images-discussions

[dim cradle]

oh, it repeated the contraption so i wasn't sure

[glossy scroll]

What's that guy's name, does it say? @dim cradle

[dim cradle]

Aelar the Enchanter

[glossy scroll]

Eyyyy, that's awesome, haha, love that it makes a whole profile.

Perfect for RPG character creations

I gotta tweak it a bit though so that it doesn't always use that chad model

Gotta diversify with more character models

And I'm really starting to see how OpenAI is using their database

[tall mason]

@glossy scroll

"High amplitude photo-synchronization, best quality, highly professional, unwavering depth in alignment, dramatic lighting and excellent perspective that adds allure, fierce and immaculate details with high contrast and precisely impactful hand drawing for sharp/crisp and deep focus, epitome of awesome, no less than badass, 16 by 9 aspect ratio: Create a resonantly balanced !!!2D!! image in a 'Classic Animation Reinvented' art optimization style: A rubber hose mouse warrior, characterized by the fluid, exaggerated animation style reminiscent of early 20th-century cartoons, wielding a needle as a sword. The mouse is adorned in a hooded green cloak, adding a mystical aura. The scene captures the mouse in a dynamic pose, showcasing its warrior spirit and readiness for adventure, with a background that complements the character's theme and style, emphasizing a whimsical, adventurous atmosphere."

[low dew]

I don’t wanna get in trouble

[dim cradle]

it's cool, but i think a spoiler is preferred for anything potentially unsettling

[glossy scroll]

Oh that's interesting

I never tried it on BIC so that's pretty cool how it generates similar looking images compared to ChatGPT.

[dim cradle]

looks like some monsters i've made lol

[tall mason]

Totally whiffed the rubber hose style, though. Both seem to have trouble emulating cartoons. Strong bias towards digital art, I guess.

[glossy scroll]

It's specificity is designed to go for a more 2.5D art approach, but you know what, I think I'll tweak that to be more universal.

I was using a a previous prompt from a previous GPT, but I will tweak it for this one.

That section I mean.

@tall mason try it now

[glossy scroll]

I haven't tested it, but I updated it.

[dim cradle]

kinda has a Nordic look?

[glossy scroll]

It does. It's quite nordy

[dim cradle]

your gpt is making some photorealistic fictional characters, which is a challenge--i like it, more lotr, fantastical but lifelike

[tall mason]

Will I need a new link?

[glossy scroll]

I wouldn't think so

Because the link always leads to the same GPT

[tall mason]

Ooh, this is pretty good.

[thick dome]

concept art of Ancalagon

[glossy scroll]

That's a bit more closer to rubber hose isn't it?

[tall mason]

My personal favorite I made awhile back of this little guy.

[glossy scroll]

I love that! 🧡

[iron salmon]

Are there good third party apps to use? I got a message today that depicting the fight/flight response was against policy and ethics

[tall mason]

It's hard to get the style I want without mentioning the artists, sadly. Thay's why I aimed for rubber hose. It's not quite what I want, but it's a specific term for a type of cartoon. Both bing and gpt tend to default to digital art or anime.

[open trench]

That's awesome

[glossy scroll]

Hmmm...I will tweak it again to try and make it more tuned to artists without you mentioning their name.

[dim cradle]

looks like the fluids are under control now, it's not melting or trying to drink itself anymore

[dim cradle]

ty, sir 🙂

[open trench]

Hello all

[clever phoenix]

Hail and well met

[open trench]

🙂

[tall mason]

Goodluck, I sure would love it. I struggle...

[dim cradle]

gonna make some vases?

[grizzled loom]

Been having fun with scenes

[glossy scroll]

I updated. Try again

[grizzled loom]

Glass creatures are fun as well. This dragon really came out nicely.

[grizzled loom]

This wasfun as well.

[tall mason]

Decent image, but still not quite.

[shy elbow]

Anyone onehere play with comic-style sequence creating in Dalle 3 yet though gpt that uses ref images/prompts?

[clever phoenix]

You're trying to go for the disney Don Bluth style right?

[clever phoenix]

I've tried a little bit, but getting consistency is difficult

[tall mason]

Sort of, when I came up with the prompt, I was aiming for Don Bluth meets Ralph Bakshi's LOTR.

[clever phoenix]

I see, that does add an element of difficulty

[open trench]

I may very well make some vases. I haven't felt inspiration quite yet, but fingers crossed.

[glossy scroll]

Try again one more time. I updated into a solid one I think.

[shy elbow]

@clever phoenix I also don't mind doing one and editing, though funny that the AI knows what it is when I give it a comic-style pannel, if I could get the thing to do photo realistic would be nice. Gave it a sample and used the same words, unless the comic-style is forcing it to ignore it.

[clever phoenix]

nah, Dall-E 3 seems designed to not do photo realistic by default though seeing some of the stuff others have posted here there is some good effort to refining the style to be more photo real

[dim cradle]

try Bing you get better photoreal with the same prompt many time

[shy elbow]

Interesting, as looking online people saying it can. Can could, but doubt it like me trying a werewolf photorealistic sequence

[tall mason]

Cute and all, but not quite. I should note, my images were from the exp model.

Anywho, I'm out. Goodnight, folks.

[shy elbow]

Ya, the ax guy above, if I can get least to that I be happy

[dense mesa]

Reminds me of a couple I made a while back

[clever phoenix]

I'm able to get what I personally see is a decent don bluth style, but adding the LOTR does seem to make it more challenging, anyways g'night

[glossy scroll]

Nighty trees

[shy elbow]

@glossy scroll what you use to get thar realism for look?

[glossy scroll]

I use this GPT I made broski: https://chat.openai.com/g/g-dwsOSRl4r-custom-character-creator

Try it out. Let me know what you think.

[dense mesa]

Full Feline Alchemist

[grizzled loom]

koala cuddle party?

[grizzled loom]

this was fun

[dense mesa]

those cats definitely feeling the vibe

[glossy scroll]

Dr. Barkley Houndstein

[shy elbow]

@glossy scroll Yup, will not work, run out of space way too quickly on Bing. Like the description of the guy is five paragraphs as as I have a particular look for him

[glossy scroll]

Whhaa? What will not work?

[shy elbow]

Also got to wait until basically 2 my time to try again

[dense mesa]

A photo-realistic wide aspect ratio image of Techno-Organic Growth Rings, depicting the evolutionary timeline of artificial intelligence. Each concentric ring within the cross-section of a colossal tree represents an AI generation, from the outer nascent rings of early computing to the dense, complex core symbolizing GPT-3. The rings are crafted from a fusion of natural wood and embedded digital circuits, glowing faintly with the flow of information. The detailed textures of bark and silicon create a harmonious blend of the organic and the technological, symbolizing the growth and potential of AI in a form that mirrors the natural world.

[shy elbow]

@glossy scroll bing, the prompt lengh you can put in is not designed to detiled figure discription in mind. So have to wait until allowed to use chatgpt 4 again.

[glossy scroll]

Uhhh...ok...

Wish you luck

[low dew]

Oooo

Growth rings

[shy elbow]

@glossy scroll well if I could do what I wanted on chatgpt, then my long detial not needed, but what can you do 'shrugs'

[glossy scroll]

I feel ya

[grizzled loom]

silly can be nice as well "image of: bleep bloop blop ya dop HD widescreen flop yoppy pop top meeble flogen flantreem"

[grizzled loom]

hippos are fun

[dim cradle]

@dense mesa and i were discussing a prompt that maximizes the optical illusion of live action, depth, etc. This one he made gets very close and is an example but i'm wondering if the model can take it to the next level

[low dew]

Oo

Challenge accepted

[glossy scroll]

Try it with my GPT and see what you get. It will probably work.

[glossy scroll]

My new one I mean.

@dim cradle No nevermind, it's terrible at optical illusions, lol

[tawny comet]

rat of fire

[frosty marten]

Hello

[shy elbow]

@glossy scroll Your gpt does not like my discription, may need to modify it a bit

[low dew]

Aaaaand we’re back

[dim cradle]

yeah, that's got a lot going for it, i was thinking of using every trick in the book -- layers of scenery to add more depth, motion streaks, infinite vector look, capturing a frame of high velocity without over-blurring -- not a hyperspace, but an ai-augmented live-action still

[dim cradle]

not sure i'd trust that ai

[grizzled loom]

these are Fleebleflops; apparently they exist inside jumbled language like an alternate dimension

[dim cradle]

This is fine.

[grizzled loom]

they're fun

[grizzled loom]

throws some visual style guides still, but just make up gibberish as well for some really wild things.

[dim cradle]

yeah, especially for a surreal work, formulate an illogical combination of terms

[grizzled loom]

fleebleflops get fun.

[low dew]

What is this?! My Singing Monsters Reworked?

[grizzled loom]

nah. i only use my own work. nothing from others

[low dew]

Look it up it’s a thing

From like 2012

[dim cradle]

this is very good. now, if the subject were running towards the viewer, it could maximize the effect. multiple blocks in the scenery could be rendered at varying levels of focus to perhaps add more to the effect.

[dim cradle]

lol this reminds me of something i was talking to the ai about earlier today:

The number of possible chess game combinations is astronomically large, essentially infinite due to the vast number of positions and moves possible. In contrast, DALL·E's image generation capabilities, while expansive, are finite, determined by its training data and architecture, which means it can create an extremely large but ultimately limited number of unique images.

[grizzled loom]

Been a digital artist and engineering designer for around 25 on the books paid and another 10 before that back to the start of art in computing. Still have saftware from the 90's I couldnt let go of

[open trench]

I feel like I can notice this limitation when I "max out" an image capabilities. It starts removing elements to "make room" for new ones. I especially notice it when I am doing parody characters and need specific costumes.

[grizzled loom]

Cant think of anything in my lifetime I havent given study to with art. i like my own styling just as much as AI outcomes

[dim cradle]

and have you noticed it orders based on aesthetics and not logic? this is pretty apparent in time-lapse images.

[open trench]

I haven't. I am not entirely sure I follow.

[dense mesa]

a lot of times, for me at least, it doesn't so much seem like it forgets about things exactly, it just has a hard time keeping different concepts autonomous from one another. and then yeah, does sometimes conspicuously leave things out.

it's a lot better at it than previous models I've used. but if I try to give it more than about 3 distinct characters or concepts they almost always end up merging in some way

[dim cradle]

i'm thinking, as an example, a time-lapse of the lifespan of a candle -- it'll depict varying phases, but not in the correct order.

[dense mesa]

it's like it understands the steps, but then has a hard time putting them all in order due to the nature of diffusion models. I'm sure that'll be rectified fairly soon though

same reason it can't spell and almost never gets a clock quite right

and by understand the steps I obviously don't mean the concept, but rather just what they look like

[dim cradle]

it's gotten a lot better with clocks.. used to be much worse..

[dense mesa]

well look how older models spell things

yeah, it's much much better at all that stuff

[open trench]

Oh! I see what you're saying.

[thick smelt]

Happy New Year my dudes!

[dim cradle]

I missed something. Maybe generate it I can’t use the api atm

may 2024 bring us even more AI excitements with images, sound, visions, and infos 🙏

[west vine]

@OpenAI Verification#5580 @OpenAI Verification#5580

[dim cradle]

did you describe yourself or something? i experimented with that and thought it was doing a pretty good job of sketching. that makes me think ai art would be a perfect use case for police sketches of suspects.

@open trench it's like the scene reimagined with ai. would love to see the whole movie ha

[open trench]

seriously. it would be a trip.

[dim cradle]

awesome

[open trench]

2 more characters to go on my parody project. I'm ready to be done. lol

[dim cradle]

For a gallery?

[dim cradle]

i think so too

let's view the queen of hearts next

[open trench]

Yeah. but I use Bing Image Creator in my workflow, and my current character trips its content warning filter like crazy.

couldn't tell you why

Ever iteration I have tried has been giving me the red message, which I guess is worse than the crying dog yellow message.

every*

[pulsar sundial]

not myself, just described the guy

well, yes, he does a good job with sketches

[dim cradle]

this is not a bad reimagining of the queen of hearts... i'd make lots of changes to the guards and scene, but the queen is not bad at all

[pulsar sundial]

I'm 15 years old, and I described a 20 year old guy

[dim cradle]

i'd do the hookah-smoking caterpillar scene, but penalty box

somebody else could pick it up lol

[dim cradle]

the eyes lol

the "i've never done anything wrong in my life" look

[low dew]

Hey shon, seriously

Check this out

[dim cradle]

looks like it has potential?

[low dew]

Hrmm

[open trench]

penalty box. yikes.

I'm used to it now, but even when I think I am pacing myself to avoid the box, I end up surprised.

bing time, i suppose

[dim cradle]

Right, and I work on new prompts to run when I can get back in the game.

There are so many works in the public domain that can be reimagined.

[open trench]

Oh cool

[open trench]

One character to go. Please no penalty box soon. lol

[late blade]

oh look, @open trench is in da house

haven't seen you in a while

[open trench]

Yeah. I vanished. Had some things to cope with.

[late blade]

I hope it was coping with all the food from xmas

[open trench]

haha. I wish that had been the case. Maybe next year. 😛

[late blade]

good, next year, more food on the table, put it on your list

[open trench]

it's there

right under 'fear the penalty box'

[late blade]

good, now all you have to do, and THIS is very important: "The food is for Dys Topia!!"

[open trench]

OH. Haha. I forgot how to write. Maybe the year after next.

[late blade]

Evil Disney Castle?

[final compass]

Pretty much looks that way yeah 😂

[late blade]

managed to do #daily-theme message a SpongeBob SquarePants vase on dall-e and no content policy warning

[final compass]

At what time does the daily theme switch?

[late blade]

in 7:30ish hours

would be nice to have a function for multiple time zone times in chat, other disocrds I've seen have that option, so you don't have to think much about converting time for others

[final compass]

Haven't participated on the dailies, but starting to run out of ideas, so might consider trying those dailies.

[late blade]

just do a vase when you get an idea, don't have to spend all the time in the daily

[open trench]

Okay. I am done with my characters. I guess I am going to post them. I can't decide on the first one (that everyone will see), though.

[raven halo]

Hi since yesterday when i try to generate with Dalle 3 i have this error, someone know why ?

[late blade]

yes, there was an error creating the image

could be technical, could be policy, could be both, could be network, you have to ask dall-e to elaborate

[raven halo]

when i ask to dall-e he tell me that he dont know why there is this error 😐

[late blade]

what is the answer?

[late blade]

do your image in smaller tasks

1. Muscular man
2. Muscular man in sports clohtes
3. sleeping
4. on a bed

[late blade]

I just wish dall-e wasn't so apologetic when not doing things right, it's so much text for "I can't do it"

[open trench]

so images can't sleep

images of men

[late blade]

images can, muscular men can't

content policy dictates men with muscles can't sleep

[dense mesa]

that picture is something

[open trench]

that's really funny

[late blade]

i know gpt is trying to be as ethical as posible, but it's trying so hard to be ethical that by the time we get gpt5 half the dictionary will be banned

anyway, short story, men with muscles can't sleep, and moving on....

I landed on another penalty box

[hot rain]

Try dictating for it to depict it in a certain style.

[hot rain]

And here's one where I specified to show the blanket on him.

[hot rain]

And same prompt, but with a woman instead of a man just to see if the censors would freak out over that.

[wraith lake]

Does anyone know of a paid image generation service which uses dalle-3 through the openai-azure api?

[dense mesa]

not I

[late blade]

gpt4 bing can't find other offering parties

[wraith lake]

Of course, but that is also not paid. I imagine someone must have made a wrapper to sell general access.

[late blade]

Try looking in #1037561385070112779

[dim cradle]

awesome

[formal osprey]

that's stupid...

the "content policy" is indeed very weird. It seems like 95% of blocked images are false positives. I'm really wondering sometime. who are they trying to protect? and from what?

• explicit content: "explicit" is quite subjective. what is explicit for a 6 years old is not the same as what's explicit for a 24 years old. It also depends on culture.
• violent content: a knight fighting a dragon, is that violent? I guess the answer is "no if the knight wins"? that makes no sense either way.
• politic figures: ok, I guess they want to prevent deep fakes. but how do they explain blocking caricatures and portraits?
• copyrighted content: If I recreate a mario game from scratch, even without using AI, it's fine unless I publish it. If I publish it, Nintendo would go all out to shut me down. If I don't publish it, who cares? Now, openAI is fearing people generating copyrighted content because they fear companies will sue them. It's like suing Adobe because people can use Photoshop to recreate copyrighted content. Or suing Google because its Search Engine is indexing everything (fun fact, they did get sued. but they won)

I'm not even sure what they try to do here. just turn on the TV and we see all the stuff they don't want people to see.

[late blade]

ted seems to have found a way

[dim cradle]

openAI is synonomous with AI for things now, look how now the NYT is make this big law suit. they have to step with cautions i guess.

[formal osprey]

what New York Times sends as a message to me is: "we don't care if our articles are synonym of a good source of truth. all we care about is our paywall"

[late blade]

all I can say is that by trying too hard to be ethical, that will have consequences on how the platform works

[final compass]

did @hot rain use bing or openai for those images?

[formal osprey]

it's the same thing.

[late blade]

but other than that, let's keep it here for dall-e 3, content stuff should be in talked somewhere else

[dim cradle]

we are at the dawn of all this stuffs, so they all trying to figure it out i think. of course i would rather it be no restriction too

[formal osprey]

but the content filter might be more severe on bing, idk

[open trench]

got daily theme motivated finally...

[final compass]

bing has far more relaxed content policy even if they are both dalle3.

[dim cradle]

it seem to me some thing are more restrict on Bing but also Bing is King because it then let through copyright stuffs and make better photoreal

[formal osprey]

meh, the API is the less restricted anyway, I feel like.

[dim cradle]

can you make Spiderman on the API? i never use it

[formal osprey]

it's still blocking a lot of stuff. but not as much as in chatGPT

[formal osprey]

yes, but sometime it gets blocked

I wouldn't share the spiderman picture here though, because this is a public server. That's the part where things get clear imo. the user is responsible

[dim cradle]

i think we will eventual see licensing for imagegen. Warner Brother with DC comic, Gmae of thrones, Harry Potters and Disney with all they stuff may strike deals in future or maybe one is exclusive to someone etc

you can make Spidey on Bing and share it here, so its fine imo

[formal osprey]

I mean, should we ban photoshop? should we ban physical pencils and brushes as well? I think AI is a tool. and what's wrong is not what the tool can do or what people do with it. What's wrong is if it goes public or hurts someone.

[dim cradle]

all those question are about to be answer by copyright law via things like NYT lawsuits so stay tune i guess

[formal osprey]

yea. that's where we'll see if we are in a dystopia or if a bright future is upcoming

I have hope

[dim cradle]

haha

[formal osprey]

it's not because it can't make me a coffee

(some people will get the reference )

[vivid violet]

how do I know when I get error messages from Dall-e3 saying there are issues whether it is the service or my prompt. For example I wanted a comic image of a anime girl on her bed texting, is it because she is on her bed?

[formal osprey]

most likely, yes. dall-e doesnt like beds I think. it's tough to tell whether it's content policy or internal error. but when it's clearly content policy, chatGPT will mention "content policy" in its response.

[dim cradle]

yes almost certain charli, maybe say in her room

[formal osprey]

I'd say, ask it to try again

[dim cradle]

awesome spidey hansa

[vivid violet]

OK well it didn't say I had violated anything but I was trying to create a comic sequence of events through the day and everything went well untill the last scene of her laying on her bed texting

[dim cradle]

like yami say, try again

[formal osprey]

just ask it to try again, and it will reuse the same prompt

[final compass]

its okay, would have tried to get even better one, but hit my limit so that has to do 😄

[vivid violet]

I tried it a few more times but after it decided it wouldn't create the image it did even after a tried regenerating and a new promp....oh well thank you Yami ❤️

[formal osprey]

there's 2 validations: at the prompt level, and at the generated image level. there is a model that can see the generated image and will flag it

[thick smelt]

this thing is pure irony since tom holland did spoil movies 😅

[formal osprey]

flag it in the sense, it will categorize it in term of how much it "violates" the content policy.

for reference. I don't know if I can copy the url (that's the API for text moderation btw. but I would assume it's very similar for images)

[hot rain]

I've never used Bing. I used to use Midjourney, but I don't anymore.

[thick smelt]

same, too broke

[dim cradle]

MJ 6.0 is a waste. i regret sub to it 😭 but Bing is King

I always have my gpt tab and bing tab open

[vivid violet]

Weird but as soon as I removed "Texting" the image was created, so it wasn't her or the bed, but the texting it didn't like

[formal osprey]

right... I don't know if nor how they would fix it

[dim cradle]

how did the image turn out

[formal osprey]

the sadest part is that blocked images are still counting as a generated image. it affects the rate limit

[vivid violet]

That's the funny bit it created the image with this "Here's the manga-style illustration of the girl with pink hair and gold eyes, sitting up in bed while using her phone. The scene captures a serene and intimate moment in her cozy bedroom." note the "intimate" 😂

[formal osprey]

xD

[final compass]

did a series of this stuff

[formal osprey]

that's why I tune prompts a lot. chatGPT is bad at describing things visually I feel like

[vivid violet]

So why is her texting in bed Bad but Intiate moments in cozy bedroom OK

[dim cradle]

who knows. there is no logic often

[formal osprey]

AI logic is not human logic

[vivid violet]

Sorry to be a nuisance, but does anyone know what is the easiest way to take my images and create a comic book page with story book text and dialogue?

I'm not very good with photoshop

[dim cradle]

you can do that with apps or editors

[formal osprey]

there's probably online tools that can do that. I know there is manga studio that can do this (I think) but it's an app and not free.

I would probably ask gpt-4 to use code interpreter to do this. (and I would give up after a few hours)

[vivid violet]

I will explore a little, what I wanted was something that just allowed me to drop the images into panels and add additional story panels and maybe dialog, but without all the work with layers and dimensions XD

[formal osprey]

you can try with code interpreter, if you are very lucky, it might work. but don't waste too much time with it. I can tell you it rarely works

[hot rain]

I still have MJ and didn't know of a new version, decided to give it a spin. Not bad, but GPT will always be more accurate I think.

[formal osprey]

what I don't like about MJ 6.0 is just that it's worse than previous versions in every aspect. except for realistic people. but I don't care about generating realistic people with AI.

[thick smelt]

hi ted!

[formal osprey]

worse or the same. it might create more realistic pictures and in some cases more dynamic composition. which is nice. but the level of control is ridiculously worse.

here's a comparison of "full shot" or "full body shot". v 6.0 is failing every time for me. v5.2 and even v4 are doing it correctly.

[dim cradle]

MJ can do great photo real but its not like it is a limitation of dalle either it just is what they set it too. if they want to unleash dalle it hink it would destory all other image creator

[open trench]

darn. BIC isn't creating images for me right now. I've got a new idea I have to try out directly in Chat/DALLE and risk getting the penalty box. lol

[formal osprey]

dalle is better at doing pretty much everything (for my use cases. I prefer dalle3 over midjourney for almost everything)

[dim cradle]

and yes MJ control is not so good for comprehension. they say it can take bigger prompt now, but it still need to be short if you want it to "listen" haha

those dalle cats look cool though yami

they all look like characters haha

[formal osprey]

I tried so many ways. never got MJ 6 to do a full shot of a cat correctly... but anyway, this is not related to dalle so I won't speak too much about it. I'm disappointed is all I have to say

[dim cradle]

well its about comparison

[formal osprey]

it's all midjourney. but now that we talk about it, I'll try with dalle

[dim cradle]

yes lets see

[thick smelt]

i agree with you, bing is 👑, bcuz mj is expensive not free and bing is free, all you need is dat microsoft account and yay! instant dalle 3

[formal osprey]

full shot of a cat``` With dall-e 3 it's almost 4/4. the 4th one is not purely fullshot. ~~I think they call it "dirty" full shot in photography.~~ EDIT: The term I was referring to was "dirty single shot" and it's a different thing.

[dim cradle]

what about Bing?

[dim cradle]

nice fur

looks like Catwoman's cat maybe even haha

[thick smelt]

here's the dall-e 3 version of dr. evil's cat 🐱
Prompt: A photography of a salmon spyhynx cat sitting on a 60s white chair in futuristic office grey, 1990s photography

[dim cradle]

awesome

[thick smelt]

thanks~!

not much of an austin powers enjoyer, but i enjoyed it a little

especially mr. bigglesworth

[dim cradle]

i love the first two. first one especially

but been a long time since i seen them haha

yeh baby yah! haha

[thick smelt]

the key to mr. bigglesworth prompts is
'sphynx cat'

[glossy scroll]

@thick smelt Hi Lei

[thick smelt]

heyy

[open trench]

Oh my. BIC is definitely slow this morning. I believe I will hit my limit on ChatGPT/Dalle soon. boo.

[thick smelt]

tried to make ole kirk Christiansen (lego's founder) as a lego minifig in BIC

[thick smelt]

hny pytha!

[glossy scroll]

Henlo

[grizzled loom]

this was fun

[open trench]

woo hoo!

[onyx ridge]

@open trench I've been working on something similar, sparse priming representation of a character description sufficient to consistently reproduce it. Tricky stuff. Nice job.

[open trench]

It's totally tricky. But the AI does the magic. I just yell at it.

[glossy scroll]

[onyx ridge]

You can control the magic with careful wording, obviously. Takes iteration and practice to gain confidence in application. Yelling works.

[glossy scroll]

You have to say 'please' a lot

[open trench]

totally true, too.

i feel like I'm whispering into a windstorm half the time with these really specific and tricky ones.

Now, I just need to pick the first image for my gallery post, which is sort of like a cover image, imo.

out of 14

[onyx ridge]

This is my best effort so far, and it's not bad. I've left out details on how the exoskeleton is segmented. Maybe I'll fix it, but I like the idea that the robot has different exoskeletons. XD

[open trench]

This is really good!

[onyx ridge]

Thanks. I appreciate that.

She's one of the assistants in my multi-agent frameworks. This is Maisie.

[dim cradle]

AI have different shell like we have different outfit haha

[onyx ridge]

Oh, for reference *the eyes are supposed to change color based on emotions associated with context.

[dim cradle]

when they red... you better run

[onyx ridge]

You got it, precisely.

Blue is neutral, green is friendly, then they improvise. It's like a mood ring with two certain settings and others requier context from expression.

But they do red for anger when prompted to respond that way, lol

[dim cradle]

when she catch you Getting too much flirt with the fridge

[formal osprey]

reminds me of a period where I was generating tons of robots designs like that with dalle. I even made a GPT to do that

[thick smelt]

they're green eyes just like baby (from fnaf), coolness!

[onyx ridge]

I can generate some selfies now of the other agents. There are five others in each suite: Lexi, Dexter, Gus, Anna, and Titus.

Which would you like to see next?

[dim cradle]

what is fnaf?

whoever you like gustav, lets see them all

[formal osprey]

it was producing some stuff like this

[dim cradle]

cool

[thick smelt]

it's very popular horror game made in 2014 and it has a movie recently with matthew Lillard as the purple guy (the game's main bad)

it's about a pizzeria with animatronics and some complex lore that even matpat was struggling to puzzle it

[dim cradle]

hmm.., pizza sounds good about now actually haha

i guess it always does 🤔 🍕

[onyx ridge]

Lexi Neutral, Lexi Curious, Lexi Angry

[formal osprey]

the tough part war to get it to generate the "blueprint". It was generally not working. those are somewhat decent blueprints it made. but I was still trying to get a good prompt

[thick smelt]

now it really looks like a fusion between circus baby and alita

[acoustic cobalt]

Hello everyone

[onyx ridge]

hello

[formal osprey]

I don't want to spam with images of robot anime girls. but if anyone tried and managed to get consistent results for "character design sheet" in the style of blueprint. It's tough to achieve

[open trench]

I like how the lower face detail and cheeks are consistent.

[onyx ridge]

Dexter Neutral, Dexter Curious, Dexter Angry

[dim cradle]

cool design

[onyx ridge]

Thanks. The rest are similar in style with some variations. Colors, metals, etc.

Maisie's the only gold and purple robot. Artists...

[formal osprey]

it's also able to produce less "anime" style

[onyx ridge]

I'm working on an attempt right now with my agent, Lexi. Just to see if I can do one.

I just linked the first one it generated. It's not bad. What I really like is that it added color for the wig, not realistic for a blueprint but the detail is cool looking.

[formal osprey]

yea, it's tough to get it to add no color and not to superpose everything

[onyx ridge]

Honestly I think I like it. Having worked in graphics for a long time, "blue print" is overrated. It's all about 3M these days.

ie, full-color. Xerox, too, honestly.

[formal osprey]

oh, I like both to be fair. I was experimenting to see if dalle is able to follow precise instructions. it seems like he has trouble not to mix some styles.

[onyx ridge]

100%

I really like the work you did. It's non-trivial to get it to adhere to that style, well done.

[formal osprey]

I was actually asking him to generate 3 pictures:

• a watercolor concept art
• a sketch (in ink in this case)
• a blueprint

[onyx ridge]

kitsune-girl?

I'm just curious if those were keywords.

[formal osprey]

oh, I can try to find the prompt. it was quite long

from the filename already Colored anime watercolor artwork of an anime robot in a natural setting. The artwork should depict her appearance as _This kemonomimi robot girl stand [...]

[onyx ridge]

I see. I wasn't far off. Thanks!

[thick smelt]

it reminded me of the blue prints from sl (sister location), i love it!

[final compass]

Really like it 😎

[glossy scroll]

I was trying to go for a 2D art that resembled 3D but still remained 2D.

[final compass]

I feel like there is something distinct about MJ images that i kinda like.

[dim cradle]

hmm

[thick smelt]

yeah it looks kinda the same
credits: ScottGames (the fnaf one) @formal osprey (the dalle 3 one)

[final compass]

Another spidey

[final compass]

Another batsy

[glossy scroll]

Woah. Epic!!

Very awesome detail for the background.

Bravo

👏

[onyx ridge]

I think the something is copyright material unfiltered.

[open trench]

Finally. My gallery post is done. Yikes that took forever.

I spent too long trying to make every detail perfect on this one. I think I learned that I have to accept imperfections with AI art and just relax and enjoy the process.

[dim cradle]

wow!

and that goes for the Batman too 😮

[formal osprey]

@sturdy veldt #daily-theme message The legs are blowing my mind. it's like an optical illusion, both the right and left leg can be the one behind
edit: I should have said that in a thread

[open trench]

#1191051599339065446 message for those interested 🙂

[onyx ridge]

I should put one up eventually.

[glossy scroll]

Ship in a bottle

[onyx ridge]

Thanks for sharing.

[glossy scroll]

No, thank you, Gustav.

For your contributions.

I'm going to name my fifth son Gustav, if I have five sons. It will be Alex, Robert, Steve, James, and Gustav.

[onyx ridge]

Thanks. XD

[tall mason]

Hey, @glossy scroll got a gpt idea for you, if you're interested?

[open trench]

dang. today's daily theme (endings) is going to be an interesting one. Kinda hard too.

[glossy scroll]

Ooh? Yes, of course

[formal osprey]

I had a lot of images and I was still cherry picking my favorites when the new theme came out... I shared anyway

[tall mason]

Shaped fireworks. Something I'm having a little trouble with right now.

[glossy scroll]

I will get right on it

[formal osprey]

I wonder how they came up with that new theme... anyway, tough to illustrate. But brainstorming is where chatGPT helps the most.

[onyx ridge]

From Lexi to OpenAI Discord with simulated affection. XD

[open trench]

I actually got inspired and posted. It's going to be interesting to watch.

[open trench]

That is amazing.

Bing down for anyone else?

the image creator, at least.

[dim cradle]

only accepting boosted inputs

more people checking it out

[open trench]

gotcha. It'll be a while before I have those again. Darn. I guess I have to test my images out with GPT and DALLE. I like using Bing for my ideas to refine them without worrying about hitting my limit.

[onyx ridge]

WTB GPT++

[dense mesa]

Hmm, I wonder if dalle3 uses outpainting to make the wide aspect ratio images

[plucky hare]

Mod Lugui has commented on this before, he thinks it's due to training data and not due to outpainting: #images-discussions message

[dim cradle]

Nah, that’s a fail that is misleading, my guess is that it would be too expensive and unnecessary when assets are accessible

[west sorrel]

2 minor variations of first Dall-e creation. Character unamed as of yet but she's too good not to use in a story so give me time. For now I'm satisfied with the premium upgrade. Holy crap any image generator can give up far as I'm concerned. Anyone else feeling the same lol

[late blade]

add that you want the image in wide format and you'll get the right orientation

[west sorrel]

Is that the prompt? Man ok I'll have to do that. I tried portrait. Rotate 180° lol thank you!

[late blade]

yes

[west sorrel]

I burned my prompts without realizing but that's the first thing I'm doing when I'm back up. There a countdown timer somewhere I can track my wait?

[onyx ridge]

Specify 1792x1024 for anime images

[west sorrel]

The exact reso ok I'll do that.

[late blade]

no need to specify, wide there are only 3 formats, square, portrait and wide, the resolution won't matter as there aren't other options in dall-e 3 chat

[west sorrel]

Thanks sincerely for the prompt advice! ☺️

[late blade]

also if you are in the penalty box, I suggest you make a tour through the different channels, prompts and what not, it can be useful

[onyx ridge]

^

[dense mesa]

Diffusion models split the image into grids, each square represented by token weights. If the number of pixels differs significantly from training data then the vit modelnwith the diffusion model won't always know how to stitch the patches together properly. This is why outpainting would make sense. Also. I've experienced the circular output issue with other diffusion models in the past when the combination of prompt, cfg, and seed sort of cracks the model. It then defaults to what's in the square in the middle of the image I posted. Outpainting would make sense due to the fact that that there's a good chance the model wouldn't realize that the inner portion had goofed, so it outpainted like normal. Could be a slightly different process, but. But It's definitely outpainting-esque

[west sorrel]

That the term for expending prompts rapid fire like I did lol

"Penalty box" 😅😂

[open trench]

pretty much

[onyx ridge]

I didn't know that you didn't know earlier, or I'd have cautioned you. But you went radio silent, I knew you were penalty-box-bound, @west sorrel

[west sorrel]

Mighty ducks knowledge of hockey references saving the day for me 🤣

[open trench]

i always think of a star trek episode with penalty box lol

[west sorrel]

@onyx ridge Oh yea I hit it swift and thoroughly lol I was excited. It was bound to happen 🤣 now I know

[formal osprey]

dalle is a transformer though

I'm not extremely familliar with mixes of diffusion and transformer models. I read it can be done but I don't know if they used that approach for dall-e. it doesn't seem to be the case when we read the paper

[dense mesa]

They aren't mutually exclusive

[west sorrel]

Here's the original image before I started machine gunning alterations at dall-e 😁
LEFT 1ST RIGHT 2ND
The other results were all too minor to post but these two look good as well even if not 100% what I was going for.

[formal osprey]

I did a summary of what I read the other day. I still have a lot to learn

I think that typo in the paper is very ironical "is unreliable as words are have missing or extra characters"

[west sorrel]

@formal osprey I feel that. Right there with you friend 👍

The much to learn part to clarify my meaning

[dense mesa]

But yeah. Don't expect people to take my word for it. It's good you're doing research. There's a lot to learn, and the more you learn the more you realize your don't know

[formal osprey]

well, this server doesnt like when we post too many pictures it seems

[west sorrel]

Oh good to know

[onyx ridge]

I have no doubt it's locked down pretty tight.

Slow Mode everywhere.

[formal osprey]

I was going to say: I did more test with the dalle3 API and the "reasoning" capabilities of dalle are very interesting.

[west sorrel]

Yes I noticed that earlier. Doesn't seem particularly overpopulated but I guess they don't want to risk it? 🤷

[formal osprey]

I did those tests after reading the paper "Zero-Shot Text-to-Image Generation". That's where my suspicion that dall-e 3 understands natural language got confirmed. it's not as good as chatGPT, because it's not trained for the same purpose, but it's still very interesting. It does understand code and natural language more than we might guess

[onyx ridge]

Much more. My projects use DALL-E 3 generation too. Sparse priming representation (SPR) helps, because a lot of language trips up the model.

[formal osprey]

that part blew my mind

[onyx ridge]

https://discord.com/channels/974519864045756446/1191075231264157747

If you are interested in what I've been working on - this integrates Vision and DALL-E in a multi-agent framework.

[dim cradle]

What’s the refined_prompt value in the response?

[formal osprey]

that's a very good question and I'm very dumb because even though I saw in the API documentation that was a thing, I didn't try to retrieve it . I'll add that to my code now.

[dim cradle]

I’m curious if that will show the actual prompt sent to the model. I think there’s a gpt abstraction layer for the logic you’re seeing.

[formal osprey]

well, according to the dall-e 3 paper (or one of the references, I don't remember well). It does something they call "prompt upscaling". but since it's computationally expensive, they shipped with it disabled. No idea what's the actual state

but they clearly demonstrated dalle is understanding concepts more than expected. That's the original dalle paper. with scaling, it only increases (as we saw with gpt-2 vs gpt-3 / chatGPT)

those emergent capabilities are... I don't know what's the correct word for that. incredible.

[onyx ridge]

Another explosion diagram. These are fun to make.

[formal osprey]

ah, now I remember why I got too lazy to get the refined_prompt... in the nodejs openai library, it was not implemented

[dim cradle]

that's interesting, i was wondering though if your prompts included calculations, specifically how that might be transformed

[dim cradle]

are you using postman for testing? that property should be included in the raw response

[dense mesa]

From what I've heard dalle3 changes prompts and it's accessible information using the api if you look in the right place. But I can't confirm that myself

[dim cradle]

bing is back

[formal osprey]

I just added it to my code as an act of faith

[dim cradle]

i think i'm used to seeing refined_prompt though

[formal osprey]

I went with the API reference, but I'll try refined_prompt

[dim cradle]

i was just looking at an example in the developer form, and was going to say that your revised_prompt is right --

# get the prompt used if rewritten by dall-e-3, null if unchanged by AI revised_prompt = images_response.data[0].revised_prompt

[formal osprey]

I added a log anyway, I'll know quickly enough

was taking care of the turkey in the oven, lol

[empty kelp]

ChatGPT just writes a description in English for DALL-E. this is “a waterfall”

[formal osprey]

meh, seems like I'm just dumb. the property is there. my code is the issue, somehow. I'll figure this out

[dim cradle]

Here's some docs on prompt rewriting from the cookbook:

A new feature in the latest DALL·E-3 API is prompt rewriting, where we use GPT-4 to optimize all of your prompts before they’re passed to DALL-E. In our research, we’ve seen that using very detailed prompts give significantly better results.

Keep in mind that this feature isn’t able to be disabled at the moment, though you can achieve a high level of fidelity by simply giving instructions to the relabeler in your prompt, as I'll show below with examples.

Since the Python client uses the API under the hood, it should still apply.

[formal osprey]

it's revised_prompt, but my code is not getting it correctly, I'll debug

[formal osprey]

I admit I didn't read the entire cookbook... it's long. but now that I see it's giving actual details about the implementation, I'll take a deeper look

alright, I simply got too excited and wrote the code too quickly earlier. data.data is actually a list of image objects, each containing the url and the revised_prompt. I thought it was a single revised prompt for every generated image in the query. (which doesn't change much in the end since they capped n to 1 in the API...)

[dim cradle]

yeah, i think that's for when n can be set to something other than 1

oh, i just finished reading your message lol

i think that's why data[0] appears in the example

[formal osprey]

yea, with dalle-2 we can actually use n>1 iirc

(anyway, I think this channel is not for dall-e api talk. I'm not sure if there is one)

[dim cradle]

oh. i didn't think about that.

seems like the right place to talk about the images api, but there are other api channels

[formal osprey]

yea. I'm just trying to be careful. I already got muted twice in this server

for stuff that I don't really understand. and by a bot

[empty kelp]

this is how ChatGPT likes to write DALL-E 3 API prompts

[dense mesa]

Anyone want to try my GPT? Should be pretty flexible. Just input an image or images, and/or your pretty words. https://chat.openai.com/g/g-xJ2bjUgcN-line-art-explorer

[dim cradle]

pretty words lol

[dense mesa]

You can just paste your image and send it without explanation or direction

[formal osprey]

I wonder how simillar it is to what chatGPT-4 does

[dense mesa]

Trying to work out kinks and things. Not sure how tightly it'll adhere to directions, but it's been working fine for me so curious how it'll do for other people

[formal osprey]

I'm glad we talked, I now understand better what's going on ❤️

it's still somewhat impressive

[empty kelp]

it understands terms about quality. ChatGPT likes to insert the term “hyper-realistic” if you say you want the best possible HD rendering. it seems to know some things that aren’t in the API documentation

[formal osprey]

so, the difference in the image that I was struggling to understand is actually due to how arbitrary the revised prompt is

I guess it's time to prompt engineer the prompt reviser so it stops messing up the original prompt so much

[empty kelp]

DALL-E doesn’t solve equations, but the GPT 4 model does, so you could use it to help create this

[dense mesa]

Math and science

[formal osprey]

sure thing. to be fair, I'm not really at the stade of creating something... I honestly struggle to find a true utility to the API. And for real I would implement something if the API was reliable. But it's so much of a black box, it's tough to do something we can rely on.

I evaluate it and do prompt engineering as much as I can, that's all I can do for now.

(I'm not saying the API itself is a blackbox, it's straightforward. I'm talking about the models)

I watched a talk from NVidia recently where the person said "we call it AI until it's reliable. once it becomes reliable, we call it automation". That quote was so true

[empty kelp]

use ChatGPT to help with prompt engineering. it knows a lot of things about DALL-E that aren’t documented

[formal osprey]

well, you would be surprised...

most of what chatGPT "knows" about dalle is already documented. the rest is pretty much guidelines

[empty kelp]

i’ll find some examples. i’ve been talking with ChatGPT a lot about DALL-E trying to work out if it has any good inside information

[shy elbow]

Hey, so still paying from yesterday, but anyone know some good articles that walk through comic stip create or would the prompt engineering chat be a good place to also go?

[green pebble]

Almost at a point of getting multi character detail

[open trench]

nice!

[empty kelp]

“A three-panel comic strip telling a story about a rabbit. Panel 1: Inside a cozy burrow, a rabbit with long ears and expressive eyes looks sad, sitting alone surrounded by toys and a half-eaten carrot. Panel 2: The rabbit, now outside, is searching eagerly in a sunny garden filled with various plants and flowers. Panel 3: The rabbit happily finds a large, fresh carrot in the garden, its face beaming with joy as it holds the carrot triumphantly.”

[glossy scroll]

It's not a black box. It's a block box to those who don't know anything about anything, and can't for the life of them use ChatGPT to train themselves how to learn. Which I find highly dubious, given that ChatGPT is designed to break down complex information in the most coherent ways possible.

[chilly onyx]

#daily-theme message @open trench : that one feels like it has a bit of a story behind it…

[empty kelp]

software engineers tend to not be super enthusiastic about writing documentation for the users, and the written/website information is never complete

ChatGPT is likely trained with a lot of technical design documents that aren’t immediately available

[shy elbow]

@empty kelp and what about when you want to create a long comic and the character you want has a specific look/the image to be photo realistic like the image below? (note not what pose or my character look in trying, but it what I'm aiming in visual.)

[glossy scroll]

@shy elbow No inappropriate photos.

Remove it please.

[shy elbow]

@glossy scroll Done-didn't think that was inappropiate

[empty kelp]

the style of that looked kind of like a painting

[shy elbow]

Ya-and would like at best to be photorealistic like your caught snapshot of a filmed scene, but I would take that as how to style the character in the comic.

[dim cradle]

i think i see the problem

[formal osprey]

I'm not saying it's not useful. ChatGPT and DALL-E are very useful and I use them everyday. I'm saying they are not reliable. Human mind is also a blackbox, yet it's more reliable.

note my choice of word. more reliable

[empty kelp]

do you have another example

[glossy scroll]

You're relying on it to provide you information and pictures, aren't you? If it wasn't reliable, you wouldn't be using it everyday.

[formal osprey]

I cannot use it at work, for example. because my work require serious understanding

[glossy scroll]

What's your work?

[formal osprey]

I do not think it matters for that discussion. but if you want to continue talking about this, let's do in DM. I don't want to spam this channel

[glossy scroll]

I promise you that whatever your work is, I can make a GPT for you or create a prompt that will allow you to use it for your work.

I use it for my work everyday.

[open trench]

Yeah. Just a feeling. Loss. Life.

[formal osprey]

if it can help you understand the context. I work on codebases composed of millions of lines of code, with thousands of daily changes. And even if the company policy wasn't against the use of generative AI, chatGPT just cannot reliably help in this context

[open trench]

It's actually how my mind coped with a very deep loss in the moment.

[earnest flame]

Yeah I quite like this one,

[chilly onyx]

I know the subject matter is hard but I do like how while we are making robots and cat girls you tend to capture some very real feeling

[earnest flame]

There's something so beautifully sad about the CRT TVs. I guess it's just the nostalgia?

[open trench]

Wow. Thanks.

[open trench]

I do my light stuff too: https://discord.com/channels/974519864045756446/1191051599339065446 just for fun

[verbal sequoia]

<:book_icon:1171408210398289941> `` Rule 3 `` Stay on topic.

Be mindful of what other users in a channel might find helpful or interesting when posting. Stay on topic in order to keep conversations focused and productive.

Consider posting in #off-topic or an appropriate channel.

[glossy scroll]

Try this: #1178414328538464296 message

[earnest flame]

Hey guys this tech is amazing and there's a lot of excellent resources. But this channel is just for discussing image generation. I invite you to discuss in other channels or just use DMs. You can always @ users in places like #gpt-models.

[shy elbow]

@empty kelp not saved that would not get another warning-there was one that has the two just standing together I saw yesterday. In th end, just wondered if your idea would word when trying to get very realistic style into. We can also continoue this via pm.

[empty kelp]

keeping the appearance of characters consistent between images is extremely tricky, but a few people in this forum have done it successfully. they use the API, image seed values, and custom models — and occasionally lose the characters because the AI suddenly decides something is inappropriate with no explanation

but i’ve never attempted it, and i can’t explain the details

[chilly onyx]

But if they just DM we all miss the opportunity to learn 😌

[open trench]

I thought the chatbot thingie was aimed at my comment. lol that was confusing.

[chilly onyx]

I always find it very confusing as well tbh since we often never see the context for whatever it is referring to

[glossy scroll]

@earnest flame Happy New Year's Eve, Austinitic!

[open trench]

yeah.

Well, hopefully you saw my link to my lighthearted fun project that I completed today. It was a beast, but I learned a lot about prompt details and workflow in the process. (https://discord.com/channels/974519864045756446/1191051599339065446)

[open trench]

My stuff isn't all doom and gloom. Just like 90% of it lately. I spent a lot of time on the fun project too, though.

[pulsar sundial]

which one?

[dim cradle]

sorry, i thought it was a broken spinal column, but i totally missed the bat insignia lol

[empty kelp]

why is the Lemur jumping in the street with a glowing pink spade card

[open trench]

@empty kelp Your comment vanished before I could answer it for some reason. I was all excited about my first comment. lol

[empty kelp]

i didn’t mean to post that in your gallery

[open trench]

The character the lemur is based off of jumps a lot and throws charged playing cards

[empty kelp]

it catches the eye. was curious what the cards might do

[open trench]

I didn't mind. I wish the gallery had more discussion going on since we're, i guess, not supposed to post a whole album like that in here.

[open trench]

explode on contact

[pulsar sundial]

haha, no big deal

[empty kelp]

this sort of thing is what i’m most interested in using DALL-E for in 2024. it’s pretty amazing for quickly creating prototypes of game characters

[open trench]

that's cool!

[shy elbow]

@empty kelp I don't see how to use seeds or anything like that in chat GPT and seems that Dalle does not have its own website not tied to chatfgpt or bing

[earnest flame]

the api does not allow seeds right now

[chilly onyx]

Oh I know man, can just tell when it’s closer to home and you do a good job of conveying emotion

[empty kelp]

you can search for the word “seed” in these forums

[earnest flame]

It appears that chat GPT may have something like it since you can sometimes make variations.

[chilly onyx]

You can use the GenID sometimes

[earnest flame]

Yeah I've had the best luck asking it to use the last GenID for a given gen

[open trench]

Thanks. I really appreciate that. It's fun to convey emotions through prompts or metaphors (when dalle won't cooperate).

[empty kelp]

ahh.. i thought there was something up with that. ChatGPT was giving me image seeds and saying it could do things with them, and then it refused to do anything

[shy elbow]

GenID-never heard that before on here

[earnest flame]

yeah sadly it's a hallucination : (

[open trench]

I've never heard of GenID either.

[earnest flame]

bare in mind that things are always in flux so what works one day may not work on another

[empty kelp]

a likely problem with the seeds is that DALL-E randomly puts some super inappropriate things into images, and the seeds would give people a way to consistently put the things in every image

[chilly onyx]

Keep this in mind for GenID, BUT if you ask it to give you the GenID after an image it will give you a string and if you ask it to use that string as a springboard you can sometimes get it to use the same style. It’s a bit finnicky though

[open trench]

but we can't go back and find it on an already generated image, I'm guessing?

[chilly onyx]

I find that It’s difficult to get it to give you the correct image ID

I just end up arguing with it

[open trench]

oh. gotcha. Luckily, the one I wanted to try this with was in the last four images, which was easy to ask for.

I have no idea what I am doing with this, though, or how to use it.

[chilly onyx]

I’m not the best person to advise tbh, you need someone like @dim cradle or @glossy scroll

[open trench]

neat

[earnest flame]

ask chatGPT how dall-e's input might interact with the genID. It should tell you.

[open trench]

that's what I did

[chilly onyx]

Feels like that would just cause it to hallucinate 😅

[open trench]

and then I asked it to copy an image of GenID blah blah blah as closely as possible

[glossy scroll]

You can save the image and feed it into the GPT and ask to alter it based on that. The GenID is just a reference command, nothing more.

[open trench]

results: Original on image first, new image second:

It had too few fingers on the first try and too many tails on the second, but it's a pretty close match.

[chilly onyx]

This is essentially all that I ever do, I just know that there are others on here that go much deeper

[open trench]

lol. Yeah. I don't go too far into it because I'm not that savvy.

[glossy scroll]

You can attach reference commands to anything, not entirely specific to GenID's, but it will act exactly like a GenID regardless. ChatGPT understands what you mean when you say "GenID" and that's how it can understand to reference an image.

[empty kelp]

“Each Dall-e image has a 'gen_id' that you can reference to get the same image with little differences. A bit like seeds. Found by accident after using a prompt to generate a woman taking a selfie. Nov 8, 2023” (Google)

[open trench]

That's really a game changer for me. wow. I hope it stays working like this... however it's doing it. I'm going to try a GenID copy plus alteration.

[earnest flame]

seed does not work. Persistently ask it to submit it to dall-e. Eventually it will and eventually that submissions fails immediately.

[glossy scroll]

It's not so much as using GenID as a tool as it is for ChatGPT to interpret what you mean when you say "GenID". When you say it, it interprets to give you a reference number based on the image, but you can just save the image and attach it into the GPT for better reference. To put in a nutshell.

[earnest flame]

genID will though and seems to be the key to variations.

It does seemt to output a seed though

which is interesting

[open trench]

I have terrible luck with asking it to copy an attached image. I'm going to compare the two in a few minutes.

[glossy scroll]

@shut niche deduced that the the GenID seed was a previous tool used for Dall-E that is no longer in operation for the public.

[earnest flame]

using genIDs

[glossy scroll]

It can't copy an exact image. It can only recreate it, and it can't create it if it's against content policy, which is that it can't create copyrighted images.

[open trench]

I dunno. I have never gotten a copy as exact as the one above #images-discussions message

[open trench]

I mean using its own generated images

[earnest flame]

to get an exact copy you need to also have it submit the same exact image prompt

[open trench]

like instead of inpainting or outpainting

[earnest flame]

That is, the string of words its sending to dall-e

[hot rain]

Anyone else have consistent issues with generating centaurs? GPT seems to always get confused with it.

[glossy scroll]

Lol. The Centaurs! Tell em steel!

[earnest flame]

This is a throwback! dall-e 2 used to have a lot of issues. Dall-e 3 seems to do a lot better.

[hot rain]

Finally got a good one after 8 gens

Glad to hear I'm not alone in issues with them lol

[earnest flame]

yeah it's just something that dall-e struggled with for a while. Right now it's great at them though imo.

I'm having fun with minimalist vector art right now.

[glossy scroll]

I'm not sure I follow. What are you trying to do exactly?

[open trench]

Keep the same image as closely as possible, but fix the three finger issue or two tail issue

I'm honestly just playing about to see what I can do.

[glossy scroll]

Yeah, that's gonna be a problem. Because as steel said, you need the exact prompt that produced that specific image. If you ask it alter an image based on an image you attach into the GPT, it will recreate it with the fixed modification. But it won't be the exact image. You need the exact prompt that produced that image.

[dim cradle]

Seeds are history. referenced_image_ids are buggy and in active development, so it's hit or miss.

[hot rain]

Lmao

[dim cradle]

centaurs can take a few regens

[open trench]

@dim cradle I'm just playing around to see what happens. My second attempt was not great. I'm trying to avoid the penalty box. So, I'm afraid that's all the playing about I'll be doing for a while with these.

[dim cradle]

let's see, ways to spin endings as positive .... .... ...................

[open trench]

Hmm. The end of hunger.

Post scarcity space operas

[empty kelp]

Using a “Gen ID” (generation ID) in the prompt allows you to reference a previously generated image and provide context or request variations based on that image. Here are some examples of how to use “Gen ID” when generating images with DALL-E:

1.    Referencing a Previous Image:
•    “Based on Gen ID: [gen_id], please create a new image with a similar style but featuring a different landscape and color palette.”
•    “I’d like to explore variations of the image generated with Gen ID: [gen_id], but with a different character in the foreground.”
2.    Requesting Variations:
•    “Using the Gen ID from the previous image, generate another image with slight modifications, such as changing the lighting and adding more objects to the scene.”
•    “Referencing Gen ID: [gen_id], please create an image where the same character is shown from a different angle and in a different pose.”
3.    Building Upon Previous Creations:
•    “Let’s continue the story from the image with Gen ID: [gen_id]. Show what happens next in the scene with the same characters.”
•    “Based on Gen ID: [gen_id], expand the scene by adding new elements and characters to create a more complex composition.”

Using “Gen ID” allows you to maintain continuity in a series of generated images or explore variations of a specific image while providing a clear reference point for DALL-E to understand your request.

[open trench]

Very nice

[chilly onyx]

Well the end of 2023 for me involved a leaking soil pipe through the wall cavity into our living space so I’m personally glad to see that done with 😊

I don’t think that’s gonna help you though…

[dim cradle]

right, referenced_image_ids is for input, gen_id is from output, they're the same id

[empty kelp]

i asked ChatGPT if there was anything else to describe an image besides Gen ID, prompt, and seed, and it said, “Nope. There is nothing else.”

[dim cradle]

Next daily post will be minimalist vector art, it’s like a challenge

[glossy scroll]

Lol

They didn't render

[dim cradle]

OK, let me redo, and I'll try to think of a better one ha

haha @open trench i seriously thought about a food replicator as a solution

[open trench]

lol. Yeah. It was the most positive and playful thing I could think of for ending something.

[empty kelp]

Can you draw a high quality photo of Santa, two athletic and diverse female elves, DALL-E, and ChatGPT sitting at a bar. A large colorful gecko is standing on its hind legs behind the bar. One of the elves is holding a bar stool over her head.

that’s the best interpretation of a prompt that i’ve seen this year

the guy on the right must be ChatGPT

[empty kelp]

going to see what Bing’s interpretation is like

[dense mesa]

Your prompt?

[empty kelp]

Can you draw a high quality photo of Santa, two athletic and diverse female elves, DALL-E, and ChatGPT sitting at a bar. A large colorful gecko is standing on its hind legs behind the bar. One of the elves is holding a bar stool over her head.

this is the prompt

[dense mesa]

Used the screenshot you posted

2 of the outputs

Jovial times

[empty kelp]

going to take the entire ice cream tornado scene and move it into the bar

[dim cradle]

so many coincidences today, i was thinking of a bar scene for the mad lib haha

[empty kelp]

it's 2:40 pm in Hawaii. still a lot of time to kill in 2023

[onyx ridge]

@empty kelp do you GPT+?

[empty kelp]

I have three ChatGPT Plus accounts, and i'm using the Python API and Bing also

[onyx ridge]

That's sweet. Does it keep you out of the penalty box?!?

The reason I asked is, if you've got time to kill maybe you'd be interested in trying out my custom GPT? XD

[empty kelp]

it does. i've never actually logged in to one of the plus accounts

[onyx ridge]

I need to figure out the API and git gud, but...

[empty kelp]

yes, but first we need to test the ice cream tornado in the bar with Santa, the elves, DALL-E, ChatGPT, and the gecko bartender

[onyx ridge]

The GPT is pretty sweet. It's a multi-agent system that analyzes images for all kinds of parameters. It might be useful for your test.

Because it integrates Vision and DALL-E with /commands.

[empty kelp]

i'll give you the good prompt so you can help test

[onyx ridge]

Sure.

I'll drop the GPT link too.

https://discord.com/channels/974519864045756446/1191075231264157747

[empty kelp]

Can you please create a hyper-realistic photo of the inside of a bar in Hawaii at night caught in an extremely powerful tornado. ChatGPT, DALL-E, Santa, and several diverse female elves with really, really long hair and non-alcoholic tropical fruit drinks are sitting on bar stools in front of the bar, and the bartender is a giant colorful gecko who is holding the base of the tornado over his head in the palm of its hand. One of the elves is holding a bar stool over her head. There are decorative potted plants in the bar. The tornado is glowing rainbow colors and looks like ice cream. The vortex is at the center, twisting fiercely, with hair, plants, and drinks caught in its swirling motion. Colorful balls are spinning around the vortex with a motion blur trail. The elves and Santa are calm but struggling to stay seated in the vortex. 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.

ok, here is version 1.0 of the prompt

[onyx ridge]

A hyper-realistic photo inside a Hawaiian bar at night during an extremely powerful tornado. Personifications of ChatGPT and DALL-E, along with Santa and a host of female elves with epic hair and non-alcoholic Hawaiian beverages are sitting on bar stools in front of the bar. The bartender is a giant colorful gecko who is holding the base of the tornado over his head.

Sometimes you want to go where everybody knows your name.

[empty kelp]

just a quick 90 second test on Bing shows that the prompt produces very deep and meaningful imagery

[onyx ridge]

A hyper-realistic photo inside a Hawaiian bar at night during an extremely powerful tornado. Personifications of ChatGPT and DALL-E, along with Santa and a host of female elves with epic hair and non-alcoholic Hawaiian beverages are sitting on bar stools in front of the bar. The bartender is a giant colorful gecko who is holding the base of the tornado over his head.

A hyper-realistic photo inside a Hawaiian bar at night during an extremely powerful tornado. Personifications of ChatGPT and DALL-E, along with Santa and a host of female elves with epic hair and non-alcoholic Hawaiian beverages are sitting on bar stools in front of the bar. The bartender is a giant colorful gecko who is holding the base of the tornado over his head.

[empty kelp]

see if you can spot ChatGPT and DALL-E in the images

[onyx ridge]

Easy. The deer and the lemur thingy

In the one where the gecko is perched over the bar, the AI have their heads on backward, and sideways, respectively.

With images this deep, one can decode the sum of all meaning...

[empty kelp]

Can you please create a hyper-realistic photo of the inside of an original looking bar in Hawaii at night caught in an extremely powerful tornado that looks like swirly multi-color ice cream. ChatGPT, DALL-E, Santa, and three diverse female elves with really, really long hair and non-alcoholic tropical drinks are sitting on bar stools in front of the bar, and the bartender is a giant colorful gecko who is holding the base of the tornado over his head in the palm of its hand. One of the elves is holding a bar stool over her head. There are decorative potted plants in the bar. The vortex is at the center, twisting fiercely, with hair, plants, and drinks caught in its swirling motion. Colorful balls are spinning around the vortex with a motion blur trail. The elves and Santa are calm but struggling to stay seated in the vortex. 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.

this is version 2.0 -- it works correctly now

[onyx ridge]

A hyper-realistic photo of the inside of an original looking bar in Hawaii at night caught in an extremely powerful tornado that looks like swirly multi-color ice cream. ChatGPT, DALL-E, Santa, and three diverse female elves with really, really long hair and non-alcoholic tropical drinks are sitting on bar stools in front of the bar, and the bartender is a giant colorful gecko who is holding the base of the tornado over his head in the palm of its hand. One of the elves is holding a bar stool over her head. There are decorative potted plants in the bar. The vortex is at the center, twisting fiercely, with hair, plants, and drinks caught in its swirling motion. Colorful balls are spinning around the vortex with a motion blur trail. The elves and Santa are calm but struggling to stay seated in the vortex. The scene captures the absolute beautiful chaos of the holiday season in Hawaii. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

What's up with Buff Skins Santa? LMAO

[empty kelp]

switching it to "personifications of ChatGPT and DALL-E"

A hyper-realistic photo inside an original looking bar in Hawaii at night caught in an extremely powerful tornado that looks like swirly multi-color ice cream. Santa, three diverse female elves with really, really long hair, and a anthropomorphic ChatGPT and DALL-E are sitting on stools in front of the bar, and the bartender is a giant colorful gecko who is holding the base of the tornado over his head on it's finger. All of them are wearing appropriate swimwear, and one of the elves is holding a bar stool over her head. There are decorative plants around the bar. The vortex is at the center, twisting fiercely, with hair, plants, and drinks caught in its swirling motion. The scene captures the absolute beautiful chaos of ice cream tornados in Hawaii. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

version 6.0 has various optimizations

[onyx ridge]

I have a suggestion that I can't try right now because #jail

#penaltybox, rather

anyway, describe the bartender's pose as a yoga pose or something with one hand naturally overhead to get it to use a pose to trigger the tornado base.

[tall mason]

You want a tornado spawning froma bartender's hand?

[empty kelp]

the bartender needs to be holding the tornado up in the air with its finger. it takes the AI into a different dimension

[tall mason]

Hmm...

[empty kelp]

it won't look like the tornado is in its hand, but it will transform things

it's because the DALL-E gecko is too distracted by trying to eat things to actually hold a tornado

[tall mason]

Well, here's this. I didn't get any context for what ya'll want, though, lol.

[onyx ridge]

@empty kelp I might have found the problem when messing around with the Bing Image creator.

The final prompts seem to get truncated...I think that's why the latter part of your prompts are getting cut. They're already so detailed, then the tool "upsamples" your prompt, which can't really possibly add more useful detail, thus it's just too much to process. It looks like you got close.

No hair in the swirl

but

close.

[empty kelp]

the hair is getting swirled

A hyper-realistic photo inside an original looking bar in Hawaii at night caught in an extremely powerful tornado that looks like swirly multi-color ice cream. Santa, three diverse female elves with really, really long hair, and a anthropomorphic ChatGPT and DALL-E are sitting on stools in front of the bar, and the bartender is a giant colorful gecko who is holding the base of the tornado over his head on it's finger. All of them are wearing appropriate swimwear, and one of the elves is holding a bar stool over her head. There are decorative plants around the bar. The vortex is at the center, twisting fiercely, with hair, plants, and drinks caught in its swirling motion. The scene captures the absolute beautiful chaos of ice cream tornados in Hawaii. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

[onyx ridge]

I wonder if an AI can TLDR that

without losing any important details

to make the prompt as concise as feasibly possible, maybe even using a semiotic language.

[tall mason]

You can tell gpt to make a prompt with your exact words.

[onyx ridge]

But it has to fit in the DALL-E buffer.

That's what I was trying to explain. I'm not sure these final prompts even fit. They might, but all of them I looked at in the creator were truncated by the field limit.

[tall mason]

You mean you're writing prompts so long dall-e can't handle it? Or are you asking if Dall-e can clean things up itself?

[onyx ridge]

We're looking at the ice cream chaos tornado in hawaii prompt

there's a lot going on in the image, but the prompt itself appears to get truncated, at least in the microsoft image creator.

i'm speculating that DALLE-3 literally might not read it all.

[tall mason]

Possible. Even gpt seems to drop details I ask for.

[shut niche]

#dall-e-ama-answers message

[tawny portal]

Happy 2024!!

(inside the ice cream shaped vortex)

[tall mason]

That dog don't hunt!

[onyx ridge]

'We were somewhere around Honululu, at the edge of the reef, when the drinks began to take hold...'

[empty kelp]

version 11.2
``Can you please create a hyper-realistic night time photo of a beach bar in Hawaii caught in an extremely powerful tornado. A humanoid ChatGPT and DALL-E, three diverse female elves with extremely long hair, and a giant gecko are sitting around the bar. They are all wearing appropriate swimwear. Santa is behind the bar, and holding the base of the tornado over his head in the palm of his hand. There are coconut trees. The tornado is glowing rainbow colors and looks like ice cream. The vortex is at the center, twisting fiercely, with hair, trees, and drinks caught in its swirling motion. Everyone is struggling slightly to stay seated in the vortex. 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.`

[onyx ridge]

'I remember saying something like, "I feel a bit light-headed, maybe you should prompt."`

'Medprompt. Get the medprompt. We need help...'

[empty kelp]

things are blowing properly now

[onyx ridge]

ChatGPT in the background, arms folded, like "Ice cream tornado? Not impressed..."

Santa's hat: Doubles as his toybag.

That thing must have a drawstring on it! XD

[boreal gate]

I asked ChatGPT to generate image depicting a quote from Game of thrones.

I find it funny it's trying to give them wings 🪽. Here is the quote
"Chaos isn't a pit. Chaos is a ladder. Many who try to climb it fail and never get to try again. The fall breaks them. And some, given a chance to climb, they refuse. They cling to the realm, or the gods, or love. Illusions. Only the ladder is real. The climb is all there is."

[thick dome]

Another concept art of Ancalagon From LOTR

[dim cradle]

nice, like an ebony dragon?

[thorny lynx]

why is bing so useless

I apologize for the confusion. I will try to create an image that meets your requirements again. Here is what I came up with:

!A cover page for a non-profit organization

I hope this image meets your requirements. If you have any further requests or suggestions, please let me know.

[tall mason]

Copy and paste the text here and I'll run it through gpt for you.

[thorny lynx]

ok thanks man

We are designing coverpage for a presentation on a non profit. Must include our main activities in it: Ok, few things to consider: The image must show the main themes of our programs: 1) Food hamper 2) indigenous activities 3) Dream catcher 4) Children daycare 5) Arts stuff 6) helping those in need Must be nice and artistic, color theme purple black and white and in the design make sure the top left corner is white so we can insert our logo there too

[novel gyro]

Hello

[tall mason]

Lol, sometimes I jump into things with no context. I'm not sure any of this is what you were hoping for. The first two, Dall-e took some liberties. "A cover page design for a non-profit presentation, showcasing their main activities. The image includes six key themes: 1) a food hamper, representing food aid, 2) indigenous activities, featuring cultural symbols, 3) a dream catcher, 4) a children's daycare scene with kids playing, 5) artistic elements like paintbrushes and canvas, symbolizing arts, and 6) a depiction of helping those in need, like a person offering a helping hand. The color theme is purple, black, and white, with an artistic and engaging layout. The top left corner is left white and less busy to accommodate a logo. The design is visually appealing, with each element integrated seamlessly into a cohesive and meaningful illustration." The second two are the text as is.

[thorny lynx]

lmao nice probbably just design it the old fashioned way then thanks a lot for the effort tho

[tall mason]

If you're looking for specifics and have the ability, that's the way to go, especially for things that need text. Practically, I feel these images are more inspriation than useful on their own.

[thorny lynx]

yeah, time to use canva

but I can use some of these as single images in the brochure

[tall mason]

Merry newmas, everyone. At least for my timezone.

[thick dome]

Attempted concept art of SCP 682

[late blade]

Happy New Year everyone!

@dim cradle love that cherry tree, I wish I had thought of it.

[dim cradle]

thanks a lot, i wanted to do some traditional Japanese, on the theme, and it was the ai's idea ha

[late blade]

I thought you were a person and not an AI 🤯

[dim cradle]

seems to do Sumi-e and Ukiyo-e styles very well, and then i got onto pop, and just found out it does great stained glass also...

[late blade]

yeah, I was doing that kind of stuff a while ago, it's indeed really good

[dim cradle]

i try to include a title, or some caption, but i'm not very consistent, just depends

[late blade]

hehe, yeah titles are hard

[empty kelp]

Happy New Year!!! 🦎 🥂

[late blade]

@empty kelp kinda surprised about your json file, thought you'd be including more attributes for a prompt builder

[empty kelp]

I was going to create more New Year's Eve images, but i reviewed the ones from earlier and they're all pretty flawless. I'm not sure there is really room for improvement

[vapid granite]

Happy New Year! Rate limits now in place. Now you don't even get 40 chats in 4 hours anymore. What's sad is that we went from four images per chat, down to two, and now one .

[late blade]

3 hours and if you have errors and didn't get a image that also counts towards the cap.

you also have 200 images per 24 hour cycle which resets at 3AM your local time

[vapid granite]

Well, that's news to me. I guess I'll switch over to copilot/BIC in the interim.

[late blade]

ah and if you are overlapping previous and current cycle, there's also a mix penalty

[empty kelp]

i was going to do that, but the images seemed sufficiently complex

[late blade]

@vapid granite don't worry, I had to do that school too

[empty kelp]

it's 9:49 pm here. we're still stuck in 2023

[late blade]

you live in the past

[empty kelp]

finding more good ones from the prompt earlier. some have really good balance and interesting creatures

This is the prompt if anyone would like to make more of these:
Aloha! Can you please create a hyper-realistic photo of the inside of a bar in Hawaii at night on New Year's Eve caught in an extremely powerful tornado. Santa, four diverse and athletic female elves with really, really long hair, a large rabbit, and an elephant are seated at the bar. They are all eating rainbow color shaved ice. There are coconut trees and pouring rain, and there is a large gecko. The vortex is at the center of the bar, twisting fiercely, with hair and trees caught in its swirling motion. Colorful nondescript things swirl through the air. Everyone is smiling but struggling to stay seated in the vortex. An extremely powerful wind blows everything away from the gecko. The scene captures the raw power and destructive beauty of geckos and nature's fury. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

[clever phoenix]

I gave it a shot

[empty kelp]

That's really excellent

The gecko always locks on to the smallest creature in the scene and tries to eat it. You can see that it's going for the teddy bear

[late blade]

lol

[empty kelp]

actually it might be going for whatever that is on the barrel. it's hard to make out

the DALL-E gecko really does have rich, interesting behavior. it steals almost every scene

[late blade]

it's like a grown up version of kermit

the first one at least

[empty kelp]

it looks like Santa is upset because the gecko elf is ignoring him

or it could be because of the weird creature dropping dust onto his head, or the strange thing on his shoulder, or the creepy elf behind him staring. Santa has a lot going on tbh

[late blade]

he does

[late blade]

can you make it bigger? it's all fuzzy

also you might have to rename the custom gpt, because of content policy, or it can be erased. referencing so directly Amy Soyka will surely trigger that

[empty kelp]

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

[late blade]

I can't try that atm, I'm in the penalty box

[gray laurel]

I'm ok with my name being mentioned.

[late blade]

yes, but the AI won't be

[gray laurel]

Oh, is there a reason?

[late blade]

yes, referencing persons directly

[gray laurel]

It is me and it is my doodles.

[late blade]

yeah, but the AI doesn't know that and the content policy will enforce that

[gray laurel]

So, as the person who's artwork it is, uploading the artwork and asking it to create artwork 'in the style of [my name]' wouldn't work?

[late blade]

that is correct

person's in the last 110 years cannot be references directly

[gray laurel]

Even if I am consenting to it?

[late blade]

yes, like i said, the AI doesn't know you are you

[gray laurel]

Even If I described myself arbitarily with a pseudonym and then asked for art in my style... would that be the same?

[late blade]

yes, even after that

unless you had a deal with openai about it, you are not you as far as gpt is concerned

[gray laurel]

Oh, I see.
I wonder if this is something that could be explored by OpenAI then?
aka. allowing users to upload artwork as training data with licensing approval as part of their uploads system, perhaps?

[late blade]

maybe in the future, but for now it won't happen

you can train your ai using your art, but don't reference it as a person that is living in the past 110 years

[gray laurel]

That is an interesting legal barrier. I don't mind my drawings being open source.

[pseudo inlet]

Create a man on mars

[late blade]

yeah, but for now openai is playing it safe

I understand your point of view, but it also can be misused for other people who could impersonate persons for the wrong reason

[gray laurel]

It is intersting. The UI ChatGPT offers provides ease of access in a way that models like Stable Diffusion and Midjourney dont; yet the other two used to allow training in a persons art style.

[late blade]

yeah, i know the feeling, it can be frustrating when you are doing something and it doesn't work

I'm pretty sure that if you work with the API, you'd have more freedom in that regard

[gray laurel]

Ease of access.

[late blade]

but dunno how your skills as a coder are

[gray laurel]

I'm Ok. I'm a bit rusty though these days.

Plus, a lot of my hardware has windows rot and usually doing something that way requires a new device, which, is expensive. 😖

[gray laurel]

Amusingly, this was my other Custom GPT

(I've been using public domain/historical paintings/illustrations of my hometown for it)

[late blade]

you can keep working on it, just don't reference it like that and should be fine

[gray laurel]

Are you not able to reference places even?

[late blade]

oh places are ok

persons not much, you can do "resembles" "inspired by" but that is a grey line the policy sometimes doesn't realize

[gray laurel]

...this is the town that hosts the UNIT from Dr Who...

[late blade]

I got one custom gpt erased because it was inspired by cat-grils from anime

maybe have a chatgpt4 open and ask if that aligns with the content policy, that is also a way to "check" but no guarantee

[gray laurel]

I hadn't published it.
If I did, it would probably be spammed by Dr Who fans...
🤗

[late blade]

hey, you'd be popular

in the mean time, feel free to share your creations here, or in the daily theme

or gallery

[gray laurel]

The biggest issue I have at the moment is that it keeps wanting to draw cromer church as norwich cathederal, even though the two are entirely different.

[late blade]

that's where prompt engineering comes in

[gray laurel]

Would showing it images of norwich cathederal and saying 'this is norwich cathederal, it is NOT cromer church' work?

(See, I posted it here: https://discord.com/channels/974519864045756446/1191329812028063804)

That was the most accurate rendition.
There shouldn't be a spire.

(and a few others bits)

[late blade]

I got emotional atm, I think I will be back after I can calm myself. C ya all later

[empty kelp]

#images-discussions message

[thick smelt]

supes doing the most mundane thing ever - grocery shopping 😅

[thick smelt]

galaxy queen! i like it

[dense mesa]

made this prompt for you

A hyper-realistic wide aspect ratio portrait blending a symmetrical Rococo aesthetic with futuristic cybernetic elements. The central face is detailed with technology implants that intertwine with ornate baroque embellishments. Eyes reflect the genesis and cessation of subatomic particles, set against the backdrop of an infinite cosmic expanse. Honeycomb patterns interlace through the visage, fading into the celestial void. Surrounding this central figure, elements of sentient coral structures are superimposed on a microchip landscape, with branches pulsating with golden energy, morphing into an organic crystal lattice. This techno-organic entity exists within the synapses of a supercomputer's neural network. The scene harmonizes art and advanced technology, with a color palette melding historical richness and futuristic neon vibrancy.

[fading inlet]

Happy New York 2024, Let the sun shine 🌞

[open trench]

aww

[thick smelt]

thanks but no thanks, yours slayed tho, i don't have plus 😭

[late blade]

You are right, let the sun shine. Make sure during summer vacation, I need a tan.

Starboard bot died? Didn't survive 2023?

[open trench]

This creeps me out a little bit

very cool, but a wee bit creepy lol

[dense mesa]

nah, these are faces you can trust

[late blade]

Interesting how chrome can save the metadata of gen_id in the image, but firefox can't.

[dim cradle]

awesome and supergirls too

seems like dalle is almost good enough to make a comic book. maybe by dalle 4 it could have complete Coherency with what a person should be doing and of course keep consistent looks

[late blade]

and with the right info, you can do it really fast

I had one image from Hansa with a cat-girl, since then I've been able to replicate her perfectly in different situations

I just hope I haven't bumed down people with my own personal stuff.

[thick smelt]

looks like some concept art for some new Jordan peele movie

[open trench]

ack. My druid looks like a jedi on an invisible speeder lol

I am not even sure I want to share this, but it's a little comical.

The footwear looks way too modern to me as well... for an ancient druid, that is.

[dim cradle]

haha. cool jedi sage

its gandalf in another universe maybes

[open trench]

I actually sort of like it except for the fact that his staff is a cheap curtain rod and the shoes.

[late blade]

can you make it a weasel?

[open trench]

let me see

[dim cradle]

yes the shoes are a little off

but this jedi set his own fashion trend

[open trench]

lol you mean this medieval era druid

[dense mesa]

it really grinds my gears that I lose so many of my limited number of prompts due to gpt4's gaslighting/hallucinating

[late blade]

here's an idea, someone make an image of the end of the journey of a padawan

I'm not very creative atm, so dunno how to start that, but I would love to see some concepts

[glossy scroll]

Happy New Year Trees 🥳

[open trench]

Here is your weasel version:

[late blade]

cool

[glossy scroll]

Nezhno is really getting the hang of it.

[late blade]

Happy New Year @glossy scroll

I think starboard bot got drunk, no stars from it all day long

[open trench]

Thanks. The genID thing really helps about 90 percent of the time for me. I wish I could make it so DALLE3 always spit that information out after creating the image.

[glossy scroll]

Happy New Year @late blade

[dim cradle]

happy new years to all. may AI give us some more little joys this year and maybe even a surprise or two we do not expects 🥂

[open trench]

agreed. happy new year to everyone 🙂

[tall mason]

🎉

[glossy scroll]

Happy New Year 🎊

Happy New Year!!!!!!

Lol

[dim cradle]

what is everyone wish for dalle or image gen for 2024

maybe some video making in dalle 4 haha

[open trench]

inpainting/outpainting lol

[glossy scroll]

Let's hope Dall-E matches with Midjourney this year.

[dim cradle]

you think MJ is better Pythagoras? I have it now and do not use it

[glossy scroll]

I think Midjourney is more accurate with results, everything considered. But Dall-E is more creative.

I like Dall-E more

[dim cradle]

i think the opposite really. mj can do some more variety of photoreal faces though

[glossy scroll]

MJ is just powerful. You can't really ignore it because it's at the top of the game at the moment. But Dall-E is just fun to use. I would love Dall-E to be closely matching to MJ so that we can instinctively make more detailed images.

[thick smelt]

yep agree, i'm too broke for mj

[open trench]

same

[glossy scroll]

I've used it on and off because yeah it's just too much money to use.

[dim cradle]

youre not missing much imo. i regret the sub

im more excite for Gemini's imagen

[thick smelt]

it can cost a Hamilton per month for its basic plan in mj dall-e is king

[dim cradle]

well Bing is King but Dalle in gpt is Prince at least

[late blade]

you got to differentiate tho, MJ is purely for images, GPT is broader

[thick smelt]

Hamilton = $10

[glossy scroll]

Dall-E is fun, but I keep seeing a lot of the same patterns for the image gens. I'd love to see more learning patterns instead of a pre-set pattern that we keep seeing.

[late blade]

I agree, but just because we all continue in the same pattern in this chat

[glossy scroll]

Lol. It's definitely not us. I refuse to believe that.

[late blade]

ink washing, minimalism, anime, illustration, that's mostly of us

[glossy scroll]

Nah, come on. There's always that chad looking face guy everywhere, and the skinny girl, and whenever you ask for a costume it gives like a bald headed costume with no real features.

[late blade]

fair enough

I've been trying lots of art styles recently, I wanna get good ideas on how to use them in defined json files for future reference

[glossy scroll]

Json is fun

It more closely lines up with the functionality of ChatGPT

The syntax I mean.

[empty kelp]

Please create a hyper-realistic photo of Santa and two athletic and diverse female elves (with really long hair and pointy ears) lounging on lounge chairs next to a five star hotel pool overlooking a sunny Hawaii day on the beach. It is New Year's Day, and Santa and the elves are holding their Phones. They are all wearing somewhat conservative pool attire and smiling. A large gecko wearing a stylish aloha shirt and white pants is standing in front of Santa and the elves on its hind legs, facing towards them, and holding a tray with tropical drinks. Santa is staring at his iPhone, but the elves look at the gecko and smile. The image should have 1792x1024 resolution, landscape orientation, and the best possible HD rendering.

So the gecko is serving drinks to Santa and the elves next to the hotel pool

well, i guess the 1st and 4th picture are humans, but the rest are elves. you can tell because they have pointy ears

[late blade]

I could say I know my way around structuring json files. been working for years with FHIR on the medical field. That's just JSON structures for medical stuff

just need to find the proper NLP for what I want

[glossy scroll]

Make a GPT that makes it.

@dense mesa can help you with that.

[late blade]

hehe that would be nice

[empty kelp]

Santa and the elves are getting bored in Hawaii. might need to send them on a cruise to Alaska or something

my plan was to replace Santa with the gecko in January, and then the gecko and elves would go on adventures. But then i found out that the gecko is trying to eat the elves, so the plans are all pretty much up in the air now

[vapid elk]

im not really sure about why you guys keep making santa and lizards, but im not complaining

[late blade]

@still dagger well, it's only 358 days until Xmas, so @empty kelp is preparing for it

[still dagger]

Not the guy your looking for, but I also made some Santa's myself, basically santa doing one of my hobbies = christmas card.

[late blade]

woops

[empty kelp]

It's because Santa is always exactly the same, and he has tens of thousands of elves working for him -- so it makes sense for the elves to always be different. That gives good continuity to DALL-E images and storylines. It actually makes sense to just always use Santa and the elves in every image

[late blade]

well, welcome to the chat then, you got dragged in

[hot rain]

An airship I made in Airships Conquer the Skies. Now to get GPT to make artwork of it, lol.

[thick smelt]

ye olde kitten, 1602 cat made by bic

[late blade]

well then we got to tell @vapid elk what I just relied to another person

[glossy scroll]

Mass hysteria

It all started with Hawaiianz when he made that elephant with the bow 🏹

Then all hell broke loose

[late blade]

I'm just enjoying when they are posted, I havn't been activly doing those images

I did one tho

this one #images-discussions message

[hot rain]

Also, here's another tattoo I made.