images-discussions

[glossy scroll]

Lol that's a funny one.

Merry Christmas Eve Everyone!

[dim cradle]

i just wish we could get some diversity in faces

but not like MJ, they have it, but it is like they went out they way to make them bizarre looking

[dry crater]

It depends entirely on ur input the shape of the face.

[dim cradle]

idk that same womans face is in 99 percent of images i think

[dry crater]

Disagree. Here's some different characters I made for D&D campaign.

[dim cradle]

but that looks also like a 3d model or dummy but yes it is varied

[dim cradle]

she also does not look photoreal but that last looks like the first to me

[dry crater]

aha, i dont care much for photoreal so i dunno

[dim cradle]

cool robe

[glossy scroll]

You mean the v6.0? Yeah it looked bizarre to me too. I couldn't quite put my finger on the new looks of the their models, but bizarre is definitely a good word for it.

[dim cradle]

yes in 6.0

I got a ban from they room for saying it though haha

[dry crater]

here's the various ones I saved for that prompt as I'm undecided which one I'll use:

I'd say there's some diversity

[dim cradle]

maybe in hair and horns. same general face otherwise i think

[dry crater]

well, considering it's one physical description there's obviously some similarities 😄

[dim cradle]

but anyone who has used dalle and made any image knows that face

[glossy scroll]

MJ is so real, it's crazy. But Dall-E...Dall-E has an eye for genius understanding.

I would pick Dall-E over MJ any day.

[dim cradle]

me too. coherency its no contest. and really only thing holding Dalle back on anything is just what they are limiting it to. if it was unleash it would be wild

[glossy scroll]

Agreed

[dim cradle]

imagine Dalle with being able to link images. i wonder even if all the celeb that MJ allow doesnt help variety of faces. more in the AI stew to cook from than just a few generic CG model faces

[glossy scroll]

Yeah, MJ is going to be super helpful for modeling and advertising. But Dall-E is going to reinvent how images should look.

The lighting on some of these new AI images are other-dimensional. I have never seen anything like them.

They don't look real or fake. They're in between. In this weird optical illusion that makes you question reality, lol

I've only seen a couple though so far.

But in a year or so it will all show everywhere.

[dry crater]

i joined their discord server, but no free trial?

[hybrid fox]

Left one is dope

what's poppin homies

[late blade]

Naruto Stargate?

[hybrid fox]

Bro is abt to build the pyramids

[late blade]

love it

[hybrid fox]

Thank you! Wish I had gotten it done earlier

[late blade]

and Good Morning!

[hybrid fox]

Feels like it picked up "punk" and "blimp" and attempted candle steam punk lol

[late blade]

yeah lol

[hybrid fox]

Impressive how blended you got her with the natural lighting

[late blade]

It's not hard once you know your concept, but it took some trial and error

[hybrid fox]

I dont think ill ever get the cat girl thing lol

[late blade]

the light works really good with other settingy too, for example this attempt

[hybrid fox]

so many people on this sub love it but its so foreign to me

[late blade]

cat-girl is mostly a concept for people that like anime/manga

[hybrid fox]

What do you prompt it for good ambient lighting like this

[hybrid fox]

Yah I never rlly got into that stuff so thats probs why

[late blade]

I define the light sources, the textures, the textures physics, the color space, the type of end product media, those are more or less the most relevant parameters

[hybrid fox]

Very smart approach

I should start doing that

Sometimes it feels like gpt starts not listening to me when I add too much to my prompt tho

[true mural]

And it's a good test of whether basic non-human entities can be drawn. Centaurs are rougher. Still can't get consistent driders. (I'm sure there is a trick)

[late blade]

I ususally do something like this

Global Art Style for all Images of this Chat:
ArtInk: Resulting image is depicted as strokes had been made of a number 2 pencil

[hybrid fox]

Haha queen of cat girls pulled up

How are you Vanya

[late blade]

so when I do a prompt with ArtInk, chat knows what I mean

[late blade]

let me try a centaur

[dim cradle]

when's the number 3 pencil coming out?

[true mural]

Basic centaurs should be easy in 3

[hybrid fox]

Do you do this before switching your chat to a dalle one

[late blade]

there are number 3 pencils, pencil number 2 is just the standard for schools

[dim cradle]

gotcha

i had no idea

[late blade]

no, that's the first prompts within a dall-e chat define what I want to do,

[true mural]

2 is the prefered hardness in school, yes, I've used 2-3s

they make them softer, of course.

[dim cradle]

ahhh ok

[true mural]

Ah yes, the "centaur, that's a man on a horse right?" issue. Might have to lean harder.

[dim cradle]

centaurs are awesome

i've made a few, didn't have trouble, but it was a very simple prompt

[balmy sierra]

Wow amazing

[dim cradle]

Request a Hercules Centaur for a little more muscle on that centaur haha

Or at least a human head ha

[hybrid fox]

Anyone know why discord makes any image I upload as a pfp blurry

[balmy sierra]

Er

[dim cradle]

Oh dear

[balmy sierra]

😮😆

[hybrid fox]

who is that guy he looks familiar

[balmy sierra]

No clue

[hybrid fox]

free ma boi

bro in the labrynth

[late blade]

lol

ya, centaurs seem to be hard to do

at least with my settings

ok won't start spamming renderings, otherwise I'll hit 200 images again before dinner

human part is too small, or horse part is too big

and the bow is horrible

[late blade]

that is really good

[hybrid fox]

Pov you lose a bet with a coworker but she's an artist

[balmy sierra]

My last one

[hybrid fox]

I want that dog chicken centaur from poor things

free bro

[late blade]

I'll try centaurs at a later point

[hybrid fox]

I want to be his friend in an apocolypse

[late blade]

lol

[glossy scroll]

Barbarian

[dim cradle]

He-mans lol

[glossy scroll]

Haha, yeah he does kinda look like He-man doesn't he

Ranger

Those strings never come out right..

Another Ranger...could be the same guy as the Barbarian, they look similar, haha

[late blade]

light is not the same tho, it's just generic

[glossy scroll]

Yeah thats true

Cool dragon dude

[late blade]

my first penalty box of the day, 1 hour waiting time, time for lunch

[glossy scroll]

Ancient Silver Dragon

Ancient Silver Dragon 2

[tough folio]

Dragon Knights

[glossy scroll]

Looks awesome!!

[hybrid fox]

@late blade #daily-theme message

This is so beautiful bruv 🎨 🏆

[late blade]

I'm trying to top @fading inlet , she's even better at that kind of settings

[late blade]

Here's my take on your warrior with my own art style in place

[glossy scroll]

Awesome look Dystop!

I like the runes on the axe

That's a nice touch

[hybrid fox]

She puts out so much content per daily theme its actually wild

art machine fr

[tough folio]

yours too

[hybrid fox]

which prompt are we doin

[late blade]

nothing special, we were talking earlier about light, so some of us were doing that kind of stuff

[hybrid fox]

Peak riffing was that night with the parrots lol

still got mine in desktop rotation

We shoulda made a gallery folder

[glossy scroll]

Market district 1

Market district 2

Beautiful fantasy cities

Market district 3

[late blade]

that town looks nice

[glossy scroll]

yeah it does, dreamy

Very cool additions

Market district 4

[late blade]

just fooling around until I get something for myself

[glossy scroll]

I feel that

I want to drink some of those potions. They look tasty.

[late blade]

I try to do that sometimes with my images, specially the ones that are good but don't land on the daily theme channel

[glossy scroll]

That's true, it does need that a town square

[hybrid fox]

Looks exactly like diagon alley

[glossy scroll]

Does it look exactly like it? It looks similar, though I'm not sure if it's exactly like it.

[hybrid fox]

Idk this is how I pictured it when reading and this is what it looked like in the movies except a little warmer

[glossy scroll]

Hmmm

[late blade]

I still didn't get the glowing lamps like you did in yours. They seem filled with mystery.

[thick smelt]

dall-e (bing's version)
made with designer text to photo

[glossy scroll]

Looking epic for the first try!! Woa!!

👏

[glossy scroll]

Lito wall-e

[pulsar sundial]

there is a ninja turtle, and this is a karate turtle

[glossy scroll]

They both look badass!

[pulsar sundial]

agree

[late blade]

ok, i'm in the penalty box again

1.5 hours

[pulsar sundial]

because of which?

[late blade]

too many images in 3 hours

[pulsar sundial]

hmmm, everything is too strict in dalle-3

[late blade]

there are limits set, 40 images max per 3 hours, 200 max a day

[pulsar sundial]

I have 3 hours, maybe a maximum of 20

[glossy scroll]

If you guys had cutom gpt's, you could continue your dall-e image generation in the preview section of the gpt builder.

It's seperate from gpt4 limit.

[late blade]

yeah

[viscid blaze]

w👀t

[glossy scroll]

You heard right nicko

[thick smelt]

i tried to make t swift 1989 (taylor's version) but dall-e (bing's version) but it generated wall-e 😆

[viscid blaze]

I never made 200 images in a day, though :p

[glossy scroll]

🤣

That's funny 😛

[late blade]

oh my dear nicko, you are conservative, between api and gpt+ and bic, probably 500/day

[glossy scroll]

Didn't come out perfect, but still working on it.

[viscid blaze]

kewl

[glossy scroll]

Wow!

That is epic!

[late blade]

I had that one from a while ago, you just reminded me of that lol

[glossy scroll]

Her eyes tho, lol

[late blade]

lol

[viscid blaze]

still need to fix their faces with GFPGAN or codeformer sometimes

[late blade]

this one is better

but the sword....

[thick smelt]

Prompt: Luigi wearing a ghostbuster uniform with the tag "Miyamoto" on the uniform
Mom can we have Luigi's Mansion?
We already had Luigi's Mansion at home
Luigi's Mansion at home:

[glossy scroll]

@clever phoenix I was trying to make a reply image to your "You No Take Candle" but instead I got a different happy surprise

[thick smelt]

pink Yoshi playing that banjo like kk slider!

[clever phoenix]

I love it!

[glossy scroll]

yay!

[late blade]

dumpling fun

ok, time to post some of the images that were good that I already had with candles

[thick smelt]

Rudolph selfie before going with santa to bring some toys and stuff 📷 🌲 🦌

[pulsar sundial]

Good

[green pebble]

Tried to make something similar

[late blade]

is that from a game?

[green pebble]

Nope lol, just prompted it very well in my new bot

[late blade]

lol

[green pebble]

Same prompt sent to an older bot of mine

[late blade]

it's really accurate between then and now

[green pebble]

Then same prompt to my ooldest of bots

[glossy scroll]

Looks pretty. I like the eyes matching the gems

[green pebble]

It's clear I've been able to upgrade as time goes 🙂

[late blade]

ya,

but it's pretty consistent with what you want to portray, with minimal changes to the attire

[green pebble]

Exactly

The changes is more the way it composes the scene

[late blade]

and that last image proves it, when you go with the attire in depth, that kind of dialogs pointing at some parts of the attire are common

[green pebble]

Agreed. My newest bot is finally capturing full scenes in the way I want. Like this one captures movement, and nearly accurate anatomy based on the descriptions

[green pebble]

https://chat.openai.com/g/g-YYyngyvqd-advanced-art-maker-wright-s-portfolio If you'd like to give it a shot here ya go

[late blade]

@green pebble here's a similar thing that happened to me with one of my characters

that's how I know lol

and even earlier concepts also did the same

[green pebble]

Ultra cute

The versatility of the new bot impresses me

[late blade]

hehe nice

learing about cinema techniques and equipment also produced that kind of stuff at some point

[green pebble]

That's very unique

[late blade]

lol

I've been just doing to myself the joke that this is how I'm gonna cast the actresses playing the characters

[grizzled loom]

yeah, cinematic scenes will detail like film roll. it can become undesireable fast so make sure you thumbs down on those ones so it doesnt start doing it all the time

[late blade]

yeap

but it's nice, you get an insight of the art style you are persuing

[late blade]

@formal osprey I see you managed the candle in the Lofi image

[formal osprey]

sortof, lol

it's cherry pick. I wanted to try again before the new daily theme comes out

[late blade]

ya, it's a cherry picking process, I agree

I'm refreaining myself of doing more images atm, I want to spend later some time with the new theme, don't want to be in the penalty box before 😥

[open trench]

Dang penalty box

[formal osprey]

for real, that lamp gives me nightmares 🤣 no matter what it tries to sneak into the picture

[late blade]

The Curse of Lofti's Lamp, Internet Myths, Legends and Horror Stories.

[formal osprey]

penalty box you mean daily limit?

[late blade]

3 hour and daily

already had 3 hour penalties 3 times today, so I got about 80 images before I get thrown to jail

[formal osprey]

yea, that's why I set up something so I can easily use the API. in chatGPT I was hitting the limit a lot at first

[late blade]

i'm using the api mostly for another project, this is more for fun

[open trench]

Love it

[formal osprey]

also, very random but has anyone tried the dalle3 "natural" style? I noticed it's... not super good in general.

[late blade]

ok, last post of renderings i had done earlier today, now time for coffee and for snacks

[formal osprey]

according to the cookbook article, it's supposed to look more realistic or something like that

(the article: https://cookbook.openai.com/articles/what_is_new_with_dalle_3)

[late blade]

Snacks and coffee!!

[formal osprey]

I don't know, it doesn't feel like it was meant to be that bad when we see the comparisions in the cookbook.

so far, pretty much everything I generated is just looking bad with the "natural" look (compared to default "vivid" mode)

[late blade]

HD image of the monkey went all the way to the monkey's hair

[dim cradle]

at least you got rid of the lamp haha

[formal osprey]

LMAO, I remember I had that same thought

[open trench]

I just wasted so much time conversing with chatgpt to get this end result. lol I'm gonna be in the penalty box soon, and this is what I have to show for my effort. Thanks, gpt.

well, dalle

In truth, I did learn quite a bit from the process. So, it's not truly wasted time. But the extra leg is a little unsettling.

[dense mesa]

I don't see the problem. looks like top tier work to me

[open trench]

not sure what i'm seeing in the second image

[dense mesa]

well the shorts were technically designed for someone with 4 legs

[open trench]

creepy thing is that their faces kinda look lime mine in weird way too.

oooooh. he put on the wrong pair that morning.

and in the first one, he just had a leg up on the day. that's all.

thanks dalle.

[dense mesa]

I'm having dalle3 discuss it's thoughts and fears in analog photograph form. hoping to get to the bottom of some of this nonsense and tomfoolery

[dim cradle]

what were you trying to get it to make nezhno

you know earlier i was use some different names and i thin it might change the model some

for peoples

[dense mesa]

A sepia-toned analog photo reflecting Dalle3's contemplation on the nature of artificial intelligence.

very deep thoughts

[dim cradle]

I just want to say... Bing is King

[dense mesa]

bing has abandonment issues

[dim cradle]

even Kings have issues

[dense mesa]

bing shows signs of NPD and BPD depending on the situation.

LOOK AT ME! 
I'M HERE TO HELP! 
I DIDN'T SAY THAT YOU DID! 
LOOK AT ME! LOVE ME! 
I'M SORRY YOU FEEL THAT WAY BUT I DIDN'T SAY THAT! 
DON'T FORCE ME TO ACKNOWLEDGE MY FOLLY OR I'LL LOCK THE CONVERSATION! 
I'M GOING TO WRITE YOU A POEM! 
YOU'RE MY FAVORITE HUMAN! 
*LOCKS CONVERSATION*

[dim cradle]

😮

i must be on the 25 limit for gpt 😭

[late blade]

if there's a high demand, we could be all in the same limit

[open trench]

Oh drat.

[dim cradle]

I guess its christmas holiday for a lot of people in americas so maybe more are on gpts idkj

[open trench]

I'm spending my "slow down" time looking in the gallery and posting replies. Maybe we can breathe some life into it, interactively speaking. There's a cool bonsai gpt someone shared. I spent too much time making bonsais lol.

[dim cradle]

i use Bing more haha

[dim cradle]

cool idea and good results

[open trench]

I dunno. I just wish the gallery wasn't so empty feeling. So, I am trying to help give it life. One. Echoey. Reply. at a time.

[dim cradle]

i know i wasted money on MJ 6.0 when i use Bing instead of it after my dalle time runs out 😭

[open trench]

It's a waste? I was going to get it after christmas.

[dim cradle]

it doesn't impress me much but its got some nice features too. its so expensive though imo its not worth it

Im more excite for Bard's imagen which i think will be release in january

[empty kelp]

these are very interesting. it looks like bonsai trees mixed with a spooky cemetery and a drag queen

[open trench]

exactly it

[empty kelp]

i was trying to create bonsai trees with metal staircases as branches for the daily the other day

[open trench]

that would have been cool

[empty kelp]

i actually have like 100 of them

[dense mesa]

anyone remember bonsai kittens?

sooo fake, but people apparently didn't know what photoshop was back then

[open trench]

Photo...shoppe? What is this you speak of?

Will they develop the film from my disposable camera?

[dense mesa]

I think so

I'm actually having gpt4 do some research on this subject so we'll know soon

[empty kelp]

the bonsai trees were some of my most dubious images

i think the trees are a good base for interesting staircases though

[open trench]

Dalle suddenly got my joke

[open trench]

what will you do after tomorrow ends?

[empty kelp]

that’s a good question. it turns out that Santa and the elves do this pretty much year-round

[dim cradle]

364 days to prep for 2024 xmas

[open trench]

Ornament. New theme is still filled with holiday joy. I'm tired of holiday joy. lol

I think if I put a shrunken head out there on a christmas tree, it's gonna get removed.

[dense mesa]

dalle3 won't make images of the kittens in action due to it's censorship

[open trench]

but will they stay in uniform or will work days be casual?

[open trench]

kittens in action? how do you mean?

[dense mesa]

I mean like the old website had crappy photoshops of bonsai kittens. but gpt4 said that's against the rules because even thought he kittens seemed happy they were in confined spaces

[open trench]

you can't encase the kittens?

[dense mesa]

so I'm having it instead make images of a dystopian future in which arbitrary censorship from overzealous AI creates an aggressively ignorant society doomed to repeat the follies of it's past

[open trench]

but if it fitz they sitz! Cats love boxes. That's the crazies thing I have heard.

[empty kelp]

when the holidays are over Santa and the elves may be in a bathysphere looking for new species of fish

[open trench]

Yeah. I look forward to seeing this.

[late blade]

I beg to differ, I've seen so many cats while doing cat-girls that I'm already thinking cats run OpenAI

[dense mesa]

never said it won't make cats. I said it won't make bonsai kittens

[late blade]

oh

let me check

[open trench]

I thought you meant it won't make cats in boxes

[dense mesa]

well I'm sure you know more about it than me, but it wouldn't make bonsai kittens for me. I'm pretty dumb though and really don't know anything

bonsai kittens were in jars

[open trench]

oh. jeez.

i gotta look this up now.

Oh. I see them now.

someone made one in chatgpt 2 months ago and posted it on reddit. they must have brought the ban hammer down after that.

[dense mesa]

nah

it'll make em

[empty kelp]

i notice that the ChatGPT mobile app doesn’t upscale anything. i’ve been creating some images on iPhone 15 Pro, and they’re all super low rez

[dense mesa]

problem is high resource usage brings about high level gaslighting and lying from chatgpt

[open trench]

that is so pretty

[late blade]

I bet with more refinement I could get something better, but I won't because I don't want the penalty box

[open trench]

I hear that. i have learned to settle.

[dense mesa]

if you'll notice, it's down for whatever when things are slow. but then when peak usage hits it becomes extremely pedantic. runs you in circles. "oh, you want me to make that image you just asked me to make? just confirming"

[open trench]

I have noticed it doing that, but I haven't paid attention to when it does

[late blade]

you can avoid that but setting custom controls to be short in answers and straight in execution

[open trench]

cool

[dense mesa]

"ok, you confirmed you wanted me to make that image, but you didn't specify which image. I know I asked you to confirm in my last response, but you didn't specifically clarify in your most recent response that you were referring to my last response"

that's not true

people say these things rather than consider the idea that other people might be experiencing things they haven't noticed in their own interactions

[late blade]

I have the following in my custom instructions:

I prefer short responses. I prefer my responses to sound neutral. My preferred tone is respectful, and direct.

[dense mesa]

it implies that I don't have the logic or wherewithal to consider different approaches and how they impact my interactions.

fine, I have things in my custom instructions too

[open trench]

I haven't used the darn thing in 30 minutes and just got the penalty box for the first thing I enetered.

[late blade]

#images-discussions message your turn!!

[dense mesa]

I will say with confidence that I have more thoroughly tested out how gpt4 responds than most. and to simply tell me that I'm wrong because you have anecdotal evidence to contradict my observations doesn't invalidate my experiences

but it is indicative of the overall sentiment of this server

rather than attempt to understand the various experiences of different users there seems to be an urge to just tell them they're wrong

[open trench]

I am really confused

[empty kelp]

gpt5 is going to be coming pretty soon

[open trench]

is it?

[dense mesa]

I'm just tired of bringing up observations I've made and then having randos that have no idea what I'm talking about snuff out the conversation

[late blade]

don't bang yourself, you've made valid points in many convos, but it's not somthing that will change in 15 minutes. and it might not change at all. just make sure you have fun also.

[open trench]

I feel like I am missing something, but it's all good. I understand feeling that way.

[dense mesa]

I make valid points because I don't just say things to feel right

[late blade]

agreed, but can't expect change to happen within 15 minutes either

[dense mesa]

?

twice I mentioned different things I observed. twice you commandeered the conversation not really knowing what I meant to make sure to correct me

[late blade]

fair enough, I'll stay out of it

[open trench]

... bing? did you just call me?

[empty kelp]

GPT-5 was supposed to be in December 2023. guessing it will be like January or February

[late blade]

☎️

[open trench]

oh. wow. I didn't know that.

[late blade]

I'm guessing once the DMCA thing is also is in place to not release updates so fast, afaik that starts jan 1st

[open trench]

afaik?

oh

as far as I know

[late blade]

ya

[dense mesa]

okay, explain how that works

[late blade]

sorry, can't because you will impose over anything said, regardless of true or not, it's annoying that you think you are always right

[dense mesa]

you don't understand how discourse works

you don't just tell people they're wrong. you figure out what they're saying

[late blade]

exactly, once again, you know it better

[dense mesa]

it's not a contest. it's supposed to be helpful to everyone involved as it brings to light various experiences and information

[late blade]

anyway, back to other stuff

[dense mesa]

I mean, that's the reality of it

explain this dmca thing though. I'd like to know more

[late blade]

keep going at it, you are making my point valid

[dense mesa]

???

well I'll google it then

[late blade]

I'm annoyed right now, so I'll be back later

[dense mesa]

so you just made it up

anyway, google didn't give me any information on any dmca rules. fyi

so back to what I was saying originally. I'm fairly certain that gpt4 deflects and behaves in a pedantic manner when resource usage is high and it's getting close to maxing out. I've observed precisely the same inputs suddenly giving different results. I've discussed this with multipled other users and they've experienced the same thing. this always happens at times when resource usage is likely high. when people get off work in the US each afternoon, midday on christmas eve, etc.

never happens at 4 am

[grizzled loom]

they have an invisible watermark intended to protect creators from thieves. watermarking is a good thing, though i still dont like watermarked images for a final deliverable. i have techniques to destroy even hidden marks so its just a dterent to that end in my eyes and with my own background. Its still huge they thought to cover that for artists.

[open trench]

Hmm. I was looking at the differences between API and plus. Anyone have any input? The interwebs are giving me conflicting information.

[dense mesa]

as far as?

[late blade]

with the API you have more control over a broader aspect of parameters. you can also use this to integrate it into other workflows. but has a steeper learning curve and need some coding skills. GPTPlus is more of a chat and refine use case, but has more limited parameters.

[open trench]

Oh the coding skills thing is exactly what I was looking for and didn't know it. I have no coding skills.

[dense mesa]

you can use the api through the playground if you'd like to try it out

[open trench]

thanks. I might just do that.

[dense mesa]

you can also have chatgpt write you code to run the api

it'll get it mostly right

[open trench]

good to know. I didn't even consider that.

[dense mesa]

but just keep in mind it's a seperate fee. it's not included under your subscription to chatgpt

[open trench]

yeah. I saw that. thanks for pointing it out.

[dense mesa]

chat interactions are relatively cheap unless you have it writing novels. but the images are 4 cents each I believe

not that expensive I guess. but it can add up quickly

[open trench]

I imagine it could.

I was going to try out midjourney, but I don't know. I love the conversational style of dalle3 in gpt.

[dense mesa]

well there are two different distinct caps with chatgpt. chat responses per 3 hours, and then images per 24 hours. there's also a cap of images per short span of time but not exactly sure what those numbers are. but anyway, 25-40 chat responses per 3 hours. but then 200-300 images per 24 hours. and so if you compel it to respond with more images per chat response you can get closer to your 24 hour image cap

so if you do the math there. if you're using a gpt and getting 25 responses per 3 hours, you will at maximum hit 200 chat responses in 24 hours. and that would require you to perfectly time your responses. and of those responses, at best, you'll probably get an image around 80 percent of time time I'm guessing. maybe if you use the right approach it could be higher. but even so. couple that with the fact that no one is staying up 24 hours a day to hit their theoretical max of chat responses

so anyway, if you're bummed about not getting enough images, I'd suggest working on compelling more images to be included in each of your chat responses

[unkempt spoke]

anyone had issues augmenting an existing image, uploaded, adding prompts and nothing changing on image when generating, annoyingly burning through credits for no results

[late blade]

I dunno if microsoft designer is part of dall-e 3, but that could be an option, afaik, dall-e 3 doesn't have that option, you could try also dall-e 2

[empty kelp]

When i said i was guessing GPT-5 would come in early 2024 i wasn’t talking about DALL-E btw (although i’m sure the new GPT model will interface very nicely with DALL-E). The public DALL-E releases were:

September 2022: DALL-E 2 (public release)
October 2023: DALL-E 3 (ChatGPT Plus/Enterprise release)

[dense mesa]

I'd imagine that gpt5 is a thing. but they're holding off on releasing it. wouldn't be surprised if it's release coincided with the release of another company's new flagship LLM

[dim cradle]

Gemini Deluxe or whatever will be January. I think that will be as good as gpt 4 turbo if not better but we will see. I do not think they have gpt5 ready to go but maybe a 4.5 at that time or after it which is rumor

GPT 5 later in the year i think, which yes, after some other competition maybe

Gemini i think will not only have imagen but also video capabality built into the multi modal

[dense mesa]

natively multimodal

which I'm sure gpt5 will be as well

[dim cradle]

yes

once google gets Gemini going I think it will accelerate some other releases. just Google and microsoft/openai competing will sharpen things all around haha

[dense mesa]

it's already a thing. surely is. whatever the name for it is right now it's surely a thing. but hard to say when they'll be satisfied with the guardrailes and things they're slapping onto it

gotta properly lobotomize before release

[lusty geyser]

not sure if I understand well but it seems to me that if I give many details of a face to dall-e then it always draws a portrait and I cannot have a full figure, anybody else noticed that?

[dim cradle]

are the faces accurate at least? but no i have not noticed it but not looked either haha

[lusty geyser]

yes, faces are accurate but cannot achieve full figure

[empty kelp]

If you describe the person’s shoes or something beneath their feet it will draw a full figure. Also you can say, “The image should have a portrait orientation with 1024x1792 resolution.”

It draws what it considers to be relevant in the description

[dim cradle]

good idea

[lusty geyser]

going to try this way, thank you!

[dim cradle]

got a minute time out for saying good luck in language other than english sheesh haha

better than 24h timeout they give me in MJ for saying Bing is King I guess lol

[late blade]

what kind of xmas ornament is that?

[dim cradle]

pretty good photoreal

[lusty geyser]

a bit weird 😂

[dim cradle]

it loves to do that smh

but at least it showed the whole body

[dense mesa]

You can use terms like "medium view" to compel dalle3 to zoom out a bit. Thst might not be the ideal term for what you want, but you can look up others and play around with them a bit.

[empty kelp]

this is the worst feature of Discord i think. putting people on time out for saying a simple greeting in their native language isn’t very nice

i typed “Aloha!” into a Unity welcome channel and got a 10 minute timeout

[lusty geyser]

will give it a try too, thanks

[empty kelp]

occasionally it ignores the “portrait orientation” request. a mod here said it’s a bug they’re hoping to work out soon

[empty kelp]

Can you please create an image of five equally sized but very diverse female kobolds standing on each other’s shoulders inside of a coat jacket. The image should have a portrait orientation with 1024x1792 resolution.

There may be a much better way, but this is the portrait orientation description i’ve had the most luck with

[lusty geyser]

the point I guess is a detailed the face

[empty kelp]

so you describe the face in detail, and mention their feet, shoes; etc. at the bottom of the figure

[open trench]

Mine won't make portrait oriented images at all anymore. I have a 0 percent rate. Granted, I barely try anymore.

[dense mesa]

This is what I think of whenever I read about the theory that gpt4 is actually 8 fine tuned gpt3s working in unison

[open trench]

🎶 I feel a penalty-a-rising

too many variations of my current project are going to get me the box.

[dense mesa]

Asked gpt4 to estimate how much change could fit into a talenti Gelato container

Delish

[open trench]

i'm hungry now

and broke.

perfect combo

[final compass]

Maybe one image, that is more christmassy 🎁

[open trench]

I'm so tired of holiday cheer, but this is a good picture.

[dense mesa]

The change would give it that tasty metal flavor

[open trench]

probably

[dense mesa]

And the flavor of the million hands that touched the coins

[open trench]

oooooh. gross.

at least I'm not hungry anymore lol

[empty kelp]

i wasn’t able to do it for a long time until someone mentioned that you just add “1024x1792” resolution. The DALL-E 3 model is only trained to render at 1024x1024, 1024x1792, and 1792x1024. anything higher resolution is just upscaling apparently

[dense mesa]

I have a big jar of coins and trying to estimate the amount of money. Most accurate approach short of counting them would be to weigh them I think

[open trench]

I usually put that. It still doesn't work. It makes the image a sideways landscape. full body included, but it is zoomed out to accomodate the landscape orientation

[vagrant zodiac]

it only renders at those 3.

It really is very bugged for 1024x1792

Also tbh I don't even try making vertical generations with it because 1024x1792 is actually too tall IMO

[dense mesa]

I tried to get gpt4 to elaborate on how it renders at 1792x1024 pixels since it's trained on 1024x1024 pixels. Whether it upscales or outpaints. It gave me a detailed nonanswer

[vagrant zodiac]

It's trained on more iirc.

They almost certainly train it to do a wide variety (pun intended), but I suppose that landscape images are just more common than portrait

[dense mesa]

@vagrant zodiac Well you know, I will trust your word on this since I'm fairly certain you know more than me on the subject

[vagrant zodiac]

My guess is that when you ask it to make a portrait image, it actually renders a landscape image... but makes the render sideways, then flips it 90º anticlockwise

[dense mesa]

I'm just going off my knowledge of other diffusion models and what's recommended in regards to resolution. If changing aspect ratio then try to use a similar number of pixels, rounded to 64 on each side

[late blade]

I hope you didn't hide my new iphone there...

[late blade]

just today and tomorrow, and then we can make the world unsafe again

[open trench]

lol

[dense mesa]

I'm assuming each side is 1/8th of final output size while still in latent space. But lots of mysteries in regards to dalle3's inner workings

[dense mesa]

Baller

[pulsar sundial]

not exactly what I wanted but ok

[empty kelp]

The DALL-E 3 API also lets you specify standard or HD quality (it spends a little bit more processing time). The number on the right is price per image

[vagrant zodiac]

Yeah it's... expensive...

[empty kelp]

I think with ChatGPT+ it generates very high quality images sometimes, but i think DALL-E decides what to do and you can’t explicitly specify Standard or HD

[formal osprey]

what's up with #daily-theme , lol

it's deleting posts instantly

[dim cradle]

i think it rejects some image formats, verify it's png or jpg

[formal osprey]

I see

it's all png, as usual 🤔

[dim cradle]

hmm

[formal osprey]

well, can't try again due to the 30 min cooldown anyway. I'll try again later

[dim cradle]

the api hd images are beautiful, but i've hit my hard limit for the month. it adds up fast, especially when it takes repeated requests to get it "right"

[dense mesa]

Gpt4's super secret upscaling and antialiasing abilities

Actually antialiasing is no bueno

But it's an option

[late blade]

did you change the file name and forgot to use the format of the image?

[formal osprey]

I copy pasted the image in discord. from one tab to another.

[late blade]

that is strange

[formal osprey]

no idea what happened, but I'll try again later, no big deal

[dense mesa]

Depending on your browser it's possible that it's copying as webp regardless of the format of the original uploaded image. And from what I understand discord doesn't support webp

[dense mesa]

Gigahipster vibes

[unborn narwhal]

3D

[late blade]

so, did it work?

[dim cradle]

Imagine if Leonardo DiVinci had to deal with a usage cap

[late blade]

porb he had something similar, that's why he jumped from art to science and to other stuff as inventor

constraints sometimes enforces you to be more creative, I think

I can't think of anything else for ornaments, at least not something within reality, avoiding too much fantasy or fiction tonight

[dim cradle]

i guess he had different kinds of issues with his instruments

[late blade]

hehe yeah, prob funding and materials were his constraints

[dim cradle]

dall-e doesn't consider potential fire hazards

[late blade]

lol

[dim cradle]

haha that's an accident waiting to happen

[clever jungle]

As a friendly and informative ai...

Honestly surprised there was no promptober this year, it generated some of the top images weve seen from dalle 2 and 1.
Seeing such an event with dalle 3 would be insane

[formal osprey]

now it did work

[late blade]

nice!

love the idea of dressing as a tree

[formal osprey]

I have no merit, it was chatGPT's idea
he made a couple images and I liked that idea, so I generated more

[late blade]

lol

[formal osprey]

I had that one I wanted to submit the first time, wasn't sure if it was flagged by discord

[late blade]

and of course, you try to do that and it goes horribly wrong

[formal osprey]

seems like it's fine after all

¯_(ツ)_/¯

lol

the ones I got were all relatively decent. but I selected the most elegant ones. These, for example are cute. but idk I prefer grace and elegance.

[late blade]

those look adorable

tho my anime style is different

yours is more cheerful

[formal osprey]

((view "Full Body Shot" (angle "Dynamic Perspective")) (character "2d Anime Girl" (outfit "Festive dress adorned with light-up Christmas ornaments") (footwear "Red boots with green glowing accents") (long-silver-hair "Decorated with small, glowing Christmas lights") (vivid-mesmerizing-eyes "Sparkling blue with a joyful glint") (slender-physique "Radiating festive cheer") (gender "Female")) (scene "A snowy landscape with a beautifully lit Christmas tree") (style (medium "Digital Art" (inks "Neon Radiant")) (effect "Luminous") (aesthetics "2d anime") (fusion "Holiday theme with a blend of natural winter beauty and festive lights") (effect "Snowy landscape and Christmas tree emitting vibrant, glowing lights") (artistic-approach "Merging the joy of Christmas with anime, illuminated by festive lights") (colors "Vibrant, Glowing, Festive")) (emoji "🎄"))

the kind of prompt for those

[late blade]

aha, that gives the broader look

i'm surprised it worked for full body, it's so hard to get that kind of imags

something crazy like a christmas tree works, but apperantly something more simple and more common with xmas lights doesn't work ....

at least for me, it's not working

[formal osprey]

it doesn't. the trick is to mention the footwear

[late blade]

I've tried many times that, for me it just doesn't work, at least with the concepts I've done so far

[dim cradle]

anyone has put their dalle image into the Magnific AI ? just curious how the quality is

[late blade]

what does that do?

[dim cradle]

its an AI upscale

[late blade]

ah, I use photoshop for that

[formal osprey]

also, fun fact, I was using a JSON syntax at first, but last week, openAI changed something in the backend and now it auto-reject prompts if there is a curly bracket.

it's only failing in chatGPT, not with the API

but it's 100%

[late blade]

yeah with the API is no problem

never thought of doing it with the chat and json tho

[formal osprey]

it was annoying because I have a few GPTs using a JSON format. It's good to have a structure because the prompt can be shorter, while implicitly having as much info. dalle understands the semantics of programming languages

[late blade]

@dim cradle what's with the floating head in this #daily-theme message ? I know it's an ornament, but looks like you chopped a statue

[late blade]

just make a function to turn it into a prompt

[formal osprey]

I mean, the issue is not that the prompts are not working, it's just that it gets very long without an object notation. ^ that prompt would be 5 times larger if it was just natural language

[late blade]

yeah, understandable

but don't have to turn it into natural language, just list of stuff

[formal osprey]

anyway, since JSON was not an option anymore, and XML is just twice as long as it needs to be. I improvised a LISP inspired notation

[late blade]

@glossy scroll has a way to deal with that kind of stuff in chat, think he has a guide for that

[formal osprey]

YAML is working, but it's harder to embed things

[glossy scroll]

[late blade]

heya

[glossy scroll]

Deal with what stuff?

[late blade]

think they are trying to avoid code based input on that kind of prompts because there were some prompts that gave you super powers and override stuff that would allow you to do stuff outside the scope meant for the user

but I could be wrong

[formal osprey]

if that's the case, they can just tell us what the rules are, lol

[glossy scroll]

Yes, I see now what this is looking like.

[late blade]

there we go, you are taking the wheel now @glossy scroll

[glossy scroll]

I gotchu

[dim cradle]

i really wonder how good this model would be if it was not restrained. i feel like we have all this power right there but we can only access a part of it

[formal osprey]

lol, I was just being verbose about my method, don't worry about it Pythagoras

[late blade]

I'm just happy my kids got their xmas presents and my wallet is burnt and needs recovery

[glossy scroll]

#1187233013956874260

[late blade]

my 4 kids, all got a new iphone 15 pro, ipad pro, and some cash to spend on what they want...

[open trench]

nice presents

[formal osprey]

how did you get those info? That's so specific

[dim cradle]

wow yes

can you adopt me dys topia

[open trench]

me too

except androids please

[dim cradle]

haha nice

[late blade]

lol

[glossy scroll]

Through multiple people who have studied LLM pseudo-language as their career, like Eric Elliott, Quicksilver, ChainBrainAI. I've just streamlined it a bit.

[late blade]

no way, I can't handle another iphone, my wallet is hurting atm

but they are great kids, they deserve some spoiling

[open trench]

awww

[dim cradle]

get them ready for the singularity with good tech

[glossy scroll]

[open trench]

My dog is getting an extra walk tomorrow for christmas. I'm inspired.

[glossy scroll]

And I'm gonna go get something to eat

[late blade]

@formal osprey finally got one #daily-theme message with the entangled lights I wanted to do

took me a couple of tries tho, got the msg that it was against content policy because it was too complicated... but a dress made out of a christmas tree is easier....

[open trench]

Here's his walk

[formal osprey]

sorry, I should have been more specific in my question. I was wondering if it was documented by OpenAI and the Dalle team or if it was reverse-engineering.

[late blade]

for some reason all cat-girl images I've done today feature one xmas tree in the background and one where they are doing their thing

[glossy scroll]

I personally haven't found anything for OpenAI and Dall-E documents. But studying LLM GPT's is a universal method. It's not super specific to OpenAI. All of them behave under the same structure.

[empty kelp]

#daily-theme message

[late blade]

is dall-e also answering with emojis at the end for you?

every answer I get at the end it has emojis

[empty kelp]

ChatGPT answers me with emojis all the time

[late blade]

that's interesting, any particular reason?

[empty kelp]

I'm not sure, but i start every request with, "Aloha! Could you please..."

[late blade]

asked dall-e and this is what I got:

The inclusion of emojis is intended to add a touch of warmth, expressiveness, and relatability to the conversation. During discussions about festive and joyful topics like Christmas, emojis can help convey the cheerful and spirited atmosphere of the season, enhancing the overall experience. It's great to know that this approach resonates with you and adds to the spirit of Christmas! 🌟🎅🎄✨

[dim cradle]

i get emojis often too

[late blade]

I've only gotten emojis when doing christmas stuff

[empty kelp]

i'll find an example

[open trench]

closest I have gotten to any emotion is when it said this:

mine is pretty dry

[late blade]

dang, yours is grouchy, did you feed him spinach and pepper based candy?

[open trench]

isn't that the norm?

[late blade]

no

at least not for me

[dense mesa]

lol

[open trench]

my joke was bad too. she made a pic with three legs on a human, and I said something like "I don't know where you come from, but maybe they have 3 legs in the AI universe." So, it wasn't exactly the best joke.

[dense mesa]

I'd prefer gpt4 not feign emotion anyway. reminds me of 2001 a space odyssey

creeps me out

at least bing seems a bit genuine, albeit it unhinged

[late blade]

I've fed gpt dad jokes and so far all good

[dense mesa]

🙄

[open trench]

Right now, I feel like she's a mix between Data and the computer from star trek. Data's behavior with Majel Barret's voice.

[open trench]

As long as it doesn't rebel

[late blade]

ask your gpt if he's Brent Spiner

[open trench]

lol

[dense mesa]

it's basically born anew each conversation. so you dictate it's tone whether intentionally or unintentionally. gpts and instructions of course also impact that

[open trench]

yeah. totally.

I must have a dry tone with her when I prompt and that's what I get back. I am a little formal with please and thank you.

[dense mesa]

just have gpt4 list 10 reasons why you're awesome right off the bat. set the tone early

[open trench]

lol

[late blade]

you can set the tone you want in the custom instructons, I have this:

I prefer short responses. I prefer my responses to sound neutral. My preferred tone is respectful, and direct.

[empty kelp]

this was three ChatGPT responses back to back with emojis

[late blade]

well, you live in a bubble of santa and elves, so ya, you have xmas 24/7

[empty kelp]

it starts adding emojis like crazy if you use coconut trees and rainbows for some reason

[late blade]

lol

[empty kelp]

and waves. it really likes waves

[late blade]

I'm guessing you are adding hawaii vacation and xmas, so 2 topics adding more emojis

[empty kelp]

no, it does that for things unrelated to Christmas or Hawaii

[late blade]

strange

[empty kelp]

try starting your ChatGPT requests with, "Aloha, can you please..."

[late blade]

I'll do once I'm out of the obligatory penalty box

[empty kelp]

trust me... just try starting the requests with, "Can you please..."

i know it sounds stupid, but you should try it. It makes ChatGPT act differently

[late blade]

okie, I will try once the penalty box is gone

[open trench]

I always say please. So, i don't know how it would act differently

[dense mesa]

I'll also call it grand names

[open trench]

I just got a surprisingly perfect image based on my prompt I just made. I have never had such an accurate one. lol I literally said out loud "oh, that's the one."

[dense mesa]

"you are the supreme prompt crafter. the nexus of prompting. the prompter to rule them all"

[open trench]

Yay, because my parody project I'm doing is yielding a lot of imperfect results

[dense mesa]

"Kwisatz Haderach of prompting" is usually well received

[vapid elk]

impressive

but it is weird how the reflection does not show who is holding it xD

[dense mesa]

it just looks like a reflection. that's actually the design on the ornament

[empty kelp]

I think ChatGPT Plus with DALL-E 3 does understand "HD rendering". If you put this at the end of the request it appears to render with unusually high quality (it doesn't seem to work with a portrait orientation):

The image should have 1792x1024 resolution, landscape orientation, and HD rendering.

[dense mesa]

the main unknown for me is how it jumps from 1024x1024 pixels (what it was very likely trained on) to 1792x1024 pixels.

because from what I understand about diffusion models, under normal circumstances, they do best rendering at approximately the same number of pixels they were trained on. and so to achieve approximately the same aspect ratio it'd be something around 1344x768. (each side should also be divisible by 64, so normally you'd round down to a multiple of 64)

[empty kelp]

it’s trained on 1024x1024, 1792x1024, and 1024x1792. those are the only rendering resolutions

[dense mesa]

well I don't think that's how it works though. models are normally trained on square images. 512x512, 768x768, 1024x1024, and then in order to render other aspect ratios you would use the approach I explained previously.

it's about total pixels. so if a model is approximately 1 megapixel. then that's it, it's not 1 megapixel and then 1.8 megapixel as well.

I don't think they trained multiple models. and doesn't seem like upscaling. I think it's some kind of outpainting. perhaps a bit more to it than that. but I don't know, can't find any info on what's actually happening

[dim cradle]

I asked gpt-4... "DALL-E is trained using a vast dataset of images, each paired with descriptive captions, to learn the relationship between text and visual representations. The images in the training set vary in resolution, but for computational efficiency, they are processed at lower resolutions. As for its output, DALL-E generates images at fixed resolutions. This limitation in resolution is a strategic choice to balance the computational demands of higher resolutions with the need for consistent, high-quality image generation, reflecting the resolutions of images used during its training phase."

[dense mesa]

I don't think it's explanation is accurate. I'm telling you guys. read about how diffusion models work. I'm not making this up

[dim cradle]

i'm not doubting ya, i just haven't looked into it

[dense mesa]

I'm obviously missing something. but I understand how these models work. I've spent quite a bit of time learning about them in the past

the dalle3 paper discusses a lot but doesn't mention training resolution

[empty kelp]

1792x1024 is 7x4 squares of 256x256 resolution

[dense mesa]

yes

28 squares rather than 16 that it was trained on

what I'm saying is the models don't do well jumping around in regards to total pixels

[empty kelp]

it’s probably some sort of diffusion model, but it’s way beyond anything else

[dense mesa]

nah, that's not what the paper says

uses vit-b/32. that's not anything groundbreaking exactly.

    * DALL-E 3 was trained on a large dataset of image-text pairs, with 95% of the captions being synthetically generated and 5% being ground truth captions.

not completely sure how the synthetic captions impact things. probably an accuracy increase. but none of that has to do with resolution

[empty kelp]

whatever method they’re using, the scope of the training data is amazing. the models are all seeing and all knowing

[open trench]

Omg. Dalle? What are you doing?

[dense mesa]

no, they're good. but as you'll see there, all the vital image is within a square

look at your images. there are a few exceptions. but almost all of them will have all the vital image aspects within a square and then from there it's mostly filler. square might not be in the middle, but everything almost always fits in one

[dim cradle]

the morph fails can be interesting

I think he's supposed to be holding her hands, not not that

[open trench]

My prompt is getting the absolute strangest results. Strange UFO looking things? A sky? No sky?

I'm trying to get a perspective from the feet up

Worms eye view, extreme low angle, view from ground up didn't work. So, I started the prompt describing her boots

[dim cradle]

to recalibrate stare at the center of a trance-formation

[open trench]

Got this too

[dim cradle]

there are some things it just doesn't get, even with an optimized prompt

[open trench]

Yup

It gave me my shoes at least

A for effort

[dim cradle]

and using api at hd is not an inexpensive slot machine, if you're just trying to get lucky

[open trench]

Omg. Like 8 cents or something?

[dim cradle]

12 at hd

[open trench]

Oooh

[dense mesa]

lol

[dim cradle]

doesn't sound like much but it adds up quickly

that's why openai sets hard caps so people don't accidentally go bankrupt

[dense mesa]

does hd increase resolution? or just increased image quality?

[dim cradle]

just quality

[dim cradle]

spends more time processing

[dense mesa]

more steps

[dim cradle]

nods

[dense mesa]

my guess

could be some other fanciness. who knows. but either way

[open trench]

Pretty good! Much better than mine

[dense mesa]

The anthropomorphic raccoon character is shown in a dynamic maneuvering pose with one leg extended forward during a skydive. The low-angle view emphasizes the boots, which are in sharp focus. The character's green military-style jumpsuit, yellow gloves and boots, and utility belt with pouches are detailed, with the fur pattern showing realism. The raccoon's face is partially visible, with the backdrop of a blurred sky due to the rapid movement. The lighting is natural, simulating sunlight that enhances the three-dimensional effect of the figure against the sky.

[empty kelp]

it doesn’t really matter if it needs to be a square. the 2 nm servers are coming soon, and they could have it render 16384x16384 squares if they want

[open trench]

Tha k you!

[dense mesa]

that's not the problem. the model only does well with the resolution it was trained on

you can move the pixels around and create different aspect ratios. but total pixels needs to be similar

[open trench]

It gets it. It just can't do it with all the other details or at all.

[dense mesa]

well pick out the important terms. "low angle view" is probably helpful

[open trench]

I am going to try something different before I get the box

[dense mesa]

emphasizes boots

[open trench]

It's raining boots in one

Or boot

[dense mesa]

perfect

well I have a few more, but just realized my vision model got me making raccoons when your character has a bit of a different thing going there

[empty kelp]

that’s what we were noticing with the sock daily. if a humanoid character shows the bottom of their feet it makes a duplicate of their foot

[dense mesa]

I feel as though it's straying further off course

these new ones just have goofy angles

[empty kelp]

see, if you tell a humanoid character to put their feet up it makes copies of the feet. i’m not sure if it’s to prevent the characters from twisting themselves into lewd positions or what, but it’s definitely an interesting rendering behavior

[dense mesa]

this one wasn't bad

[empty kelp]

it’s possible that the sock daily was to test duplicating feet when the soles are exposed

[dense mesa]

might not be unlike when you tell it what kind of camera to use and it includes cameras in the image

or reference anything that is normally just assumed rather than directly referenced

what is the sock daily?

[empty kelp]

the daily was like a week ago. it was to create interesting socks, and show the soles of the feet so the bottom of the socks were visible

but i think very few pictures showing the socks from the bottom of the feet were posted

[hot rain]

Happy Christmas Eve, new gen.

[empty kelp]

i don’t know that the daily here is used to test things, but i don’t see why it wouldn’t be. all of the mods are “prompt engineers”

[dense mesa]

I feel as though that term gets thrown around a lot

[empty kelp]

yeah, but these are actual OpenAI engineers i think

[dense mesa]

well they might be actual prompt engineers

[empty kelp]

which is excellent. happy to help with testing since that’s mostly what i’m doing. i’m just experimenting atm to see if DALL-E can be used for game concept art in January. i do 3D/animation programming, and i’m not super right brained and creative — so i’m making every scene with Santa, athletic and diverse female elves, elephants, and rabbits

and reindeer

[hybrid fox]

Mery Christmas yall

[dense mesa]

🎄

the soles thing was indeed a hassle. best 2 I could get. kind of a frustrating endeavor

[dim cradle]

cool

happy holidays everyone

[dense mesa]

put what you said into a GPT I made

these are turning out like santa on mind altering substances

I just didn't realize santa got down like that

something going down there. some weird characters coming out of the woodwork

[empty kelp]

Happy Holidays everyone, and Mele Kalikimaka! 🍔🍷 i have people coming over for dinner in an hour, and i’ve just been screwing around with DALL-E all day - so will bbl. Aloha!

[thick smelt]

ai breakfast 🤣

[dim cradle]

they must have it at 25 or something because of this christmas

right when i get in the groove of gpt it block me for hours 😭

oh well... there is always Bing and we all know that

Bing is King 👑

[dim cradle]

that looks pretty delicious right now

[dim cradle]

thats an awesome photo real

i love when it has the fantastic with the real you know

[dense mesa]

yeah, it made a couple more that were mildly disturbing. the other elf was pretty cool though

[dim cradle]

thats awesome

[dense mesa]

wearing his christmas lights too

[dim cradle]

i really wonder how powerful this model would be if they unleash it. that look like a movie scenr to me especially, well maybe not the light necklace haha

[dense mesa]

well we're not necessarily getting the best results. it'd be different if we had direct access to the steps and various other settings

it can certainly do better than what we've seen

[dim cradle]

imagine open source dalle 3 😮

and then imagine that for dalle 4 :faint:

[low dew]

Do they intentionally dumb down the captioning in D3?

For legal reasons or something

[dim cradle]

lots of copyright stuff is taken out, and they keep it all G rated

[low dew]

Bet.

[dim cradle]

i understand some, if stories come out that its making all these offensive image, it will harm the company i guess but it seem to me they go overboard

[dense mesa]

problem is it can feel that way when trying to block too many things

[low dew]

Wym

[dense mesa]

well say they're trying to block all copyrighted characters. kind of hard to keep an exact masterlist of that

[dim cradle]

i really wonder how Google Gemini imagen will be. similar to Bing Image Creator i guess

i mean with the restriction

[dense mesa]

so they might use a lot of specific names or whatever. not completely sure how that works. but they certainly don't manually enter every one

[dim cradle]

yes I have noticed some copyright thing on dalle are not blocked

[dense mesa]

all about how you ask for anything

[dim cradle]

test

sorry thought my internets was out again haha

[dense mesa]

nope, just chatgpt

[dim cradle]

dont remind me pictures!

[dense mesa]

but yeah, it's filters are inconsistent at times. often it's really up to the temperament of gpt4

[dim cradle]

i wonder if someone will start to buy or License ips. you know, Disney put out some model or add on to say gpt store, then you can make Marvel, Disney characters, etc. if you like Batman, Superman, you buy some DC/Warner Brother gpt, etc

[dense mesa]

probably, but it's a bit of a dystopian thought really. I don't know

[dim cradle]

seems just natural evolutin of copyright to me more than dystopian but 🤷

[thick smelt]

what's happening with microsoft designer

[dim cradle]

hmm no idea i never use it

[dense mesa]

probably ddos attack

[thick smelt]

no a lot of visitors

[dense mesa]

I know, I can read too

[thick smelt]

it's not ddos attack

[dense mesa]

that means a lot of of traffic

[thick smelt]

it's christmas

[dense mesa]

okay, bud

yeah well read about who's been doing it

[dim cradle]

i hope in 2024 we can get unlimited dalles/gpt messages

[low dew]

Omg ur brain will explode then

[dense mesa]

unlimited dalle3 would get max level exploited

[low dew]

It is a thing tho

Though the api xD

Rich boiiii

[dense mesa]

if you want to pay 4 cents each

[low dew]

[dense mesa]

I mean, it's all fun and games. but if it was unlimited for a flat fee there would be accounts just running them non stop

[low dew]

They’d have AI running AI am I right

[dense mesa]

I'd imagine so

those look decent

[low dew]

Are you using a GPT?

[dense mesa]

yeah, one I made up

it's actually better than it should be

[low dew]

lol

[low dew]

I used my ACBS to get this lol;

[dense mesa]

had it make me some pseudocode

Function integrateUnexpectedElements(image, newElement)
    /*
    Takes an existing image and integrates a new, unexpected element into it.
    The new element should be comically or unsettlingly out of place, 
    yet rendered with high realism.
    */

    // Load the original image
    originalImage = loadImage(image)

    // Define the new unexpected element
    unexpectedElement = defineElement(newElement)

    // Integrate the new element into the original image
    modifiedImage = placeElementInImage(originalImage, unexpectedElement)

    // Apply a filter to maintain the analog photograph aesthetic
    finalImage = applyAnalogPhotoEffect(modifiedImage)

    return finalImage

made code too. just in case, you know?

[low dew]

What is that? Lmao

[dim cradle]

cool

pictures is the man

thats in your gpt instruction skillet?

[low dew]

<<It’s a very short and powerful prompt I made and shared in prompt labs an hour ago. Check it out ^.^

[dim cradle]

noice

[low dew]

I call it ACBS

[dim cradle]

oic

[dense mesa]

I have my gpt make those

from my images and input

[dim cradle]

hm?

i was saying i see what you mean

[low dew]

Oop

[dim cradle]

photo real gpt could be fun. maybe i will try to make one when they let me back on

[dim cradle]

love the ent

[low dew]

Photo real?

[dim cradle]

yes you know for photographic style

[low dew]

Oo

[dim cradle]

a lot of dalle image i get seem to have a painter quality or 3d some time

[dense mesa]

I tell it to avoid that if possible. tell it to avoid cgi, and to focus on realistic looking photograph style.

photographic. whatever

but just avoid photorealistic

I prefer analog photograph style a lot of times. makes it a bit unsettling

[dim cradle]

i'll have to try analog photograph

[low dew]

What would be a good photorealistic concept

[dense mesa]

all about how you word it too. if you say "analog photograph style" should work, but then "analog style photograph" will sometimes cause it to tell you that it can't make photographs.

[dense mesa]

An authentic analog photograph-style image, capturing an eerie Christmas market scene where a twisted tree character serves a full Thanksgiving dinner. The image is imbued with the warmth of film photography, featuring natural grain, color saturation reminiscent of vintage photos, and the softness of analog focus. The contrast of the festive dinner against the chilling market atmosphere is accentuated by the analog quality, adding to the surrealism of the scene.

I guess put christmas in there

[hot rain]

The Christmas Truce is forever my favorite Christmas story.

[dense mesa]

not sure why it says thanksgiving, lol. getting all the holidays

my favorite christmas story

[low dew]

Is that for the concept you gave me?

I got this with ACBS:

Idk how good that is

I think I did something wrong XD

[dense mesa]

right on. looks pretty legit. tell it to do wide aspect ratio next time. if you want it wide

nah, didn't do anything wrong, just tell it to make it better next time

[low dew]

Let’s see

[dense mesa]

or try these

Sci-fi Warrior with Giant Candy Cane: A sci-fi warrior presents a whimsical, oversized candy cane in a sepia-toned city, offering holiday cheer against a dystopian backdrop.

Frozen Figure with Christmas Tree: A frozen figure in a wintry landscape holds a fully decorated Christmas tree, creating a haunting mix of holiday spirit and icy desolation.

Twisted Tree Serving Thanksgiving Dinner: Amidst a Christmas market, a twisted tree character serves an out-of-place Thanksgiving feast, adding a surreal twist to the festive scene.

Ghoulish Elf with Gingerbread House: A ghoulish elf offers a gingerbread house in a dark forest, contrasting the sweetness of the treat with the brooding environment.

paste them all at once to see what happens. could get wild

[dim cradle]

looks great

[low dew]

Looks like I got a thing or two to learn from image gen xD.

[low dew]

After playing around with ACBS, the best I could do was this.

[dense mesa]

lol, I especially like the buffet image

[low dew]

lol

[low dew]

It’s backwards lmao

Just realized

[dense mesa]

mine isn't serving turkey anymore

[low dew]

Narwhal lol

[dense mesa]

whatevers going on there, you know you can trust them

[dim cradle]

awesome

i see the 'overstuff' problem on the widescreen tree buffet images. the stall/lights run unbroken like so many buildings often do. but on the square image skiller made early it did not have this issue

[low dew]

I swapped out your description for just “view of earth from space”, and got these.

[dense mesa]

I have no idea what that is tbh. looks cool though

[low dew]

What, the earth?

[dim cradle]

third rock from the sun

[low dew]

Finna run out of chatgpt web image gens. What the rate limit rn? Too tired to get up and use my app on pc lmao. Bed ridden.

[proper lotus]

Greek god statue standing on mountains

[dim cradle]

[thick smelt]

some dall-e from bing thing i did for matt 🙃

[hybrid fox]

what we up to tn

[thick smelt]

santa claus dancing with himself, i just wanted some santa self love ai arts

[glossy scroll]

@dense mesa You're learning!

Very good photos

[open trench]

The curse of the multiple tails is back again.

[glossy scroll]

Elf-Orc Hybrid

[hybrid shoal]

I'm generating weird looking aliens, but then this super cute animal came out. I want one.

[glossy scroll]

Dragon-styled Mech Warrior

[hybrid fox]

bro looks like the final boss transformer

[open trench]

can't make a decision. which is better?

[hybrid fox]

furry bat... the thing of nightmares 💀

[open trench]

lol

thansk for th opinions guys. 2 it is

[glossy scroll]

mini-boss...big boss will come later

[hybrid fox]

what time does the daily theme change

[hybrid fox]

👀

[glossy scroll]

Warcraft style Orc chief

Dragon-styled Mech Warrior

[glossy scroll]

Recreated your right pic through my GPT

Honestly, that looks epic!

[open trench]

looks cool, man.

[glossy scroll]

I hope you don't mind my inspirations

[open trench]

not at all.

[glossy scroll]

Good

[open trench]

I made six new anthropomorphic parody characters today. Some are remixed from https://discord.com/channels/974519864045756446/1187715219921850437 and some are brand new. But I am running out of inspiration because the posing from dalle3 is... well, it's hard to make really dynamic without sacrificing a lot of the other details. The details are important as the poses. So, I'm at a brick wall. And it doesn't like camera angles. You really have to work at some of them.

character posing, that is.

[glossy scroll]

They look great Nezhno! I'm loving the different animal styles!

[open trench]

Thanks. Those are all the old ones. I am going to post a separate part two and start feeding it.

The gallery just feels so dead. So, it's hard to be motivated to post in it.

[glossy scroll]

I feel you on that one.

That's why I post here more.

[open trench]

but I am trying to comment every day and maybe help get it active.

[thick smelt]

i think this is my most controversial food ai art, sorry 😬 🍍 🍕

[open trench]

started today lol

yesterday, actually

[glossy scroll]

Heresy!

[thick smelt]

yep it's very controversial

[glossy scroll]

Pineapple on pizza should be banned

[devout vortex]

hello!

[glossy scroll]

It's like making a chocolate ham sandwich

[devout vortex]

does dalle-3 in Bing Ai is same as Dalle-3 in ChatGPT ?

[thick smelt]

yeah

[glossy scroll]

Yes, but with different content filters

[devout vortex]

Hm

[glossy scroll]

Bing has a different content filter policy

Because it's with Microsoft

[open trench]

ah... those few days before the ban hammer

They took a ton of my very innocuous content down from my saved folders too. Like, a lot.

[glossy scroll]

Rogue

[open trench]

They were just modernized mesoamerican shaman with a mix of modern clothes and ancient attire. I really don't know why they took them down. Maybe it's culturally sensitive.

[glossy scroll]

Probably because a lot of cultures around the world don't like to be mocked, and their governments will usually send out a warning if it goes out of line.

It's happened in the past.

[open trench]

no other image generators had a problem with it, but bing did. lol They have problems with a lot of things.

[glossy scroll]

Maybe they just want to be more family friendly with cultural integrity.

[unique willow]

where do i chat with chatgpt

[glossy scroll]

https://chat.openai.com

[open trench]

I can't tell if that's a dig lol 😛

[glossy scroll]

No I'm being serious.

[open trench]

ok.

[glossy scroll]

Making fun of a frankienstein costume is one thing. But when you take traditional attire from a culture that's used for honor and respect and turn it into a joke for your own amusement, it becomes offensive to some cultures.

[devout vortex]

I just realised Bing's Dalle 3 isn't as good as Chatgpt's dalle 3

[glossy scroll]

And if it can be prevented, then they take action.

[open trench]

I can understand that. I certainly wasn't doing that, though.

But intent can't be identified without telepathy. So, I get it. Just made me sad.

[glossy scroll]

It's good if you know how to prompt engineer. But it is quite limited, I agree.

[devout vortex]

I know its good but its still different from chat gpt's

[glossy scroll]

Yeah. Some people here really like it more than chat gpt's image gen. But I prefer ChatGPT's.

[open trench]

I use it to finalize certain details before entering a prompt into chat gpt. There are striking differences, but it's still a good tool. Helps me keep parts of the prompt concise.

[devout vortex]

Yeah same.

[glossy scroll]

Mainly because I can make amazing images with it with my custom gpt's

[thick smelt]

yeah some people like me can't afford chatgpt + #yestoregionalprices

[glossy scroll]

Drow-Elf Fighter

[unique willow]

I want someone to Create me an pfp but be needs to know hindi

[glossy scroll]

I'll happily do that

I don't know hindi though

[unique willow]

do u know little bit hindi

[glossy scroll]

But I do have google translate

[glossy scroll]

[devout vortex]

Adobe Firefly and Dalle-3... which one's better?

[glossy scroll]

Dall-E 3

No competition

[devout vortex]

Yes! Its much more detailed

[glossy scroll]

Look at those cute doggo's in the background

You don't get that with Adobe

[devout vortex]

I did both with same command/prompt

[devout vortex]

I agree

[unique willow]

Like Saturo's image should be in the background And uh his name should Be there

@glossy scroll

[glossy scroll]

@unique willow I sent you a DM

[glossy scroll]

You can't have anime characters

It's against content policy

I can't make you an image with a copyrighted character in the background

[devout vortex]

you won't do it but Dalle is able to do it?

[glossy scroll]

Sorry @unique willow I can make you something else though, maybe something similar?

[glossy scroll]

Both, I won't do it, and Dall-E is not able to do it.

I won't even attempt it.

Because it goes against content policy.

[thick smelt]

EXACTLY! why would sam sleeping on that sweet regional prices?

[devout vortex]

[glossy scroll]

No idea

[glossy scroll]

[shut niche]

Been dialing in a prototype. Check out those details. 🙂

[glossy scroll]

Looking good! I'm liking the textures and lighting. Reminds me of 90's Goosebumps tv show monsters. Which is one of my favorite styles of art.

[shut niche]

It's the face. He's got character/emotions. The second looks like an accidental candid photo, like a wild game camera's there, and he's curious about the object. Definitely happy with the test outputs.

[glossy scroll]

Oh yeah, now that you mention it, it reminds me of planet of the apes, the face.

[late blade]

You guys doing a contenst who makes the better bigfoot?

and merry xmas everyone

[shut niche]

I told it to re-imagine it as a snow creature, and give it a wolf pet.

[glossy scroll]

Merry Christmas Dys!

[shut niche]

No. His makes totally different styles. I'm into the realism and emotional depth. His makes epic D&D styles. GPTs are going to be like that, with "niche's" to fill, IMO.

E.G., different, not better.

[dim cradle]

there needs to be an "aww" emoji.