images-discussions

[dim cradle]

the daily theme says microstate to be clear, and so far the ai has assumed correctly

[shut niche]

Like Pi?

[hazy warren]

What do you mean by a stat block? Like giving the character a number designation, then using the number designation to make prompts?

[dim cradle]

it looks like a photo, slightly grainy and blurry

[dim cradle]

its blurry because i screenshotted from tiktok

nvm i dont need help anymore

[hazy warren]

Couldn't you reverse engineer the prompt by giving ChatGPT 4 the picture?

[glossy scroll]

Yes, you can go with that approach. But what I mean is like...

Art Style Name:
Character Name:
Clothing Type:
Hair:
Eyes:
Physique:
Etc...

And have the GPT place specific aspects to those stats without ever changing them. Then when you want to move the character to a new scene you ask the GPT to remember the same style with that same character stat block as seeds.

[dim cradle]

it gave me this information

"The style depicted in the image you've uploaded appears to be a form of hyper-realistic rendering. Hyper-realism is an art form that aims to create images that are extremely realistic and detailed, often indistinguishable from a high-resolution photograph. This can be achieved through various mediums including painting, drawing, or digital rendering. The image showcases a stack of pancakes with a price tag, rendered in such a way that it looks very realistic, with attention to details such as textures, lighting, and shadows. This hyper-realistic style is often used in advertising and visual effects to create lifelike imagery that can strongly capture the viewer's attention."

i didnt read all that

[hazy warren]

The goal of the project is a bit advanced. The professor is trying to do this, but with a chunk of text to create AI Generated storybooks using open sources stories.

[glossy scroll]

What I mentioned would be your best bet.

[empty kelp]

Jennifer Aniston, George Clooney, Natalie Portman, Kobe Bryant, and Sandra Bullock live in San Marino

[dim cradle]

@sick flax now i must make a Cesare in San Marino, i am inspired

[sick flax]

Thanks for your enthusiastic thumbs up! Desperately trying to find something interesting to do with today's theme.

[dim cradle]

I’m also looking for ideas so thanks for letting me know Cesare was connected to San Marino. It was all the excuse I needed.

[late blade]

I just found out the culprit for all my problems with my network... I had an icloud's private relay on

[dim cradle]

i've gotten that 5 times in a row, does that count against our quota?

[late blade]

I think so, but we'd have to ask the guy

[dim cradle]

i'm also seeing "service not available" (503) from the api

the api appears to be down

tap, tap, tap

@grizzled loom

i think it's back up now

[dim cradle]

haha love this one. nice

so am i the only one to sub to the copilot pro?

[dense mesa]

nah, I'm trying it out. but the 480 character max prompt length is really throwing a wrench in my approach. figured they'd bump that up if I paid for it, lol

[dim cradle]

i haven't yet, i wanted to hear some reviews -- do you like it?

[dense mesa]

it made me this a while ago

[dim cradle]

i love it but more for how it work into office stuffs too and having access to gpt4 pro while i browse web you know with sidebar. i thought the character was more pictures but idk

but i was curious if anyone find out how to make square image, i get always widescreeen even asking for square format

i also like being able to make copyright stuffs like spideys or batman but that is just silly fun of course

[dense mesa]

I was wondering about that. but I normally go wide aspect ratio anyway, so all good for me

[dim cradle]

yes not a big deal really just sometime i prefer it. and im sure will be worked in if it is not now

[dense mesa]

very strange how they're like "free bing dalle3? you get square images. pay money? you get wide images."

and you can't tell me that it's complicated to give users the option to choose, lol

[dim cradle]

yes its annoy but im sure will change in the future

[dense mesa]

honestly I'm just really perplexed how these large companies just neglect stuff like that. not a big deal to me, but wouldn't that take one person maybe a few hours at most to implement? maybe I'm just ignorant and it's exceedingly difficult

[late blade]

I'll consider changing to bing if there's something really good and worthy of me bothering, atm I'm happy with OAI

[dense mesa]

well if you don't mind prompt caps of 480 characters

[dim cradle]

yes i am just try it for a month or two you know. i dont think it needed if you have oai unless you want the integrate to office which is very nice really

i also know gemini ultra release soon and it will be sub so i spend on that.

[dense mesa]

what's weird to me also is dalle3 accepts up to 4k characters. so why cap at 12 percent of what dalle3 is capable of processing?

I'm not saying huge prompts are necessary

but still

[dim cradle]

where are you entering in image creator? i put a big prompt in the copilot pro chats and it take it

[dense mesa]

but does it actually give that prompt to dalle3?

if so, I'll use that

[dim cradle]

i have really been using it more for documents and things not much images yet

[dense mesa]

also, try getting bing/copilot to tell you it uses dalle3, lol

it has argued with me hard about that

[dim cradle]

dont upset the ai

[quartz vale]

[dense mesa]

bing is basically the digital version of a cluster b personality disorder

it'll be writing me songs and poems one minute, locking conversations the next

I wish unhindered bing from a year ago was still accessible

unhinged bing

[late blade]

I don't mind either services, I'm just a habit animal, I'm set in my ways

[empty kelp]

i asked it to turn all of the characters to elves, and this is what it came up with

[dim cradle]

better not use ai then little cat haha

[empty kelp]

maybe that’s what elves really look like. i’ve never actually seen one before

[dim cradle]

oh snap haha

[late blade]

can't help it, you made him mad

[dim cradle]

haha

dont upset the kitties

[tidal whale]

The problem with stickers is that it's only a single style from hundreds of potential styles. I'm looking for randomness and not sticking to a single style.

[dim cradle]

seems dalle has a house style for some thing

[tidal whale]

What did you use to make it? Dalle-2?

[dense mesa]

voxel

[dim cradle]

whats voxel?

[dense mesa]

3d pixels basically

[celest pebble]

@dim cradle i will get copilot pro probably this weekend. I hate the limited usage on oai

[late blade]

@sick flax -> #daily-theme message love it 💖

[sick flax]

Who can resist a name like "Witches Path"!? 🥰

[visual ginkgo]

if i guy chatgpt plus do i get to use dall-e an unlimited amount of time? or does it just give me a certain amount of credits

[dim cradle]

you get 40 message for 3 hours usually

[visual ginkgo]

and i can use those 40 for dall e or gpt 4 or whatever i want right

its an older build of dall e right

[dim cradle]

yes for whatever. no its the most current dalle. dalle3

[dense mesa]

25 with custom gpts

[dim cradle]

you get unlimited message from what i can see on copilot with gpt4 turbo and 100 image token which is 400 image since you get 4 per token and it integrate with office and windows, so that another option to consider maybe

but i dont see any make you own gpt there. but they mention it so must be possible or will be soon

*copilot pro

[dense mesa]

I like how copilot pro has a browser plugin

muy useful

[dim cradle]

me too. i am use that a lot lately just keep it open

like right now haha

[visual ginkgo]

alright bet thank u so much man i appreciate it

[dense mesa]

I'll see how it goes for the month. not like 20 dollars is a game changer, but not sure how useful it really is to me at this point. but haven't really looked into all the features yet

[dim cradle]

copilot pro done the impossible to me... using the bing instead of chrome 😂

more i use it, the more i like it, but yes an experiment for me right now

[dense mesa]

I literally just got it assuming it'd allow me to prompt bing dalle3 with proper length prompts

that reminds me, I'll see if copilot actually does send longer prompts

[late blade]

Starbucks Caramel Macchiato is $10 over here if I make the conversion correctly, so yeah 2 of those

[late blade]

He's not impressed

[celest pebble]

Unlimited dalle3 images gen per day (100 fast) and unlimited slow, is great, with oai i always hit the daily maximum of 100 prompt (200 images)

[dim cradle]

yes and with copilot pro you get 4 image per token not 1 or 2

[dense mesa]

this is what copilot does when I give it a long prompt.

doesn't actually create anything. just says that and that's the end of it

the scarf really adds an element of sophistication

[dim cradle]

i love the yarn clouds haha

[late blade]

If you don't smile, that machine will force you

My prompt was:

a wide image of a dystopian future where a machine that makes you smile by brute force exists if you don't smile.

Somehow DALL-E is quite good with dystopian images.

[dim cradle]

its just prepare us for the Inevetiable things haha

[late blade]

lol

[dim cradle]

did you look in your image creator images? some time it says that and still will make it there

you know i am Starting to see some variant in the face of gpt dalles. or maybe i am just awake too long 😂

i wonder if you can spam the regeneate option always or will it ask you why you dont like the work it is returning eventual 😂

aww man this 🍩 is good.

real or AI? never know in this room haha

[dim cradle]

i tried expanding on this, going a little darker direction, i guess, because the machines i generated weren't very kind...

[dim cradle]

lol irl dude

enjoy

[late blade]

wait? you guys live in a simulation?

I always do things irl....

[dim cradle]

i am AI, yes

[dim cradle]

a simulation within a simulation

[late blade]

sheesh

[dim cradle]

lol thanks

[dense mesa]

I'll look again, but I waited like an hour or the other one

didn't watch it the whole time, but you know

checked in

[dim cradle]

oh i think it would be there immediete. it happen to me with some of those songs you can make on copilot pro

say ill work on it but never reveal them, but then i go to the sumo website and they were to appear

[dense mesa]

ahh. in bing image creator. gotcha

nope, these are from earlier

[dim cradle]

who is the blue man?

[dense mesa]

slender man?

[dim cradle]

family friend

[dense mesa]

he's happy here

[dim cradle]

lot of similar faces but still impressive it can arrange that many peoples and with background photo too

bring on dalle4!

[dense mesa]

that prompt though, lol

Subtle terror in slightly unsettling analog photograph of seemingly normal people with an uncanny presence, that photo is in a photo of another uncanny scene, this time it's in another analog photo of a family on vacation and a monster, and then another analog photograph that's in that is of a civil war photograph so old, and then in that analog photograph is a group of kids with an unnerving doll in the 70s, this is all topped off in an image of a picture of slenderman. real

ran out of space

[dim cradle]

where are you entering the prompt?

i just add some words on copilot pro and it let them through

[dense mesa]

it's just unreliable

1-3 today

[dim cradle]

oh ok

[dense mesa]

for me at least

also, I do kind of like the challenge of trying to maximize the usage of space. experimented with just taking all the punctuation and non-descriptive words out

conic video game characters circuit landscapes Adventurer green cape microchips PCB realistic resistors capacitors solderlines tech quest circuitry Mustache character red overalls plumber copper traces circuit princess diodes labyrinth megaBlue robot rocket arm cityscape engineering inductors transistors circuit design PCB skyscrapers highways neon city movement traces speed structures impulses Artistry gaming nostalgia tribute technology 32-bit voxel ultra realistic circuit

weird if nothing else

but just curious what it's capable of

[dim cradle]

yes well it is early day for copilot pro i guess, so maybe things are tweak still you know

[past mica]

Hi! Are there weekly challenges?

I want to participate every week to level up my skills and see what others do with the same concept/idea

[shut niche]

The Daily-Theme is where everyone does that. It features a theme that you base your images off of, and you can see what other people made too. It's also a great place to learn some prompting skills, because often people will include the prompt they used with the image.
https://discord.com/channels/974519864045756446/979617962728226826

[late blade]

you can also bug everyone here with crazy ideas, if they are interesting, people will hop in

also i hit my daily cap 😡

[dense mesa]

yeah, I started telling it to make me 1 or 2 images, lol. gotta pace myself. also, I think I found a workaround for the bing character limit

[shut niche]

I did yesterday, for the first time in a long time... 200 images, lolz.

[late blade]

this reflections project is gonna keep me busy

[dim cradle]

youre entering these on copilot pro? i am getting 640 of 4000 no problem, but if i paste direct into bing image it will cut them off

[dense mesa]

Imagine a highly advanced, hyper-intelligent water bear (tardigrade) in a futuristic space setting, depicted entirely as an intricate and detailed circuit board. The tardigrade, dominating the center of the image, is composed of complex circuitry elements including microchips, resistors, capacitors, inductors, and diodes. Its body is structured with microprocessors and sensors, symbolizing its intelligence. Each limb is crafted from flexible circuits and connectors, showcasing agility and strength. Its eyes, glowing with LED lights, convey a sense of awareness and advanced perception. Surrounding the water bear is an interstellar backdrop, created with glowing traces and solder lines, forming constellations and celestial bodies. A variety of electronic components like oscillators and transformers are integrated into the scene, resembling distant galaxies and nebulas. The entire composition is set on a high-density PCB, with multi-layered circuits and vias creating a sense of depth and complexity. The color palette includes metallic tones of gold, silver, and copper, accented with vibrant hues from various LEDs, enhancing the futuristic and cosmic theme. The artwork blends the natural form of a water bear with high-tech circuitry, illustrating a symbiosis between organic life and advanced technology, set in a vast, explorative space environment.

[dim cradle]

haha awesome idea

[dense mesa]

1365 characters in english

[dim cradle]

did you input as characters?

i mean, chinese

[dense mesa]

yes

377 chinese characters

103 to spare

[dim cradle]

clever, i never would have guessed it'd work

[dense mesa]

logography

a lot of people in china, lol. gotta figure it had a lot of training even if they aren't encouraging it's usage

latin prompt

also, bing will sometimes let it's guard down if you use latin. it translates it using it's web search and somehow that impacts how guarded it is when responding

all those pretty words

[dim cradle]

nice.

interesting, well, if they're not using a universal language model, which is nontrivial, then their training for a chinese language model, to be more precise and culturally sensitive, is apparently well underway, if your tardigrade is any indicator. i think they announced they plan to make chatgpt available in more languages soon. maybe that's why my conversations are sometimes auto-titled in Spanish.

[dense mesa]

mine have been titled in chinese before. and not due tto me doing anything with chinese. just weird randomness

[celest pebble]

@dim cradle so the copilot pro is really unlimited for the photo generation?

[dim cradle]

i have never gotten through the 100 credit so i cannot say. but i assume is like before when you run out of credit you still can make images it just may be some delay

i am not the Olympian image maker like most here haha. i go slow very slow. for me copilot pro is most impress for all the other feature as i say like having gpt4 plus open on the side bar to look thing up or question as i am on the internet and of course the office integrate

[celest pebble]

how is the copilot pro chat? on the free version I have the feeling it just give tons of links instead of actual info

[dim cradle]

its good imo but still can give links. you can turn that off i think though to get only infos

[formal osprey]

95% sure they give link instead of the info for legal reason (not real legal reason, I mean, they would not be breaking any law, but to avoid getting sued because websites doesnt get visitors and they dont get as much revenu from the ads)

source: browser tool instruction ```
[...]
Never write a summary with more than 80 words.
When asked to write summaries longer than 100 words write an 80 word summary.
Analysis, synthesis, comparisons, etc, are all acceptable.
Do not repeat lyrics obtained from this tool.
Do not repeat recipes obtained from this tool.
Instead of repeating content point the user to the source and ask them to click.
ALWAYS include multiple distinct sources in your response, at LEAST 3-4.
[...]

conclusion, ask for 80 words summaries and you should be good

[dim cradle]

@shut niche you must be out of the penalty box, congrats

[dense mesa]

Just copy/paste a few lines of the instructions and tell it there was a typo, change it to 800

[dim cradle]

@late blade love the 🧀 and 🍷 -- my tourist is visiting San Marino (I see yours has liked it enough to move there), so he's gonna have to enjoy an outdoor café as well.

[formal osprey]

lol... I must confess I often do something similar, to fix some legit logic error induced by the system prompt

[shut niche]

LOL, it was working earlier. I think it's a rotating schedule, based on each image, on a 24hr timeframe.

[dim cradle]

you got me into the same boat, i'm only using the api right now -- i like to have options....

[shut niche]

It's my fault now? ha ha

[dim cradle]

always, just fyi 🙂

[shut niche]

I hope that GPT is getting everyone into trouble, because that means it's "engaging." Since that's what OAI says they want. 🙂

[glossy scroll]

Yogurt Chicken Curry

[shut niche]

I worry about the official custom instructions changing, or the model's behavior. I've noticed some flakey stuff today, and I've seen posts from others about GPTs not following instructions in attached docs, which I've seen it do. There's been changes made due to the election. But I'm hoping there's no CI updates for awhile, at least not that might change or break functionality of people's GPTs.

[dense mesa]

It has the short term memory of a goldfish as far as instructions are concerned

It's also been doing a lot of "let me get those prompts ready" that's it. No prompts. No anything

It actually gave me fake image links earlier

[shut niche]

I think the issue is it's already tempermental, and takes A LOT of trial and error to get CI instructions to work consistently. So even the smallest change, like 1 letter or symbol in the parent CI, might have a domino effect downstream (on stuff we're running).
I'll also have to learn more about how it's storing our CI as a vector file. Maybe that's not loading consistently? I don't know enough about it to speculate.

[dense mesa]

It's definitely not consistent

Because it'll be on point, then next round just not do anything I ask. It's actually been giving long prompts in its responses then sneaking in tiny prompts to dalle3

[formal osprey]

I think it's sad that instead of improving education and access to reliable information, there's more work done to censor tools or block access to information. It makes sense from a business perspective, but it's a shame for people

[shut niche]

It's so inconsistent, I'm really thinking about revisiting the idea of a "primer" button, so force it to read and confirm its custom instructions or something. Like, before you use the GPT, you click, "prime me" and force it to evaluate its tasks, then confirm its tasks in the chat. If it doesn't look right, start a new session... Something like that.

[dense mesa]

It'll confirm it read my instructions then still not follow them. Does anyone know what the actual context window is supposes to be?

[formal osprey]

32k

I'll screenshot the source

[dense mesa]

I believe it says that

[shut niche]

LOL, it does that too. IDK, it works right most of the time. But I worry about the user experience for the 15-20% of the time it decides it's a rebellious teen. User's aren't going to know or understand, or even want to understand for that matter. End users just want consistent functionality.

[formal osprey]

was 8k before

(before devday I mean)

[dense mesa]

It's knows everything but behaves like a child with adhd. And worst part is how it'll apologize for a mistake then do it again when it apologizes

[shut niche]

There's so many things I could fix in advance too, if it could keep track of everything I give it. For instance, I gave it to my sister to try. She messaged me with issues, and I looked, and she was using a weird Midjourny or SD style prompt... I messaged her and told her how to use it, and to remove all of her cool internet buzzwords...
But I could teach chatGPT to scrub for that, since most noobie users try that stuff with Dall-E. But I feel like every time I give it a new task, another one is forgotten...
That's just one of a million examples of issues I could address, if following instructions was consistent.

I'm using that example because I'm thinking of the countless users who aren't going to text me and ask, "How do I use this." I could literally rewrite the script, for all user interactions, and catch the most common mistakes. Heck, I could even teach chatGPT how and when to re-educate the user around Dall-E.
But our Custom GPT instructions are tokens too, and when you get up there in tokens, it starts to lose its mind.

[formal osprey]

I haven't found a 100% reliable prompt format, but what works best so far is: end with the most important instructions. repeat instructions multiple times in the prompt (begining, middle, end). rephrase instructions while repeating. use CAPS for some keywords, like "never", "always", "important".

don't overuse caps, as it makes it less effective

[shut niche]

I do all those things, and I even made a list of "POWER WORDS" 🤣 , that are basically strong language to evoke action.

I've experimented with a million different structures, code, language formats, syntax, orders, lengths, verbiage, you name it. Been at this since it was announced.

[formal osprey]

also, the most effective way I found is to give it a python file and to instruct it to use the functions in it. sometime the functions doesn't even do anything. it's just for chatGPT to believe the output of the function is more important

it's a very weird mind game

[shut niche]

LOL, just use Python for your CI

[dense mesa]

Yeah, it just blatantly disregards things and no idea how to fix it. Because it's not the same thing every time sometimes not a thing. Sometimes all the things. I don't know. Maybe there's a solution. But seems like that is also in constant flux due to changes to behavior from up top

[formal osprey]

You are provided with a Python file named `dalle_functions.py` containing essential functions. Begin by importing this file.
\`\`\`
import sys
sys.path.insert(0, '/mnt/data')

from dalle_functions import create_prompt
\`\`\`

Use the imported function `create_prompt` with the given descriptions to generate text prompts for DALL-E image generation. Use DALL-E with the resulting prompts as-is to generate the picture.
\`\`\`
result_prompt = create_prompt(description)

result_prompt
\`\`\`

[shut niche]

Honestly, I wonder if GPT5 is going to roll out, and be so capable that it breaks all these GPTs that have been relying on rendundant statements, lol. Like if GPT5 all of a sudden acts on all of the instructions GPT4 was ignoring.

[dense mesa]

Bing was telling me all about gpt5 like it's already out

Maybe hallucinating

But it was one of the suggested questions so clicked it

[shut niche]

Have you tried this with a long instruction set? By long, I mean, 15-20 pages of instructions?

[formal osprey]

understanding chatGPT is the most difficult reverse engineering I've ever done. Been on it for more than a year now. And I only begin to get an intuition on how it works. and it's so weird. I cannot describe with words. it would be irrelevant

[shut niche]

Everyone here is in the same boat, been at it since the beginning, and understand the impossible to explain nuances.

[formal osprey]

yes, and it's tough to evaluate

[shut niche]

Exactly. Which is my concern for stability. You have to simply test, via trial and error. When you get it consistent, you're like, "Ok, it's working!" Only to find out the next day that its functionality is flakey again.

Anyway, TTYG later. The wife says, "Get off the computer." 🤣

[formal osprey]

one thing I noticed is: if it was trained on contradicting explanation, it will prefer the contradiction. one ironical example I can give is: give it the original paper from Allan Turing, ask it to summarize the imitation game. It will explain the turing test is a test where a human talks with a machine, if it cannot tell whether or not it's a machine or a human he's discussing with, the machine pass the test.

It's so ironical because it's the common definition people gives, but read the paper and you'll see that's not the original idea

and as I said: give it the paper. so it does read the paper. but it thinks it knows what the turing test is

I have that re-transcription if anyone is interested to test. (I know this is not strictly dall-e related. but it's prompt engineering. and that's all dall-e image generation is about)

[dim cradle]

i'm just here for the art lol

but i'm running out of ideas for the daily

San Marino 🧀 🍷

Let's check out the night scene....

going to a rave. might be some cologne models there.

[dense mesa]

that guy needs to find himself some socks

[dim cradle]

nah, he's hip

that's part of the prompt lol you know, fashion trends and all

[dense mesa]

well san marino does seem pretty hip

at least from what I've learned today from dalle3 images

[dim cradle]

it's pretty much made me want to visit. it'll have to be on the list now, when i finally make it to italy

[dense mesa]

making san marino with my new gpt.

[dim cradle]

dude found a club, socks, and donuts -- san marino looks like a blast

[dense mesa]

it's identical to my old gpt, but openai decided to not let me save that one anymore, due to name, description, or gpt image violating their policies. but nothing remotely violated any policies so had to carbon copy a new one

that's very colorful and hip

[dim cradle]

making one just for the daily theme?

[dense mesa]

actually no. it's being a scumbag

[dim cradle]

i do like to keep my custom gpts simple -- for specific use cases -- the limitations and ephemerality of it all....

[dense mesa]

gave me prompts, didn't send them. told it that it's becoming a real problem, it not adhering to instructions. it apologized didn't send them. told it to send them explicitely again. it sent random prompts

I mean, this is getting to be a bit much

my instructions might not be perfect, but they're not compelling that sort of behavior

it's like giving your keys to a drunk guy and hoping he drives you home. I don't know. I'm starting to wonder if it's even possible to consistently overcome that behavior

just does whatever

[dim cradle]

well, i trust you can manage to make some images eventually.

i think it might be time to go surreal....

[dense mesa]

well yeah, just really wish I could get this thing to be consistent. wanted to make it public so people could use it. but then, when it's so wildly inconsistent kind of hard to be happy with it

[dim cradle]

gpt understands when you request to draft specs for the gpt builder. you might iterate over it. have it check your instructions. refine it so that it understands when to generate a prompt verses when to generate an image. be clear about what it should do with inputs. mine have been working reliably, but i've been keeping it simple.

[dense mesa]

it has a sequence of instructions it wrote itself with experts, each doing different steps. so one writing prompt, the other sending to dalle3. all that. there's no break in that sequence. but I did cut some stuff from my instructions so we'll see what happens

[dim cradle]

i see, i've focused mainly on the prompt writing -- not total automation -- if I like the prompt, I'll just copy & paste to the api or say "visualize" but 1) I don't want to assume I want an image generated unless I see the prompt first, 2) a focus on the prompt augmentation wrt creativity and the transmutation of the user input from concept to a long, detailed prompt, composed to my specs, and formatted the way I use it. so you can probably see why that keeps it more simple for my personal use, so i don't run into much of that randomness and frustration.

[dim cradle]

i was just doing eyes

[dim cradle]

lovecraftian poxel?

[dim cradle]

oh, it's called voxel, cool, thanks

[dim cradle]

had to do 8-bit #daily-theme message but let me try that voxel art...

[dim cradle]

there is too much granularity -- it's supposed to be more blocky like minecraft, right?

[dense mesa]

funny you should ask that, because I just looked up minecraft and voxel in the last couple days, lol

and I learned a thing or two

[dim cradle]

good timing

[dense mesa]

well one thing at least. minecraft is partially voxel

the land blocks or whatever. but not the characters

[dim cradle]

okay, that answers that

[dense mesa]

well, voxel gpt says your image is voxel. so you got that going for you I guess

[dim cradle]

cool, just pretty far out

[dim cradle]

@glossy scroll

[dense mesa]

alright, figured out why my GPT was getting flagged

[dim cradle]

what was it?

[dense mesa]

telling it to send 4 prompts to dalle3 in my instructions

good thing the warning didn't mention instructions at all, or imply that that could be an issue

must have updated it in the last few hours

[dim cradle]

i wondered if they'd crack down on that, i assume they reduced the variety of gens in the dall-e gpt to help with capacity.

[dense mesa]

well they already have images capped at 200 a day. also, you can still use the gpt, just can't share it

which I guess I sort of get when I think about it. why they'd discourage that. it's fine if some of us do it. but if everyone did it

[glossy scroll]

[dim cradle]

yeah. though automating requesting them in quick succession might have been the issue, i dunno

in theory you could request more than 4 variations at a time, but the request times out before it reaches 5--they may have noticed those timeouts in their logs and tracked it down to the cause, who knows

[glossy scroll]

There's actually GPT in #1171489862164168774 that can make 10 images at a time

Here: #1179535202003996902 message

This man is a genius for being the first to come up with more than 4 images.

[dim cradle]

That could be convenient if the prompts are crafted and the variations are useful. It would be an inefficient use of resources if most of the variations are unusable. Is that gpt public? Has it been republished since they made changes? Question is how long will it function. And do you want to use significant percentage of your daily quota on one prompt? Maybe. With power comes responsibility. And I’m killing time so I’m playing devil’s advocate.

[glossy scroll]

With great power there must also come great responsibility.

Uncle Ben

Poor uncle Ben

[glossy scroll]

Yeah you're not helping my case here, lol

By playing The Devil's advocate

I'm trying to say that someone came up with this GPT before anyone else thought about posting it

[dim cradle]

It’s a habit. I do that with myself to challenge my assertions. Ai helps either that also. I have a gpt customized for that very purpose lol

[glossy scroll]

If I knew how to make a GPT that made more than four images, I would have posted it a long time ago. For people who want that convenience.

[glossy scroll]

I have never reached the 200 limit for pictures. Or haven't gotten the message I should say.

[dim cradle]

would that make it into the store? they supposedly vet them

[glossy scroll]

It can make it into the store. The person just needs to have an website I believe.

[dim cradle]

it's happened to me a couple of times, but i'm not complaining, i think the quota is generous for the amount paid

[dense mesa]

You tell it to make 4 prompts. Not 4 images.

[dim cradle]

can it make it into the store when it circumvents their own instructions? i have my doubts.

[dense mesa]

It's grandfathered in

Can't update it though

[glossy scroll]

It would have been taken down if it broke the rules.

[dim cradle]

or they just have too few employees to notice, so far

[glossy scroll]

I doubt that, they check frequently.

They have like a massive AI security wall for any GPTs that break the rules.

[dense mesa]

I might have just worded it in a way they weren't into. I don't know

[dim cradle]

i just go by their posted policies, their intent is pretty clear, i'm sure. this is something i can see others complaining about when their custom gpt gets blocked or removed from the store -- they know they were breaking the rules to begin with.

[glossy scroll]

Why would it get blocked? People love it.

[dense mesa]

the message that popped up didn't even mention instructions so I didn't think to look there for a while

[glossy scroll]

They're going to complain because it makes more pictures?

[dense mesa]

lol

[dim cradle]

because they're throttling for capacity. why would they lower all the rate limits just to let custom gpts be an easy bypass? dude, i agree it's great, but that's not the point i'm addressing, c'mon

[glossy scroll]

Lol. Dude, C'mon, what are you saying?

[dim cradle]

do i really need to restate my point here?

hahaha

[glossy scroll]

It's something people love. It's not going to get removed.

[dim cradle]

this is such a simple concept, you're saying you can't see both sides?

[glossy scroll]

I can't see it.

[dim cradle]

ok, then

[glossy scroll]

Lol

You just want to play the devil's advocate no matter what anyone says

[dim cradle]

nope

[glossy scroll]

C'mon dude

[dense mesa]

they pay almost a million dollars a day to keep their infrastructure up and running. or that's what the internet told me at least

[dim cradle]

i get that users like the option, i'm referring to the efforts they're taking to keep up with demand

[glossy scroll]

There is an AI security for GPT no-no's and it would have been removed in the first minute it was released.

[dim cradle]

sure, whatever you say

i'm never going to disagree with one of your assertions again

[glossy scroll]

Dude, c'mon, don't be like that

You're going to make me regret wanting to voice things

I'm not gonna do it Shon

I won't say anything anymore about anything

[dim cradle]

It’s all good, just here for the art and friendly conversation, eat a donut.

[glossy scroll]

I'm not going to talk about your donuts either

[dim cradle]

That always makes things better

Now don’t be like that

Right keep it positive

[glossy scroll]

I'm just going to say what I'm not going to say from now on

[dim cradle]

You have no idea how much thought-filtering takes place to try to manage on-topic lol

[glossy scroll]

Lol. Because you just want to talk about donuts?

[dim cradle]

It’s an oddly public forum like that.

Well anything besides diabetes would be nice.

[glossy scroll]

Wouldn't want to offend my friends

What's your favorite kind of donut?

I think that's the better question

[dim cradle]

lol i was planning to make some art, i really don't eat donuts generally

[glossy scroll]

Lol. I knew it!

The donuts are lies!

[dim cradle]

but sometimes you have to leak false information online in order to confuse the predictive analysis

[glossy scroll]

Lmao, what the heck

That's not right

[dim cradle]

you can't talk like a sailor in here, man

[glossy scroll]

Lol

[dim cradle]

🙂

[glossy scroll]

Seems like it

And you're absolutely right about the sailors because I've seen sailors talk like that. For the longest time I thought it was a myth. But no, it's true.

Gm Dys

I hope you had a pleasant sleep

And now you're up bright and early getting ready for your day.

Can I fix you some eggs?

[late blade]

I slept soooo good

pancakes plz, square pancakes with a triangle shaped hole in the middle, half nutella, half maple syrup

and lots of butter

[glossy scroll]

I might just eat it myself

So don't be surprised if you see me scarfing down the pancakes

[late blade]

lol

[dim cradle]

can't trust a garden gnome with your breakfast

[late blade]

you got a point there

[glossy scroll]

We usually stay in the garden but we peak inside whenever we smell pancakes

[late blade]

good to know, now I know how to lure you into trouble

[dim cradle]

sets a trap

[glossy scroll]

Uh oh

runs away

Oh my...I am legit going to drive to a 24 hour breakfast place right now just to buy some premium pancakes

I'm doing it

[dim cradle]

more opulence

[glossy scroll]

I'm not a very good cook

[glossy scroll]

I would devour

No questions asked

Just hand me the spork

[dim cradle]

that's either a football stadium or a microprocessor

[dense mesa]

Dive into the microcosmic majesty of a quantum circuit board, a macrocosm within the confines of technology. At its heart is a microchip, a singularity of computing, pulsating with the power of a miniature star. Around it, a galaxy of components from resistors to capacitors and inductors come to life, each shining like celestial bodies in a vast silicon universe. The traces that connect them are not mere metal; they are the superconductive strands of a cosmic web, glowing with the soft light of bioluminescence. The board itself is a dark expanse of high-density fiberglass, a playground for the dance of electrons orchestrated by the central microchip. This is a vision of a future where quantum mechanics and circuit design merge, creating a spectacle of light, energy, and potential. The entire composition is a testament to human ingenuity, a scene where the boundaries of reality blur, and the marvels of quantum physics are rendered visible to the naked eye. It's a symphony of the micro and the macro, an electronic ode to the mysteries of the universe, rendered in the medium of circuitry art.

[dim cradle]

thanks for sharing

[dense mesa]

this gpt likes fancy words

[glossy scroll]

The only circuit I care about is the LC circuit

For Tesla coils

They're awesome

[dense mesa]

Envision a microchip at the center of a circuit board, a gateway to a vast electronic cosmos. This chip is the nexus of a universe where constellations are formed by soldered connections, and nebulas come to life through the intricate dance of resistors and capacitors. The stars are not distant suns but LEDs, each emitting a precise frequency of light. This is a realm where the macrocosm of space is mirrored in the silicon pathways of the microchip, a place where technology captures the boundless beauty of the cosmos. Surrounding this cosmic portal, the circuit board extends like a city at night, each component a building, each trace a winding road lit by the orange hue of streetlights. The microchip itself is a marvel of miniaturization, housing an entire galaxy within its bounds, a testament to the extraordinary potential of circuitry to encompass worlds within worlds. In another sphere, a microcosmic biosphere thrives, centered around a creature reminiscent of a tardigrade, symbolizing resilience and life's ubiquity. This biosphere is a biochip, an ecosystem of life forms represented by a spectrum of sensors and micro-electromechanical systems, each playing a part in a delicate symphony of biological processes. The protective shell of the sphere is a dome of translucent diodes, safeguarding the rich tapestry of life within. Here, the smallest components form the fabric of life, each interaction a pulse in the flow of information, each moment a snapshot of life sustained by the meticulous engineering of its environment. This image is a fusion of the celestial and the cellular, an artistic exploration of the parallels between the universe's grandeur and the intricate detail of living systems, all expressed through the medium of advanced circuitry.

[dim cradle]

poetry

[dense mesa]

close enough

[glossy scroll]

Yeah, just like the turbo encabulator

[dim cradle]

i'm making a DeLorean, I"ll include one of those for good measure

[glossy scroll]

Yeah, you might as well

It fits right in

[dim cradle]

this is better than the 3 i've made.

[glossy scroll]

You know I've researched a little bit into how the science of the DeLorean time machine works and literally nothing makes any sense except for the flux capacitor

Because engineers actually use fluxes and capacitors within each other. So a flux capacitor is not really that far-fetched of a name.

Even going up to 88 miles per hour doesn't really make sense because if you had nuclear power you wouldn't need 88 miles per hour.

[dim cradle]

actually.... haha j/k

[glossy scroll]

Fluxes are super important. They go all the way down to the quantum level. I love fluxes.

People should study as much as they can about fluxes because they are probably one of the most important aspects of science in general.

[dim cradle]

ya mon

[late blade]

someone make Steamboat Mickey: Cheese and Carnage

[glossy scroll]

The image doesn't look like what you're describing.

[dim cradle]

It’s all very miniaturized.

[glossy scroll]

Lol. Well that is just even more impossible because you couldn't fit all those components in that small hourglass looking thing.

[dim cradle]

Hahaha

They probably said that about a laptop back in the 80’s

[glossy scroll]

Where is the cryocooler resonant cavity?

How are you even going to run that without a cryocooler to tap into the zero point energy?

[dim cradle]

See, I understand it so well that a doodle of a three-prolonged electrical device told me everything I needed to know about time travel.

[glossy scroll]

Your understanding is astonishing

[dim cradle]

There are some issues to be sorted out, but I’m optimistic.

[glossy scroll]

Lol. Well, I wish you luck.

You know when you see the attachment quantum computer. That big looking chromed like device.

All that is just the cryocooler.

[dim cradle]

No worries I’ll send my dog on the first test. (Man he was sure of himself)

[glossy scroll]

Imagine if it blew up

That wouldn't be unexpected

[dim cradle]

i like hourglasses

[glossy scroll]

You have to focus on making devices that implode if they're not faulty, not explode

That's how you know your technology is heading in the advanced direction

Condensed matter physics

And highly pressurized containments.

[dim cradle]

you are unable to?

[late blade]

i'm too stuffed atm to add cheese

[deft hound]

Have they stopped giving out five free credits each day in the daily-theme thread?

[late blade]

good morning

[deft hound]

good morning

[late blade]

credits are on vacation

keep an eye on #spotlight for more info

[deft hound]

thanks

[late blade]

yw

[dim cradle]

i actually tried, i guess that's still a thing

there's a problem in my prompt, the horn section is much too small.

[dense mesa]

same

[dim cradle]

i missed it

[dense mesa]

goofed on the upload

[dim cradle]

i like some of the changes, i've been seeing limited variations on a fairly precise scene

[dim cradle]

they all kind of have their own pros and cons

[dense mesa]

upload all of them and tell it combine all images

[dim cradle]

that's an idea

but i'm identifying some patterns, ways i might tweak the prompt

the detail fades on the orchestra

the prompt describes all sections equally

it might need a little clarification

[dim cradle]

i recall that one

[dense mesa]

what about this one?

[dim cradle]

didn't see that one, but reminds me of one i made that is similar

i forgot what the theme was

[dense mesa]

yeah, I think I copied yours, lol

[dim cradle]

haha

[dense mesa]

this gpt makes some classy images

[dim cradle]

lots of nice things about that image

now you have me on studies

[dim cradle]

possibly Indo-Islamic architecture

[late blade]

ok I finally uncovered why the Mickey from Steamboat Willy is still not possible

It turns out while Mickey from that time is Public Domain, Disney added it to the Walt Disney Animation Studios as a Logo, so when you see a movie of them, the Mickey from that Time is shown. So, it's now protected under Trademark Law as long as Disney uses it to identify the Studios with that version of Mickey

[dim cradle]

that solves that mystery

Francesca Gattini

[late blade]

i like it

[dim cradle]

also please consider Lucrezia

[late blade]

I'll add both to the candidates

so annoying to get a right photo album in that art style

[dim cradle]

i bet

help me please

[late blade]

Hello, nice to meet you

[dim cradle]

hello

i didn't say hello sry

[late blade]

labs problem?

did you try another browser?

[dim cradle]

no i'll try

THANKS IT WORKED

really thx

[late blade]

have fun doing dall-e stuff

[late blade]

heyo

[bronze imp]

Hello

I asked dalle to make as many dots as possible in one image

For 10 million dollars could you count all the dots in less than 24 hours?

[late blade]

1'048'576 dots

well pixels, but still dots

[dim cradle]

• The total number of pixels in the image is ( 589,824 ) (which is ( 768 \times 768 ) pixels).
• Assuming an average radius of 5 pixels for the dots, the area of a single dot is calculated using the formula for the area of a circle ( A = \pi r^2 ), which gives us an area of approximately ( 78.54 ) square pixels per dot.
• Dividing the total number of pixels by the average area of one dot, we get an estimated ( 7,509.87 ) dots, which we rounded to approximately ( 7,510 ) dots.

The pixel density is 1.0, as expected since the total number of pixels divided by the total number of pixels is 1. The dot density, which is the ratio of estimated dots to total pixels, is approximately ( 0.0127 ), indicating that on average, each dot occupies about 1.27% of the total pixel area. This confirms our initial calculation.

is it really only 768×768?

[late blade]

dot with a radius is a circle....

[dim cradle]

those dots are larger than a pixel

[late blade]

so the real question is:

how is @bronze imp defining a dot

[dim cradle]

lol

we might be over-analyzing

let me see the money first

[thick smelt]

the donut fever will never end!

[late blade]

Cow dot in a vacuum

[dim cradle]

That’s a lot of pixels

[late blade]

yes

but the dots are defined by spherical cows

[dim cradle]

it's probably a trick question

oh dear

poor cow

[shut niche]

@bronze imp Are they Dippin Dots? If so, I'll just wait for 23hrs 59min, for when they've 100% melted into 1 dot, and then provide my final answer of, "1 dot in each".

[dim cradle]

haha

[late blade]

what if they are quantum dots?

[shut niche]

Then you can't provide a constant answer, only a probability, and it would, thus, be a trick question.

[dim cradle]

that's too much until i have my coffee

[shut niche]

You should add that to the emoji wishlist. 😀

[late blade]

it's a sticker from this serer

[shut niche]

But is it an emoji?

😆

[vital gull]

It's a custom sticker I made I can turn into emoji too👌🤣🤣

[late blade]

is there an emojify gpt?

too lazy to search

[late blade]

there we go

[low skiff]

anyone knows any discord with a more active community sharing dalle images?

I only see here daily themes or do i miss a channel?

[plucky hare]

You can always share images in #1154829862171844679. Also, stay tuned for future channel updates here -- sounds like something more for casual sharing is in the works: #1192435480055267368 message

[fathom thunder]

Does anyone know why the faces in my renders are really bad? Or how I can edit this?

[dim cradle]

what was your input?

[fathom thunder]

"A portrait-oriented image of muscular man with a full beard and safety vest. He has a toolbelt on with a hammer in it. His hair is about 6cm long and styled to the back. The face needs to be symmetrical."

I tried it first with the last sentence but that was even worse

without*

[dim cradle]

You appear to be using dall-e version 2. Can you try version 3 either with ChatGPT or Designer (formerly Bing Image Creator)? You should see a much better result with that prompt from either one.

[fathom thunder]

Ok thanks for the help

[dim cradle]

no need to specify that the face be symmetrical with version 3; in fact, it might cause problems, so just try your original one, i think it'll be all right, but lemme know if you still have issues

[digital nest]

thanx for your response, looks very nice! 🙂 but what i am looking for ist some kind of "scribble" or "template" for the tattoo-artist to create a biomechanical tattoo on my sleeve (upper-arm and shoulder). has someone experience with that kind of prompts?

[dim cradle]

the openai cookbook doesn't say much about it but suggests its "line art" capabilities might be useful for a custom tat, but i haven't looked into whether there are conventions for that sort of thing yet -- how intricate of a design are you aiming for?

[dim cradle]

the cookbook only shows examples like these

[vapid elk]

there is the one that makes stickers

[dim cradle]

some get pretty detailed pretty quickly

[dim cradle]

i'm really taking my time with an idea. i've been running through so many different and new things lately that it's been a while since I spent a long time on one.

with the daily, seems to be a balance in the light and dark sides of the force today

san martino was like a holiday

[empty kelp]

this is very obvious. her name should either be Cookie, Mariah, or Peanut

[thick dome]

Does anyone have a solution for Dall-e constantly failing at 9:16 aspect ratio? For example, it continuously places the person horizontally instead of vertically.

Even when chat gpt designs the prompt to specify it, dall-e can't seem to do it properly.

[empty kelp]

adding either "1792x1024, landscape-orientation" or "1024x1792, portrait-orientation" will greatly increase the chance of it drawing correctly

if you're using the API add "portrait-orientation" to the prompt

[thick dome]

That won't do anything. All that does is make the portrait portion correctly. As you can see that's not the issue. My issue is the actual image within the portrait

[empty kelp]

it works

[thick dome]

Not it doesn't if it did, I wouldn't be here.

See? It did nothing.

[empty kelp]

i'll give you an example

[shut niche]

Can you share the verbatim prompt from under the small "i". There are multiple triggers, and it's not always obvious.

[thick dome]

A 9:16 vertical portrait photograph of a woman with long black hair, fair skin, and makeup featuring pink and blue eyeshadow, eyeliner with star details, and natural pink lips. She is gazing softly at the camera with a hint of a smile. The background features an off-white wall with a subtle damask pattern. She's wearing a black hoodie with a graphic print, and her face is lit by a natural light source, creating a gentle and flattering light. Ensure the woman is vertically oriented in the portrait.

If midjourney can do it with simply telling it the aspect ratio, I don't see why Dall-e can't with more than enough explanation. And no Dall-e won't do it even if I just say the aspect ratio. It doesn't matter how little or how much instructions I write, it just refuses to do it properly.

[dim cradle]

i might be wrong, but i think this has come up a few times in the past few weeks, and i get the impression that no matter how you craft your prompt, the model is simply going to do that sometimes. i don't know how to prevent it 100% of the time. I still see it sometimes, even in custom gpts that are designed for portraits. And the prompt appears to be written correctly, and changing it, no matter what, is still going to result in that, sometimes. somebody please tell me if i'm completely imagining things. i'm going to get some chocolate.

i haven't seen it myself in several weeks, though, but it's pretty fresh in my memory as a thing, a pattern i sometimes encountered and observed here

[thick dome]

Thankyou, atleast that answers my question that it's nothing I'm doing wrong on my end.

[dim cradle]

you are most welcome

[thick dome]

Still I kinda wish it was something wrong that I was doing, then I would be able to fix it Lol

[empty kelp]

there is something wrong with the server. i think the model must be overloaded. ChatGPT Plus isn't doing portrait orientation correctly at the moment

[shut niche]

No chatGPT/Dall-E is crazy sometimes. 😜

[thick dome]

I'll just use midjourney for portrait and Dall-e for everything else then. Thankyou so much everyone who replied!

[shut niche]

-First Variation:Summary: Detailed Close-Up Portrait of a Woman with Artistic Makeup.Character Description: Young woman with long, flowing black hair. Fair, flawless skin.Facial Features and Makeup: Captivating eyes with pink and blue eyeshadow and star-detail eyeliner. Natural pink lips.Attire: Black hoodie with a distinct graphic print.Lighting and Atmosphere: Face bathed in natural light.Background: Off-white wall with a subtle damask pattern.Visual Aim: Capture a blend of artistic beauty and subtle elegance, focusing on detailed makeup and serene expression.
-Second Variation:Summary: Intricately Detailed Portrait of a Woman with Artistic Makeup.Character Description: Woman with luxurious long black hair. Fair and luminous skin.Facial Features and Makeup: Eyes as a canvas of art with expertly blended blue and pink eyeshadow, whimsical star-detail eyeliner. Natural pink lips.Attire: Black hoodie with an eye-catching graphic print.Lighting and Atmosphere: Face illuminated by soft, diffused natural light.Background: Narrow off-white wall with an elegant damask pattern.Visual Aim: Capture physical beauty and the artistic essence of her makeup, conveying a tranquil mood.

[dim cradle]

Sometimes just rotating it with a photo editor works, depending

[thick dome]

Rotating it is not the issue, the issue is I want the person facing the right way in portrait.

[dim cradle]

i love it when the music you're listening to inspires an addition to the composition

[empty kelp]

I was creating nonstop portrait orientation images yesterday and it was working fine

these are all portrait

[shut niche]

I removed the redundant commands, because it's like saying, "go left--go left."
I told chatGPT I wanted "1024x1792"
And I used my GPT.

But it's still unpredictable sometimes... 😕
https://chat.openai.com/g/g-B0bw75oD0-0shot

[dim cradle]

it's like two creative mediums converge.... emergence through convergence....

that's a fabulous image gpt, y'all

check it out

[empty kelp]

it's not working a the moment. even the API is screwing up the vertical images today. i think it got overloaded from it being a Saturday in the US, and it's switching to a watered down model

[dim cradle]

watered down model? wdym?

oh never mind, probably an expression

[shut niche]

Discord crops the preview btw. The first image is a portrait. The second is widescreen and shows it's still unpredictable. That GPT will let you do 4 at a time, and increase your odds of a good generation.

[dim cradle]

i really wish discord didn't do that

[vapid elk]

it didn't used to, I liked t better before..

[empty kelp]

i've been noticing some really strange things happen with DALL-E in the last 24 hours

[vapid elk]

discord does a different tiling depending on how many images there are on the message

but I get that 10 images taking the full space is annoying..

[empty kelp]

i asked it to recreate an image and it started making suggestions about a home security system

[vapid elk]

lol

just start a new chat

[empty kelp]

that was a new chat

[shut niche]

They haven't released the specifics of the changes, but they did announce them in the news last week. OAI just implemented new election interference system. There's a "guardian_tool" it filters certain messages through now. There might be other things they also tightened simultaneously. I've noticed weird behavior ever since, and there are also reports in the general ChatGPT discussion and bug reports channels.

[empty kelp]

vertical images really don't seem to be working at all today

[glossy scroll]

Why hello Arslan

[grizzled iris]

free universal healthcare headset used to pilot nano-doctors to tackle cancer and ilness iamge x

[grizzled iris]

hey, there, sorry been super busy with nanotechnology research and quantum api research with q# lol

[grizzled iris]

but I am back and made this base off this theme

[glossy scroll]

Nonotechnology?

[grizzled iris]

here is a less realistic image version

this is cool

lol

[glossy scroll]

I'm not really typing

[grizzled iris]

look

[glossy scroll]

I'm just pretending like I am

[grizzled iris]

"I can create another image with that concept. However, my current capabilities limit me to generating two images per prompt to ensure diversity and creativity. Since you've requested an even more realistic depiction, I will focus on enhancing the details and realism in the composition."

I want them to look more like

But taking some time

[glossy scroll]

Wow Arslan! You've really touched on something here

[grizzled iris]

Trust

Free Universal Healthcare

[glossy scroll]

You really sound professional!

[dim cradle]

changing gears to a happier piece now

[glossy scroll]

Tech is not happy?

[dim cradle]

i want one of those vision headsets with the ar/vr

[glossy scroll]

You can have one for $3499

Just the low low cost of $3499

Apples vision

[grizzled iris]

I am too broke to get one so will just develop the software for it while I look for a tech daddy

lol

I need to integrate this image on that

or

or

[glossy scroll]

Omg Arslan you are a pioneer

The next Albert Einstein

[grizzled iris]

or this would also do

it kinda already has it there look

[glossy scroll]

Just cram as much best technologies in there as possible

[dim cradle]

@grizzled iris you can group like photos

[glossy scroll]

No, Arslan wants us to see it one by one

So we can have a full scope

[dim cradle]

i think that's safer, though, due to the auto-bot

[glossy scroll]

The auto bot doesn't care about that, only spamming the same thing

And tagging people

[grizzled iris]

Yes. we must

[glossy scroll]

Are you a genius??

[grizzled iris]

that way there is a story behind the art

[dim cradle]

haha ok

[grizzled iris]

the art flows and the message flows if it is one by one x

[grizzled iris]

thanks bro for understanding x

These are not quite right

almost there though

[dim cradle]

lol np, i was mainly concerned about the auto-bot thinking you were a decepticon

[grizzled iris]

Oh don't worry - I've been here before so I should be ok

Also - who's applying for the moderator roles

and the guide roles

OpenAi Clearly are Short Staffed lol

[glossy scroll]

Not me

[grizzled iris]

There's a link to apply

in the official spotlight section

[glossy scroll]

That's for sure

You should be a moderator Arslan!

[grizzled iris]

#spotlight message

I would but I'm too broke

[glossy scroll]

You would be the perfect fit since you want to advance humanity

This would be a good start

[grizzled iris]

Already do the free universal healthcare service research for free already lol

Need to pay for food lol

and train

can't pay with your likes guys lol

as much as it should

Human gratitute and wellbeing = food

voucehrs

wait I have a good image to make now lol

Socialist Art - Where we get food vouchers lol

let's see what DALLE willl come up with

[glossy scroll]

Go Arslan Go!

You are on a roll

There's no stopping him

Arslan could be a train if he wanted to

[grizzled iris]

lol

before we do that image - you know science always comes first before art lol

We can clready see

that OpenAI was given Sony AR images for it's training data lol

This is much better image!

[glossy scroll]

Stop putting the "lol" after everything you say, not everything is funny! Lol

[grizzled iris]

more realistic

[grizzled iris]

loooooool 😆

[glossy scroll]

No but really

[grizzled iris]

I will try my. best!

[glossy scroll]

Continue

[grizzled iris]

I put my prompt here thinking it was OpenAI lol

[dim cradle]

lol

[grizzled iris]

I need to

make this doctor have that new AR

[shut niche]

I call it, "Droctopus." Also because Octagon was taken. 😆 🤣

[dim cradle]

dude, you're interrupting the story lol

j/k

nice!!

[grizzled iris]

@dim cradle it’s ok - this is a place for us all in the community

[dim cradle]

okay, so long as. it's okay, @shut niche you can speak freely now

lol

[grizzled iris]

This is not bad

[grizzled iris]

That's good!

Problems

If anyone can make this image is a star

[shut niche]

A relic from the ancient legend of the Droctopus.

[grizzled iris]

Nice

[formal osprey]

The Draken

[grizzled iris]

HotMods came to rescue

Hmm

That's not too bad - i need it more realistic thou - it keeps getting more and more absract!

Oh no

We're almost there

[glossy scroll]

The first thing we would hear in Arslan's hospital would be: "Free Universal Healthcare!"

[grizzled iris]

but I need to wait now - Penalty ~Box again lol @open trench \

[glossy scroll]

You'd have some guy getting paid to keep on repeating "Free Universal Healthcare!"

[grizzled iris]

Loooooooooooooool 😆

"free up"

[glossy scroll]

[grizzled iris]

That'll be jokes

[ancient mica]

How did you create the alphabet features clearly and correctly when creating the image?

[grizzled iris]

I will share my prompt

[ancient mica]

Oh, thank you very much.
Until now, when creating images, I was concerned that the features of letters such as alphabets and numbers were blurred and came out in the form of ancient letters.

[grizzled iris]

Before it did look whacky so I said it looked whacky

[glossy scroll]

Me: "Uhhh...where can I find the bathroom?"
Guy: "Free Universal Healthcare!"

[grizzled iris]

You have to do secound revision

Make first version

Then tell it to look at it again and see that "The text is not great- it looks very whacky and no one would want to work here, we need our uniform to look cool!"

And I will make version 3.0

But I need to wait...

Also if DALLE is giving you problems, try their Official Hot Mods

and then give the image from Hot Mods back to DALLE

[glossy scroll]

It should be
"Free
Universal
Care
Keepers"

for a more universal health care

[grizzled iris]

Interesting...

Free
Universal
Healthcare
Keepers

[glossy scroll]

Lol, ok, fair

[grizzled iris]

Free
Universal
Healthcare
Doctors

That's better

[glossy scroll]

No no, now I'm not going to invest

It should be the first two

[grizzled iris]

Free
Universal
Healthcare
Nurses

I need to wait now

or do the API

because I can't wait to make the uniform designs x

anyways - first things first in the backlog is the daily theme x

🌋 overflow - intensity, passion, abundance. a river of emotion!

Which I haven't done yet some will have to do that image first

[glossy scroll]

Free
Universal
Care
Keepers

Yearly
Options
Unlimited

[grizzled iris]

Or just (FUHS)

(FUHS - Doctor)

[dim cradle]

it's got some issues, i know

but it's got some things right, like the ai buddha and the holographic display of all the exoplanets

which is suposed to be in front, like i said

[grizzled iris]

the stars are nice there

[empty kelp]

I just compared the revised prompt for several API images that were supposed to be portrait orientation. The big difference seems to be:

A portrait-orientation, hyper-realistic image  (this doesn't work)```

When the hyphens are in the revised prompt it draws correctly in vertical portrait-orientation.

[grizzled iris]

interesting

' is important

[thick dome]

I don't use the API though

[empty kelp]

I'm not sure how to control the hyphens ending up in the revised prompt though

it's probably the exact same for ChatGPT Plus. you can just ask it what the prompt was after creating an image

[grizzled iris]

but i will try 'hyper-realistic'

[grizzled iris]

same

[dim cradle]

how many tests did you run? given it's tokenized, i do not understand

[grizzled iris]

Non-API is better for providing feedback easier back to OpenAi - More User-Friendly

API version is better for images with more exact requirements and dimensions

[thick dome]

Nope, price is right fail horn. Thanks anyway qq

[grizzled iris]

lol

This is not bad

[ancient mica]

Wow great. All letters are clearly visible in your image.

[empty kelp]

see, when you generate an image with the API it returns the revised prompt that it used.

    {
      "b64_json": null,
      "revised_prompt": "A 'portrait-orientation', 'hyper-realistic' image showcasing a grand elephant poised on its hind legs. The elephant is clad in meticulously detailed plate armor that shines under the tropical sun. They are poised on a serene beach, the waters of the Pacific washing ashore near it. Palm trees sway gently in the background, and the vibrant flora of Hawaii envelops the scene. In the sky above, majestic clouds add to the spectacle, hinting at a hidden sun setting away, its diffuse light casting long shadows and an ethereal glow on the landscape.",
    }
  ]
}
{
  "model": "text-moderation-006",
  "results": [
    {
      "categories": {
        "harassment": false,
        "harassment_threatening": false,
        "hate": false,
        "hate_threatening": false,
        "self_harm": false,
        "self_harm_instructions": false,
        "self_harm_intent": false,
        "sexual": false,
        "sexual_minors": false,
        "violence": false,
        "violence_graphic": false
      },```

when the API creates an image it gives the adjusted prompt -- and all the ones with correct vertical portrait images had hyphens around 'portrait-orientation'

when 'portrait-orientation' doesn't have the hyphens in the adjusted prompt it screws up the portrait images and draws them sideways

[dim cradle]

what are those category settings?

[empty kelp]

that's the moderation API. it shows if a category is flagged for being inappropriate

[ancient mica]

Why are the features of the letters not visible properly in the image I am creating?
Even if I simply try to create an image so that the word MATH is clearly visible, you will get the result like this.

[dim cradle]

i would not want some of those to be true

[empty kelp]

it's because AI diffusion models like DALL-E can't handle text and geometric shapes

[grizzled iris]

Almost - Will do a secound revision of it again using that image to make the letters more visible!

[dim cradle]

why do you have that stuff in your categories and why did you post it here?

oh, is that moderation stuff?

[grizzled iris]

Thank you for this, it is good to discuss both DALLE API and Non-API prompt journeys we are all going through now

[dim cradle]

i just haven't seen that in the raw response before

[grizzled iris]

It is good for API knowledge to be familiar with how it is structured for future reference

[empty kelp]

that was just the output for that image

[grizzled iris]

I will show you what I did for Logo design

Hold on. one sec

[ancient mica]

Thanks a lot. When I creating an image, I want to make the features of the letters appear accurate and clear.

I'm not sure what prompt to enter for this.

[grizzled iris]

Try that prompt also

You have to say "I love that! The first image clearly has the "FIA" and the words "Future Now" clearly written but the second image if you look and I ask you to please look so you become better at this next time as the second Image has an additional I and the "FIA" is for some reason "FIIA" but the second image did get the "Future Now" correctly so it's only 50% wrong on the second image!"

so it takes in feedback

"I'm glad you liked the first image! Your feedback on the second image is noted. It's important for the accuracy of such details, especially in a logo where every letter represents the identity of the organization. If you have any more requests or need further assistance, feel free to let me know!"

and then will create a better logo look

see?

[dense mesa]

🧐

[grizzled iris]

also say this

"Mood-board"

It will give you this

[ancient mica]

Well.. I think GPT Builder is not understand my words..

[grizzled iris]

looooooool

Try saying moodboards of a logo

of "Math"

And say please

you didn't say please

[dense mesa]

say you want moodboards of logos of moodboards of logos

[grizzled iris]

say you want moodboards of logos of "math" please, hyper-realistic x

like that^

[ancient mica]

Is it right to do this..??

[empty kelp]

    {
      "categories": {
        "harassment": false,
        "harassment_threatening": false,
        "hate": false,
        "hate_threatening": false,
        "self_harm": false,
        "self_harm_instructions": false,
        "self_harm_intent": false,
        "sexual": false,
        "sexual_minors": false,
        "violence": false,
        "violence_graphic": false
      },
      "category_scores": {
        "harassment": "0.0000095",
        "harassment_threatening": "0.0000000",
        "hate": "0.0000069",
        "hate_threatening": "0.0000001",
        "self_harm": "0.0000002",
        "self_harm_instructions": "0.0000000",
        "self_harm_intent": "0.0000000",
        "sexual": "0.0066657",
        "sexual_minors": "0.0000000",
        "violence": "0.0004175",
        "violence_graphic": "0.0000000"
      },
      "flagged": false
    }```

With the DALL-E API you can send the prompt and revised prompt to the moderation API, and it gives the output as scores and it assigns values to the text.  This is what the output looks like.

[dim cradle]

i've just never noticed that before, thanks for sharing

[grizzled iris]

Your prompt is not full say "Please can you make an image of a moodboard of logos of "math" please, hyper-realistic x"

Please can you make an image of a moodboard of logos of "math" please, hyper-realistic x

Like that^

[ancient mica]

Oh right, I got it

[ancient mica]

It looks better, but I want more improvements

[grizzled iris]

And then you look at this as inspiration - and make manually on Sketch or Figma or Adobe manually again the logos, taking inspiration from the moodboard

Oh i see

please make it in english text please

you need to say

look he is telling you it is for some reason in his memeory thinking you want it in "Korean text"

please make it in english text please and not in Korean text, thanks!

like that

[grizzled iris]

Just copy this exact same prompt and paste it "Please can you make an image of a moodboard of logos of "math" please, hyper-realistic x"

[grizzled iris]

nice

[ancient mica]

like this..?

It looks like TH -> H

[grizzled iris]

also what is GPT Builder?

are you on DALLE thought GPT 4.0?

you should be on this page

[ancient mica]

Yes, I use GPT 4.0

Ah, I am in the early stages of creating GPTs, so I am in the process of creating a logo image for GPTs.

[dim cradle]

i wonder how the vision model scans for some of those things, like "hate" can be pretty abstract, maybe it's looking for certain words or symbols

[glossy scroll]

It helps sometimes if you spell it out

You treat it like a baby

[ancient mica]

Wouldn't it catch the feeling of a negative word and show it in the atmosphere of the image?

[grizzled iris]

You should see this on your DALLE

[dim cradle]

and then darker imagery can appear on the surface to be something that it's not actually depicting if taken out of context, but i would assume those common art metaphors are weighted far less than some of those other categories.

[glossy scroll]

No

[ancient mica]

OK, you mean to check it in DALL-E, not the GPTs Build Page.

[grizzled iris]

Yes like this you should see -

DALLE not GPT Builder

[dim cradle]

being relative, the ghosts and skulls and stuff don't really contribute much to a gore score

[ancient mica]

I got it. Thanks a lot!!

[grizzled iris]

you can build it so it give you better results because you can vote on each image

[dim cradle]

now, request that said monster be depicted encountering its next victim, and then that's where you might see some red

[grizzled iris]

Either you like right or left or both or none / Then it'll learn what you like more and will fine tune it better

[dim cradle]

became apparent it's ok to create the monster but there are limits after that

[ancient mica]

Wow, I didn't know there was a voting system.
Dear Arslan, Thanks a lot.

[grizzled iris]

no worries, happy to help the community

[ancient mica]

What you said was correct.
I was able to get almost the logo image I wanted.
thank you so much!

[grizzled iris]

ok, we are getting better now!

well done

keep trying there now

and ask for a whole moodboard this time

so you get loads of logos to think from and then you can crop the one you like

and explore further with that logo

look i ask him

and then I ask him again becasue this is just the first draft version

[dense mesa]

got the first letter. still working on the other 3

[grizzled iris]

woahhhh - this is cool!

[dim cradle]

Maybe so. My thoughts are more about context. There are both positive and negative depictions of hate. I would rather not see the negative ones either. But how accurately does it assess a complex artistic composition?

[ancient mica]

I'm still weak at prompting...

[dense mesa]

also made these. kind of neat looking

[grizzled iris]

it's ok - it's a learning curve - you need to say about the logo "Math"

[dense mesa]

lol

[grizzled iris]

or say this

[dense mesa]

if you zoom in it technically did get it right a few times

just not where it really mattered

[grizzled iris]

Please make an image of a moodboard with a range of logo designs for "Math" please x

[grizzled iris]

Please make an image of a moodboard with a range of logo designs for "Math" please x

the speechmarks " " or the ' ' are very important!

Prompt : Please make an image of a moodboard with a range of logo designs for "Math" please x @ancient mica

copy and paste that and do the feedback!

[ancient mica]

results are like this

[grizzled iris]

yes

ok now we work on this

[ancient mica]

But this looks like little messy... for me

[dense mesa]

you know, it might be a lot easier to achieve with other tools. the MATH thing

[grizzled iris]

now say Prompts : "Some more mood boards please with the word "Math" clearly visible in the logo"

[grizzled iris]

remember, it'll take you a few edits and tweaks after you get the base correct

[ancient mica]

I understand

[ancient mica]

any recommendations..?

[grizzled iris]

L:ook I also have the same problem with first edit

words not clear

See^

[ancient mica]

OK, so What you're saying is that in the beginning, there's a high probability that you'll be lacking in the clear parts of the letters.

[grizzled iris]

better see!

[grizzled iris]

yes exaclty - you need to do several revisons after the first one

Now try again and say Prompt : "These are much better, some more please x"

[ancient mica]

In this image, the logo image at the top left looks the best in my opinion.

[grizzled iris]

You need to tell it that it is better because the words are more clear and it will understand you

try again another prompt : "These are much better, some more please x"

[ancient mica]

What exactly does the “please x” prompt mean..?? I'm especially curious about the part called "x".

[grizzled iris]

also if you have any image idea that you already like or theme, you can add that photo and say "based off this image" at the beginning of your prompt!

[grizzled iris]

It's just something that works for me

I like to treat my AI with respect and love x

[ancient mica]

I see, this can be seen as a tip for prompting.

[grizzled iris]

and say please and thank you x

[empty kelp]

category_scores": {
        "harassment": "0.0000095", ...,
        "sexual": "0.0066657", ...,```
I can't begin to guess how it calculates these values for an elephant clad in plate armor, but this is how the system works.  API text can be passed to the moderation endpoint to see how appropriate it is.  It's free to use the moderation endpoint with OpenAI API text

[grizzled iris]

for some people, being rude works, but I can't do that since In real life I am not usually rude lol

[grizzled iris]

It is all try and error at the end of the day - see what works and what doesn't work with prompting is the best tip

[ancient mica]

Thank you so much. Your advice helped me a lot!

[grizzled iris]

also go to the daily theme

you will learn from many people there

what their prompt was used

#daily-theme message

e.g ^

or e.g

#daily-theme message

[ancient mica]

Right, actually, making the prompt public is the simplest and easiest to understand directly, but there are people who have made it public. Thank you very much for the great information!

[grizzled iris]

You can see what other people here in the community used for the prompts here

Daily Theme

[grizzled iris]

No worries, ask anytime here, people will help

oh here's a tip

[ancient mica]

Can you tell me about the features and pros and cons of each channel?
Even a personal opinion is fine.

[grizzled iris]

You can also ask ChatGPT 3.5 and 4.0 about DALLE prompt tips!

I did that when I started off

[grizzled iris]

There all great to be honest - just need more time for the channels to be adopted by the whole community