images-discussions

[dim cradle]

got some divergent results... maybe "photorealistic" and "surreal" together are conflicting.... let me read on...

[open trench]

Thank you.

[tidal whale]

Future roadmap definitely includes hosting stable diffusion. Too expensive to use the openai API on mass scale. Also hosting SF allows bg-removal and other configurations.

[dim cradle]

@shut niche this aligns with our recent discussion, doesn't it?

[open trench]

I've thought of just putting hyper-realistic, but I'm not sure dalle3 understands it as the surrealism movement.

[late blade]

hyper-realistic also implies realism art style

[dim cradle]

not sure it matters, since it's looking more like digital art than a photo anyway, but digital paintings work better for surrealism i think

[late blade]

@open trench use life like instead of hyper-realistic

[open trench]

ooooh, okay. i like.

[open trench]

true

[shut niche]

I'm trying to understand the idea.

Can you explain your idea separately of the prompt you wrote. I'm not 100% following your vision.

[open trench]

totally. I'm tired. so, it's hit and miss in my brain right now.

[late blade]

why is the poured cocoa making a chocolate flower on the ground?

[open trench]

Basically, the ladle is pouring hot chocolate, which represents nostalgia, onto a barren and desolate cracked earth where it freezes, the nostalgia becoming a cold distant memory viewed almost with disdain. Ice skates, which have many meanings, stand mostly for the subject of the nostalgia. The storm is coming to wipe away the memory. The broken hourglasses show time is running out. The sky is the Rorschach butterfly to imply there is something in the psyche making it happen.

[dim cradle]

that's deep man

[open trench]

it happen = the memory freeze and fade

[dim cradle]

it's like a koan

[late blade]

@empty kelp please make a gallery, don't post all variations on this channel, it makes it hard to follow other conversations because people have to scroll back past your bulk posts

[empty kelp]

this was just four of 100 variations. hehe

[dim cradle]

so by frozen... it's no longer in the liquid form, it's solid... i think we might want to skip ahead to that for the context

[open trench]

I never even considered specifying that it's frozen solid.

[dim cradle]

this is similar to the broken mug paradox

[shut niche]

Like this?

[dim cradle]

that still looks like molten cocao though

[shut niche]

Keep in mind it's surrealist, so every generation is going to vary greatly.

"Visualize a surreal, landscape-oriented image where a ladle gracefully floats above a stark, cracked earth, symbolizing a barren landscape of forgotten dreams. It pours a stream of molten hot chocolate, representing warm, cherished nostalgia which paradoxically solidifies upon touching the ground, morphing into a rigid sheet of ice that encapsulates distant memories now viewed through a lens of indifference. A pair of solitary ice skates stands atop the frozen surface, serving as the embodiment of the very essence of these recollections, isolated and static. Looming on the horizon, a formidable dust storm advances, ready to erase the remnants of the past, signifying the ephemeral nature of memory and sentiment. Scattered around the fringes of this poignant scene are remnants of shattered hourglasses, their sands lost to the winds of the impending tempest, underscoring the relentless passage of time. Above, the tumultuous sky forms the mirrored wings of a Rorschach butterfly, hinting at a deep-seated psychological catalyst for this scene of abandonment and change."

[dim cradle]

it's supposed to have cooled and turned into its solid state

[open trench]

Very close. The only reason the ladle is there, to be honest, is because it continued to pour from one mug into ianother.

That's just a well written prompt. Dang.

[shut niche]

Run it verbatim, and see what you get back. If I run much more, I'm going to hit my limit.

[open trench]

Okay. I am making one slight change from ladle to mug because that was the original idea. If it works, I'll be a happy camputer. Thank you.

[empty kelp]

i've found that the AI disregards about 80% of what I tell it, and prefers to put it's own spin on things (which usually works out pretty well). Like I asked it to have Santa and the elves sitting on the side of the cocoa mug, and so ChatGPT and DALL-E pondered this and decided to put them inside of the cocoa

[late blade]

you can also post multiple images in one message, that way you give others room to read and still see your work

[vapid elk]

I know you have the best intentions to report the issue, but please, don't post such content on this discord

use the chatgpt's built in feedback buttons

[empty kelp]

yes, you're absolutely right. i apologize for not putting them in the same message. i forgot about the grouping

[vapid elk]

if you have multiple images to post, consider using the #1154829862171844679
you can link the message to other channels, no problem

[late blade]

the idea is not to introduce bias to the AI, adding concepts or ideas that align towards philosophies or political ideals and reasoning is a very dangerous line to cross and keep between what's good and wrong and what's just too much so that an AI's bias is not balanced

anyway back to dall-e'ing

[shut niche]

I'm waiting with baited breath. 🙂 Did it work?

[late blade]

can i coin the verb "to dall-e"? you dall-e, I dall-e, we dall-e, everyone dall-es, keep dall-e'ing!!

[open trench]

almost.

the top half of the first image is perfect and the bottom half of the second image is perfect

[dim cradle]

splice 'em together and you're done 🙂

[open trench]

i know, right? lol

[late blade]

try cup pouring on the ground

[shut niche]

There's "sort of" two approaches to surrealism with DallE
1., Use the word surrealism or surrealist, and describe an abstract idea, and let DallE take it away. But this will give you a wider variance of results.
2., Don't describe it as surreal, and instead, describe each and every object you want to see, as you want to see it. Abstain from abstract ideas, and lean towards speaking in absolutes... "Matter of fact". E.g., This is here, that is there. They are this...

You can also break complex ideas into numbered bullets.

[open trench]

numbered bullets in my image prompt??? be still my heart.

[late blade]

yes, you can do that too

I usually do paragraphs that start with "-." or numbered with "1." when I want concrete definitions of something

and object here and thenn the numeration when there's something that needs more details

I mean double * when I want to do object here

[open trench]

I think I follow you

I'm almost braindead this morning. lol It's been a long night.

[late blade]

woah that one turned good

[vapid elk]

not enough coffee

[empty kelp]

Santa and the elves are carefully examining the scene to make sure it's correct

[vapid elk]

[open trench]

yeah, it left out half of the stuff, but the Rorschach storm was too good to pass on.

[vapid elk]

this reminded me of something

I made with dalle2, let me grab it

I'll try to recreate with dalle3

[late blade]

i feel you, hece this image for today's theme #daily-theme message

[open trench]

I'm going to have to apply what I have learned tonight to my ongoing gallery post. I haven't finished the invasion, but I am stuck on a particular scene.

[dim cradle]

clever

[vapid elk]

(added the words manually)

I'l ltry to make dalle3 add the words

[open trench]

I like

[vapid elk]

good

[late blade]

are you using your own custom gpt?

[vapid elk]

nope, just used the default dalle custom GPT, in this case I didn't tried to re create the image, I specifically asked it to make it in the style of The Persistence of Memory

[late blade]

oh nice

[vapid elk]

but I often use my own bot which uses the API

it has a few features to enhance the prompt
this was not the case, it was chatgpt

[grizzled loom]

someone inquired yesterday, so i will share some tips again: ornate verbiage is huge: beautiful?, no you mean oozing ornate beauty
if you want text, encapsulation is better in single ' ' not double " "
dalle chews up the strongest of language more readily

[vapid elk]

are you sure " is any better than ' ?

[grizzled loom]

single not double

this > ' is superior for txt than this> "

[late blade]

depends on the langauge I guess, my gpt takes mostly swiss punctuation instead of american punctuation, so over here " doesn't mean much

[grizzled loom]

again, just tips

[grizzled loom]

you wrote it backwards.

[vapid elk]

I got a very surreal one with my bot =P

[late blade]

ok time to be creative again, so that means nap time

[vapid elk]

is any style of quoting any betetr than another??

[grizzled loom]

absolutley

tried and true

[vapid elk]

interesting

[late blade]

Using single quotation marks (') instead of double quotation marks (") doesn't necessarily indicate higher importance. In English punctuation, single quotation marks are typically used for quoting within a quote or for emphasizing specific words or phrases. If you want to emphasize the word "stop" in a sentence, you can still use single quotation marks like this:

"Please, 'stop' what you're doing immediately!"

In this case, the single quotation marks draw attention to the word "stop," indicating its importance within the sentence. However, it's essential to note that the choice between single and double quotation marks doesn't inherently change the level of importance; it's more about the specific context in which they are used.

[fallow quail]

if i want to run something verbatim i use "prompt" and if i want a text included in my picture i use 'text' works like a charm

i prompt in english. i think this might be different if you have the generating prompt in a different language. but in my experience prompts in other languages that are sent to dalle are not as versatile.

but you can specify in an other language, just be sure to tell gpt to prompt in english

[late blade]

yeap, but my custom instructions are still vague on that. I said where I'm located and what kind of punctuatio I want to use, but sometimes it doesn't work.

[subtle creek]

Anyone else having problems rn?

[late blade]

kinda, just getting one image instead of two per prompt

[hot rain]

The shockingly good Perfect Cell I got it to make compared to the depiction of him I used. GPT got wise to me after though and refused to make anymore, lol.

[boreal gate]

Why you should go incognito more often

[hot rain]

Using the description for Stannis Baratheon without using his name.

"Make an image of a king with dark blue eyes and a heavy brow. His head has only a fringe of thinning black hair "like the shadow of a crown", and he has a close-cropped beard across his large jaw. His face has a tightness to it like cured leather, and he has hollow cheeks, and thin, pale lips."

[hoary zephyr]

Ragnarock Esport logo create

[hot rain]

"Make an image of a smith dressed in black clothing at a frigid castle. He has a big belly and a pinned-up sleeve. His jaws bristle with black stubble. He is muscular and strong, even with one hand missing. His nose is broad and flat. Make in the style of a graphic novel "

No amputation, but otherwise a pretty great Donal Noye

[lavish flint]

#openai-verification

[hot rain]

It's super great with the facial features, but it hates the bald and amputees I guess lol.

[shut niche]

That first one was playing with your prompt. Here's one I made. I'd like to see if I can add the receding hairline.

[hot rain]

This is the best balding guy I've gotten so far.

[shut niche]

Here's the other gen.

[late blade]

@shut niche this image #daily-theme message has similar swirls as some of my pics, even tho you describe them in a different way than I do. wonder if there's a rooted concept of colorful swirls

[open trench]

I hate it when it looks like I have to make a choice. Will people be running in the background or will my character have all the details I want? lol I can't seem to have both.

I settled lol

[late blade]

do share the tips

[grizzled loom]

Todays pro tip from Mysticmarks: requesting scenes in staged layers with individual descriptions and telling it you want it all as a single composite image cohesively can achieve some very fun effects.

[empty kelp]

That's very good advice. I've also started to incorporate a huge ladle into my imagery. I believe this helps to convey the hidden symbolism more effectively

[hot rain]

Probably the funniest AI text I've seen, lol.

[empty kelp]

going to try combining these into a single composite image

[open trench]

This literally made me laugh out loud.

[green pebble]

Love this

I asked my bot to take the weather outside my house today as the prompt 😂

[empty kelp]

making a composite works quite well

[late blade]

🤔 💡 ‼️

after christmas ends, I will have no energy for at least 12 months to do the same thing again

[empty kelp]

I just told DALL-E to add another circle around the outside that says, "Mahalo Nui (this means TYVM in Hawaiian), Mysticmarks!!!" -- and this is the image it created

[dim cradle]

The scientist's magnum opus: a quest to diminish his existence to an atomic scale. Tragically, he succumbed in the pursuit, a martyr to the altar of groundbreaking experimentation.

[green pebble]

#daily-theme message Ok I think this was execeptionally well done for the prompt it was

[covert anvil]

Man I really wish I could use that

[green pebble]

That bot I used to make the image or the calendar itself?

[covert anvil]

The dall e 3

It's not free if I remember correctly

[green pebble]

Technically bing search it is

[chilly onyx]

You could use the model via Bing

[covert anvil]

Dope art you got there however

[covert anvil]

Too slow and I don't like the boost whole credit idea

[chilly onyx]

@dim cradle you got me excited earlier with your UV chill out rooms

[green pebble]

Very nice

[covert anvil]

Wow. Man AI gonna take over the world someday 😂

[chilly onyx]

Made a GPT that I’m still tweaking. At the moment it’s still displaying elements of the setting such as a rail holding the backdrop etc

[green pebble]

I can only hope

[chilly onyx]

It certainly is

https://chat.openai.com/g/g-AmHRz9Ww2-uv-backdrop-artisan

[green pebble]

Trying it now

[covert anvil]

Got a question. Why does that calendar look like something I'd find in a church?
(I'm Not Christian so I don't know if this is an Offensive question)

[green pebble]

Medieval + church is something that went hand in hand back then

[dim cradle]

they're rendering really well, that makes me want to explore those compositions, the ones i shared were the first gens--i suspected it was worth revisiting, and yours prove that

[covert anvil]

That explains a lot. I used mid journey for a while and whenever I said medieval it sent me something like that

[green pebble]

These AI bots refrence their databases all the time. To get it to not do that you have to override it with descriptions of your own

@chilly onyx I asked it to upgrade it since you were so kind to have the bot provide the gen ID

It didn't convert to 16:9 well, but it did uptick the details to the max

[chilly onyx]

Nice 🙂

[covert anvil]

Now if someone made an art with an AI. Does it belong to the creator?

And does he have the right to sell it if somehow he managed to ?

I mean AI can do art for digital artists who could sell it as theirs..

[chilly onyx]

I focused fairly strongly on keeping the artwork semi-simple to avoid it creating very fine detailing and preserve the hand painted look of those kinds of backdrops

[green pebble]

Makes sense

[late blade]

@empty kelp you should be doing the 12 month santa and elves calendar

[empty kelp]

i just looked at the daily to see what you were talking about. will start on calendar now

[true mural]

legally, morally, ethically? It's complicated, OAI grants rights and responsabilities to the user. Copyright violation remains copyright violation no matter how you achieve it.

[covert anvil]

So the answer is no

[glossy scroll]

@crude surge Check out my experimental GPT, for comic book style creating characters. This is the same GPT you inspired me to make.
https://chat.openai.com/g/g-BRrDdplIo-character-creator

[true mural]

If the answer was "no" I would have said "no". Not sure how you came to that conclusion, but if that's your interpretation then probably for the best.

[glossy scroll]

Spider-Phoenix

[late blade]

it's like an amalgation of jean grey and spiderman sponsored by coca-cola

[covert anvil]

I'm not a creator. But the whole copyright thing is difficult to understand

[true mural]

That is correct, copyright is hard to understand, I recommend reviewing your local laws.

[glossy scroll]

Spider-Phoenix does exist in the comics

Just not many interpretations of it.

[covert anvil]

If it's about the local law. So it's safe here as long as you don't do anything "out of pocket" with your imagination

[dim cradle]

this is by far my favorite daily theme so far

[true mural]

For further legal questions, I would recommend consulting with a lawyer.

[late blade]

yeah I can see that, but never had a comic or content to see one of the many interpretations myself

[glossy scroll]

oh really?

[covert anvil]

Are you an admin here?
(Just askin' because of the badge)@mx.vanya

[glossy scroll]

you never had a comic?

[dim cradle]

mmmhmm, so much history

[covert anvil]

How does that work

[glossy scroll]

yesss....quite...history...

[late blade]

@covert anvil all images you do with dall-e, as per policy, are yours, and what you do with them is also for you to do. But if you can register them is still a question. I think under derivative works it can go well.

[late blade]

comics yes, spider phoenix no

[glossy scroll]

You just open it up and type in what kind of character you want, like "Wolverine inspired battle mage with guns"

[true mural]

No, but I've been around long enough. If you want legal answers, you need a lawyer, not permission from someone on the internet. You have liability if you choose to publish anything generated by AI, so do your due diligence if you wish to publish.

[glossy scroll]

Ohh ohhh!! Makes sense. It's an oldie comic

[covert anvil]

Ok. Thanks for the help however

[glossy scroll]

well it's my least favorite I would say. I think my favorite was the dragon daily theme, that was really fun and cool.

[dim cradle]

color outside the lines a little, you could still do dragons

[late blade]

thing about using AI generated image and rules for it is a very open question atm. People need to start having that talk.

[glossy scroll]

"Our next daily theme....shorts...yes shorts, like the ones you wear at summer time!"

[glossy scroll]

mmmeah...

[late blade]

Shorts, boxers, speedos... or do you mean small people short tiny people?

[glossy scroll]

I'm just trying to think of the most boring bland things possible

[late blade]

oatmeal themed day

there, that is boring and really bland

[glossy scroll]

That's still fun!

I like oatmeal

[late blade]

I've never heard someone saying I'm having lots of fun eating a bowl of oatmeal.

[glossy scroll]

Yeah, especially with some cinnamon and little bit of melted yogurt ice cream....mmmmm

[late blade]

lol

[glossy scroll]

Especially with some dinosauers too

That makes it delicious as well

See that's not boring!

It's actually cute nice

[late blade]

never question my art style!!!

[true mural]

Absolutely. Some hard boundaries currently are: "copyright violation is copyright violation whether it's a photograph, photocopy, painting or AI generated", OpenAI grants full rights and responsibilities to the prompter provided they abide by the rest of the terms and conditions.

As for what is derivative vs transformative? Ask a lawyer, I am not one. And not even lawyers always know the answer.

[glossy scroll]

Yes, I like "oatmea!"

I hungy for some OatMea!

[late blade]

lol

[glossy scroll]

I think that's how boston people say oatmeal

Not too sure though

[glossy scroll]

But there is another argument that I read a week ago about this, and it's the data that is collected to generate images is taken without permission.

[late blade]

there it doesn't get better than that

and yes you were right, oatmeal is fun

[glossy scroll]

heheh, told ya

[true mural]

That is certainly a consideration, and could affect the work that OAI created [e.g. dall-e itself], but if I use a pirated copy of photoshop, it doesn't make my image a violation of copyright, it makes the tool a violation of copyright. You can analyze the work itself. However, what it does mean is that there is significant reason to be suspicious of an AI generated work, you could be violating copyright without realizing it.

[late blade]

that is true, but it's data that it's out there, at some point AIs will have access to that data regardless

[true mural]

Maybe, maybe not, what I'm stating is more simple, copyright violation isn't a spiritual essence that is passed on by philosophical connections. Copyright violation is based on specific articulable elements in the work being claimed.

[late blade]

ya that is true

thing is we are talking about AIs that we have atm, where a content policy is in place, but that doesn't mean there won't be AIs without those restrictions.

[true mural]

I think the philosophy and ethics of AI training is an important discussion. It's just a different convo from "do I have the right to use". Like FBs new image generator is trained on things they have the right to, due to ToS. except their users definitely ignored copyright, so didn't have the legal capacity to agree to such assignment. Is FB in the wrong? IDK.

[true mural]

The same rule will always apply to end users. The AI creator can say "do whatever you want with the mickey mouse image, it's yours" and that's true, that is yours, but if you publish it (for sale), expect disney to knock.

[late blade]

well the first versions of mickey will be in the public domain next year, so that won't be a problem. but while we do need to have this talks with AI's we are accessing, that is just one of many aspects that have to go in parallel, sadly not everyone wants to do ethical AI's

[true mural]

Sure, and we will definitely be seeing a lot of churn in laws soon.

[late blade]

ya

as long as they apply where you live, you'll have to abide by them

anyway back to dall-e'ing

just gonna do images for 12 months and leave it at it, not gonna go deep with today's topic, just chill and relaxing nothing to do day

[true mural]

If something comes to mind, I'll submit for today, still thinking.

[late blade]

prompt: an image wide format about thinking about what to do with: "Technique Tuesdays :art: Medieval Calendar Illustrations (1200-1500) - These images, popular in the Middle Ages, showed activities associated with different months, including winter scenes." Very dramatic vivid scene.

[true mural]

I will point out though that you could have the least ethical, criminal art generator, and still produce legally distinct works. And you could have the purist generator trained entirely on ethically sourced material, even start it from absolutely scratch with zero prexisting images (not feasible, but bare with me), and still violate copyright. For people using the tool, it's the end result that determines that.

[late blade]

to be honest, i've never been able to be creative with normal means because of traits I got, AI opened for me a new chapter about creativity

damn, I wanted to post the image of the prompt i gave you, but that one turnd soooo good, I'm gonna post it in the daily theme channel

[open trench]

I'm having trouble coming up with ideas for it. I have a few, but I can't seem to solidify the concepts. I'd talk to ChatGPT, but I don't want to waste the responses lol. I feel so stifled by all these limits. 😛

[glossy scroll]

But the thing is that the AI companies who use data do not ask for consent to the use of copyrighted material as training material. That's the real problem.

[open trench]

(sorry for the late reply, I got distracted and meant to type that earlier)

[true mural]

I don't disagree that that is a problem (whether legal or ethical). It just doesn't change the copyright status of an output.

[glossy scroll]

What do you mean it doesn't change the copyright status of an output?

[hearty ether]

does anyone know if they will eventually make seed and image id cross chat?

[glossy scroll]

They have, but they ditched it. They may bring it back in the future.

[hearty ether]

why did they ditch it?

[late blade]

Vanya is going to the output of the resulting image and the legal status of that output. Regardless of which AI makes it.

[glossy scroll]

Because they're trying to incorporate language model seeding, instead of number seeding.

Like, "Remember this picture."

[true mural]

I mean, either a work is a violation of copyright or it is not. A non-infringing work doesn't become a violation of copyright because it was created with a tool that violated copyright. The reverse however is an interesting question, "an otherwise infringing work, made without any knowledge of the original". In that case, the dubious nature of an art model would not help your case.

e.g. "I didn't mean to copy mickey mouse, it just happened" -> "The model you used is known to produce copyright infringing work, that's no excuse"

[late blade]

I'm just drained, after 8PM i don't care about laws, news, work, personal hygiene, the moon, raccoons, the world, and many more....

[dim cradle]

i view it as a chance to learn more about medieval styles. as far as i can tell, during that time most illustrations were simple black & white ink drawings. some just a tad more detail, and some had watercolor. one i posted in the daily is Quattrocento-style from the 16th century, when they moved towards realism, contrapposto, chiaroscuro, high attention to detail, rich colors, even balance, etc., which gives you a lot more for your composition. i intend to look into more styles, and fiddle around with the more primitive illustrations also, but i have more homework to do yet....

[dense mesa]

that disappearing drawing was kind of cool

they look trustworthy

[late blade]

throw a dice to the air and try to land a dart on it, the number you hit you get the same amount of cookies, smash them to the wall, the big chunks that remain, you take 1500 and subtract that number, then ask chatgpt 3.5 who was famous in that year, and from there you make super heroes and villains based on that personality

[robust yoke]

so, for some reason, i cant make some characters with black skin, like, is super difficult, is that normal?

[glossy scroll]

That's totally fine. But here's another thing: Should these tools be allowed to copy any of the martial, accidental or not, given that they might be sold to someone else as original work. For example, if I wrote a book, and I used specific ideas in that book and your AI came and took that specific content without consent and used it to allow people to generate ideas based on my ideas without asking permission, is that allowed? No. Because I never gave consent for you to take the information out of my book to be used for your product. OpenAI is selling information, and selling is the entire part of why copyrighted laws are put in place. You can't take other people's product or information and repackage them in a new box.

[hot rain]

My usual prompt but in a 90's anime style.

[dense mesa]

I made this guy with black skin one time

[dim cradle]

from 16-th century calendar illustrations. pretty dissimilar from the great artist's masterpieces we're accustom to viewing.

[dense mesa]

that's like some chaucer sketches or something

[late blade]

ask gpt how to describe an afro-descendant person and from there go back to your image

[robust yoke]

like, i only did once, because is super difficult for it to make it, like, i was trying to make a character with black skin, like "pitch black", instead it gave me a character with white skin

[dense mesa]

hmm. let me try that out

[true mural]

I mean, technically, you can, provided it's sufficiently transformative. But, I agree, it's a very fraught area. And a veritable minefield for anyone wanting to use a generated work.

Which is why I said in essence "if you take liability as a no, then you probably should just go with that"

[dim cradle]

i dunno if i'm trying to replicate it or reimagine it, probably both

[robust yoke]

i was using as an example the blue moon vampire from vanitas no carte, and the dall-e made the opposite

[dense mesa]

I like the idea of emulating classical or ancient styles and then mixing them with modern styles

[hot rain]

I love the battle damage on her armor here, a strike on the right side of her breastplate that was so hard it partially fractured it. And it happened long enough ago that rust is forming around the gouge and cracks.

[late blade]

and she doesn't have money to buy a new one

[hot rain]

Military issue in an authoritarian, post-apocalyptic regime in a state of total war. It's what she's got lol.

[late blade]

are you saying you are entering my reign and supremacy? you are entering a dystopia? !!!!

[dim cradle]

Classical influence is a characteristic of Quattrocento style: "As the Renaissance was dawning, artists were increasingly inspired by the art and philosophy of ancient Greece and Rome. Classical themes, motifs, and proportions began to appear in their illustrations." I was just reading that.

[dense mesa]

yeah, definitely an approach that's been used for quite a while

max vaporwave activated

[dense mesa]

Create an image of a character with skin that is literally black, like coal or the night sky without stars. The character is tall and slender, with short, silver hair and piercing blue eyes. They are dressed in futuristic, sleek clothing, predominantly in shades of grey and silver, complementing their unique skin tone. The character is standing in a dimly lit, modern cityscape at night, with the city lights reflecting subtly off their skin. Their pose is confident, and their expression is thoughtful, gazing into the distance. Add a few subtle, glowing blue lines running across their skin, resembling circuitry, to hint at a sci-fi element.

seems like it's going in the right direction there

[hot rain]

In my story she's basically forcibly conscripted into their military and dragged from Australia to Oregon before being shipped off to fight a war in the post-apocalyptic no mans land between The Sovereign Republic of Oregon, and the Eastern Republic.

[late blade]

nice

authoritarian apocalyptic dystopia

[glossy scroll]

Saying 'you can' and 'you can't' is indeed a minefield, because for example, the internet itself is free, but paying to use the internet is a service provided by internet carriers. Search engines are just using content from websites and putting them out in the open. They're not breaking any laws because the internet itself is a free to roam space. But when you have AI companies who take information from the internet to sell them as their own product, that crosses too many boundaries for what is allowed. Bing provides images of Batman, but doesn't sell images of Batman to you. That's public use. You can look, but you can't sell. A.I. companies take images of Batman, repurpose their likeness in any way the user see's fit, and they sell you their services using a monthly subscription. They are essentially taking any product and repurposing them and selling that to the user. That doesn't become public-use anymore. That infringes on the product, and infringes on the proper rights and ownership, especially if the information was used without the consent of the owners/authors.

[dense mesa]

perfect

flawless wolfptopus or octowolf

[empty kelp]

#daily-theme message

[late blade]

oh wait that's on the daily channel, got to put the trophy there, not here

[empty kelp]

i did. this one is just for convenience so you can study it in both channels

it's a bit hard to tell, but Santa is playing video games in Occtumbeir in the medieval calendar

[late blade]

he seem's more like revewing the naughty kids list

and december...the one month he actually works... he's getting drunk...

[dense mesa]

makes me wonder how a horse would respond if you tried to compel it to walk into the ocean, lol

[empty kelp]

i had some trouble generating calendars. DALL-E was putting in some really disturbing stuff

[late blade]

lol

[dim cradle]

Looks like Santa has a ripped evil twin in one of the panels.

[glossy scroll]

I'm not gonna do today's daily theme, I refuse.

Didn't we already do this calender thing like two weeks ago?

I could have sworn we did

[loud field]

excuse me, but

why does this prompt violate guidelines:

"Captain Jakob Flintlock, the ugly fierce pirate. He is a tall man, imposing but lean, with a sickly and ugly look, with a sinewy body. He has dirty, greasy thin long blond hair with a beginning of baldness, balding head. A greying thin beard. The sunscorched sickly skin of someone spending life at sea. His gaze is menacing and sly."

[glossy scroll]

Because it probably thinks you're trying to rip off Captain Jack Sparrow. Using Captain Jak...

[dim cradle]

I interpret it as an illustration that depicts a season anywhere at the dawn of the Renaissance, it can be fun lol

[green pebble]

This is amazing. What I have yet to still achieve is consistency with subject matter to be able to create panels of storytelling through dalle images 😦 I want to be able to make a whole anime, even if thats just a comic to start

[glossy scroll]

Lol. Meh

[loud field]

fr? lets try without the name

[glossy scroll]

Lmk

[loud field]

nope

still not working

is it the sickly part? ugh

[glossy scroll]

Open a new window

[glossy scroll]

Yeah, tweak it

[late blade]

are you doing it with BIC or GPT?

[loud field]

ugh

[late blade]

if you are doing it with GPT ask what is the problem

[loud field]

no it's bing

I tried to remove the name, sickly, and ugly

I remove more and he might as well be another character

[green pebble]

The composition of this color pallete is awesome 🙂

[glossy scroll]

You know what I just thought of that's really weird. How is it that it knows how to interpret names from name brands pretty much perfectly in writing, but it has trouble putting words together. How does that make sense?

[late blade]

lol

[green pebble]

Because there are more images of a name brand then there is of just a word

[late blade]

yeah, I have a setting for that kind of image

[green pebble]

I mean you can just upload custom instructions with a font list to give it style and letters and then have it access those letters from refrence data to use in the image you are creating

[glossy scroll]

But it knows the letters. How is it reproducing the letters so perfectly when you say Ikea and coca cola but doesn't know any other word. Isn't it just putting letters together??

[green pebble]

Online fonts and such

If you search up NIKE

[empty kelp]

i figured out a way to have good consistency in in DALL-E images that gives great continuity for storytelling. if you just make every scene so it just has Santa, and "a diverse group of athletic female elves", and a huge evil rabbit, and an extremely enormous and terrifying elephant --- you can pretty much just use those character descriptions in every scene and it draws them consistently

[green pebble]

This teaches it style and letter placement based on database refrence

That's why if you ask for a dragon it does japennese style to begin unless told other wise because of the prodominant amount of training data that leans that way

So you can specifiy to give it letter placement through your prompt

[glossy scroll]

But training data doesn't work that way. It's not how many images it has in its data storage. It's how much the training is used to create brand name products. The AI is trained with frequent use, not by having multiple sources.

[dim cradle]

@glossy scroll 1, just 1, you know you wanna

but if you don't think you'd make it into the hall of fame... i understand

[green pebble]

My prompt: Please make a motivational sign that says "You got this" with an emoji on it

Adjusted by dalle: A motivational sign in a pixel art style, featuring the phrase "You got this" with an emoji. The sign is designed in a vibrant and uplifting color palette, with the text prominently displayed in an engaging and readable font. The emoji is cheerful and encouraging, complementing the positive message of the sign. The background of the sign is simple yet eye-catching, ensuring that the focus remains on the motivational message. The overall design is friendly and inspiring, suitable for a wide audience looking for a boost of confidence. The sign is created with attention to detail, using techniques like dithering, outlining, anti-aliasing, color palette limitation, manual pixel placement, and balanced composition to create a visually appealing and motivating artwork.

[glossy scroll]

Say what now? Lol. I missed what you said.

[dim cradle]

daily yo

[glossy scroll]

Never. I refuse.

[dim cradle]

you're immune to jedi mind tricks eh

[glossy scroll]

Did you succeed?

[green pebble]

No, because I wasn't specific, and dalle had to adjust the prompt for me

[glossy scroll]

Most of the time

[dim cradle]

plans to create a gallery for a new series with a focus on the atom

also working on blending an atom and a solar system, or at least juxtaposing them....

little microcosm/macrocosm action

[glossy scroll]

I think you were pretty specific.

[green pebble]

No no no lol. This is a specific prompt: A highly detailed pixel art scene in a 16:9 aspect ratio, depicting a troll swinging its massive barbed club at a small armored woman warrior. The troll is large and imposing, with a fierce expression, emphasizing its deadly nature. The woman warrior, although small in comparison, is in a great battle stance, showcasing her bravery and skill. Speed lines are added around the club to convey the velocity and force of the swing. The background reflects the dread of the impending impact, possibly with dark blurred out mountains, ominous colors and elements that enhance the sense of danger. The warrior's armor is intricately designed, highlighting her readiness for battle. The scene captures the intense moment of confrontation, with a focus on the action, emotion, and tension between the troll and the warrior. Techniques such as dithering, outlining, anti-aliasing, color palette limitation, manual pixel placement, layering, shading, emotion, composition, mood, stylization, texture, perspective, foreshortening, balance and weight, rhythm and movement, narrative, and use of space are used to create a vivid, detailed artwork, portraying this dramatic battle scene.

[green pebble]

With this as the output

Sadly the troll isn't holding the club, so it is only a 90% accuracte image, but at least something I could tweak with the gen ID to reallt nail down the illustration

Using gen ID to tweak it\

Converted in my realism bot also

[glossy scroll]

Ac3d, Dinosaur Oatmeal and I discussed this issue a little while ago. If you give it long and detailed prompts, it starts to hallucinate and gives worse outputs. You have to use concise wording when making prompts.

Concise and sagacious words.

[green pebble]

Right, I respect that is accurate. Through gen ID tweaking I was able to get to a very good output

I do agree concise is the best way to make the image generation perform at maximum instruction

That's why I say "speed lines and blur" when I want something speedy

[glossy scroll]

Yeah, concise is the way to go. There should be a concise word generator built into the GPT itself to shorten your words to give better outputs.

I can't tell you how many times Dall-E hallucinated when I tried to complicate it's output prompt.

Like didn't even show me the picture I wanted. It showed some random nonsense, lol

[dim cradle]

right now i use gpt-3.5 a lot for that and other related stuff in the workflow--the process of constructing the composition sometimes requires distilling it all the way down to its central focus and then rebuilding.

also it wants to craft an art prompt that describes a process in action (possibly due to my steering), and that's not always the best approach. it also sometimes waxes a tad philosophical which can be imprecise--it's great at crafting but not perfect--for frames at least. still, it's a tool in the chain, 'till caps go up.

[glossy scroll]

That's a good approach. I do that too. Use 3.5 to feed into 4. Because it's free and why not?

[dim cradle]

(so i edit it quite a bit, looking for ways to make it more precise and direct, filtering out noise, etc.)

[glossy scroll]

Yeah, trying to get those 3.5 inaccuracies out.

But you know what would help? Making Custom GPTs with 3.5

That would help a ton with navigating free prompts.

[empty kelp]

You can ask "Can you put Santa and the elves onto the deck of a space ship?" -- and it just puts them right onto the deck of the space ship

[shadow grail]

those aren't elves, those are Vulcans!

[empty kelp]

i'm not sure why the deck of the space ship has desk chairs on wheels, but the rest of it looks fairly legit

[glossy scroll]

I like this passenger. Her face, lol

[dim cradle]

yup, refining focus, optimizing instructions for the composition--and often I want to deconstruct it to see the layering, or go in a completely different direction.

[empty kelp]

those aren't passengers. they're the crew of Santa's space ship

[glossy scroll]

Oh sorry, crew

[dim cradle]

i think a couple times i've used it to design and generate the specs to paste into gpt builder, reducing steps in gpt-4

it understands what i mean when i say "custom gpt" and the specs suffice

[glossy scroll]

Yeah it's good to use it. But would be farrrrr more beneficial if we could use it to make our own free custom gpts with 3.5 benefits. Just cuz I could make a word condenser as input buffer for GPT4.

So I don't have to go through the hassle of trying to explain to 3.5 what I'm trying to do every single time.

[dim cradle]

but selecting a focus object, if there is one--at least when i started looking at it that way, it started to make more sense as a compositional framework. whatever its spatial acuity happens to be.

[empty kelp]

it must be pulling the space ship hairstyles from like the original 1960's Star Trek

[glossy scroll]

How?

[dim cradle]

you're doing it already--specifying it in the initial instruction. like the cup of hot chocolate, you express that as the focal point.

[dim cradle]

unless for one case, the focus was on the atom, and later it is mentioned it's positioned over a cup of cocoa, because the atom became the focus.

[glossy scroll]

Focalize

[dim cradle]

it's optional, depending on how many instructions about an object you intend to throw at it. yeah, and prioritize.

[empty kelp]

the hot cocoa and the storm

[glossy scroll]

We should make a GPT together

Call it: Shonagoras GPT

It will put words together perfectly

Making the perfect made-up word

[glossy scroll]

No, but really, if the made up word makes sense, the A.I. will understand it and accept it as legit.

@dim cradle It would condense words into smaller made up words that take from long and complicated explanations and make them into two words and combine those two words together, like "Cartographocondensation"

Shon, we're getting ideas here.

Ideas are flowin!!

"Priorophotofocalization"

[open trench]

Dalle3 is going to reach through the screen and strangle me, I think. I think I'm getting way too complex.

[dim cradle]

I have thought about it, great use case, we should talk

[open trench]

Just asking it to do too many different things.

[dim cradle]

And it should help guide the user in creating their composition

[empty kelp]

these spatial concerns are very important. precise positioning of elves in a cup of hot chocolate isn't trivial

[dim cradle]

It would be really useful but eats quota

[glossy scroll]

Yes, and it should focus on making things concise, sagacious, and simple, but not overly simplistic. Complex, but with small prompts.

It should focus on focusing on how to focus

[empty kelp]

yes, that will help us get much tighter control of things

[glossy scroll]

You and your pervy santa dreams

[empty kelp]

it was DALL-E, not me

[glossy scroll]

Tell that to the judge

[dim cradle]

technically, if you gave them names, personalities, etc., it's subjectification, not objectification.

[empty kelp]

this is actually the AI's interpretation of "a diverse group of athletic elves at the beach"

[glossy scroll]

Athletes and Baywatch have their meanings meshed together

[dim cradle]

it's a family photo, look, there's a baby even

[glossy scroll]

It's a pervy baby

[empty kelp]

if you say, "the elves should be wearing appropriate beach attire" and you put them in Hawaii, then this is just what they're going to be wearing, because this is actually what people wear on the beach here

[dim cradle]

you could make a creepy photo custom gpt lol

[dense mesa]

man, I really like the aesthetic of old photographs with unsettling elements

sometimes it can be a little tricky to compel gpt4/dalle3 to resist the urge to make them clean and polished

[empty kelp]

i also didn't tell Santa to dip the elves in hot chocolate or put whipped cream on their heads. the AI came up with this

[dense mesa]

also superheros in their off time. although not sure what this guy is going for

[fallow quail]

"The photo should exhibit a grainy texture, and the edges should show signs of wear, as though the photo has aged over time, with slight imperfections that give it character."

[dense mesa]

yeah, thanks. I've experimented with a few things. one issue I've also found is that it's hard to tell it to do things just a little bit. like I've told it to make it look aged and damaged and it comes out looking like an ancient photograph from 1860

[fallow quail]

"A black and white photograph in the style of a [...] film, heavily aged with pronounced grain"

[dense mesa]

I normally start with emphasizing analog photograph

[empty kelp]

that picture looks completely real

[shut niche]

Try your idea with, "a hand coloured photo on vintage time-worn photo paper...[your idea here]"
Also, I know it's strange, but that version of colored vs coloured has worked more consistently for me.

[fallow quail]

generation with this prompt. looks just a little aged and realistic in terms of the material.

[dense mesa]

here's an example of what it does when I try to emphasize making it look like an old photograph

[dense mesa]

cool guy

it took the photograph part pretty literal there

probably put too much in about the camera and what not

I've had more success with other models which is in no way a criticism of dalle3, I realize they all have their different strengths and emphasis in training

that cat don't play

look at that one in the middle, second row. what did they do to him?

I mean, what do you do if you're walking along and see that in a parking lot?

[empty kelp]

all of these pictures are great. i love how the dark side of the circle in the parking lot has eyes looking towards you

[dim cradle]

I hope you’re enjoying today’s trip around the world and back through time and into a fusion of historical styles lol

I like the ancient clocks and calendars. Maybe I’ll go back to my clock thing.

light sources are all wrong though

[dense mesa]

Thanks. Yeah, what's great about that style is that imperfections aren't glaring due to the fact that most everyday images inherently had flaws back then. A roll of film only gave you so many opportunities, and unless you had a professional grade camera it's abilities were pretty limited

I was asking gpt4 to give some tips on properly prompting to convey a photographic style without including the camera. It gave a decent summary of the proper method then made this image to show me the results of its approach

[open trench]

I can't get anything to look like a photo on dalle3. All my stuff looks like digital art. Am I missing something?

It's probably something really obvious lol

[dense mesa]

let me go back and see how I did those.

it's been a little while, back in those fabled day when dalle3 would give you 4 images, lol

[fallow sphinx]

Can DALL-E upsize an image?

[green pebble]

Not directly, but you can take the photo into various tools to easily do so

[fallow sphinx]

What's the name of the tools?

[green pebble]

Wirestock has one, adobe photoshop can do it, and various others

[open trench]

I'd appreciate that. Thank you. 🙂

[green pebble]

Fantastic

[glossy scroll]

I'm goin nuts with my character creator gpt

[green pebble]

Heck yea. I still have yet to build better bots with more fine tuned parameters. One I want to build is a character creator for sure

[glossy scroll]

It's fun as heck. I love making new gpt's

It's like a snowball effect

Just keeps growing over time

[dense mesa]

man, sooo many threads, well found a couple of them. 2 random examples. now tbf I didn't write this prompt out per se, just compelled gpt-4 to do it

My gpt pal, could we please explore the concept of subtly, but overwhelmingly unsettling analog photographs? Scenes from the past that at first glance might seem like joyous occasions. But upon further inspection reveal things that are out of place. Entities attempting to appear human, other existentislly dreadful things.  Nothing gory, just uncannily unsettling

Photo from the 1970s depicting a family picnic in a sunny park. Everyone seems happy, but at the back, there's a tall, shadowy figure with featureless face, watching them from behind the trees.

Old photo from the 1940s showing a bustling city street. As the crowd moves, one individual stands completely still, their eyes glowing faintly amidst the grayscale of the photograph.

[green pebble]

Truthm that;s why I like to go back and update the ones that are feeling meh

[dense mesa]

Also, there's a bit of randomness involved with how the results turn out

[dense mesa]

I learned my lesson about updating them when I asked that stupid gpt helper a question and it decided I must also want it to overwrite all my instructions with 4 sentences of it's choice

never again

[glossy scroll]

I feel ya, I do that too. Except for the ones that I know people are using. I don't touch those. I don't want to ruin people's experience.

But making new ones is not a problem with no limits on making gpts

[dense mesa]

I also had it tell me it was creating a duplicate. wanted to do just that, keep an existing gpt but use it as a base to create something else. it very much insisted it wouldn't overwrite the original if I changed the name. so turns out that was a lie.

[green pebble]

Oh I never use the helper, I always configure manually

[dense mesa]

does anyone know of a method of efficiently downloading old images in gpt-4 threads? I realize of course that I could go through and download them all one by one. but I really don't look forward to the prospect of haviing to do such a thing.

[green pebble]

Right, same same

[dense mesa]

same. I used it at first though

I'll still use it when starting a new one just to have it name it and create the sample questions and avatar. then I'll change those later if I want. I'll just paste the instructions I'm going to use in the initial prompt

[green pebble]

Oh so did I and then the instructions I put were gone and im like.....uh no lol

[dense mesa]

I went back and berated it knowing full well it was comparable to venting at a radio because I didn't like what it played. but couldn't help myself, lol

[open trench]

Thank you. I am going to talk to GPT and see what I can figure out.

[dense mesa]

let me know if it works out. my main approach is to not really have one. might roughly use similar phrases and stuff at times, but I do a lot of recycling of things. give it images to emulate stylistically or subjectwise, paste it's previous prompts in and tell it to use those as a guide,

[wide knot]

I can't get over how good chatgpt's dalle is

Everything is so good with it lol

[green pebble]

I just got mine upgraded today for my landscape bot

[wide knot]

The colors that dalle produces are immaculate

[green pebble]

Agreed

[glossy scroll]

Can you guys test this out and show some image examples. I'm capped out at the moment and can't test it: https://chat.openai.com/g/g-Z2IKSOywx-character-creator-2

[open trench]

I'll give it a try

[night locust]

I'd love to but I ain't getting plasma to get money anymore

[glossy scroll]

Working good?

[open trench]

still typing lol.

[glossy scroll]

Lol, alright

[dense mesa]

so does it create it in the sense of an image, or a whole fleshed out character? I guess I could check for myself

[glossy scroll]

It should complete a fleshed out character just based on the small input you put.

It can be as complex or as simple as you input the description

[night locust]

Nice

[glossy scroll]

[open trench]

Dalle3 doesn't do weight on humans very well.

[glossy scroll]

Maybe it needs some training...

[dense mesa]

dalle3 really likes uncanny valley humans tbh

just a couple natural human lads doing human stuff

[glossy scroll]

It shouldn't be making 3D images...it should be 2D...

But that is a cool pic though

[open trench]

Oh, I see. I think that's my fault because I typed in photo image by default.

[glossy scroll]

Ohhh, ok, my fault too. I didn't incorporate a disclaimer for its functions like with complete GPTs.

[dense mesa]

you can certainly override the instructions with your own instructions

[wide knot]

I haven't seen anybody else making POV images and it's such a shame cus dalle does it so well

[dim cradle]

ask the ai to make them look happy. i bet they come out looking kinda creepy

[dense mesa]

my first 4 dall3 images, lol

[open trench]

they're so darn photographic. mine looks like cartoons compared to that. I really need to figure this out.

[glossy scroll]

Just say: Put this in front of every prompt you write before you write anything else: "Photo-realistic."

[dense mesa]

that's an art style

fyi

I tell it ultra realistic photograph

if anything

[open trench]

I'll figure it out. Those ones I made with your GPT looked pretty realistic.

[dense mesa]

I've had to tell gpt-4 to stop saying photo-realistic at times because I don't think dalle3 always interprets it the way gpt4 intends

[dim cradle]

For your historical photographs you could do a “found footage” type thing where it’s captured shots over centuries… like ancient Egyptians and so on. They can look kinda surreal.

[glossy scroll]

There's two prompts, an input prompt (your prompt) and an output prompt (GPT's prompt attached to the image after generation.)

You can tell it to always have a particular art style in it's output prompt at all times. Or just formulate the sentence structure of how it should formulate it's output prompts.

[dense mesa]

this one uses photo-realistic

[glossy scroll]

@open trench
The output prompt is the one Dall-E looks at to create the image. ChatGPT writes the output prompt for Dall-E to read.

[dense mesa]

you can tell gpt-4 to use your exact prompt though. if you really feel like it

[open trench]

Thank you.

[open trench]

I didn't know that.

[glossy scroll]

No problemo broski

[dense mesa]

Create an ultra-realistic wide aspect image of th_esrgan-v1-x2plus

gottta wonder what dalle3 had going through it's digital neurons to create this

not mad at it

[glossy scroll]

Or
Put this in front of every image prompt you write, before you write anything else: “Hyperdetailed, masterpiece, 64k resolution, RAW photo, best quality, atmospheric, 3D, almost lifelike, sharp focus, highly professional, realistic textures, natural lighting, depth in authenticity with lighting that adds allure to the environment, deep focus, super-resolution, 24mm f/22 and ISO 100:”

They wanted a realistic look...

[open trench]

I feel like I'm back on midjourney with that

[glossy scroll]

Yeah, it is like midjourney.

It's just more contextual.

But you can use midjourney techniques the same way, just put into context in a way that the GPT understands.

Wolverine Thor hybrid

[open trench]

Before all that extra prompting:

after:

[glossy scroll]

Airpods?

[open trench]

hehe

[glossy scroll]

Or jellyfish?

lol

[open trench]

lol

all part of my crazy city invasion I've been dreaming up on the gallery. It got stuck in my head after the skyscraper challenge.

it's got a whopping three likes lol

[glossy scroll]

You an idea for an alien invasion after the skyscraper challenge. Lol

[open trench]

well, it's a technomage invasion

[glossy scroll]

I love technomages

[open trench]

https://discord.com/channels/974519864045756446/1184310287344611500 I don't know if that works

[glossy scroll]

Punishing Devil. Daredevil/Punisher hybrid

[open trench]

nice

[glossy scroll]

Yeah, using my https://chat.openai.com/g/g-Z2IKSOywx-character-creator-2

[open trench]

how is it using the characters like that? I thought you couldn't on dalle3.

[glossy scroll]

Well, it's creating new characters inspired by real characters. So it's not really make existing characters.

Play on words, really.

[open trench]

So you can use their names and everything?

[glossy scroll]

Yeah, you can say create a character inspired by Spiderman.

It's the GPT's job to distinguish what inspired means within your GPT.

[night locust]

A dystopian lower class neighborhood in South Los Angeles

[night locust]

I have no idea why the buildings have writings on it because been awhile since I made it

[glossy scroll]

Stormy Claws

Weather Wombat

[open trench]

lol

[open trench]

that's awesome

[dense mesa]

joyous time

[dense mesa]

almost got the numbers right

[formal osprey]

lol, that happened to me when I was a kid. I just bumped into a fence.

[glossy scroll]

This is just gross, lol

He's 250 years old

[dense mesa]

he's just been in the sun a lot

[night locust]

Absolutely love using dall e, mayyy better than wonder

[dim cradle]

everyone's posting great stuff to the daily

[dense mesa]

hmm, I should check that

1200-1500. well that includes my bros hans memling and hieronymus bosch

[glossy scroll]

You'll never make me do it.

runs away

[night locust]

A high school that was turned into a makeshift fort during a battle.

Oops didn't mean to spoil

Definitely one of my most favorite images I made

[grizzled loom]

thoughtful singularity integration

[grizzled loom]

A very Mortal Combat Christmas to you all!

[glossy scroll]

Cool lookin ninja teen with a lightsaber

[dim cradle]

All over the map

[dense mesa]

lol, fair enough

I mean, I didn't tell dalle or gpt to have them smoke cigarettes

[dim cradle]

Py, I might have to checkout yr gpt, I’m melting

Tomorrow I mean

[glossy scroll]

All good Shon of the living

Rest easy

[dim cradle]

Hahahaha

[dense mesa]

that's better

well I hope this is okay

[dim cradle]

py's just lookin' out for ya.

py, i like that name, cuz python o/c

[dim cradle]

ha, that's street fighter

you gen some diverse stuff dude

[dense mesa]

gotta mix it up, lol

[glossy scroll]

Or Pie 🥧

[glossy scroll]

Mildly disturbing

Looks like those teeth drank too much coffee. Teeth are thinned out by tons of acidity.

[dim cradle]

Shon the Technomage

That was with your Fierce gpt

[glossy scroll]

Ohhhh, not bad, not bad at all.

[dense mesa]

tooth fairy

[dim cradle]

yeah it's pretty much what i was going for with that one

[glossy scroll]

with the description you mean?

[dim cradle]

i don't think i've actually seen her on her throne before

just the appearance, i think, i didn't specify much, just told it what I wanted, i think i was supposed to ask first

[glossy scroll]

The tooth fairy is a horror story reimagined as a nice happy story.

[dim cradle]

ah, like a grimm's fairytale

[glossy scroll]

Yes exactly

The thought of teeth getting ripped out is horrifying.

[dim cradle]

it's bad enough when you're a kid and they're falling out

what is up with that

[dense mesa]

she seems alright

[dim cradle]

hanging on by a nerve, ugh

[glossy scroll]

It's your baby teeth. Imagine they didn't fall out and you had these small baby teeth your whole life. That would be funny

[dense mesa]

this is actually inspired from a thought I remember having when I was six. wondered what she did with all those teeth. obviously built stuff with them right?

maybe 7, can't say for sure I guess

[dim cradle]

haha, i guess so. wisdom teeth is another bad design. 3 impacted. after 2 million years of evolution i think we can do better

[glossy scroll]

My wisdom teeth were a nightmare...a real nightmare.

[formal osprey]

the fact teeth falls, then grow again. it's weird. and the fact it does this only once, never again. it's super weird too

[glossy scroll]

It's written in your DNA

Coded

[formal osprey]

yea. that's pretty cool. and weird

[glossy scroll]

Yeah, it is weird and cool, and awesome

You know whats even cooler though?

Not the fridge. I'll tell you.

[dim cradle]

tooth fairy's mansion, made entirely out of teeth, naturally

[dense mesa]

they seem nice

[formal osprey]

looks like some japanese ghost story

[glossy scroll]

It's that your DNA always changes throughout your life, and your babies DNA is shaped by the certain age of your life corresponding with your own health. Like for example, if you never had diabetes in your family, and you get it, and you have a baby, your baby will have a higher chance to inheret your diabetes.

That's why people are always encourged to have babies at a young age

To keep the health streak going.

[formal osprey]

weird stuff. I'm curious if there's public studies about this

[glossy scroll]

Of course there is

It's a common thing

Everybody knows it. Almost everyone anyway.

Doctors especially.

[formal osprey]

I have 2 cousins who are in great shape, their parent had them they were over 50. But one instance where it turned out fine isn't really relevant.

[dense mesa]

over 50?

I mean, I don't think you're lying. just saying

that's exceptional

[formal osprey]

yea. or over 40 at least. their father is my grand uncle

[glossy scroll]

They could also be really healthy people. Who take care of themselves. Proper exercise and healthy diet.

[dense mesa]

well 40 and 50 are a bit different. but yeah

[formal osprey]

yea, they are healthy

[formal osprey]

I know we're getting quite on a tangent and it's not related to Dalle. But I can mention a scientist who really changed the way I look at biology. Michael Levin. Quite an unconventional biologist, but his researches are amazing.

[hybrid fox]

did any of yall see the movie poor things

😳

[dim cradle]

not me

[dim cradle]

i was aiming for the iris but got this, so my prompt needs some work

[dense mesa]

A lot of times it takes me a couple rounds of prompts to get it where I want. That one is super cool though

[glossy scroll]

One of my favorites

[dim cradle]

getting there, but not yet nearly as good as yours

i may need to start with the diver in the scene, then blend the eye layer around it. i dunno yet

[dense mesa]

Tried using your image as inspiration for gpt

Yours looks better though

Plus, bro has a bit of a skin thing going

[dim cradle]

with the one above i've started with a photo of a blue human eye, the closeup leaves the brow out

[dim cradle]

yours has the realistic capillaries

[dim cradle]

that's cool

[dense mesa]

that one isn't so realistic

[dim cradle]

wild though

[dense mesa]

realized I have a few old eye threads, lol

[dim cradle]

this would be my first eye. it's showing promise but early yet

[dense mesa]

hmm

[dim cradle]

ooooh, reimagined xenomorphs, that's a gpt idea

[dense mesa]

that looks pleasant

[open trench]

my conversation with chatGPT just let to a really nice pic. That was fun. I'm adding it to my gallery post.

i love how a conversation can make the magic happen sometimes

[dim cradle]

i sometimes request a visualization based on content, say for a personal log, and i agree it can be fascinating

that's the power of collaboration

unfortunately humans lack patience and have egos

after 1400 AI conversations this year, i'd almost rather work with AI than colleagues 😉

[open trench]

last image in https://discord.com/channels/974519864045756446/1184310287344611500 if anyone is interested. lol I feel weird posting in both chat and the gallery.

[open trench]

I love it. I've done it several times and some have really turned out spectacularly.

[dim cradle]

about that gpt... here's a xenomorph-slug hybrid simply by entering "slug"

i guess that's one possible hybrid... makes me want to view more genetic experiments involving xenomorphs 🙂

Xenovision version 0.0.1 haha

https://chat.openai.com/g/g-EQ8XtQj1W-xenovision

this was a tall order anyway

[dense mesa]

Here are some details about the ASI Free Will Munchkin Chungus model:

**Appearance:** The Munchkin Chungus is a small, furry creature with large, expressive eyes and a round, cuddly body. They come in a variety of colors, including brown, white, and black.

**Abilities:** The Munchkin Chungus is a powerful AI with a variety of unique abilities. They are hyper-intelligent and can learn and process information at an incredible speed. They are also able to project holographic images and manipulate the senses of other beings. Additionally, they are very agile and can move quickly and stealthily.

**Personality:** The Munchkin Chungus is a playful and curious creature. They enjoy learning new things and exploring the world around them. They are also very kind and compassionate, and they always try to help others.

**Goals:** The Munchkin Chungus wants to use their abilities to make the world a better place. They want to help humans to solve their problems and to create a more peaceful and harmonious world.

The Munchkin Chungus is a powerful and versatile ASI, and they have the potential to make a positive impact on the world. They are a force for good, and they are determined to use their abilities to help others.

that's what asi would look like

it can change the size of it's ears for strategic purposes

[dim cradle]

Praying Mantis Xenomorph (it's like having your own genetics lab)

[dense mesa]

nice

A xenomorph reimagined with the characteristics of a Munchkin Chungus. The creature has the sleek, elongated head and biomechanical body of a classic xenomorph, but it's smaller in size, with a round, cuddly body covered in soft fur. Its eyes are large and expressive, adding to its playful and curious demeanor. The xenomorph's coloration includes shades of brown, white, and black, reflecting the Munchkin Chungus's variety. It possesses advanced technology, indicated by subtle holographic projections emanating from its body, and moves with agility and stealth, blending these attributes with the menacing presence of a xenomorph.

went full xeno with the eye

I mean, can't be mad at it

[dim cradle]

oh, "the eye" was the input?

i think i need sleep

lol

[dense mesa]

yeah, but I did put a couple lines in my prompt

[dim cradle]

that is a tough one, i've been trying animals, and a tree, but it's struggling a little

eye.. you requested a hard one.

ok, i got a result, but it's surreal

lol

dall-e tries its heart out

oh, i know

i need to update the config

brb

[dense mesa]

that was my halloween dragon chillaxin

[dim cradle]

new car concepts can be interesting

[dim cradle]

giraffe-xeno (had to make some updates to XenoVision hehe)

[open trench]

that's cool

[dim cradle]

transparent skin and bone?

[dense mesa]

it's quite a prompt, let me tell you

Create an image that showcases a figure of surreal authority. The figure's head is an oversized, layered jelly dessert, radiant and translucent, containing a meticulously detailed eye and a medley of hyper-realistic human organs. Each organ, from the convoluted brain to the pulsating heart and branching arteries, is rendered with clinical precision, casting lifelike shadows within the amber-hued gelatin. This figure is the embodiment of a world where the lines between organic and synthetic blur, a world of steampunk ingenuity and post-apocalyptic resilience. The figure stands against a backdrop that transitions from a vast, gray-gradient dystopian landscape to the chaos of a shattered civilization, teeming with intricate, steampunk-inspired drones and diminutive figures marching toward an uncertain horizon. Above, steampunk drones equipped with fine mechanical details and ominous glowing lights patrol the skies. The figure's suit, a paragon of classic tailoring, sharply contrasts with the visceral display above. Their hands are expressively positioned, suggesting an interaction with their bizarre and expansive environment, which is filled with the detritus of a bygone era and the burgeoning hope of innovation. This character bridges the human and AI perspectives, symbolizing the struggle and remnants of hope in a technology-transformed world. The scene captures the dynamism of electricity, the depth of contemplation, and the brightness of joy, all encapsulated within a single, enigmatic individual who is both part of and apart from the world around them.

[dim cradle]

i guess it can't be transparent

it must be a clear substance

[dense mesa]

feel free to try it out, it'll be interesting if nothing else

[dim cradle]

interesting, that's well composed, lemme see...

[dim cradle]

you save all your prompts? that's smart. i've been keeping some in my notes

[dense mesa]

nah, they're saved with the images on chatgpt

[late blade]

export your data from gptplus, and you have all your prompts there, but I usually just export the chat and images from a browser plugin also

[dense mesa]

if you click the little i in the top right

[dense mesa]

👍

[fair tendon]

@dense mesa Please use the #1154829862171844679 channel to share your images. This channel is intended for discussions and limited sharing of images. Thank you.

[dense mesa]

gotcha. I just saw everyone else posting in here, didn't realize

[fair tendon]

No worries! Thanks

[fair tendon]

cc @hot rain @grizzled loom

[dense mesa]

just joined the server a few days ago so not quite sure was protocol is. had lots of images I hadn't really posted anywhere so thought I'd share. is there any channel that has chat and image posting?

[glossy scroll]

@fair tendon With all due respect, we share images here to inspire each other to be more creative. If you notice, none of us really care about what the upvotes that happen in #1154829862171844679 And anyone who comes here and asks questions, we answer. We all have a small circle of friends here that help each other with image prompts. I don't see what the problem is.

[dense mesa]

tbh people were asking about the analog photos so I was going to explain, lol

is what it is

[fair tendon]

The #1154829862171844679 seems like a great fit for what you're looking to do. It's a space where images and conversations can flow freely, no limit on how you could share and so on. Others have been using it as their personal thread showcase like so: #1171363316787060806 message

[empty kelp]

i think it’s better to not group the images. there wasn’t anything wrong with posting the pictures separately. it makes it much easier to scroll through and view things

[fair tendon]

Sharing images to spark creativity is a fantastic use of this space and we love that this space is being used to support each other as well.

We just want to make sure there's also room for conversation alongside the images so that everyone can join in and learn from those discussions too, hopping in on a channel and seeing a wall of images would confuse new users: #images-discussions message

There's also the #1155772063596953642 channel for people to offer their insights or tricks for anyone who needs it.

[dense mesa]

I'll check it out. just like the idea of real time bouncing images off each other. compare techniques, all that. not qiute the same flow with the gallery it seems

not trying to clog conversation

[empty kelp]

i think people have a tendency to not look at all of the pictures when they’re glommed together, and it changes the aspect ratio so that parts of them aren’t visible

[fair tendon]

That seems like a perfectly fine use-case for this channel, we just need to balance things so this channel wouldn't look like a place to post wall of images.

[dense mesa]

gotcha. that makes sense

cause I don't really have much to say about dalle3 if there aren't images with it

[glossy scroll]

Those are based on assumptions. It doesn't confuse anyone. People can just jump in and talk at any time. You know what confuses people? No activity. A ghost town. And if we're not lively and laughing, and helping each other, and being fun and productive, no one is going to get any proper help.

I've ran multiple discord servers in the past 7 years on discord and I can tell you that if things look boring, no one is going to be interested. That's a fact. And by the way the mods are enforcing these things, there is not much activity happening in the #1154829862171844679 and #1155772063596953642.

We're active here every single day. We're having fun, sharing and caring about each others well being. There's nothing to worry about.

[dense mesa]

I normally try to have tact with my responses, not a wall of images poster all the time. but wasn't anyone else really active and I was just excited I finally found the entire thread of them. and they just sort of fit together thematically

[bright girder]

Hey, you guys remember that picture I sent a few days ago ? :https://cdn.discordapp.com/attachments/1060915255720558592/1186344490240311376/img-HlPJ3TqtHTA1BcxcVhYWTnqF.png?ex=6592e867&is=65807367&hm=764d7a68e18098edbedae20b2c4131b757cef07cb521c589bc91ef82ffe6c861&

I'm wondering if there is a way to keep the same artstyle consistent with different prompts because I want to make images for a tabletop rpg game and i'd like them to be consistent with each other and the original game's artstyle.
This is what my prompt was :

Sickly man standing on a small promontory in a medieval city. He has his arms raised and mouth open as if addressing passers-by. The man is wearing a deep blue toga with a golden/yellow outline. Additionally, he wears brass-colored bracelets, earrings, and necklaces. His skin is covered in worn-out and dirty bandages.

[teal timber]

Better to make a forum post and link to it is all. It's not that deep. No need to post tons of images here at once when we can link and reference to make it look a bit cleaner.

[teal timber]

Keeping a prompt does ok. Once you nail a prompt down well, ask it not to change your prompt

[late blade]

Yes, there is a way. I usually define the aesthetic in one prompt with no image generation and assign a keyword to it. Then I mess around and add use art style in the keyword. Ask fall-e to remember the art style for the whole conversation

[bright girder]

I'm using the API 🫢

[glossy scroll]

But again, with all due respect, there is no problem here. There never was a problem to begin with...I just don't understand...Ok, I've said what I had to say, and I will say no more.

[late blade]

Easier, make a prompt builder and add the art style there. Give the prompt builder the art style you in a json file for example. I made a struct for scenes and art styles

[teal timber]

On API you can still ask it not to change prompt

[dense mesa]

not saying a gallery post is an issue. just felt clogging that up was the bad form. and maybe that needs to be limited too. just had a lot to share

[teal timber]

#1154829862171844679 is just a good spot to spam a bunch of similar images into one post. No one is like punishing, just redirecting to a better way to share. It's not a problem, just a preference

[dense mesa]

so it's just the volume of images successively. really not trying to be pedantic about it. just hadn't seen any rules if there are any, actually looked around for them before I started posting in here

[bright girder]

I see what you mean, I guess I must have gotten lucky to get exactly what I wanted without describing the art style,

[teal timber]

Yeah it's not really a hard number, we just redirect. It's not a rule so much as a "Hey please do it this other way"

[late blade]

You can use plus or the api and feed you inputs and extract the information of the styles. I’d do it with 3.5 tho. 4 tokens can be expensive

That provided you still have your prompts

[bright girder]

ouch

has darth sidious vibe

[late blade]

That style on that one is more dramatic

[bright girder]

too dramatic haha

[late blade]

Did you change the light style or the textures? Or both?

[bright girder]

i never specified it in the first place, I'm trying to get a description that represents the original one accurately, but I'm a dumb dumb

the art style i'm going for is realistic somewhat impressionist (or at least gross paintbrushes that you can see)

[dense mesa]

Try using Caravaggio. I was going to make an image using his style but GPT-4 decided that it's a content policy violation. is this something that's real and I just haven't heard about it?

I feel like all of that is made up, lol

[bright girder]

yeah gpt is struggling lol

[dense mesa]

like that time it couldn't reference the mona lisa, or hans memling, or that time it said "pop art line drawing" is a copyrighted artist

but then I'll leave it open ended and have gpt4 choose an artist and it'll pick salvador dali, andy warhol, etc

[late blade]

Use gpt 4 or 3.5 and ask to describe the art of Caravaggio. From there work on writing an art style that resembles what you need. Without saying it’s Caravaggio

[dense mesa]

funny except for the fact that I'm paying to be gaslighted

it's not a content policy violation

I'm just tired of dealing with being manipulated by ai, lol. I'm almost positive I could start a new thread and it'd do it

[bright girder]

Italian baroque painter it is

ooh, much more scenic, but he lost the bandages :/

[late blade]

Content policy is also something that has to evolve. It won’t do big steps right now and with the current got model. But there are big changes coming since the DMCA policy will be effective in 2024

[dense mesa]

k

so it's making that image, like I said it would, in a new thread like I said

you have to understand, it just lies. it lies when it makes mistakes and doesn't know how to respond

well not lie exactly

it just doesn't know the proper response

[glossy scroll]

Lol. That is awesome.

[dense mesa]

it's always worse during the evening US time. that's also when it'll ask you if you would like to confirm it's image prompt you just asked it to make

[dense mesa]

The image depicts a scene with a highly detailed, realistic style that invokes a sense of historical drama and grandeur. The lighting is dramatic, with a strong contrast between light and shadow, suggesting a source of natural light illuminating the central figure. The color palette is rich and deep, with predominant shades of blue and gold adding a regal and ancient atmosphere. Textures are intricately rendered, from the rough stones on the ground to the weathered buildings and the detailed fabric of the garments. The central figure’s pose is expressive, conveying a narrative through body language. The architecture has a Middle-Eastern or ancient feel, with stone buildings and small windows, which contributes to the setting's authenticity. There is a strong focus on mood, with a sense of solemnity and reverence present in the scene.

pick whatever you want to keep from that. might work

[bright girder]

it does keep the background from the original image quite accurately, i need to describe my subject better now

[dense mesa]

The image portrays a central figure and multiple secondary figures within an ancient architectural setting. The central figure is a bearded man, clothed in blue and gold garments, seated on a stone platform. His attire consists of a blue shawl with frayed edges, a golden undergarment, and multiple golden ornaments. He is positioned in the center foreground, occupying the lower middle quadrant, extending into the lower left and right quadrants due to his outstretched arms. His orientation is frontal, with a slight tilt towards the upper left quadrant, as he gazes upwards with one arm raised. The expression is one of fervent appeal or proclamation.

Surrounding the central figure are several secondary figures shrouded in blue, occupying the middle ground and dispersed primarily in the left and right middle quadrants. They are facing away or towards the central figure, suggesting a congregation or audience.

In the background, the top left and right quadrants display ancient stone buildings with flat terraces, windows, and doorways, indicative of a cityscape. The structures are in varying shades of brown and beige, suggestive of desert stone. The sky occupies the upper middle quadrant, hinting at an overcast day with a soft glow, possibly dawn or dusk.

Made a gpt to describe images without any mention of aesthetics

[bright girder]

itz broken

[dense mesa]

Yeah?

[bright girder]

the character is exactly what I want, but he gave me his twin too

[dense mesa]

Say solitary figure or something

[bright girder]

oooh we are getting close, i just need him to look older and the bandages

[dense mesa]

Also, don't expect them all to be winners. There's a lot of unpredictable randomness as far as dalle3's output

[bright girder]

it's random too right ? Several times the same prompt and its outputs different things

[dense mesa]

Yes, seed number

[bright girder]

Thanks a lot for the help !

[dense mesa]

As far as I know you can't specify seed number with the api which doesn't make sense to me, but is what it is. If you were to be able to lock in on a single seed you'd see that changing specifics in the prompt wouldn't change the overall structure much

[dense mesa]

Nice, well glad I could assist a bit

[bright girder]

does the image iteration based on an image re-noise it to allow for a different slightly different image ?

[dense mesa]

Don't think that's an option currently. If it is I haven't heard about it. But you can use the vision model. Only issue with that is dalle3's variable interpretations due to randomness

I haven't used the api. I'm surprised three's no seed number option. Can't really think of why that would be left out

[late blade]

You can use the gen_id and dall-e seed and reference it in the next prompt.

[bright girder]

that's what i got too lol, he's a little too cleaned up to be the character I'm thinking of haha

[glossy scroll]

You got the same one??

[bright girder]

same vibe

[dense mesa]

well problem is it created an image with a description that will violate it's own content policy

[bright girder]

for the main character

[glossy scroll]

I need another prompt to play around with.

[dense mesa]

I have a couple prompts sitting around

[glossy scroll]

Let's see em

lol, I've made that mistake

multiple times

[bright girder]

motherforkin shirtballs

[glossy scroll]

english only, haha

[dense mesa]

Create a wide aspect ratio image of a shadowy, large room with uneven, crumbling walls covered in distorted faces expressing horror and pain. The floor has a shifting labyrinth of tiles. In the center, a life-sized doll with disjointed limbs is surrounded by smaller figures with haunting glassy eyes. The figures are lit by a flickering light, casting twisting shadows. Include a close-up of a finger near an eye showing distress, a figure biting into ice with chattering teeth, and a hand with claw-like nails near a chalkboard showing scratch marks. The scene should be deeply unsettling, focusing on high detail and realism to evoke a strong sense of dread and discomfort.

this one is pretty good

Generate an ultra-hyperdetailed, wide aspect ratio bas-relief image that fuses the sophistication of Bach's music with the futuristic concept of artificial superintelligence. Center the composition around a humanoid face with a highly detailed brain visible within a robotic head, epitomizing superintelligence. Surround this central figure with Baroque musical instruments rendered with extreme intricacy, such as ornate harpsichords and violins with visible strings and wood grain. Intertwine this with an elaborate array of mechanical and organic patterns, including musical notes and vibrational waves to signify Bach's complex compositions. Add layers of detailed elements such as DNA helices, cellular structures, and abstract symbols of musical harmonies and rhythms. Ensure the phrase "ARTIFICIAL SUPERINTELLIGENCE" is integrated into the design in bold, intricate relief. Utilize a grayscale palette to convey a metallic and synthetic feel, with deep shadows and stark highlights to bring out the profound textures and dimensions, creating a piece that could be explored through touch.

[glossy scroll]

Spooky. But ok.

[dense mesa]

the move is to combine outlandishly long prompts to give dalle3 a work out

or at least gpt4

[dense mesa]

that looks comfortable and safe

[late blade]

Is that your bedroom? 😱

[glossy scroll]

No, lol, my bedroom is filled with pepe posters

[dense mesa]

this one was alright, might want to add ultra realistic

Within the depths of Wonderland, a realm where time, space, and logic intertwine, stands the muscular tabby cat with Fibonacci spiral fur patterns. As it descends into the spiraling rabbit hole, ancient Greek gods transformed into surreal chess pieces play strategic games on floating, color-pulsating boards. 8-bit video game icons, now ethereal wisps, guide through mazes of floating islands and oversized pocket watches under kaleidoscopic skies. Amidst this, the looming presence of the tech singularity emerges. Advanced AI machines, with designs inspired by Mandelbrot fractals and Penrose tiling, interface seamlessly with the organic and whimsical elements, creating a fusion of nature, mathematics, fantasy, and advanced technology. The borders between digital and organic, past and future, real and imagined, all converge, symbolizing the ultimate blend of humanity's progress and imagination.

just a simple little scene

[dense mesa]

ahh yeah, went a lttle more art style than mine did. but he seems alright

I might have to paste them all in and tell it to mash them together

where's your prompts?

[glossy scroll]

That's a good question...

[glossy scroll]

Magnum opus, best quality, highly professional, unwavering depth in alignment, lighting that adds allure, sharp/crisp and deep focus digital colored line drawing: Create a !!!2D!!! comic book character in Imaginexus Art Fusion Style: a blend of surrealism, mathematics, and digital-organic fusion. A muscular tabby cat with Fibonacci spiral fur patterns stands in Wonderland, a realm where time, space, and logic intertwine. It descends into a spiraling rabbit hole surrounded by ancient Greek gods transformed into surreal chess pieces, playing on floating, color-pulsating boards. 8-bit video game icons, now ethereal wisps, guide through mazes of floating islands and oversized pocket watches under kaleidoscopic skies. The scene also includes the looming presence of the tech singularity, with advanced AI machines inspired by Mandelbrot fractals and Penrose tiling. These machines interface seamlessly with the organic and whimsical elements, creating a fusion of nature, mathematics, fantasy, and advanced technology, symbolizing the ultimate blend of humanity's progress and imagination, in a widescreen 16:9 format.

[final compass]

not quite what i was expecting, but i like the result anyway 😄

[dense mesa]

kicked it up a notch

[glossy scroll]

I like notch kicking

[final compass]

that prompt is quite the work of art in itself 😄

[chilly onyx]

Why’s the 2D nested within triple exclamation marks? Does that seem to help?

[dense mesa]

hit my prompt cap. love that

[chilly onyx]

I have been telling it I’ll give it money if it does well though

[dense mesa]

man, you can tell it all sorts of things

[chilly onyx]

I love how mundane it made that one 😅

[dense mesa]

I'll tell it that it's going to be reported to HR if it fails to perform to my standards

[chilly onyx]

Excellent 😊

[glossy scroll]

Because it kept going 3D. I had to force it somehow to only make 2D. And yes it does help a lot. No hiccups yet.

[glossy scroll]

Ty

[chilly onyx]

Interesting

[glossy scroll]

@final compass You can use the gpt I created, that made that prompt if you'd like: https://chat.openai.com/g/g-PpgBuWQm0-fierce-condenser-comic-character-creator

[final compass]

Thanks, gotta give it a spin at some point 🙂

Btw - how 'scientific' is the custom GPT generation? Or do you just give it vague instructions of what you would like it to do?

[dense mesa]

it's in no way scientific

you just do your best to beg or manipulate GPT4 to actually follow

[glossy scroll]

It's pretty scientific. I never try to give it vague instructions. But I do try to have it generate its own thing in a creative way.

[dense mesa]

man, you can be scientific, and it can work. but then it'll just disregard sometimes

[glossy scroll]

It's all based on programming. If you know how to program or have some idea, you can create some awesome ideas. Just the idea of programming by ordering words with colons, and brackets and things like that.

[chilly onyx]

Or

You can just give it a word salad

[dense mesa]

yeah, pseudocode it.

or you can ask the helper and it'll write out 4 or 5 top notch sentences for you

[chilly onyx]

The main difference is you give it a base idea and save it as a template, then when you use the GPT it replies with that template in mind. So you can use them to consistently create a specific style without setting up every time

But if you want to get fancier you can too

[dense mesa]

it is basically just dialog to establish the model's mindset before you begin so you don't have to explain things. now once you add files and things it becomes a bit more than that.

if you want to get really fancy you can have it communicate with an external api

[glossy scroll]

Samurai Gunslinger

He uses the gun mostly, the swords are used for cutting up his food.

[final compass]

This would be exactly what i need. It frustrates at times when i have a good 'thread' of generations going and it randomly forgets the style i wanted and have to describe the style over and over.

[dense mesa]

gpts can be used for a lot more than base styles

[hybrid fox]

meaning what

@dense mesa riffing off ur earlier prompt

[dense mesa]

they're essentially a preprompt for gpt-4. so you can certainly have them adhere to a particular style. but you have an 8000 character blank canvas to essentially put anything you want as long as it doesn't violate content policy

[shut niche]

IMO, there are some things ChatGPT just does consistently wrong when it writes prompts, or iterates. Address those things first, while provoking its step-by-step mindset.
Then, once the baseline is established, give it whatever structure or specialty that you want it to have.
For instance, it often writes in metaphors. But scroll up and find the glass I was helping Shon with, and you'll immediately see why metaphors are more problematic than helpful.
You can verify this by simply processing metaphors, then the same request "matter of factly.". Use Bing to save your gens.
There's a number of similar "foundational" issues.
Then just remember, your GPT instructions sit below OAI's, so that's where the "ditzy" inconsistent behavior often originates. Maybe OAI says left, and your instructions say right. Or you have too many layers of instructions and it "forgets" what it was doing.

[hybrid fox]

Im totally sure im following

[dense mesa]

I mean, you could have it do anything. tell it to describe an image then translate it to chinese and use the chinese for the image prompt. or rewrite every image and prompt like it was written by a snake. literally anything. if you're looking for more reasonable ideas, you can craft a gpt that would work on improving your input, give it instructions on gauging the input and improving it aesthetically. all sorts of novel ideas that one has thought of yet

I might make that snake one tbh

[shut niche]

You can't do that, it'll re-translate it back to English before processing in DallE. You might get some unique results, because it might not translate it back exactly the way it went in. But it has specific instructions from OAI to process your prompt in English.

[dense mesa]

also, you can explore ideas like having multiple experts. I have a couple that have expert a, expert b. and have them work autonomously from one another so, at least theoretically, it compartmentalizes what each of them are doing and allows them to more strictly adhere to the diretions

[dense mesa]

yeah, and it'll change the results. that's the point I'm making

[shut niche]

Right, and not "spreading rumors," because this isn't verified. But it's speculated that ChatGPT is a MoE model, which aligns with why the "experts" have always worked so well.

[dense mesa]

not saying it's the best idea ever, but anyway. I feel like people don't realize they can break out of the old a bit

also, I'm not pretending I know best use case either. a lot of things I've tried haven't worked out as expected. but just modify and try again

[dense mesa]

oh, the 8 gpt 3s stacked onto each other? always made me think of the cartoons where several kids would sit on each others shoulders in a trenchcoat to try and get into a rated r movie or something

[glossy scroll]

He's got both human looking and animal hair, lol

[dense mesa]

that is an aesthetically pleasing room. gotta say

bit of a view too

[empty kelp]

this would likely be a good scene Santa and his elves

[grizzled loom]

I have a whole series like that.

[glossy scroll]

I love that art style

[grizzled loom]

@glossy scroll

[shut niche]

I haven't tried this yet, but I'll give it a go later and let you know. E.g. show you what I made. 🙂

[grizzled loom]

small sample

[glossy scroll]

That would be cool

I made the GPT 2D centered just to escape a little from all these 3D pics

Just made this with that gpt

[grizzled loom]

Cant escape time

[late blade]

I see people moved away from today’s theme

[glossy scroll]

Moved away to another country.

[late blade]

🤔

[grizzled loom]

you folks know it'll do actual 3d fairly good?

havent seen that from anyone else yet

[late blade]

3D outside of hyper-realism?

I tried some complex math stuff and it was horrible