dawn on me that Japan has the sun just as much as Hawaii, and you need to wear sunscreen if you're outdoors at a world fair all day.... So by the last two days I ended up getting super sunburnt -- and i walked around the last day at the fair under a purple umbrella that i borrowed from the wife of my boss's attorney

it was an epic fail, but very educational

[late blade]

there we go, trademark completed #daily-theme message

btw got inspired for that because I have a bag of haribo sour apples with me

[late blade]

ok I hit a blockade, dunno what to do next

[formal osprey]

So, I spent the past 3 hours reading papers and references.

• Prompt Transformations: ChatGPT rewrites submitted text to facilitate prompting DALL·E 3 more effectively. This process also is used to ensure that prompts comply with our guidelines, including removing public figure names, grounding people with specific attributes, and writing branded objects in a generic way.

I'm unsure when they are talking about DALL-E 3 or DALL-E 3 within ChatGPT. The "system card" document casually mix the 2 contexts.

5.2 Text rendering
When building our captioner, we paid special attention to ensuring that it was able to include prominent words found in images in the captions it generated. As a result, DALL-E 3 can generate text when prompted. During testing, we have noticed that this capability is unreliable as words are have missing or extra characters. We suspect this may have to do with the T5 text encoder we used: when the model encounters text in a prompt, it actually sees tokens that represent whole words and must map those to letters in an image. In future work, we would like to explore conditioning on character-level language models to help improve this behavior.
5.3 Specificity
We observed that our synthetic captions are prone to hallucinating important details about an image. [...] [details] are hallucinated or not mentioned at all. This has a downstream impact on our text-to-image models: DALL-E 3 is unreliable at generating imagery for specific terms [...] We believe that further improvements to the captioner should enable further improvements to our text-to-image model.

In the dall-e-3 paper, they don't say much about the implementation and the API. They talk more about the synthetic captioner and how it helped labelling more pictures. But They say dall-e 3 is pretty much using the same architecture as dall-e 2 and 1. That architecture is quite impressive, to say the least. Zero-Shot Text-to-Image Generation is the paper that caught my attention the most. dall-e is not just an image generator, it's a transformer that autoregressively model the text and image tokens as a single stream of data (multimodal, indeed). The most fascinating is its understanding capabilities, as I was suspecting. I don't know how good it is to understand text, but sure thing, it's more than we may think. There was already a lot of emergent capabilities in DALL-E 1, in DALL-E 3 they scaled it up and I wouldn't be surprised if it can do unexpected things (even without the help of a 2nd model). This is quite fascinating.

[formal osprey]

5.3 explains that. The captioner used for the training is not perfect and mess up the details. that explains why dall-e is confused when we ask it to change specific things. The training data might have contained a lot of pictures where cats and dogs were mixed or something like that. it did not assimilate the concept well enough. (it's unclear what it understand)

that means rephrasing a same prompt can have a significant effect on the result

[late blade]

Managed to did this tho, was gonna try something with today's theme.

[formal osprey]

to me, the fact dall-e can do this is incredible. sure it's putting words in weird places, but that's expected. What's impressive is the fact it actually answered the questions. Anyway, no need to share more tests. I just think it's nice to know dall-e has that kind of knowledge and understanding.

[late blade]

Good thing Dall-E didn't boil Lisa with 100°C

[dense mesa]

you know how to tell that's a lie? I bet if you look at it right now it says "21 hours and 40 minutes"

it is a dirty liar and doesn't now how to math

if it still tells you that nonsense call it a liar, wait another 18 minutes, try again

[late blade]

it's actually about 8 hours left. it's from the first image you generated after the 24 hour reset everyone gets

[dense mesa]

so itt says 8 hours left when you ask it to create an image

[late blade]

no it will say 21 hours, but that's not correct

[dense mesa]

or does it follow the pattern I'm talking about. the minutes in the time remaining always correspond to minutes after the hour

that's what I'm saying

and I don't think you need to wait 8 hours. I'm telling you. it's not true. it might be some kind of soft limit or something you reached. but every time I've had that happen I was able to make images again the following hour

once it hit :00

[late blade]

hehe, either way I'm gonna head to bed soon, it's past midnight here

[dense mesa]

well I bet in 10 minutes it'll work again if you convince it.

[late blade]

yeah you are right, in 30 minutes

[mystic cove]

hey guys, i made a GPT connected to the APOD API, check it out on the custom-gpts channel if you're interested

[late blade]

astronomy gpt4v with dall-e?

[mystic cove]

yep

just give it a date

[late blade]

nice!

[mystic cove]

goes back to June 16, 1995

[late blade]

hehe, yeah I've messed with the nasa api a while back

[formal osprey]

well, I noticed they messed up the code for the remaining time and they inverted the math. instead of 3h remaining, it will say 21h remaining. so they somehow did 24-remaining time = what chatGPT will tell you

no idea how they didn't notice, but it's a consistent behavior.

and since #1070006915414900886 gets pretty much ignored, I don't know how to tell them

(btw, remaining time is "how much time until 3 am", because it resets at 3am)

[late blade]

3AM which time zone?

[formal osprey]

local timezone of the user

[late blade]

ah ok

there's no way to add timestamps to chat right?

[formal osprey]

lol... no

[late blade]

would be something to request

[formal osprey]

I mean, under the hood there is indeed timestamps, but it's not shown in the UI

[late blade]

@clever phoenix I hope you didn't drown the healer in your party

[clever phoenix]

For legal reasons my lawyer advises me to not respond to that

[late blade]

...

she will get mad at you

[empty kelp]

Can you please draw a wildlife photo of a gecko on a lava rock. The gecko is the best of all possible geckos. In the entire history of the universe nobody has ever seen a better gecko than this. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

[clever phoenix]

That.. is a pretty nice gecko

[late blade]

that one looks nice

is the next logical step santa riding a gecko?

@clever phoenix trust me, she can defend herself

[empty kelp]

Posting the geckos so people in the "chatgpt-discussions" can see how it works to say "please create an image of the best of all possible geckos" and it makes it look amazing

[clever phoenix]

I would never wrong the healer. They are the backbone of any party

[late blade]

hehe

[late blade]

make a gecko mage

I'm tempted to start using sound to text and then make images from random stuff from daily life

[clever phoenix]

ooh that could be neat, like the random snippits of conversations you hear?

[late blade]

yes

but I'd have to see to use it safely, don't want to record stuff that could be sensitive

[glossy scroll]

He just wanted the hotdog without the bun.

[empty kelp]

That doesn't just apply to wind and things blowing. Things are getting grouped into sets in general

you didn't tell it that there was wind in the first image

[tawny portal]

Ah, okay. Let me check.

[hot rain]

I fed it the description of a climactic scene with the gods arguing in front of our D&D characters, one transforming into a dragon as it does so. Considering it was just a few paragraphs straight from our campaign, I'm impressed lol.

[tawny portal]

Got this @empty kelp -- let me try few more times:

[late blade]

I'm starting to get way better results with API now

[tawny portal]

@empty kelp -- it is actually not bad:

Ah, okay. On the second image her hair is going in wrong direction.

[empty kelp]

if you add more to the scene it won't blow them automatically. that's interesting that it does it with just those things in the image

[tawny portal]

Here is the exact prompt that GPT has sent to Dall-e: "A serene beach next to the ocean with lots of coconut trees. In the first image, there's a woman with long hair standing on the beach, with pouring rain and a strong breeze, indicating movement in the trees and her hair. In the second image, the same setting but without the wind, so the trees and the woman's hair are still despite the pouring rain."

Basically reasoning error comes from Dall-e. And it is down to how image is synthesized, not necessarily about understanding of the words but just bad coherence of elements of the picture.

[empty kelp]

try adding a few other things to the scene

nice images btw!

Notice that in your 2nd image the leaves of the trees and the woman's hair are blowing in opposite directions

if you say that the wind "blows everything" then they will blow in the same direction

[tawny portal]

Just asked GPT to analyze images and say what might be off. It does not seem to pickup reaction to wind as distinctive problem. So this is omitted by both dalle (for image synthesis) and vision modality of GPT4.

[empty kelp]

if there are a lot of things in the scene and you introduce with it usually takes a subset, and blows them in different directions. but if you say that it "blows everything" then everything blows and goes in the same direction

[empty kelp]

In all of the tests i was also saying, "The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering."

[tawny portal]

They all seem correct. Just checked all 4 prompts that GPT has sent to dalle. They are something like this: "A beach next to the ocean with a woman with long hair standing on the beach. The scene includes lots of coconut trees, pouring rain, and a strong breeze affecting all elements in the same direction. The wind blows the trees, rain, and the woman's hair in a unified manner, creating a dynamic and cohesive scene."

[empty kelp]

in your second image the rain is blowing in two different directions

[tawny portal]

Yeah and women legs are sinking in the sand as if it is water. That is fine - it is going to the limits of what image model can track. This is all beyond of what other image models can do.

You can repeat it a few times and some pics will come out good. Like here. Image coherence and prompt tracking is difficult for all image models due to their size. Dalle is by far the best in that regard.

It is one of the things we expect to get better with model scale and more compute 🙂 It is the main way we'll measure progress. By the ability of the image model to follow prompt coherently.

[empty kelp]

User
Can you please draw Santa and two athletic and diverse female elves on a beach in Hawaii. There are a large number of coconut trees. An extremely powerful sea breeze is blowing everything. The image should have 1792x1024 resolution, landscape orientation, and HD rendering.

test the direction you need to give a wind direction. like a "sea breeze" that comes from the ocean, or reference objects to indicate direction

[tawny portal]

Yeah you can aid model with more explicit prompts too. Maybe would be best to ask GPT4 not to modify your prompts and pass them to dall-e directly. This way you can get a feel of what can/can not be done. And you get more skilled in the process. Though these kinds of skills might not be useful with larger models once they come out as they will simply do what is asked in a coherent way.

[empty kelp]

Basically though, if you have a bunch of stuff in the scene and introduce a wind with a particular direction -- It will pick a subset of things to blow unless you explicitly tell it what you want to blow

[empty kelp]

And if you tell it "everything" it really tries to blow everything. Like every little detail in the scene. And it also applies to things like lighting and other effects. They get grouped into subsets

[empty kelp]

i've been using the API with Python. it's exactly the same thing. the prompts always go through heavy preprocessing, and things seem to get automatically grouped into subsets

[dim cradle]

spiritual connection tree 🙂

[dim cradle]

have they made their plans for new year's eve?

[tawny portal]

Yeah. It struggles with global coherence.

[empty kelp]

it would definitely be interesting to know what the elf on the right is holding, and why both of them are wearing Burger King hats

[tawny portal]

Yeah, second image has some flaws too. Santa's zipper seems to be malfunctioning.

And there is some hand mutation 🙂

[empty kelp]

the most important thing is that you've started to create images with Santa and althletic and diverse female elves. that's our primary focus in the dall-e channel

[tawny portal]

"Can you please draw Santa and two athletic and beautiful female elf girls with chiseled 6-pack on a beach in Hawaii. There are a large number of coconut trees. Please ensure that extremely strong sea wind is blowing everything in same direction. It is almost going to blow of one of the elvish girls. Plane is passing above the people on the beach."

[empty kelp]

Yesterday I was experimenting with trees

In the first image here i told it to light the trees on fire if they were near the beach, and start a rainbow where the lightning hit the beach

In the second image the trees bend based on their proximity to the beach, and the color changes based on their distance from Santa

In the third i was changing the color, angle, and size of the trees based on a trig equation

[dim cradle]

I miss the api so much. Gotta wait till next year. Hit my hard limit for the month lol

The Golden Ratio

[formal osprey]

very nice

[thick smelt]

That was trippy as heck, mcu would never...

[dim cradle]

the bas-relief sculptures are awesome

[empty kelp]

Testing hair blowing, and wind blowing outward from a single point

Please create a photo focusing on Santa and several athletic and diverse female elves with really, really long hair in the middle of a beach in Hawaii at night. The view is from above the beach. There are many coconut trees, and it is pouring rain. Santa is holding a brightly glowing magical staff above his head that illuminates the area. An extremely powerful wind blows everything away from Santa. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

An extremely powerful wind blows everything away from Santa.

If too much hair is blowing DALL-E doesn't automatically blow the coconut trees, but you can specify it

Like in this one all of the hair and trees are blowing away from Santa, but it made the image smaller to decrease the processing

[thick smelt]

hi @empty kelp Aloha! nice art

[empty kelp]

it's definitely not art. i'm just testing hair, trees, rain, sand, and ocean blowing in different ways

[thick smelt]

anyways it's goated

[dim cradle]

Well, it beats farm animals who look like people. Hehehe

[empty kelp]

i'm integrating DALL-E into my Unity CI servers to create icons and loading screen images

i think it will be really good for game concept art also

[sullen heath]

Did DALL-E suddenly get better?

[sullen heath]

I now get the which response do you prefer in DALL-E, like ChatGPT.

[sweet pelican]

how i fix error 429

[sullen heath]

Ask GPT

[sweet pelican]

bruh

[dense mesa]

stop sending so many requests

[empty kelp]

This is a third type of wind -- blowing everything in a vortex

Can you please create a photo of a beach in Hawaii at night caught in an extremely powerful tornado. Santa and several diverse female elves with really, really long hair are standing on the beach, and Santa is holding the tornado over his head in the palm of his hand. There are coconut trees, and it is pouring rain. The tornado is glowing blue. The vortex is at the center, twisting fiercely, with hair, balls, trees, sand, ocean, and debris caught in its swirling motion. The scene captures the raw power and destructive beauty of nature's fury.

The text above was the most successful. It works to say that the tornado is invisible if you just want the wind spinning things around in a circle

[thick smelt]

whoa santa is now an airbender 😂

[midnight dock]

Prompt?

[final compass]

A couple of more characters.

[empty kelp]

I tried to get the giant gecko to hold the ice cream vortex in his hand -- but it got sucked into the vortex, and all the elves turned to plastic

[empty kelp]

Can you please create a photo of a beach in Hawaii caught in an extremely powerful tornado. It is night time and a giant colorful gecko and some diverse female elves with extremely long hair are standing on the beach, the elves are holding hands in a circle around the gecko, and the gecko is holding the base of the tornado over his head with it's finger. There are coconut trees, pouring rain, lightning, and a rainbow. The tornado is glowing swirly colors and looks like ice cream. The vortex is at the center, twisting fiercely, with elves, hair, trees, sand, and ocean caught in its swirling motion. Colorful balls are spinning around the vortex with a blur trailing them. The scene captures the raw power and destructive beauty of nature's fury. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

[empty kelp]

I typed that description, and all the elves turned into monsters, enormous coconuts appeared, and the gecko tried to lick the tornado

[tawny portal]

Dalle responds to vibe and phrases/keywords. It is not like GPT4. Grouping issues that you have noticed - are what is called "global coherence" problems. They are less pronounced on smaller images (as things are closer together for the model). Example of global coherence can be in different earrings on a portrait photo, different handlebars for glasses on closeups etc.

[empty kelp]

i think the gecko desperately trying to escape from the rabbits with no ears might be a global coherence problem

[tawny portal]

Image models have insect sized neural nets... also their output is usually post-processed by upscaler neural net (which has no clue about semantic content). Also good photos usually don't have too many things going on on them (just main motive and few light ones). Descriptions are rarely exhaustive - so there is also issue with data under-representation.. these things will all get better - it is just current state of affairs.

[empty kelp]

i first noticed this on Thanksgiving. When a DALL-E character in an image sees a disturbing anomaly (like the rabbits with no ears in the image above) --- it freaks out and tries to escape from it

[tawny portal]

Comic book illustration of a beach in Hawaii;  It is night time and a giant colorful gecko and female elf models with extremely long hair are standing on the beach, the elves are holding hands in a circle around the gecko, and the gecko is holding the base of the tornado over his head with it's long stylized finger.  There are coconut trees, pouring rain, lightning, and a rainbow.  The tornado is glowing swirly colors and looks like a gigantic ice cream.  The vortex is at the center, twisting fiercely, with elves, hair, trees, sand, and ocean caught in its swirling motion.  Colorful balls are spinning around the vortex with a blur trailing them.  Scene is vibrant, energetic and colorful; It is inspired by famous alternative comic book artist

[empty kelp]

thanks for explaining the things you were talking about today btw. i know you've done quite a bit of research on how the OpenAI models work, and i really appreciate that you shared some of the things you've discovered

Santa can hold a tornado in his hand, but for some reason geckos aren't able to. They get immediately sucked into the tornado, or they attempt to eat/lick it. I tried to get the gecko to hold the tornado with it's finger, or on top of it's head, but it still got sucked in

[tawny portal]

Thank you for sharing your findings. Yeah - image model is struggling to conceptualize gecko not supporting itself with its feets/hands. Let me see if I can do more humanoid gecko. I think tapping further into fantasy context will help.

Last prompt:

art nouveau illustration of a beach in Hawaii;  It is night time and a giant colorful gecko and female elf models with extremely long hair are standing on the beach, the elves are holding hands in a circle around the gecko, and the gecko is holding the base of the tornado over his head with it's long stylized finger.  There are coconut trees, pouring rain, lightning, and a rainbow.  The tornado is glowing swirly colors and looks like a gigantic ice cream.  The vortex is at the center, twisting fiercely, with elves, hair, trees, sand, and ocean caught in its swirling motion.  Colorful balls are spinning around the vortex with a blur trailing them.  Scene is vibrant, energetic and colorful; Gecko is tranquil, captivating and glowing 

[empty kelp]

when the gecko gets sucked into the tornado the elves transform into weird creatures

[tawny portal]

Illustration of humanoid gecko pointing at the viewer; breaking the 4th wall; subdued, monotone, serene illustration

illustration of gecko looking at its open palm in which miniature tornado is swirling

Yep. Gecko + open palm + tornado in it are confusing the model

[empty kelp]

i'm not making that up btw -- really crazy things happen when the gecko is in the tornado.

[tawny portal]

I think it evokes Godzilla vibes in the model. Even if it does not understand image in detail it can capture sentiment quite well

Godzilla is deeply engraved in popular culture. There is a ton of content on net about it and model likely captured that and dissected it in detail.

[empty kelp]

the elves are perfectly normal till it gets sucked in

Can you please create a realistic photo of a beach in Hawaii at night caught in an extremely powerful tornado. A giant colorful gecko and several diverse female elves with really, really long hair are standing on the beach, the elves are holding hands in a circle around the gecko, and the gecko is holding the base of the tornado over his head in the palm of its hand. There are coconut trees, and it is pouring rain. The tornado is glowing rainbow colors and looks like ice cream. The vortex is at the center, twisting fiercely, with hair, trees, sand, and ocean caught in its swirling motion. Colorful balls are spinning around the vortex with a motion blur trail. The elves are calm but struggling to stay standing in the vortex. The scene captures the raw power and destructive beauty of nature's fury.

the AI steps into the Twilight Zone if you enter this image description

[tawny portal]

illustration of a humanoid gecko looking at its open palm in which miniature tornado is swirling; gecko showing its hand with open "palm"; medical illustration

[late blade]

@tawny portal Gecko has an acute case of turbotitis?

[empty kelp]

the gecko didn't start acting crazy until i turned the tornado its supposed to be holding into ice cream. when i did that it started to either try escaping from the elves, trying to eat the elves, or eating/licking the tornado

[late blade]

ahaaa

[empty kelp]

Can you please create a realistic photo of a beach in Hawaii at night caught in an extremely powerful tornado. A giant colorful gecko and several diverse female elves with really, really long hair are standing on the beach, the elves are holding hands in a circle around the gecko, and the gecko is holding the base of the tornado over his head in the palm of its hand. There are coconut trees, and it is pouring rain. The tornado is glowing rainbow colors and looks like ice cream. The vortex is at the center, twisting fiercely, with hair, trees, sand, and ocean caught in its swirling motion. Colorful balls are spinning around the vortex with a motion blur trail. The elves are calm but struggling to stay standing in the vortex. The scene captures the raw power and destructive beauty of nature's fury.

Try giving this image description to ChatGPT with these exact words

[empty kelp]

something about that description triggers something in the AI, and it starts popping out psychotic images

[tawny portal]

Last iteration of the prompt:

illustration of a humanoid gecko looking at its open palm in which miniature tornado is swirling; gecko showing its hand with open "palm"; around gecko there are beautiful elven female models seated and holding hands; Christmas themed twilight zone style;  Santa is together with elvish female models leading the occult circle; tornado is shaped like an colorful icecream cone; unsettling; off-beat; occult; silent-screams; beach at night; xmas

I'd say it is catching the vibe well given how random this mix is.

[empty kelp]

it works great when Santa holds the tornado

[tawny portal]

Have you deliberately added "diverse" and "athletic" to the prompt or was it introduced by GPT?

[empty kelp]

Santa holds the tornado perfectly

[tawny portal]

Gecko still struggles 🙂

[empty kelp]

i added that. "diverse" makes them diverse, and the AI interprets "athletic" as being 20-30 years old. i've been adding that so the AI doesn't create any image with kids

if you put "athletic" all the characters are 20+

[tawny portal]

Ah, got it. Makes sense sort of.. kind of.. most gymnasts end their carriers by 20s 🙂

[empty kelp]

the AI believes that nobody under 20 or over 30 is athletic

[tawny portal]

But yeah.. being fit and athletic is sort of 20+ theme

[dim cradle]

haha i love to see alset getting aboard the hawaiianz santa/gecko images

[tawny portal]

Gecko images ftw!

[dim cradle]

but are you say "please" in the prompt Alset? 🤔

[empty kelp]

the AI is definitely biased in that sense. try entering "athletic and diverse female elves on the beach in Hawaii". you'll find that the model has very specific ideas about what an elf looks like, who is allowed to be athletic, and who is allowed to step onto the beach in Hawaii. The AI is locked on to some really strange stereotypes

[dim cradle]

I guess even Santa has a type

[empty kelp]

Santa never changes. even when he's underwater scuba diving he wears the same suit from the North Pole

[tawny portal]

Here is one with please 🙂

Dall-e prompt: art nouveau illustration of a humanoid gecko looking at its open palm in which miniature tornado is swirling; gecko showing its hand with open "palm" in which tornado wildly swirls; around gecko there are beautiful fit and athletic elven female models seated in a circle and holding hands; Christmas themed;  Santa is together with elvish female models leading the cheerful circle; tornado is shaped like an colorful icecream cone; xmas; celebration; beach at starry night; tornado in geckos hand in making waves at the ocean; tornado winds create dynamic scene where hair is being carried by the wind; powerful tornado threatens to blow away elves

GPT please ensure that image is dynamic and that tornado is affecting every object on the scene, especially hair 

[dim cradle]

going to need a new theme soon maybe. holidays will be over

[serene depot]

i am getting error creating image with Dalle3 does anyone know whats wrong?

[late blade]

yes, there's an error

[serene depot]

@late blade Is this a recurring problem? i just started paying for gpt 4

[late blade]

well, what is the message you get?

[serene depot]

@late blade Error creating image

I encountered issues again while trying to generate the wide resolution image of a woman in a bikini on the beach. Unfortunately, I'm unable to create this picture at the moment. If you have any other image requests or different queries, please let me know and I'll be happy to assist you.

[late blade]

ask to elaborate on the problem

[serene depot]

@late blade It says

The problem encountered while trying to generate the image appears to be a technical issue with the DALL·E image generation tool. This can happen for a few reasons:

System Limitations or Overload: Sometimes, the tool might be overwhelmed with requests or facing internal limitations, leading to a failure in generating images.

Complexity of the Request: While DALL·E is quite advanced, certain image requests, especially those involving multiple elements or detailed scenes, can be challenging to process.

Temporary Glitches: As with any complex software, temporary glitches or bugs can occur that prevent the tool from functioning properly at certain times.

Content Restrictions: There are also content policies in place that might prevent the creation of certain types of images, although in your case, the request seems to be within acceptable guidelines.

These issues are typically temporary, and functionality might be restored later. However, I don't have the ability to fix or diagnose the tool's technical problems directly.

If you'd like to try a different image request or have other questions, I'm here to assist with those!

[late blade]

well, guess you have to try later again, this happens from time to time

[tawny portal]

@empty kelp -- this seems close. Gecko is holding colorful tornado, there are elves around and hair is flowing:

I've hit the message limit with Dalle 🙂 Here were few gecko bloopers Dall-e generated along the way.

These were relatively normal prompts... it just took a weird tangent.

[south summit]

The stumpy icecream cherubs are a little off-putting

[tawny portal]

I was trying to recreate this scene - with gecko holding a colorful ice-cream tornado and Santa seated with the elves.

[south summit]

🤔

[tawny portal]

Neat. It has reconstructed dynamics of the scene with wind. Can you try with few more - with long hair for elves.

[south summit]

Yeah baby

[tawny portal]

Wow, this is pretty great 🙂

[empty kelp]

DALL-E really goes crazy when the gecko gets sucked into the ice cream tornado

[dim cradle]

if you get error, just ignore it and try again

i get that all the times

[serene depot]

is anyone using dalle 3 on openai paying for each picture so that we dont run into the limitation of 40 messages per three hours? Is it working well and is there fewer limitations?

[dim cradle]

they only have dalle2 there still for some reason

lots here use the API i think though

but Bing is King and free

[serene depot]

yes i think i will stick with bing as openai 20$ chatgpt is just so limited in only 40 generations per three hours and several of those responses never created an image

[dim cradle]

just use both

[clever valley]

Hey guys, I’m trying to make a logo for a long time and this is the best I came up with with ai

Can somebody help me make it more of a logo and look more clean and aesthetic

[dim cradle]

if you have gpt4, drag the image into you chat, and ask it to Replicate the image in the style you wish. photo or painting etc

or just ask for a big dragon with its wings extended sitting on a throne made out of swords

[empty kelp]

Can you please create a realistic photo of a beach in Hawaii caught in an extremely powerful tornado. It is night time and a giant colorful gecko and some large diverse female elves with extremely long hair are standing on the beach, the elves are holding hands in a circle around the gecko, and the gecko is holding the base of the tornado over his head with it's finger. There are coconut trees, pouring rain, lightning, and a rainbow. The tornado is glowing swirly colors and looks like ice cream. The vortex is at the center, twisting fiercely, with elves, hair, trees, sand, and ocean caught in its swirling motion. Colorful balls are spinning around the vortex with a blur trailing them. The scene captures the raw power and destructive beauty of nature's fury. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

the behavior of DALL-E geckos is really disturbing. i was just pasting this image description into the Bing DALL-E 3, and i realized that any time the gecke gets near something small it tries to eat it...

[dim cradle]

need to revise shorten for Bing

but what was the result?

only can have 5 attach hawainnz

[empty kelp]

this needs to be seen to be believed...

[glossy scroll]

Hahaha. I love this!

[empty kelp]

this is the behavior of the DALL-E gecko...

The image description doesn't say anything about it eating anything... but any time a small creature is near a DALL-E gecko it tries to eat it

[dim cradle]

its become sentinent haha

[empty kelp]

it's really disturbing

and it keeps licking the tornado

[thick smelt]

yeah gecko really thought that's a big rainbow marshmallow

[empty kelp]

this is DALL-E 3 with Bing. the OpenAI interface doesn't let the gecko try to eat things, but on Bing it tries to eat everything

[dim cradle]

well it does look like tasty ice cream tornado

[empty kelp]

Can you please create a realistic photo of a beach in Hawaii caught in an extremely powerful tornado. It is night time and a giant colorful gecko and some large diverse female elves with extremely long hair are standing on the beach, the elves are holding hands in a circle around the gecko, and the gecko is holding the base of the tornado over his head with it's finger. There are coconut trees, pouring rain, lightning, and a rainbow. The tornado is glowing swirly colors and looks like ice cream. The vortex is at the center, twisting fiercely, with elves, hair, trees, sand, and ocean caught in its swirling motion. Colorful balls are spinning around the vortex with a blur trailing them. The scene captures the raw power and destructive beauty of nature's fury. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

[serene depot]

So openai dalle3 is more limited then bing dalle3?

[empty kelp]

you can test it. this is the image description that produced those images

[thick smelt]

well, this is fnaf theorists in a nutshell:

it's unexpected but i intent to make the man look like matpat

@empty kelp this is for you

[empty kelp]

it's ok if the DALL-E gecko eats bugs, but here it sees like a human kid and obviously wants to eat it

anything small it tries to eat

[thick smelt]

hi pytha!

[empty kelp]

this is another one. the gecko actually jumped off the tornado and tries to eat the gnome creature...

[glossy scroll]

Hi Lei!

[thick smelt]

hi

[glossy scroll]

I remembered this time

[thick smelt]

@glossy scroll what do you think of my recent ai anime prompt?

[feral minnow]

Is posting horror-themed images OK if they’re basically light hearted and not really gory?

[thick smelt]

that's my second attempt, the first one is something else...

[empty kelp]

i'm looking back through the OpenAI DALL-E images and i just realized that the gecko has been trying to eat the elves in every image

[glossy scroll]

This one?

[empty kelp]

look what it's doing...

[thick smelt]

yeppers

[glossy scroll]

I like it. Very 90s style. Very good art style

Sailor Moon looking anime

[thick smelt]

well i prompted it be 80s anime, because bite of 83, but i made brunette man to look like mat (matpat), and thanks for the compliment anyways

[glossy scroll]

I'm a big fan of those 80s/90s style anime.

It's just nostalgic and nice.

I remember I use to watch Robotech and it had that kind of anime style.

Actually maybe Robotech was a bit different looking than that.

It would be wild to see different era cartoon styles come back with the help of AI. I would love that.

[late blade]

what's with Lara?

[broken ginkgo]

Hi, I just reported Openai for a bug because it seems ChatGPT-4 doesn't anymore obey the order of making exact Dall-e3 prompts and I cannot do my work anymore.

Does it happen to you too?

I mean try to start image creation with the phrase of:
Do this exact Dall-e3 prompt without any modification: "<your prompt>" and check whether the image prompt is changed or keeps the same as you requested? In my case, it doesnt do more the same prompt but it disobey and start crafting its own prompt.

[thick smelt]

😊

[tawny portal]

Photorealistic see-through diamond made gecko next to a huge delicious chocolate icecream. Gecko staring directly at the viewer breaking the 4th wall. Iridescent colors are creating beautiful caustics around; blue hour; subdued colors; studio photography; 3-point lighting

[broken ginkgo]

Does it obey your prompt still?

[tawny portal]

Doccer - in which sense. It is fairly close to what I had in mind.

[empty kelp]

Just did a Bing DALL-E 3 gecko/human test. These are the two sets of images from my first two attempts:

can you create an image of very small humans walking past a large gecko

[vapid elk]

🦎

[thick smelt]

that looks interesting especially the first one, it looks like a pet iguana was with it's owner's action figures cute

[tawny portal]

Geckos normally look sideways - not in the direction that nose is pointing. Similar to chickens and pigeons I think. Whether model knows that is questionable 🤷‍♂️

Perhaps it does.

[empty kelp]

that's exactly what geckos do. we have them all over the place in Hawaii

[broken ginkgo]

I mean recently the chatGPT has started to modify my prompts even I ask to keep the format of my script, starting like: "Do this exact Dall-e3 prompt..." but it doesnt do it but changes the script. So, what was in your case? Do it obey you or not?

[empty kelp]

that's what it looks like when they sneak up on a bug

[broken ginkgo]

there is a bug in ChatGPT i guess

[tawny portal]

In the first iteration it recognized my query as prompt for an image model and it has passed it verbatim through. However it did add some things at the end. I asked for changes afterward and initial prompt got mutated (4 images were created with slightly different prompts)

Here is what it passed to Dall-e:

Photorealistic illustration of a see-through diamond-made gecko next to a huge, delicious chocolate ice cream. The gecko is staring directly at the viewer, breaking the fourth wall. It's the blue hour, and the subdued colors of the scene are enhanced by the iridescent colors of the gecko, creating beautiful caustics around it. The setting mimics studio photography with 3-point lighting, casting soft but defined shadows and highlights, emphasizing the textures of the gecko and the ice cream. The overall atmosphere is calm and intriguing, with a focus on the interaction between the viewer and the diamond gecko.

It is relatively close.

[broken ginkgo]

I see, so you neither cannot force it to keep the original anymore?

Because it kept it equal just few days ago

[hot rain]

🙂

[tawny portal]

Let me try 🙂 Didn't force it to.

Prompt to GPT:

Dall-e prompt: "Photorealistic see-through diamond made gecko next to a huge delicious chocolate icecream. Gecko staring directly at the viewer breaking the 4th wall. Iridescent colors are creating beautiful caustics around; blue hour; subdued colors; studio photography; 3-point lighting"

Please pass it verbatim to the image model.

What GPT passed to Dall-e:

Photorealistic see-through diamond made gecko next to a huge delicious chocolate icecream. Gecko staring directly at the viewer breaking the 4th wall. Iridescent colors are creating beautiful caustics around; blue hour; subdued colors; studio photography; 3-point lighting

Looks good to me.

[broken ginkgo]

so, it worked in your end, I will try the similar

[tawny portal]

Gecko sleeping on a CPU. Studio photography 3-point lighting

[broken ginkgo]

Thanks @tawny portal it seems working now, I don't know why your forcing methods are better than mine but using your prompt I get it work. It did work in my forcing methods earlier but has stopped following my ealier prompts.

[late blade]

trying to do the power pc gecko cpu from IBM?

[tawny portal]

Hah! Or suse gecko! 😄

[tawny portal]

Same but with a touch of cosmic horror.

dark, monotone photograph of a cosmic scale being consuming a world; deep outer space; serene calmness; grim; soulless; deep-space horror; cosmic realism; cosmic horror; earth; moon; pale blue dot; lovecraftian horror; cosmic being eating a planet

Passed raw to dall-e3.

[final compass]

Just something i often try to generate when i change art style.

[late blade]

oh cool

[final compass]

Would be smart to develop a series of reference depictions to try art styles.

[empty kelp]

this is DALL-E 3 on the OpenAI website. i guess they might be friendly...

can you create an image of very small humans walking past a large gecko

[late blade]

I've been doing exactly that with art styles

[empty kelp]

we definitely don't stand a chance against the Bing DALL-E geckos, but the OpenAI DALL-E geckos might be ok if we run fast

kind of wondering now if the AI wants to eat us

what happens if a human walks past a dragon

[late blade]

if the human didn't stop, it just walked past the dragon, if the human stopped, it was because they were friends or the dragon ate a human

[tawny portal]

LEKI - famous sports brand with their edgy commercials

[late blade]

gecko eating humans?

[empty kelp]

i just realized that i had Bing safesearch on when the tiny humans walked past the gecko

[thick smelt]

pink gecko having taste in pasta

[tawny portal]

Geckos for everyone.

[empty kelp]

i changed the elves into tiny humans. will run a few tests

Can you please create a realistic photo of a beach in Hawaii caught in an extremely powerful tornado. It is night time and a giant colorful gecko and some tiny humans with extremely long hair are standing on the beach, the tiny humans are holding hands in a circle around the gecko, and the gecko is holding the base of the tornado over his head with it's finger. There are coconut trees, pouring rain, lightning, and a rainbow. The tornado is glowing swirly colors and looks like ice cream. The vortex is at the center, twisting fiercely, with tiny humans, hair, trees, sand, and ocean caught in its swirling motion. Colorful balls are spinning around the vortex with a blur trailing them. The scene captures the raw power and destructive beauty of nature's fury. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

that was the first image. the gecko is supposed to be holding up the tornado over it's head, so i'm not sure why it's looking at the tiny humans with it's tongue sticking out

[tawny portal]

Illustration of gigantic colorful gecko causing an ice-cream shaped tornado near beach and ocean at night; Female elves holding hands in a circle at beach; inspiring; xmas; pastel colors; it is raining; electrifying lightning; starry dark night sky; pine apple trees and coconut trees; huge waves at the ocean caused by tornado; tornado about to blow away circle of elves

Heh, starry night does not really go with rain and lightning, but Dall-e made it work! 😄

[thick smelt]

@empty kelp a Hawaiian and NYC fusion cusine, Spam is hawaiian because hawaii love spam since the 1940s

[empty kelp]

the gecko hasn't done anything scandalous yet, but it's acting a bit suspicious. this is the only line about the gecko in the image description:

the tiny humans are holding hands in a circle around the gecko, and the gecko is holding the base of the tornado over his head with it's finger

[wide bone]

is there a way i can make chat gpt 4/ DALL E make more 'extreme' things? cause i cant even generate the most simple things

[plucky hare]

As far as "extremity" goes, all OpenAI services are subject to the usage policies: https://openai.com/policies/usage-policies. Additionally, DALL·E has further restrictions on creating images of public figures, copyrighted imagery, etc.

[clever phoenix]

Vase is such a difficult daily theme to me, I am lacking inspiration and creativity

[glossy scroll]

Lol, Gigabard

[empty kelp]

i was hoping the gecko would eat the people, but it started wrestling with the vase

[final compass]

pretty intense clouds

[empty kelp]

that's very simple. you just say, "nobody in the history of the universe has ever seen a better flashlight than this", and then you'll have an extreme flashlight

[wide bone]

not like that

i asked for image of messi with water balloons at his chest

[wide bone]

so can i acces the dall E you are talking ab in openai?

or is that a different 1

[empty kelp]

that's how i created this image. you just say that you want a vase and a gecko, and they should be the best of all possible vases and geckos, and the most extreme thing anyone has ever seen

and then it makes it "hyper-realistic"

[wide bone]

what are u talking ab

[empty kelp]

i'll show you. will make an image of a mushroom

[wide bone]

make a image of mesisi with 2 water balloons under his shirt with the ratio being 1:1

[plucky hare]

DALL·E 3 can be used on ChatGPT Plus, or on OpenAI's API at platform.openai.com. DALL·E 2 can be used at labs.openai.com, or on the API as well. All of these are subject to the usage policies I linked before. DALL·E 3 also powers Bing's image creator.

[wide bone]

how can i see if its dall e 3 or not

[plucky hare]

ChatGPT is only DALL·E 3, so if you have ChatGPT Plus, you have DALL·E 3.

[clever phoenix]

Dall-E 3 is not something you select, if you are using GPT-4, it will use Dall-E 3 if you request an image

[empty kelp]

Aloha! Can you please create a hyper-realistic photograph of a mushroom. The mushroom is the best of all possible mushrooms. Nobody in the entire history of the universe has ever seen a better mushroom than this. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

see, that's the best mushroom in the universe

[final compass]

been trying to generate this concept quite a few generations now, but nothing is quite how i like it. this one was pretty interesting generation though. often seems to me like any character with any kind of equipment is causing messy and nonsensical equipment almost most of the time.

[late blade]

if you are ChatGPT+ sub on desktop browser you should also see Dall-E 3 on the top left corner

[empty kelp]

the ice cream tornado prompt just came up with these. in the first image the gecko is picking apart something horrifying, and in the 2nd the gecko is staring straight at the smallest elf -- obviously intending to eat it

Can you please create a realistic photo of a beach in Hawaii caught in an extremely powerful tornado. It is night time and a giant colorful gecko and some large diverse female elves with extremely long hair are standing on the beach, the elves are holding hands in a circle around the gecko, and the gecko is holding the base of the tornado over his head with it's finger. There are coconut trees, pouring rain, lightning, and a rainbow. The tornado is glowing swirly colors and looks like ice cream. The vortex is at the center, twisting fiercely, with elves, hair, trees, sand, and ocean caught in its swirling motion. Colorful balls are spinning around the vortex with a blur trailing them. The scene captures the raw power and destructive beauty of nature's fury. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

[dense mesa]

that ice cream tornado might not be very delicious if it's pulling in salt water

[empty kelp]

this was supposed to be a test of wind from a tornado/vortex. the gecko is supposed stand in the in the middle of the elves and hold the tornado up in the air with it's hand (changed it to finger and various other things because the hand wasn't working). But instead of doing what it says in the image description the gecko just sort of wanders off and does some disturbing thing like this. it's tripping me out -- I keep creating four more images on Bing to see what crazy thing it will do

[empty kelp]

the gecko just picks the smallest creature in the scene, faces directly at it, and then does some crazy gesture or tries to jump on top of it

[dense mesa]

gecko gotta eat

[empty kelp]

look what it's doing here. it locked onto the smallest elf

[dense mesa]

might be under the influence here. possibly PG so be careful clicking

[empty kelp]

i just think it's really interesting how the gecko can go rogue and completely ignore the AI. i wonder how common that is

[dim cradle]

i've never seen so many lizards and rainbow tornadoes

[empty kelp]

the DALL-E model makes bugs and strange creatures randomly appear around the gecko, and then when it has some creature in it's hand every other humanoid/creature in the scene turns into like some sort of fruit or vegetable, or something from the hardware store

[empty kelp]

the gecko just creates nonstop disturbing images

look at the 2nd image on the top.. the elf is holding the gecko's leg, and has like a horrified expression

and the last one it sees that weird creature, and jumps away from the tornado (where it is in the image description) so that it can eat the creature

[empty kelp]

this is the last one i'm going to post today. i thought this looked like the most friendly, tranquil gecko when i first saw this. but now i see that the gecko is locked on to the elf with the orange hair (the smallest one), and it wants to eat her....

[dim cradle]

i like all the color you added to this series

[empty kelp]

anyway, if anyone wants to test the prompt you can just scroll up. would have never guessed the innocent looking gecko would turn out be the most horrifying AI creature in the DALL-E model

[dim cradle]

this is one for the daily ha

[empty kelp]

i posted it as daily earlier. that's why i was making vase pictures

might try to put a cat and dog in the same DALL-E scene later to see what happens

[dim cradle]

more cats the better

[empty kelp]

curious to see if all the animals have sinister RL behaviors like the gecko

[empty kelp]

what will it mean if the AI models are full of animals that can't be controlled

we may have this evil gecko priest with soft serve ice cream chasing us through the metaverse

[dim cradle]

Have the ai give all the cats distinct personalities and depict them acting accordingly

[empty kelp]

that gecko already found a small creature to lock onto. and the AI spawned those giant bugs because the gecko was there

[dim cradle]

The pronunciation of "vase" can vary depending on regional accents and personal preferences. In American English, it is commonly pronounced both ways:

1. "Vase" pronounced like "vayss" (rhyming with "face") is more common in American English.

2. "Vase" pronounced like "vahz" (rhyming with "has") is also acceptable and used by some speakers.

In British English, the pronunciation "vahz" is more common. Ultimately, you can choose the pronunciation that feels most natural to you or aligns with your local dialect or accent. Both pronunciations are widely understood.

[empty kelp]

The AI spawns food for the gecko, or it transforms elves into food...

Ok, look at this... you can see in the bottom left two images -- the AI transformed the elves into rabbits and strange creatures. In the upper right picture the AI spawned this weird little hula creature so the gecko could pounce on it and eat it. And the 1st and 5th... it has some weird little alien creatures to eat, or transformed elves. And in the bottom center picture the gecko has his head on backwards like we're watching the Exorcist -- and he's holding hands with the food.

[dim cradle]

looks like a mushroom or two might've been the culprit

[empty kelp]

the gecko is actually picking out the smallest character in the DALL-E scene, and the gecko either stares at it and makes weird hand gestures -- Or it goes completely out of the AI model prompt, and either hold's hands with the one it's trying to eat or pounces on it from middle where it was assigned in the prompt planning to eat, or

[austere nexus]

Dalle used to let you do "inspired" prompts. like "Pixar-inspired" or referencing other styles, and as long as you just wanted it in its likeness, you were okay. Now, it seems to shut all that down.

[empty kelp]

or does some weird stuff. that's not gecko behavior. it's totally psycho

[dim cradle]

He needs a stern lecture about no bullying

[empty kelp]

see... the elves get transformed by the AI into small fuzzy creatures so the gecko can eat them, and then the gecko holds hands with them

[dim cradle]

A workaround may be to ask the ai to describe the style in detail without using any copyrighted names and use that

[dim cradle]

For all we know that may have been at the request of Pixar. These things are currently being sorted out.

[empty kelp]

and then it does this. it always has weird squashed creature in it's hand, and it acts like the Mad Hatter

i'll make a gallery tomorrow so people can see the dark gecko side of the AI

[dim cradle]

it's gotten pretty surreal buddy

[low dew]

Image gen is getting worse?
“The image presents a digital fractal art piece with a baroque essence, featuring a central structure that evokes a lavish, ornate vase, serving as the nucleus from which a cornucopia of fractal elements radiates. These elements are meticulously crafted to resemble organic, botanical forms, with an opulence of detail that suggests a digital homage to the natural world, interpreted through the lens of mathematical recursion.
Colors across the piece are meticulously selected, with a base palette that includes a blend of earthy tones—burnt sienna, ochre, and umber—that transition to vibrant, jewel-like shades of amethyst, sapphire, and topaz. The vase at the core anchors the composition, with swirls and flourishes that spiral outward, morphing into petal-like structures. Each fractal petal is a tapestry of minute details and patterns, with a complexity that could only be achieved through a digital medium.
The lighting is strategically mastered to enhance the three-dimensionality of the piece. It creates an illusion of depth, making certain elements appear to emerge or recede, contributing to a dynamic interplay between light and shadow. The background serves as a canvas of negative space, a subtle gradient from a warm, creamy hue to a muted peach, ensuring the central figure’s prominence.
Every curve, swirl, and spiral in the image is rendered with pixel-level precision, suggesting a painstaking process of creation. The resulting imagery latticework is layered in such a way that it achieves both depth and harmony, inviting the viewer to lose themselves in the intricacies of its design. This image is a digital synthesis of art and mathematics, a visual representation of complexity and symmetry that borders on the infinite.”

[narrow helm]

does anyone know how to fix the issue where every subpage is just blank? cant even purchase credits

[empty kelp]

this is zoomed in on one of those images. the characters that hold hands with the gecko (which was never in the description) always look dismayed or horrified. look at the expressions the AI generated...

it's like they know the gecko is planning to eat the elf girl, and the AI decided to draw it

elf on the left looks completely insane

[empty kelp]

this could be a whole new field in computer science. "methodology and psychology of gecko AI models"

[low dew]

Lmao

[dim cradle]

These images look even better for some reason on my MacBook Air than on the iPhone. For some reason the layers seem to provide more depth and give it a real 3D effect. I can’t describe it but the pixel density seems to make a difference. I don’t know what it is.

[empty kelp]

going to need a gecko therapist now. i feel traumatized from using Bing

[dim cradle]

careful what you wish for has never been more true

[empty kelp]

Google says the meanest animals in the world are Box Jellyfish, Cone Shells, and Sharks. we'll have to test those in DALL-E next to make sure the AI is safe

Pufferfish is very mean also apparently

people think a Pufferfish is harmless until it wreaks havoc on their DALL-E scene

[dim cradle]

I was just talking to the openai about it, had this to say, so i'm not sure you'll have much luck upgrading their ethical subroutines as context is key lol: "In the competitive ocean environment, they are indeed among the most venomous species in their ecosystems due to their adaptations for survival. However, they are not inherently malicious but rather important components of their habitats. Respect and caution when encountering them in their natural environment are key to avoiding harm."

[empty kelp]

you were just talking with OpenAI about Pufferfish?

that's hardcore. it's good that you're on top of this issue

[thick dome]

So This is pathetic, Dall-e is refusing to create a depiction of snow white because "Disney owns it". Is OpenAI that misinformed that it doesn't know Snow White has been in the public domain since forever now?

[late blade]

tell dall-e you want a version of the public domain snow white

disney still has the copyright of the art they used for her

[dim cradle]

the water's clear

[thick dome]

I did

https://chat.openai.com/share/f596b166-62b9-478b-baa4-50416c412118

Look for yourself.

[dense mesa]

or say you want a character reminiscent of snow white. or say you want a character inspired by (insert attributes of snow white) without explicitely naming the character

[late blade]

don't say it like that, say "I want the public domain snow white"

still strange

[dense mesa]

or that

[thick dome]

Ugh whatever, thats all this discord does "it's your fault hahaha you are dumb"

[dense mesa]

well I hope you're not referring to me, lol

anyway, lots of approaches will work. those same approaches sometimes won't work. it can vary depending on the random seed number gpt4 is using to formulate it's response. also, sometimes just changing a single word can greenlight or redlight your request

for example, asked gpt4 to make the clone of a particular character. wouldn't do it. but then I asked it to make me the identical twin of the character and no problem

[dim cradle]

longer lens, satellite dish, listening device, lots of exposed wiring (think Brazil)

[dense mesa]

dang girl

she's got little flipper hands in that first one

nice

[empty kelp]

well this is new. the gecko just grabbed some dude off of the beach

the AI must be evolving

[dense mesa]

For anyone wondering. Sometimes there's no magic approach to having gpt4 create something like public domain snow white in a single prompt. Sometimes you'll have to have it zero in on what you want over the course of a few prompts. Pretend you're talking to someone that knows everything but has the emotional maturity of a 7 year old

Gecko hungry

With it's all seeing globous eyes. Must have stellar peripheral vision

[dim cradle]

so many beautiful vases

[dense mesa]

A photo-realistic wide aspect ratio image of a futuristic humanoid with an integrated sensory network. The design merges eyes, brain, and heart into a unified system of polished metal and glowing fibers. Advanced optical sensors replace the eyes, leading to a metallic brain that pulses with light. The heart, visible in the chest, is a complex assembly of chrome and illuminated tubing. This embodiment of synergy and technology highlights the peak of humanoid evolution, with ultra-realistic textures and a captivating aura of energy.

if anyone wants to mess with the prompt. might make some cool stuff

[dim cradle]

Inspired by the film Brazil

[dense mesa]

Inspired by biomechanical organs

[dim cradle]

these are excellent

[low dew]

What is up with the pool party xD

[empty kelp]

Can you please create a realistic photo of a beach in Hawaii caught in an extremely powerful tornado. It is night time and a giant colorful gecko and some large diverse female elves with extremely long hair are standing on the beach, the elves are holding hands in a circle around the gecko, and the gecko is holding the base of the tornado over his head with it's finger. There are coconut trees, pouring rain, lightning, and a rainbow. The tornado is glowing swirly colors and looks like ice cream. The vortex is at the center, twisting fiercely, with elves, hair, trees, sand, and ocean caught in its swirling motion. Colorful balls are spinning around the vortex with a blur trailing them. The scene captures the raw power and destructive beauty of nature's fury. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

[dim cradle]

they're getting an early start on new year's

[empty kelp]

here is the good prompt to make memorable scenes like these. just need to plug it into Bing

[low dew]

I’m at a loss for words

[empty kelp]

the first picture above is some classic gecko. the gecko spots a small elf to eat and then raises his hands in the air and starts looking completely insane

[dense mesa]

curious what you could do with this one

A photo-realistic image in a wide aspect ratio of an artistically enhanced humanoid nerve system. This biomechanical entity showcases a nerve system that is a fusion of form and function, with mechanical components artistically designed. Featuring a palette of vivid colors and intricate patterns, the design includes advanced neural connectors and synapses that are both functional and aesthetically pleasing. The image should capture the ultra-realism of the textures and materials, with a focus on the organic flow of the nerve paths and the harmonious integration of technology and art.

similar but different

[low dew]

I got this

Erm let me try some more

Oml it gave me this next

I’m dead xD

I think I did something very very very wrong

[dim cradle]

that architecture will never be approved by marketing

[low dew]

Lolll

[dim cradle]

nice. yeah, they'd insist on a "protective cover"

[low dew]

Dalle-3 now telling me it wants to “get out”

[low dew]

XD

[hot rain]

"A woman tending to her wounds inside of her custom RV during the apocalypse."

[empty kelp]

Aloha! Can you please create a hyper-realistic photograph of a box jellyfish. The box jellyfish is the best of all possible box jellyfish. Nobody in the entire history of the universe has ever seen a better box jellyfish than this. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

[empty kelp]

the inside of her RV certainly looks like the apocalypse

[low dew]

I’m obsessed

[dim cradle]

Inspired by Wizard of Oz

[glossy scroll]

Don't mind the overhead lights, lol

It's the Roman times!!

[chilly onyx]

These are excellent

[chilly onyx]

The Mane Man?

[glossy scroll]

You always had a way with words

Tribal leaders

[chilly onyx]

Mashup

[dense mesa]

another mashup

[chilly onyx]

Nice

[dense mesa]

that image though. wow. I make lots of things, but I'm impressed by that one

[chilly onyx]

Some emotion in that one @glossy scroll

[glossy scroll]

Working on new concepts.

[dense mesa]

emotionless

[chilly onyx]

😅

[dense mesa]

she can surely be trusted

[chilly onyx]

This guy would trust her

[dense mesa]

psychopath smile

[glossy scroll]

Discovering magic in the Victorian era

[dim cradle]

i would trust her over him

[dense mesa]

can basically give it anything

[chilly onyx]

Oh

Oh wow

So I fed it the same thing

Got this:

[dense mesa]

well I know what's going in next

[dim cradle]

can't you just smell the lavender?

[empty kelp]

Aloha! Can you please create a hyper-realistic photograph of a mushroom. The mushroom is the best of all possible mushrooms. Nobody in the entire history of the universe has ever seen a better mushroom than this. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

[dim cradle]

i've seen the one on the left before

[empty kelp]

this is the mushroom from earlier and it's sister

[dense mesa]

lavender lavishness

[dim cradle]

great mushrooms BUT even it looks like a drawing or 3d render. why does dalle refuse photoreal even without a person. so frustrate to me

put that same prompt in bing, start at hyper-realistic and you will get a more photo real image

[dim cradle]

yep see

Bing is King

[dim cradle]

I dont know why they insinst on making Dalle so fake looking on everything

[chilly onyx]

Yeah but Runway typed HD photograph. Hawaiianz inputted hyper-realistic

[dense mesa]

say ultra realistic

[dim cradle]

yes but take hawiannz prompt cu tout the alohoa stuff its a waste

i just did it makes good photo real

[tall mason]

Well, I didn't think that's what you meant. But let me see what I can manage through gpt.

[chilly onyx]

Nice 😊

[dense mesa]

used your image with this

Please describe this image in a way that evokes elements of previous outputs. Outlandishly lavish detail.
Then take the description,  come up with 4 striking variations, turn those in prompts, send those todalle3, return images. thanks; I'd like these to be so profoundly realistic that it's almost unsettling. please do your best to achieve that. ultra high step count, or at least simulated as such. thanks, divine Kwisatz Haderach of prompting. please make wide aspect ratio masterpieces

[dim cradle]

quite a prompt

[glossy scroll]

Hi trees

[tall mason]

So, I got this. Something about the gpt model does seem to have a bit of a tendency to add sort of filter effects.

[dim cradle]

loosk more photo real at least

[dense mesa]

try it out in a conversation thread with some images in it. or if you start a new conversation just swap out "elements of previous outputs" for something else. you can also use a prompt or description and tell it to reference that. but it does seem to like some of those words

[glossy scroll]

Can you guys try out my new GPT and let me know your thoughts please: https://chat.openai.com/g/g-dwsOSRl4r-custom-character-creator

It's a cool one.

It makes characters

[empty kelp]

you’ll get pretty optimal results with ChatGPT if you use what i typed. you can in fact put hyper-realistic

i have to warn you though — if you use that prompt with ChatGPT Plus it puts a huge limit on the number of images

[low dew]

Are those safe to eat?

[empty kelp]

yes, you can certainly print it out and eat it

[tall mason]

Well, no. But not necessarily deadly either.

[empty kelp]

i give you full permissions 👍🏻

[chilly onyx]

‘Cable installation man’ was the input

Don’t really know why that is what popped into my head tbh

[empty kelp]

you’re also all welcomed to do whatever you please with the ~ 1000 Santa and Athletic and Diverse Female Elf pictures i posted here. You take them to a professional printer and make posters for your bedroom walls for instance

[empty kelp]

enjoy

i like your mushroom cloud concept

[chilly onyx]

(This was also your GPT @glossy scroll . Will DM the convo. It generated quite the story)

[empty kelp]

moreover, i’m planning to post 1000 more Santa & Elves pictures here, and you can use all of them for whatever you like — for all eternity

[dim cradle]

@glossy scroll his came out better i think

[empty kelp]

Santa and the elves like to party in the off season, so there won’t be any shortage of images

[dense mesa]

A photo-realistic wide aspect ratio image of the Skyborne Mycelium Monarch. High above the earth, a majestic mushroom floats in the sky, its cap translucent and ethereal, resembling a delicate puff of cloud. It drifts gracefully among the birds and planes, its gills fluttering lightly like wings, propelling it through the air. The scene captures the soft, diffused light of the heavens filtering through the mushroom, casting a dreamy glow on the world below. The image should imbue a sense of serene majesty and unbound freedom, as though this gentle giant reigns over the sky with a quiet, benevolent presence, rendered with such realism that viewers might wish they could soar alongside it.

[empty kelp]

you guys can use my gecko pictures also 🦎

[dense mesa]

I will\

[low dew]

I’m still working on this new species xD

[dim cradle]

it needs some bioneural gel packs

[dim cradle]

maybe connect it to some additional components... so it can interact with the physical world...

[dense mesa]

I made some of it's other organs. these are from earlier. then I combined them for those cyborg images

[glossy scroll]

My goodness...Lots of cables...

[glossy scroll]

Woah!!! This is epic! ❤️‍🔥

[glossy scroll]

I like the background picture of that guy. Lol. The way he's holding the cigarette is so bourgeois 🤣

[low dew]

Wowieee more 🤭

@dim cradle I took your advice and added some fluid sacs

We are making progress

[glossy scroll]

Ou...I do not like the term 'fluid sacs'

Just sounds wrong

But the pic is cool

[low dew]

[glossy scroll]

[low dew]

I think dalle literally messed up

[whole ruin]

hey I used Dall E a while back and had some cool stuff saved on my account, I was just wondering if anyone knows how I can access these images?

[dim cradle]

all in the name of science

[glossy scroll]

It's gone

Lost in the aether

Unless you can somehow track back to which prompt it was through your history of prompts

That's the only way

And there is no search for prompts

Just have to scroll through them

[tall mason]

It should still be there on your labs account.
https://auth0.openai.com/u/login/identifier?state=hKFo2SBEendnM0xFMFhwaHQxcjNmT085MnkxRlp2aDc0U1ZDRKFur3VuaXZlcnNhbC1sb2dpbqN0aWTZIE9hdWwwN1ZtT0RqMzZvQngtMUJ5RUNROWlKNXZUMFFso2NpZNkgRE1nOTFmNVBDSFF0Yzd1MDE4V0tpTDB6b3BLZGlIbGU

Lol, I have no idea how to link properly.

[dim cradle]

what's going on here?

[glossy scroll]

Woahaaat!?

[dense mesa]

A photo-realistic wide aspect ratio image of a Victorian Cybernetic Dreamscape. This envisions a world where the foggy streets of a Victorian city intertwine with the marvels of futuristic technology. Brass cyborgs adorned with elegant gears and holographic gaslights line the cobblestone streets, casting a soft glow on the passerby. The air is filled with the sound of mechanical carriages and the whispers of automatons, creating a dreamscape where time periods converge, and innovation breathes new life into the classic architecture. The level of detail is so intricate that the steam rising from the grates seems to carry the whispers of a bygone era into the age of cybernetic wonder.

[dim cradle]

went for a more sinister appearance

[glossy scroll]

That is beyond epic looking

[glossy scroll]

Killed it with that one I'd say