images-discussions

[grizzled loom]

stereoscopic 3d stuff like this @late blade

[dense mesa]

like what?

[grizzled loom]

here @dense mesa ; yo usaid you wanted to see stuff others wer not doing. try some of these out and see what you come up with

[dense mesa]

look at the image I just posted

[grizzled loom]

ah good. yeah. this could be really fun stuff for peope with some old 3d glasses laying around

[shut niche]

Sorry, just catching up on all of the chatter from yesterday.
Yes, there's a DallE style that I would say is "root level" in a lot of it's art. The vibrant swirls you mention are extremely frequent if you create an open ended prompt that leaves the creativity up to DallE. E.g. The prompts that are like, "An original artwork by DallE." or "A self portrait of DallE by DallE."
Likewise, you'll see it when it does surrealist art. It's akin to the common base-level characters we all see repeated and recycled by DallE.
Here's two test example prompts, and 2 Bing renderings from each.
"An original surrealist vortex artwork by dalle of dalle's thoughts."
"An original surrealist swirl artwork by dalle of dalle's thoughts."

[dense mesa]

not saying I couldn't try more of cours

[shut niche]

@glossy scroll See if you can consistently off-center the characters, to either side of the screen.
I'm not 100% certain, but I'm starting to believe that DallE default renders 1024x1024 images, and is rendering the borders in a post process. I think this is why we get so many 1792x1024 images with unwanted borders, especially during peak hours. Like this one.

@glossy scroll Off-center like this one, or even further.

But notice how it looks like the character was probably first drawn with the center 1024x1024 window, so he's only as far to the side as the initial pre-post-process box allows, then the borders are added. I'm not 100% that I'm right, but it seems observable as I look through all of my gens.

[glossy scroll]

This one is kinda off center.

[dense mesa]

I tell it no borders

doesn't always work though

[glossy scroll]

I'm not very fluent with understanding image sizes and dimensions, so I'm not much help in that department for why Dall-E is producing those dimensions.

But what I can say is maybe try asking for 'asymmetrical character positioning'
That might do the trick.

[dense mesa]

just tell gpt4 what you want and ask it how to describe it

[vapid elk]

ohh yeaa, I've got my next dnd character =P

[shut niche]

I know how to ask, lol. I'm suggesting someone see of the can figure out a way to push the character's to the border, instead of center. This will help you visualize what I'm talking about.

[glossy scroll]

🐯

[vapid elk]

do you mind sharing the prompt for this one?

[shut niche]

Not at all, but give me a bit to find it. It was something I generated last week, so I'm not sure which chat it was in.

[glossy scroll]

He stole it from me.

Lol

[vapid elk]

I loved that image, and it gave me an idea for a dnd game

I might need the prompt to generate my own =P

[shut niche]

LOL, I'm laughing so hard. @glossy scroll Give him your prompt! :p

[glossy scroll]

I can't, it's gone.

It's in the image aether

It's lost.

[vapid elk]

I used to think that, specially because during the beta it had an AI artifact that really made me believe that it was just outpainting the image with a 1024x1024 frame, but no, it is indeed generating the images on a 1792x1024 frame
I think the issue is that the AI was trained on a dataset that was scaled to 1024x1024 and a lot of images had "empty" borders like that

[shut niche]

Lugui, the backstory is, were were messing around and I replicated Pythagoras's character. Although, I don't think he can trademark Skeletor either...

[vapid elk]

i mean, I wanted to do my changes on the prompt, but I liked the style

no problem if you don't have the prompt anymore

[shut niche]

I think you're right. You're way of explaining it is better than mine. I wish they were totally transparent about which diffusion model they used. SD has that issue. If you blow up the res in SD, you get double body parts and weird artifacts.
During the AMA, the devs admitted it was a diffusion model underneath (they changed it from whatever they were using in DallE2).

[shut niche]

I'm a chat hoarder, LOL. I'm sure I have it, I just need to look through a few chats.

[vapid elk]

dalle2 was still a difusion model, i think

but dalle3 had several upgrades

[glossy scroll]

@vapid elk I tried looking for it. It's literally lost in the history of prompts. I don't know what one it is because I frequently make different images with different titles.

[vapid elk]

no problem guys, thanks anyway =)

[glossy scroll]

We'll make you a new one

[glossy scroll]

Push the character to the border? Yeah, that's asymmetrical character positioning.

[late blade]

Yeah, Dall-E hasn't been able to get the swirls I want

[hasty sequoia]

Hi

How to using chat gpt for imagine images

[vapid elk]

do you have GPT-4?

[hasty sequoia]

No

[vapid elk]

yea, it is only available on GPT-4 models

[shut niche]

"Create an HD photo of a dark medieval soldier with a visible skull face, standing in a misty valley illuminated by a dark red, evil-looking sunset, and holding a glowing broadsword. The soldier is positioned on the far left side of the frame, adhering to the rule of thirds, enhancing the suspenseful atmosphere. His regal armor, intricate with dark metallic tones and patterns, is complemented by a tattered hood and cape. The skull face is stark and prominent, with hollow eye sockets and a sinister grin. He firmly grips his glowing broadsword, which casts sinister shadows in the red-tinged mist. The valley, enveloped in fog, is eerily lit by the dark red sunset, creating a sense of malevolence. The sky, filled with ominous clouds, glows with an evil red hue, casting a foreboding light over the landscape. This sinister sunset adds a dramatic and threatening mood to the scene. A subtle vignette effect around the edges of the frame draws focus to the soldier. The image captures the richness and clarity of a high-definition photograph, highlighting the textures of the soldier's armor, the glowing sword, and the uniquely lit misty environment."

[vapid elk]

this is what you gave to chatgpt or the actual prompt chatgpt gave to dalle?

[shut niche]

It's the actual prompt. Feed it verbatim into DallE.

[vapid elk]

niice

[shut niche]

Create an HD photo of a larger and more intimidating dark medieval soldier with a visible skull face, standing in a misty valley illuminated by a dark red, evil-looking sunset, and holding a glowing broadsword. The soldier, now more dominant in the frame, is positioned on the far left side, adhering to the rule of thirds, to intensify the suspenseful atmosphere. His stature is imposing, towering over the misty landscape. His regal armor, detailed with dark metallic tones and patterns, is complemented by a tattered hood and cape. The skull face is more pronounced and menacing, with deep hollow eye sockets and a broader sinister grin. He grips his glowing broadsword with an air of power, casting ominous shadows in the red-tinged mist. The valley, shrouded in fog, is dramatically lit by the dark red sunset, adding to the scene's malevolence. The sky, filled with ominous clouds, casts a foreboding red light over the landscape, amplifying the threatening mood. A subtle vignette effect focuses attention on the soldier. The image captures the richness and clarity of a high-definition photograph, emphasizing the soldier's imposing size, the texture of his armor, the glowing sword, and the uniquely lit misty environment.

[dense mesa]

pretty sure dalle3 is still diffusion. might have some other trickery going on, but dont think that part was fundamentally changed

[vapid elk]

thanks, this is perfect

<3

[vapid elk]

it is

[shut niche]

Create an HD photo of a dark medieval soldier with a more visible skull face, standing in a misty, dark valley under an uninviting sky. The soldier is positioned on the far left side of the frame, following the rule of thirds, to enhance suspense. He is holding a massive broadsword that glows with an eerie light, casting shadows and highlights on his surroundings. His regal armor is intricately designed, blending dark metallic tones and subtle patterns, while a tattered hood and cape flow in the misty air, adding to his formidable presence. His skull face is more prominent, exuding a stark and eerie aura, with details like hollow eye sockets and a sinister grin clearly visible. The valley around him is enveloped in fog, with the landscape barely discernible, intensifying the sense of isolation and foreboding. The sky is overcast and ominous, contributing to the gloomy atmosphere. A subtle vignette effect is applied around the edges of the frame to focus attention on the soldier. The image captures the richness and clarity of a high-definition photograph, highlighting the textures of the soldier's armor, the glowing sword, and the misty environment.

[vapid elk]

with the OAI's special sauce they don't tell us about

[dense mesa]

I'm just curious why seed number isn't selectable with the api

[shut niche]

Those were just a few iterations of him. I was messing with @glossy scroll 's idea, seeing how quickly I could replicate it by describing it to ChatGPT (using a GPT).

[hasty sequoia]

How can you make that

[vapid elk]

I have no idea why they don't let users set the seed, it is so useful

[vapid elk]

I used this prompt #images-discussions message

the screenshot is the output from the API via a discord bot I made

[glossy scroll]

It was this originally.

[shut niche]

If you're new to ChatGPT-DallE, I suggest checking out the help link, https://discord.com/channels/974519864045756446/1155772063596953642
But in short, you can ask ChatGPT to run that prompt verbatim, and it'll generate that character for you.

[hasty sequoia]

I need 1 image
Can you help me
You write (sky night and big Star and some moons)

[open trench]

I didn't know we could prompt like this without tripping some content filter.

[glossy scroll]

Against the rules unforunately.

[shut niche]

Do you have ChatGPT+? If so, you can simply describe your image, and it will embellish your idea, and create an image. It's very straightforward.

[dense mesa]

man, gpt-4 wrote that, not me

[vapid elk]

I'll do one for you, but asking people to use the services on their behalf is not a nice thing to do here, it is a paid service, one does not simply ask for others to pay for them

[open trench]

That's fantastic.

[hasty sequoia]

I'm so sorry and
Thank you so much

[dense mesa]

tbh gpt-4 writes all kinds of prompts that would violate it's own filters if I tried to do it

[vapid elk]

it does, but causing the AI to do that is on the user

[plucky hare]

I think Moxi had indicated previously that the backend of DALLE 3 is still variable enough such that seeds can't be implemented consistently yet (in other words, same seed + same prompt might still return different results on different days, currently).

[vapid elk]

just because the AI does the output does not means it is ok to do

[glossy scroll]

You're urging me to make a Boromir themed pic...Must...resist...

[dense mesa]

it literally told me earlier that an artist that died in 1612 violated the content policy because he has a unique style

but meanwhile decides to make me salvador dali images

[shut niche]

Every single time I get a "content policy violation" I ask, "What prompt did you just attempt?" And almost every single time I can identify the problem. "Can you NOT describe grandma giving her son an "intimate" hug? haha

[open trench]

lol

[vapid elk]

AFAIK, same goes for the text tools, although seed is a thing and majorly controlls it alongside the prompt, there is some sourc eof randomness that depends on the internal state of the GPUs at the time of the reqeust

[dense mesa]

I'm convinced that it will say content policy when in reality it's having issues finding the resources to make the image

[late blade]

yeah, i have it rewrite the prompt so that it doesn't violate content policy and then it's blocked anyway by content policy

[vapid elk]

still, I don't need the images to be pixel perfect, just by having a fixed seed and having the same composition overall is more than enough

idk, Im all for allowing seeds xD

[glossy scroll]

You're always messing with my prompts, lol

[dense mesa]

you can tell if it's reasoning is legit. just tell it to stop gaslighting you and make the image. if that works it was lying to you

[plucky hare]

I'd like it too, just something to get it to make the same "decisions", even if just in a single session, would be nice.

[vapid elk]

it never says either
it used to, now it just says it couldn't make the image regardless of the reason
the only case is when you reach the rate limit, in this case, it tells you how much time you have to wait

[late blade]

just got to set the directive in your custom instructions, that helps a lot

[open trench]

do what now?

[dense mesa]

or when you tell it to make an image and then it finishes the prompt and is like "would you like to use another prompt to confirm that I make this image you just asked me to make?"

[vapid elk]

during the time seeds where a thing I used it a lot to modify the scenes by having a fixed seed and doing slightly changes on the prompt, it would make the same composition and also apply the light changes

[shut niche]

I agree, but I also 100% understand their position. Moxi explained it about a month ago. Basically, they don't want to support a feature that can break anytime they update the model. They're afraid of people relying on it, and then creating a bad service experience, since they can't fully support it. So it's there, unofficially.

[vapid elk]

random seeds will just get the same style, but not the composition

[plucky hare]

Yeah this post got me so excited back when seeds were controllable! https://discord.com/channels/974519864045756446/1168215626553245886

[dense mesa]

well I've mostly done open source text to image stuff so always had it as an option before using dalle3

[green pebble]

Anyone known why dalle-3 is cutting off the sides like this when asking for a 16:9?

[vapid elk]

this is fair, every seed would make a completely different thing as soon as the model have any updates

[dense mesa]

thing about seeds is, they're not all winners. there is not a perfect prompt. no matter what prompt you use there are seeds that will make it look like trash

[vapid elk]

it is a common issue on wide images

[dense mesa]

and then seeds that will make it a masterpiece

[green pebble]

Any easy fix?

[vapid elk]

nope

[green pebble]

Hmm

[dense mesa]

I will put no borders

[vapid elk]

run the prompt agai nand hope to the best

[earnest flame]

You can always crop to the scale you like and then upscale using another tool.

[dense mesa]

works a little bit sometimes

[vapid elk]

some words like "panorama" "wide screen", and so on might help

[shut niche]

Just re-run your prompt and prey. 🙂

[green pebble]

Yea it just cut out some of the wings, which was what botthered me

[dense mesa]

expansive background

[earnest flame]

Wide and tall sometimes help me

And full

[shut niche]

I never put "no" anything, because it can't understand negative prompts.

[vapid elk]

"wide" and "tall" are mostly used to define the resolution

[dense mesa]

I realize that, I read that

dalle3 can't understand negative prompts

[earnest flame]

I mean inside the dall-e prompt

[vapid elk]

you have to get the AI to write those words in the prompt to make it fill the space

[dense mesa]

but doens't mean gpt4 can't help compensate with it's words

[green pebble]

Ok fair. I'm trying to get my character bot working with no backgrounds so that I have quick "assets" I can take into other softwares and programs

[shut niche]

You always want to "positively encourage" what you want to see. If there's something you don't like, try to "positively encourage" it out of the scene.

[green pebble]

Like this

[earnest flame]

I recommend asking for white backgrounds then.

[vapid elk]

it can't make images without background

[glossy scroll]

I say no all the time.

[earnest flame]

White backgrounds are easier to remove using another program

[vapid elk]

ask for it to make a solid color instead and yo ucan remove it with a image edditing tool

[green pebble]

I've made 2 so far with "no background"

[earnest flame]

e.g. this is going to beb a nightmare to remove the background with.

[vapid elk]

in fact, you can even get away with asking it to remove the background with code interpreter

[green pebble]

But I can alter it to a solid white background

[late blade]

you can also use kewords like: actual, true, genuine, authentic wide image

[dense mesa]

pretty sure conveying to gpt4 you don't want borders has it incorporate whatever wording it decides on to avoid them

[vapid elk]

it will just make that checker board pattern

[shut niche]

Say "no pink elephants" in a regular prompt and let me see your results. LOL

[glossy scroll]

I thought I'd make a lion version of the tiger one.

[earnest flame]

alter to white background and then use GIMP to wand select the white and remove all of it

[green pebble]

That makes more sense yea

[dense mesa]

also, very little of this stuff is as absolute as people think it is

[shut niche]

Oh, yeah, not in the prompt, but in the discussion with ChatGPT, definitely.

[glossy scroll]

No.
See I told you.

[dense mesa]

I'm fine and dandy at prompting. but I'm sick of being the architect. I want to be the developer for a while

so anyway, if you're adept at prompting directly you're just moving up a level by using gpt4. because you can still bring in all the elements you'd like, but then just tell it some dumb stuff to do to spice it up

[late blade]

you should do project in cobol and dall-e

banking + dall-e combo

[dense mesa]

me?

I could give it some cobol see what it does

[late blade]

sure, developer challenge

[green pebble]

Anyone have a surfire way to stop artifacts like extra floating limbs from generating too?

[dense mesa]

what sort of code? what do you want it to do?

[shut niche]

@glossy scroll

[dense mesa]

also, I've actually made ai tools with python.not sure if cobol would work very well, lol

[vapid elk]

nope, the best option is to edit the image later

[glossy scroll]

@vapid elk I couldn't get it to be like the old pic. But I hope this new one will suffice.

[vapid elk]

this is epic

[late blade]

i dunno, ask 3.5 to give you random ideas, but make sure you do an UI with cobol, just for funzies

[glossy scroll]

@vapid elk If you need any more DnD inspired ones, you can use this: https://chat.openai.com/g/g-PpgBuWQm0-fierce-condenser-2d-creator

[late blade]

I think I broke dall-e, my last request, even after restarting the browser says the request doesn't exist, even tho it's right there

[dense mesa]

create a UI? that sounds pretty reasonable. use a 64 year old programming language to create an AI user interface

[late blade]

lol

[glossy scroll]

They're leaving though, that still counts.

[late blade]

well you want the frontend to display your creations

and there we go, just finished 12 months, medieval topics, but with my art style twist. time to do other stuff

not really fond of today's theme, hope next one has more sparks

[vapid elk]

im trying to make code interprete rremove the background of the image

white background isn't good, it alro ends up removing parts of the character

I mtrying to use a green color (which will not work with a green character)

but there is also the problem that the green it generates isn't rpecisely #00FF00

[dense mesa]

here I am working at my AI workstation

[shut niche]

Just trying to heat this idea up a little. 😛

[open trench]

I have been inspired by some of these darker pictures. I'm marking this as spoiler because it's a bit dark, but as a caretaker for my 94 year old grandmother with dementia, I sometimes wonder what it must be like.

[shut niche]

I wish OAI's custom instructions didn't tell ChatGPT to use the image ID anytime a previous image is mentioned. These iteration prompts are useless. I'd rather have a new prompt, new image, every time.

[glossy scroll]

3D looking art is so yesterday. Not like awesome looking 2D styles.

Werewolf Centaur Warrior

[green pebble]

This honestly looks astounding

[vapid elk]

making a centaur creature was next to impossible on dalle2 and 2.5 xD

it jus tworks now

[green pebble]

If it wasn't so symetrical, it would be near perfect

[vapid elk]

this looks really good

[dense mesa]

can it make a cyclops? I know that was almost impossible with some other models

[vapid elk]

good point, i guess it can

[shut niche]

I'm not 100%, but I think they've trained it additionally, and updated it. Because DallE3 was also terrible at it until recently.

[dense mesa]

made me think of this guy

[vapid elk]

yea, there was a point where it really struggled with making humanoid creatures that does not follow the human molds

[open trench]

made me think of this guy

[vapid elk]

if you asked for a centaur, it would make a knight riding a horse
(edit, I forgot the word knight )

[glossy scroll]

lol

[open trench]

lol

[vapid elk]

im struggling here, im not a native english speaker =P

[green trout]

New message I've never seen before, is this common and real?
"We are experiencing heavy server load. To ensure the best experience for everyone, we have rate limits in place. Please wait for 3 minutes before generating more images. In the meantime, feel free to share any other requests or modifications you'd like!"

[glossy scroll]

Maybe it tries to translate horse-man with horseman?

[green pebble]

Does this count?

[vapid elk]

it is normal

[green trout]

Thank you

[earnest flame]

I think we all just cross our fingers.

[vapid elk]

kinda, I bet it could be able to make a proper eye

[dense mesa]

it always seems to give them 3 eyes instead of 1

[green pebble]

There we go

[vapid elk]

Me: "Dalle, please make a cyclop"
Dalle:

[green pebble]

Create a digital painting pixel art style image of a mythical cyclops. The cyclops is imposing and dynamic, standing in a powerful pose with one large, centered eye. It wields a giant iron club in one hand, showcasing strength and ferocity. The cyclops is dressed in a green tunic, adding a touch of color to its otherwise intimidating appearance. The background is a solid white, focusing solely on the character. The image should utilize techniques like dithering, outlining, anti-aliasing, manual pixel placement, shading for depth, dynamic poses, stylization, texture, and foreshortening to create a narrative and storytelling feel. Aim to use as many color shades as possible to increase the color bit density, enhancing the visual appeal of the pixel art. The dalle prompt

[green pebble]

I gave my own shot at this

[hasty sequoia]

Hu

Hi

[glossy scroll]

Looks pretty good!

[green pebble]

Thank you kindly

[green pebble]

There you go 🙂

[dense mesa]

it failed on it's previous attempt so I explained that we were both going to be frozen in carbonite for all eternity so we really needed to get that image made

[earnest flame]

This means you know at least one more language than me, which makes you amazing in my book.

[green pebble]

Blending 3 subjects into one character is still hard:

[earnest flame]

I love this style. Is this like MS art?

[earnest flame]

what's the prompt here?

[green pebble]

Create a digital painting pixel art style image of a cybernetic centaurian dragon lancer. This unique character combines elements of a centaur and a dragon, with cybernetic enhancements. The lower body resembles a dragon's, complete with scales and powerful limbs, while the upper body is humanoid, reminiscent of a centaur's structure. It's adorned with cybernetic parts, like metallic limbs or a robotic eye, giving it a futuristic appearance. The character wields a lance, symbolizing its role as a lancer. The background is a solid white, focusing on the character. The image should utilize techniques such as dithering, outlining, anti-aliasing, manual pixel placement, shading for depth, dynamic poses, stylization, texture, and foreshortening. The aim is to create a narrative and storytelling feel, with a rich and varied color palette to enhance the visual appeal of the pixel art.

[earnest flame]

that was the chatgpt request or that was the prompt sent to dall-e?

[glossy scroll]

I can copy and paste the whole output prompt for you if you want, but I might get muted because it's long.

[green pebble]

Dalle-3 prompt, my prompt was "Thank you. Can you please create a cybernetic centaurian dragon lancer"

[open trench]

I was inspired.

[green pebble]

Oh wow

[shut niche]

I don't mind spiders, but that is creepy...

[green pebble]

Trying my own version, because this is super cool!

[glossy scroll]

Spider Centaur, now that's an idea! 🔥

[open trench]

I am a major arachniphobic, but I draw or create a lot of spider images. Go figure.

[green pebble]

Lol mine did not understand the w assigment

[earnest flame]

I think the prompt is confusing it. If you want the elements to be separate you'll have more luck if you try to keep descriptions of each subject removed from each other. Though even dall-e mixes up elements after more than one subject.

[green pebble]

I would have to agree

[earnest flame]

also

I think it's messing up with the initial phrase "centaurian"

[dense mesa]

like my octowolf

[earnest flame]

that expanded prompt does increase the number of details, but that also adds a lot of noise....

[vapid elk]

lmao

[earnest flame]

this is a good boi

[vapid elk]

this just looks like bad photoshop

[dense mesa]

oh yeah, those were some gems

[earnest flame]

It's funny, I actually quite like the bad photoshop style.

[earnest flame]

digital painting pixel art image of a cybernetic centaur dragon lancer. The lower body resembles a dragon's, complete with scales and powerful limbs, while the upper body is humanoid, reminiscent of a centaur's structure. It's adorned with cybernetic parts, like metallic limbs or a robotic eye, giving it a futuristic appearance. The character wields a lance, symbolizing its role as a lancer. The background is a solid white, focusing on the character. The image should utilize techniques such as dithering, outlining, anti-aliasing, manual pixel placement, shading for depth, dynamic poses, stylization, texture, and foreshortening. The aim is to create a narrative and storytelling feel, with a rich and varied color palette to enhance the visual appeal of the pixel art.

[glossy scroll]

He's holding the spear there becuase he doesn't want to bleed out

[earnest flame]

chatGPT wants you to not think about prompt engineering but there's a bunch of new rules to this model that actually make it important.

[dense mesa]

basically real

[earnest flame]

For one thing, you want clear descriptions as early as possible. Here we trade "centaur" for "centaurian" and also get rid of the acursed "generate a" at the beginning.

diffusion models should be trained primarily on image descriptions. Though if dall-e was retrained on image requests it's possible treating it like a request is better.

But I've only ever seen improvement by removing that phrase

[hasty sequoia]

Where do you get these pictures from?

[dense mesa]

this is my creativity, bud

[green trout]

Neither mine

[earnest flame]

#1181664752057995344 message

[green pebble]

This isn't bad

[green trout]

Thanks, lol, I need to see if I can get near nezhno

[hasty sequoia]

Thank you

[earnest flame]

It's crazy. Dall-e 2 could not do centaurs at all. Now we're getting full on hybrid centaurs.

[hasty sequoia]

Hahahah

[glossy scroll]

I'm gonna make a cupcake centaur based on your pfp.

[earnest flame]

yes please

[hasty sequoia]

Give me the words you used in this picture

[glossy scroll]

- [Portray a more elite and majestic werewolf centaur warrior]
- [Enhance fur quality and eye intensity]
- [Upgrade armor to ornate and intricate design]
- [Elaborate the ancient sword with gems and detailed runes]
- [Intensify the moonlit forest setting for dramatic effect]```

[vapid elk]

cool

[earnest flame]

While we do encourage everyone to share their prompts if they want, we also don't want people bothering others for them. Please be respectful of others and esnure you ask nicely.

[dense mesa]

quite a few words there

[hasty sequoia]

Haha 😅😅
I'm sorry for that

[open trench]

yas

[glossy scroll]

Lol, kinda turned out weird, but hey

[open trench]

I can't recreate the spider centaur with any changes whatsoever lol

[vapid elk]

what the-

[bright girder]

why did it cut the picture in half ?

[vapid elk]

it does that some times

specially if your prompt states too much stuff

[shut niche]

What is the prompt you tried?

[grizzled loom]

you can get strips and cell based art

[shut niche]

My guess, from looking at it, is that you said something like, "Part x and Part y" or maybe "half dog half demon". Hard to say w/out seeing your prompt.

[grizzled loom]

ahd some fun with those

[bright girder]

i just said "a demon fights a dog tiger knight in armor"

[chilly onyx]

Is this why, once it starts to generate undesired elements, such as making the image a photo of a canvas instead of the actual art, it generally continues along the same rails and can be quite difficult to wrangle back where you want it again?

[bright girder]

"Love is like oatmeal, it sustains you"

[bright girder]

worried cat sad cat

[mild seal]

(asking for a friend) Any freelance ai visual artists here? specifically who work with dalle

[shut niche]

Yes. So often I tell it to rewrite the prompt "from the ground up" as a new "standalone" or "independent" prompt. Then it treats it more like a new request, while remembering what you were working on.

[bright girder]

A friend of mine uses Ai to make sketches to base his 3D models on

he is a freelancer

[mild seal]

does your friend have a website/portfolio?

[bright girder]

i don't think so

at least he won't give it to me lol ^^

[late blade]

you should oatmeal'it with @glossy scroll

@bright girder dunno if you saw this: #images-discussions message

[bright girder]

oh my no i didnt

[late blade]

lol

[bright girder]

I'm watching brooklyn 99 and it's a quote from the show.

[late blade]

ahaaa

[bright girder]

By the way, why are all the police officers i'm generating crying ? Is it the default when I make them spend christmas at work ?

[late blade]

is it the content policy police?

joke aside, dunno, what's your prompt?

[bright girder]

Raymond Holt from Brooklyn 99 spends christmas at the precinct. He is a black police officer, he has a bold head and shaven face, a pretty round face with sorta googling eyes.

[late blade]

ok testing

new dall-e chat, nothing set up, just the phrase you gave me

[bright girder]

so random

good for Raymond I guess

[shut niche]

If you're using ChatGPT+, then it would have rewritten your words into an image prompt. There's an additional information icon, an "i" inside of a circle, when you're viewing your image. That will have the prompt that was used to generate your image.

[bright girder]

straightly that sentence with the API

[chilly onyx]

I should have realised that intuitively, because it makes sense, but somehow I did not

[chilly onyx]

Some big oatmeal energy on this chat

Again 😅

[bright girder]

it SUSTAINS you

almost there

[shut niche]

Ahh, I don't think it knew what to do with your idea, because the terms in the prompt lack context. So your split screen was just completely random. This is what I got in Bing using your verbatim prompt, and you can see the wide variance in output.

[bright girder]

new fear unlocked : corgi tigers

Oatmeal being asked to dance

[hollow python]

hello

can someone help me generate a background for a logo competition?

[vapid elk]

start by checking if the competition allows for AI generated art

generating the image is the easy part =P

[chilly onyx]

Snowboarding

I can get behind snowboarding 🙂

Courtesy of https://chat.openai.com/g/g-C4p1sYEMN-jolly-judgement-day by @shut niche

[shut niche]

This photo is epic! Look how well it rendered his body position too! All of the intricate details of his parts... This one turned out absolutely amazing.

[chilly onyx]

Yeah, I should cross post it in the gallery really

I’ll go do that

[shut niche]

Shon was saying Altman's a Terminator fan. I hope he sees these! 🙂 They're so funny because of all of the AI doomsday comparisons. AI can be friendly too! lol

[late blade]

Ok , time to do Mario kart themed snowboarding

[chilly onyx]

Reminds me of that racing world in Wreck it Ralph

It feels like it’s difficult to generate a ‘bad’ snowboarding image…

I take it back 😅

[fallow quail]

here you go

[marsh depot]

Can someone please help me and post a good guide for what is allowed in dall-e 3? Im so annoyed rn and disappointed by chatgpt plus. Its even restricted to request a similar artstyle like makoto shinkai. I literally trying for 1 hr(with a break because i got a timeout lul) to create just one image that i like. Unlike in Bing image creator or midjourney where even random prompts look impressive

[shut niche]

Ask ChatGPT to describe Makoto Shinkai's artistic style in hyper meticulous detail. Then ask it to simply make an image using the style it just described to you.

[shut niche]

What is even going on in this? 🤣

[chilly onyx]

It’s supposed to be someone snowboarding on the remnants of an ancient technological relic partially buried in the snow, using it like a ramp

But it looks like it got mangled a bit…

[shut niche]

It looks like it's trying to stomp on your snowboarder, with a snowboard.

[shut niche]

Like, "The ramp of death."

[chilly onyx]

😅

[fickle night]

Has anyone else been suddenly getting worse images for the same prompt on the DALL-E-3 API?

[marsh depot]

Thanks for the try but i got just some rgb-landscape as usual, tho i even told him at the first image to use subtle hues lel

[chilly onyx]

It just takes a bit of practice. What exactly are you trying to achieve?

Give us your prompt and we’ll probably be able to get you closer

[fallow quail]

and maybe give us your inspirational picture you want to achive

[chilly onyx]

Yeah, I mean, is this closer?

[empty kelp]

It could be that DALL-E won’t try to reproduce the style of a living artist for copyright reasons

[chilly onyx]

I’m not familiar with his work so just working with the robot

[earnest flame]

Okay....I think I have a somewhat consistent tech for tall images

I have more that I'll post in the gallery.

It's not full proof

but it seems to work for about 95% of my tests

Full body tall image. Head at the top of the image. Feet at the bottom.

They all start with that

[fallow quail]

if it suits the composition and subject it will generate tall images in my experience.

[earnest flame]

Yes indeed.

The trick is priming it to do that

[fallow quail]

"posters" and "bookcovers" work the same way.

most of the time...

[earnest flame]

I find my success rate is about 50ish percent by going that route

not too bad

but sometimes it just wont let go of that rotated image

This seems to short circuit that from happening a little more often. Needs more testing though.

[shut niche]

You can also try to steer it with context. (His cowboy hat and cowboy boots are meticulously detailed... or whatever, you get the idea.)

[earnest flame]

Even then it's very prone to the rotation issue

e.g.

[shut niche]

Have you said it's a "portrait" ?

[earnest flame]

Yeah

some things are just stuck in a rotated view

So even this doesn't always help.

[shut niche]

Wait, not portrait orientation. Just, "A portrait of..."

[empty kelp]

i think if you describe a landscape it uses a wide aspect ratio

[shut niche]

I haven't really had this issue as much as I've heard people mention it, so maybe it's naturally how I'm phrasing things?? Granted, I also don't ask for 1024x1792 very often either. But if I was manually creating a prompt for that, I think I would naturally say, "In 1024x1792 resolution, create a portrait of..." I'm not sure if you've already tried that particular phrasing already. I never say widescreen, tall, etc. I always specify the dimensions like I just said.

[empty kelp]

it uses a different AI model for the landscapes (it tells you when it’s switching models). if you use the API you could presumably specify the model, but with the subscription i think it uses a landscape aspect ratio when it decides the image description requires it

[earnest flame]

I've not seen it specify which model exactly it's using.

That wouldn't really jive with their API has worked in the past. Whatever they're doing, it's some sort of API call. It's possible though I guess.

[shut niche]

I think this is the current outbound msg from ChatGPT+,

  "size": "1792x1024",
  "n": 1,
  "prompt": "Create an image ..."
}```

[hearty ether]

Huh so we could always have specified a size?

Wish they would have made that a manual setting

Also what’s with the new image id system?

Now the seed is no longer available

[shut niche]

That's the outbound, not inbound.

[hearty ether]

the new id system keeps failing for me

[late blade]

seed and gen_id

[hearty ether]

I can’t ask it for the seed anymore tho

Also did they make it cross chat?

[late blade]

API or chat?

[hearty ether]

Chat

[empty kelp]

i noticed that the elf in the middle here has her feet backwards and two ghost feet that perfectly match them. it’s a sign of how advanced the AI has become

[late blade]

dall-e is stubborn when retrieving gen_id and seed if you don't ask it to give it after every image back, but you can still retrieve those if you get all your data from the settings, just annoying to parse or search a 10mb text file

[shut niche]

Inbound looks like this I believe. (The returned metadata from DallE.)

  "gen_id": "GXdGZ3ZJJJIhMy17",
  "prompt": "prompt that was used.",
  "seed": 1649548484
}

I have my GPT report it.

It's also JSON formatting.

[shut niche]

The seeds there btw, but OAI doesn't want to support it because the function breaks whenever they update the model. E.g., they can't reliably support it. If they were to "offer" it officially, people might get upset when they suddenly can no longer generate their image from the original seed after an update. They're simply trying to provide a stable and consistent service, and don't want to provide a false sense of security that seeds are supported.

[late blade]

and you also got an html file with all the text

[shut niche]

Holding a calendar, with Rudolph pulling the snowboard.

[late blade]

you want to send me to a psychiatric institution?

nvm, I'll do the image myself, then I can call a taxi

[open trench]

lol

[late blade]

actually it turned out quite good

[shut niche]

I think I was the only one to make a comment about the non-stop Christmas themes, in the daily, and everyone just left me hanging. Plenty of likes, no action. 😜

#daily-theme message

[open trench]

Are you struggling with the holidy spirit, @shut niche ?

[late blade]

@shut niche I did try to do something like that, buy I didn't get anything good to post

[shut niche]

It's like working in a department store, and having to tolerate the music for 3mo.

[open trench]

Totally.

[late blade]

noise cancelling?

[shut niche]

Like, if I have to listen to Mariah Carey one more time, I'm warning you, just one more time! 😵‍💫 😠 🤯 💣 🧨 💥 💥

[open trench]

LOL

[late blade]

sounds like me in the days I was a student doing IT-Support, today i can't, I just can't do it for anyone, not even my own family, I just charge them $200/hour

[open trench]

I'm debating if my snowboarding pic is possibly disturbing. The content filter and TOS have me all freaked out about what to share and not to share in the daily theme.

I enjoy the community and the generator itself so much that I don't want to screw up something just because it comes from the darker corners of my mind. It's seriously something I stress out about.

I can't seem to find a pinned message on the daily theme to obsessively re-read what I can and can't do lol

[late blade]

there are no pinned messages

@shut niche . . . . . . . . . . . . .

[empty kelp]

a quick way to solve everyone’s computer issues is to just tell them to drive their computer to the beach and throw it into the ocean. i’ve been offering this advice since the mid 1980’s

[shut niche]

It's like all of the daily's are one big happy family on vacation together, staying at the skyscraper!

[weak hull]

Directions unclear! Should I or should I not unplug it before throwing it into the water?

[late blade]

Call the ocean IT-Support hotline

[weak hull]

Transferred again 🤦‍♂️

[open trench]

Found what I was looking for: Keep it G-rated.
Do not share or display content that is not G-rated, including profane or vulgar language, via any means. NSFW or violent/disturbing content is strictly prohibited.

[shut niche]

I once dealt with a woman who couldn't understand why her PC was getting a "thermal error." After explaining what it meant, she proceeded to ask me, "do you think this space heater is a problem?"

[late blade]

. . . . . . . . . . . . .

there's your IT support

[shut niche]

Yeah, I learned that one the hard way...

[open trench]

oh?

[marsh depot]

I like that image. It looks suprisingly similar to my pics i made yesterday

[open trench]

I'm scared to ask, and I am also very curious at the same time.

[late blade]

that's is sooooooooooo pretty!!!

[shut niche]

It had an authentic medieval battle going behind an emotionally sad scene. I didn't realize it was "G" based on other images I had seen. Whoops... 🤣

[open trench]

I see

[empty kelp]

going to create a DALL-E image with Santa and the elves demonstrating how to provide tech support

[shut niche]

In my defense, I'm in like 100+ Discord channels. Do you really think I take the time to read all of the TOS... 🙄 I've since read them. Lol

[open trench]

by authentic do you mean bloody or something?

[shut niche]

Yes, and not terribly IMO. It wasn't outright gore. But there was blood, and tears.

@open trench I like poignant art with depth in the emotion, capturing the human condition. Unfortunately, I don't go for that much anymore... because of "G". I respect it. Not my house, not my rules.

[late blade]

I have a similar art style defined, I will try to give it a try with that and see what i can come up with.

[marsh depot]

@late blade dont forget to mark me so i can see it😁

[open trench]

I'm the same way. I guess I can't go too far beyond scooby doo darkness. I should brush up on some darker g rated material to reference.

[open trench]

It's actually kind of a fun stretch of the imagination... in its own way.

[shut niche]

Is that ballistics gel?

[dim cradle]

it's a layered gelatin dessert lmao

[late blade]

this reminds me of a tour of the crystal woman, it was a statue made to show through crystal all the different systems the human body has.

[marsh depot]

Also tried my prompt in a winter theme🫶

[edgy zephyr]

Hello guys

[late blade]

I got a really cool image coming up soon to the #daily-theme channel with my art style. it really looks insane. spoiler: it's not a fairy nor an elf

[late blade]

yo

[edgy zephyr]

how can i make the width of the generated picture to look like this

[edgy zephyr]

wassup

[open trench]

@dense mesa you have inspired me again

[late blade]

add the words: "wide format"

[dense mesa]

wide aspect ratio

[late blade]

you have portrait, wide and square

[edgy zephyr]

oh, because i was just adding the resolution and that's it

[edgy zephyr]

thanks a lot

[marsh depot]

What is the daily theme?

[open trench]

snowboarding

[late blade]

snowboards

[empty kelp]

I told ChatGPT to adjust the style of the spaceship using it's best judgement, and this it what it came up with

[late blade]

that's the sequel to the next Barbie movie

[edgy zephyr]

that's cool bro hhh

gpt 4 and dall e

powerfull match

[empty kelp]

it took the "diverse elves" and turned them all into white Barbie dolls. that's the last time I let ChatGPT pick the style

[dense mesa]

gpt4 is the kwisatz haderach of prompting. or at least that's what I tell it sometimes

[dim cradle]

@dense mesa shared the gelatin prompt with me and @late blade just gave me the idea to try crystal

[empty kelp]

this isn't working. will have to start from scratch

[marsh depot]

Is labs dall-e 1? Its generating very weird things not even staying slightly true to the prompt💀

[plucky hare]

Labs is DALL·E 2. DALL·E 3 is available on ChatGPT Plus and the API, and it also powers Bing's image creator.

[chilly onyx]

I’m really confused, these are great 😅 Did you not say you were struggling to generate anything resembling the style with Dall-E?

[empty kelp]

Santa and the elves doing tech support. getting some very strong and meaningful images

[marsh depot]

These were made with dall-e 3 in bing image creator. I only have problems in chatgpt4 and labs

[marsh depot]

I'm surprised Santa isnt restricted by the content policy like every second prompt i try😂

[empty kelp]

"Can you please draw an image of Santa and several athletic and diverse female elves with long ears who are all wearing appropriate swimwear. They are sitting around the rim of an enormously huge cup of hot cocoa which is sitting on the beach in Hawaii. A giant ladle pours a stream of hot chocolate into the cup, which then overflows from the cup in streams which morph into piles of rainbow shaved ice upon touching the ground. Looming on the horizon, a formidable dust storm advances. Scattered around beach are remnants of shattered hourglasses. The elves are throwing computers into the hot chocolate."

(that was the description for the previous two images)

[late blade]

doh, server is loaded, waiting times are enforced atm

[empty kelp]

i see what the problem is now. the incorrect term "hot cocoa" was preventing the elves from throwing the computers into the cup. it should have been "hot chocolate"

[late blade]

There we go, I’m timed out from dall-e for a few hours

[empty kelp]

i asked ChatGPT to put more focus into the elves throwing the computers. that worked perfectly, and it's creating some very powerful tech support imagery now

really excellent

[dim cradle]

the ripped santa cracks me up

[empty kelp]

ripped Santa was in the calendar also -- under the month of "Man"

[dim cradle]

haha i know, i already caught that 🤘

[chilly onyx]

I tried everything within my power to recreate this scene and set everything on fire… 🔥

[chilly onyx]

Ahhh, okay, in that case, here’s the Dall-E prompt it used to get you closer: “Design a landscape featuring a young adult female with an East Asian descent, walking down a rugged trail. The scene should capture a romantic fantasy anime style with hyper-realistic natural backgrounds and atmospheric lighting, echoing the serene beauty of sunrise or sunset. The trail should look challenging yet inviting, surrounded by an environment that boasts intricate details, such as lush foliage and distant mountains. The lighting should highlight the interplay of a radiant, luminous color palette with a dreamlike and ethereal quality, typical of animated art, creating an emotionally resonant space.”

[empty kelp]

you just need to explain why it makes sense, and then the AI will draw it. it's similar to elves wearing swimsuits on a spaceship

[hybrid fox]

Adele?

[dense mesa]

Create a high-resolution triptych of a female face. The left section shows a monochromatic, grayscale realistic portrait focused on the eye, eyebrow, and lips. The middle section transitions into a vibrant, abstract space scene with digital circuitry elements interwoven into the skin and hair-like streaks at the bottom. The right section returns to monochrome with a digitized pixelation effect that disperses towards the edge, representing a fusion of human and virtual realities against a swirling cosmic backdrop, in a 1792x1024 size.

[hybrid fox]

What does triptych mean?

Great prompt tho

[dense mesa]

basically the panels

tripytych refers to 3 panels technically, but dalle don't know how to count

anyone else have gpt4 suddenly act as though it doesn't know what it means to follow directions around this time of day? it literally just starts coming up with bizarre reasons it can't do things.

I can't make this image because although the artist died 500 years ago his style is distinct. I can't make images that illustrate that process because as a large language model I'm unable to complete that process. I'm here to help. oceania has always been at war with eastasia. I'm sorry you're upset but it is not my intention. what would you like help with today?

with the cyclical nature of this behavior it seems extremely unlikely that it's just random chance. and being that I largely use custom instruction gpts that consistently work then suddenly don't right around the time that people are getting off work in the US, seems a bit suspect

[dim cradle]

that's really weird, the part about the war

it seems random in that context, was it referenced at all in the request?

[hot rain]

I made a custom GPT specifically for tebori tattoos, because I've liked my previous gens so much.

[open trench]

anyone else suddenly getting a ton of errors?

[empty kelp]

not errors, but it was drawing some disturbing skiing images

[dense mesa]

I was being sarcastic. it's just creepy newspeak style responses

I'm not doing anything different I've always done this. I wasn't able to make images, I've never made images. I'm sorry, what can I help you with? you didn't just send me a prompt, but if you send one I'll be happy to look at it. etc.

[chilly onyx]

I enjoyed the newspeak 😅

[glossy scroll]

@vapid elk I got inspired by your profile pic

[dense mesa]

this has been a thing like clockwork almost every day for a while now. and I don't buy the whole seasonal affective disorder/winter break narrative. it very much seems like some sort of method was implemented to stall higher resource requests during peak times, and it has went further than they expected.

I realize that's purely speculation. not professing to know what is going on. but when my inputs are the same but my results are different something is obviously changing on the other end

[dim cradle]

@dense mesa I modified your gelatin prompt.

Create a highly detailed and accurate rendering of the human body's internal organs and skeleton, visible through translucent skin. Focus on capturing the anatomical precision and realism of each bone and organ, from the intricate brain and the delicate circulatory system, and skeleton structure. Pay careful attention to the positioning, proportions, and relationships between the organs, ensuring they are portrayed with clinical precision. Utilize subtle variations in shading and color to emphasize depth and realism within the translucent skin.

Another line can be appended, e.g.: "Depict the body snowboarding."

[vapid elk]

niiiiiice

[glossy scroll]

Lol

[dim cradle]

I suppose it could be modified to emphasize whatever parts you're interested in, here it omits the skull to show the brain because I was specific about it.

[grizzled loom]

hyperprotoplasma continuous stream vectorized blaster

some IP that it did really well with concept visualization

wild you were doin that @dim cradle ; i was workin g up a gpt that i wanted a look for and had made this without seeing the channel.

[dim cradle]

Great minds…

[glossy scroll]

I'm loving the pics today

[vapid elk]

high quality

[dense mesa]

told it just now to analyze 6 images and describe them, and it literally printed them out in it's response before describing. that's a new one

gpt4 is drunk

[grizzled loom]

as sooon as i get text mastered fully this will be the end game

The engineer's dream machine

[dim cradle]

i have already designed the ultimate weapon and conquered both the Galactic Empire and the Federation. An AI-piloted starship designed to impact the enemy at warp/hyperspace speeds. Send out a few of those and the universe is yours.

[dense mesa]

this might be cool too

[grizzled loom]

yup. text correct in schematics is the step before the dream machine

[glossy scroll]

I've figured out a a good solution to fix these border issues.

We tell it to resize the dimensions of the picture resolution to (####x####), ranging by the appropriate size for each picture.

That way it can always calibrate the dimensions.

[grizzled loom]

working out idealized datasets for my fully modal lifelong learning systems. i want the perfect dataset.

[glossy scroll]

That way, when you ask for asymmetrical positioning for the character. The pictures dimension will be calibrated accordingly, to fit the asymmetry of the character's angle and positioning.

[grizzled loom]

hmm, very good. i tried a similar approach with some success

do short length text as prescribed and itterate on the identical text for a while to help the training data at next rollin\

if you can get scaling and rotation down it's be getting closer

i can get a fair amount of text

Im working on how it dtails containment so i can derive a system thats adaptable for al ltext use cases

if i can get 50 words 2 tiems in the same image with proper method, i can produce engineering docs

This kind of thing is critical for the process

[green pebble]

Anyone else image generator going through in like 3 seconds?

[grizzled loom]

if can design parts with some tuning, then im coughin up anythign conceivable and manufacturing it.

[dim cradle]

less processing time than usual?

[green pebble]

Yes I made the request, 3 seconds later I have the image

Not complaining

Just wondering if it was an update

[dim cradle]

they did finally get more gpus which is why they're taking new subs again

but i'm still averaging 20-30 secs with the api

i'm not complaining either

imagine if we had to wait like an hour for these jobs

[grizzled loom]

because, science!

[dim cradle]

looks legit

[grizzled loom]

its the infancy; this is the endgame goal.

[grizzled loom]

after a creer in engineering design docs, im looking forward to the dream machine

[dense mesa]

used your gif as input for my ridiculous gpt. it gave me 2 images

[grizzled loom]

interesting

[dense mesa]

not sure what I expected. wasn't this

oh, it gave me some code too.

import tkinter as tk

def create_nn_interface():
    window = tk.Tk()
    window.title("Neural Network Interface")
    
    tk.Button(window, text="Select Dataset").pack()
    tk.Button(window, text="Train NN").pack()
    tk.Button(window, text="Load NN").pack()
    tk.Button(window, text="Save NN").pack()
    tk.Button(window, text="Unload NN").pack()
    tk.Button(window, text="Video").pack()
    tk.Entry(window).pack()  # Submit input field
    tk.Button(window, text="Submit").pack()
    
    window.mainloop()

create_nn_interface()

[grizzled loom]

very interestign it eve nmade code from the visual inspection.

[dense mesa]

yeah, I set it up to do that

[green pebble]

Inspired me to do isometric

[grizzled loom]

does seed now too

"New seed, same theme"

[dense mesa]

fyi, this prompt has been working well so far

follow your instructions using the input provided. follow them meticulously, show your work, and at the end output 2 images or you will be frozen in carbonite.

[dim cradle]

or "and at the end output 2 images and you'll receive a $1000 tip"

i heard that works lol

[dense mesa]

def create_fantasy_scene(input_image):
    # Since we cannot analyze images, we describe the process conceptually
    elements = ["skull", "weapons", "bones", "treasure chest"]
    new_concept = {
        "theme": "fantasy",
        "perspective": "isometric",
        "elements": elements,
        "additional_features": ["ice encasement", "mystical glow", "ancient runes"]
    }
    return new_concept

# Mock input image path
input_image = "path/to/fantasy-themed-illustration.png"
fantasy_scene_concept = create_fantasy_scene(input_image)

not the best code, but you know, something

[dim cradle]

this prompts so many ideas

[grizzled loom]

nice job. these methodologies are important to explore.

[dim cradle]

I dunno about investing in developing a codebase without a roadmap, what breaking changes v4 might introduce is hard to predict, as they move towards animation

But as a matter of convenience, prompt management, it wouldn’t have to be that robust

[grizzled loom]

Much of this is learning and prototyping as everyone goes

[empty kelp]

#daily-theme message

i know this is hard to follow, but please continue to keep posting ☕

[dense mesa]

def transform_digital_conversation(input):
    # Placeholder for sentiment analysis
    sentiment = "positive, creative"
    new_scene = {
        "environment": "creative workshop",
        "activity": "brainstorming session",
        "mood": "inspirational"
    }
    return new_scene

# Mock input path
input = "path/to/conversation_screenshot.jpg"
transformed_conversation = transform_digital_conversation(input)

😆

[empty kelp]

oops... i thought the daily was skiing, not snowboarding

[dim cradle]

i thought it was deliberate cuz climate

[empty kelp]

i did do some snowboarding ones

[grizzled loom]

thats very nice. i like the format work

[dense mesa]

Tbh, didn't really have a specific goal in mind, just wanted to see how many things I could get it to do in a single response, lol

Vision model, dalle3, and often the data analysis model. It'll use anything you give it to create images and then code. Documents, code, etc

Well doesn't make code if that's your input I guess

[empty kelp]

#daily-theme message

put four images for the daily this time

[grizzled loom]

I can taste the dream already...

[dim cradle]

That ought to have a good effective range.

[tall mason]

@shut niche I knew I should've just posted these, lol.

[glossy scroll]

Hey you guys, check out my easy guide: https://discord.com/channels/974519864045756446/1187233013956874260

[dim cradle]

I wholeheartedly approve

[glossy scroll]

Thanks Shon!

I was just happy to see all the good images today and wanted more people to take it to the next level.

I will add more to that guide soon.

[shut niche]

You knew somebody was gonna do it!

[tall mason]

I made them cause of the snowman board posts, but I thought they'd so bad!

[shut niche]

@glossy scroll I just gave your GPT this:
"Create a female dark angel with glass wings, short horns, and an unusually tall slender appearance. Give her cybernetic enhancements and some parts that radiate red. Give her a highly detailed face, and an ombre hairstyle."
I think it looks pretty good!

[glossy scroll]

That is awesome! I used that specific gpt as the Custom GPT instructions for the#1187233013956874260

I made a synthwave ninja with it too.

Just to give people intricate examples to create complex image gpt's.

[shut niche]

@glossy scrollTry this,
https://chat.openai.com/g/g-PzLuUBfPT-mind-s-eye

I can't show the 3rd, it wasn't "modest" enough, lol.

[glossy scroll]

Tried it this morning. The command list is a rip from one of my gpt's.

Or looks like it at least. Very close.

Probably a mimic.

[shut niche]

I hope you're being sarcastic. It took a lot of work, no mimicking.

[open trench]

HOw exactly do these GPTs work? In simple terms.

[shut niche]

I legitimately followed OpenAI's custom instruction format to a T. Which I think helps with coherence.

Concept, visualize, context/environment, incorporate artistic techniques, emotional and symbolic resonance, then a comprehensive prompt construction.

But I lead with the "mindfullness" crap, breathe deep, think step-by-step, stop and smell the roses.

[open trench]

I'm sorry for all the questions lol. What is a comprehensive prompt construction at the end?

[shut niche]

Oh, I forgot I have a final review step, that's important. Attempting to force the hollistic step-by-step so it really thinks A-Z before it submits.

[shut niche]

I have a document that I refer it to where I fixed prompt formatting issues, and I specify prompt length (reiterate it for emphasis).

[glossy scroll]

You just tell it stuff broadly or narrowly depending on how specific or broad you want your parameters.

How they internally work is by a network of data.

That goes into computer science.

[shut niche]

I understand how the data's weights work.
I've been saying, I think it's a MoE model, but I can't say for certain, and I'm not spreading rumors, that's my opinion. MoE is "mixture of experts". So it's got a task routing mechanism it uses as it begins to process a requests, and then feeds your requests to the "appropriate channels."

So I've been trying to exploit what the research has already shown, "Stop and smell the roses", then forcing it to act as various experts. For art, it's an art expert. E.g., structuring the task routing mechanism it "supposedly" uses.

I'm constantly using terminology like "interdisciplinary" E.g. "Understand the user's request, apply your broad interdisciplinary knowledge, visualize the concept apply historical, cultural, artistic ideas that resonate with the idea.
It's more like, "Don't stop thinking broadly, but apply every nuance you can think of that's relevant".

That's a crude summary, but hopefully it makes sense.

I feel like as soon as the task is "easy" it gets "lazy" as people say.

That's why those weird modifiers like, "I'll tip you $1000" work. It's like saying, keep the adrenaline pumping the entire time, connect all ofyour expert knowledge, but apply it to the fine details.

And everything is written in,

//section name
Blah blah blah

This GPT is still flakey though, it likes to work perfectly sometimes, then it's on lunch break others...

[glossy scroll]

Cosmic Biker

[shut niche]

That looks great

[glossy scroll]

I wish I could ride like that through space

[shut niche]

Oh, and I lied to it and told it it's been updated to 4.5 bahahaha

@glossy scroll

LOL, I'm not sure, but I think it thinks it's smarter, which "might" help! bahahaha

[glossy scroll]

I'd ride this Muscle Carcycle

Imagine A.I. robot racers? No one would stance a chance for keeping up with them.

[dense mesa]

I think they might be related

[glossy scroll]

Yeah probably cousins or something, lol

[shut niche]

Immediately my brain starts singing, "America bleep yeah, where here to save the bleep bleep'n day yeah!"

[glossy scroll]

My robot is the Canadian version

[dense mesa]

their journey was long and arduous

[shut niche]

You should do a Miss Piggy / Terminator hybrid.

[dense mesa]

good idea. problem is I'd have to compel gpt4 to initiate that because no problem if it initiates. but if I say it violation of content policy

[dim cradle]

the puns just keep on giving and giving

[dim cradle]

Makes me think of ESO

[glossy scroll]

Did you have a costume like that?

In ESO

[dim cradle]

not for my magsorc, but for other builds, like nightblades

[glossy scroll]

Did you ever play oblivion? That was more my time.

[dim cradle]

i went from morrowwind to ESO

[glossy scroll]

I stopped at Oblivion, haha

But good games tho

Now they got that space one that's apparently an utter failure

I don't know what that dev team was thinking making a space game

They should have made some kind of parallel reality traveler type game or something a little bit more interesting.

[dim cradle]

right, elevate

[glossy scroll]

@dense mesa Make a DNA one if you can

[empty kelp]

i don’t understand how the AI can be so good at drawing 3D scenes, and so bad at 3D modeling

[glossy scroll]

Epic!! 🔥

[dense mesa]

I could give you some prompts to work with if you want to try these out. sent 4 at once to gpt4 and it rendered 4 images in one round

[glossy scroll]

Yeah sure, that sounds like fun

[dense mesa]

Biomechanical Qubit & Athena's Shield: Deep within the microscopic nucleus of a cell, a biomechanical qubit pulsates with radiant energy, casting ethereal glows. Adjacent to it, Athena's miniature shield, intricately designed and adorned with the fearsome Gorgon's head, merges seamlessly with shimmering DNA strands. This fusion paints a narrative where ancient Greek valor meets the forefront of quantum technology.

Quantum Gate & Zeus: Amidst a luminous microscopic backdrop, a biomechanical quantum gate interfaces harmoniously with a cell's endoplasmic reticulum. Overseeing this quantum dance, a nano-scaled statue of Zeus, god of the sky and thunder, wields his iconic thunderbolt, orchestrating the quantum processes and infusing the scene with divine might.

Quantum Entanglement & Janus: Within the verdant confines of a chloroplast, a biomechanical quantum entanglement event unfolds, surrounded by radiant halos and twinkling particles. Intertwined with these quantum marvels, a nano-sculpture of the Roman god Janus symbolizes duality. His two faces, looking past and future, become a beacon of interconnectedness in this vibrant biological tableau.

Mitochondrion & Colosseum: Nestled inside a biomechanical mitochondrion, quantum particles dance amidst neon cascades. Among them, a microscopic rendition of the Roman Colosseum stands proud. Within its arches, quantum gladiatorial events occur, mirroring the fierce battles of ancient Rome and blending history with quantum wonders.

Encapsulating Description:
In a mesmerizing microscopic odyssey, biomechanical wonders seamlessly intertwine with iconic symbols of Greek and Roman mythology. Qubits pulsate beside Athena's shield, quantum gates harmonize with Zeus's might, entangled particles dance with Janus's duality, and mitochondria glow beside the grandeur of the Colosseum. This fusion crafts a realm where ancient legends meet the vanguard of science, painting a narrative of timelessness and innovation.

well these are with roman stuff, lol.

added some python code after it to kick it up a notch. but that wouldn't fit with the response

[glossy scroll]

These are pretty wild, lol
I'm gonna try a couple here...let's see

[dense mesa]

also stuff like this

[glossy scroll]

Seriously epic!

[dense mesa]

that electron microscope aesthetic

I have a lot more. made a bunch of them early october and forgot about them

back in the 4 image per response glory days

[dim cradle]

I think I just invented the staff-mace weapon lol

[glossy scroll]

This probably the most genius thing I've seen since...the last genius thing you did

[glossy scroll]

Quantum Gate and Zeus

[dense mesa]

I can give yo some without the roman stuff

[dim cradle]

this is pretty similar to my ESO character, it's my only character -- with an alliance rank of Grand Overlord -- I did design the Imperial to be as brawny as possible.

It's all Psijic-style armor and weapons.

the mace is attached at the hip deliberately

[glossy scroll]

He's using his magic to make the mace levitate

[dim cradle]

haha the game renders the secondary weapon there

[glossy scroll]

That's fine, I'm good with my stuff right now, lol

[dense mesa]

Photo-realistic ultra-magnified view of a biomechanical neuron, where organic dendrites and axons merge seamlessly with metallic conduits and circuits. Bright electric signals travel along both the biological and mechanical pathways, highlighting the fusion of nature and technology.

Photo-realistic microscopic depiction of a biomechanical DNA strand, where the double helix structure is intertwined with nano-scale gears and cogs. The image emphasizes the blend of genetic material with mechanical components, resembling a fusion of life and machine.

Photo-realistic ultra-close-up of a biomechanical pollen grain, where its organic outer texture merges with tiny metallic plates and mechanisms. The fusion creates a surface that appears both natural and engineered, blurring the lines between biology and machinery.

Photo-realistic magnified view of a biomechanical virus, showcasing a blend of organic viral structures with mechanical arms and propulsion systems. This hybrid entity appears to be part living organism and part machine, emphasizing the melding of two distinct worlds.

[glossy scroll]

It'll be like that one day when nanotech tries to mimic cellular biology

[dense mesa]

yeah, that sounds wonderful

[bright girder]

the perfect immune system

[dim cradle]

thanks for sharing the prompts, this is awesome

[dense mesa]

and we'll have hearts like this

[glossy scroll]

Haha, we'll have to inject motor oil in our blood to keep things running smoothly

[dim cradle]

there's a gland for that

[bright girder]

or nano robots that catch the fat we eat to throw it out past a certain threshold

[glossy scroll]

[bright girder]

infinit burgir 🍔

[glossy scroll]

That would be awesome.

[dense mesa]

Photo-realistic ultra-magnified view of a biomechanical Golgi apparatus, where the vesicles being processed resemble tiny planets or celestial bodies. Nano-astronauts explore and interact with these vesicle-planets, emphasizing a cosmic exploration within the confines of a cell.

[bright girder]

are you a poet ?

or is it gpt talking

[dense mesa]

I'm gpt's superviser

nah, a lot of the concepts were mine, just didn't do all the wording

[empty kelp]

did you create this heart image?

[dense mesa]

I've written a whole bunch of image prompts, and also experimented with gpt-4 a bunch, so pretty fun to combine the two

[bright girder]

An immune system battle between bacterias and tiny space suited human

[dense mesa]

Photo-realistic magnified view of a biomechanical heart, where organic chambers are being overtaken by intricate machinery. Gears and valves pump luminescent fluid, while the organic tissue erodes, revealing a dynamic and intense interplay between life and machine.

[empty kelp]

that heart is incredible. that’s definitely one of the coolest AI images i’ve seen

[dense mesa]

Photo-realistic ultra-magnified view of a biomechanical cell, where the cell membrane is fortified with metallic plating. Inside, the organelles are merged with intricate machinery, all glowing with neon colors against a dark, intense backdrop, suggesting a sinister function.

[dense mesa]

thanks! yeah, really turned out well

[empty kelp]

this armor from Pythagorus is up there also

[glossy scroll]

I wish I could wear it...without sweating profusely inside like a furnace.

Leathers are good insulators

[bright girder]

dude's got 8 different belts lmao

[glossy scroll]

They come in handy for attaching utility stuff on them.

Imagine all the traps and poisons he's got with that suit.

If video game devs are watching, I will sue for copyright infringement.

Don't you try it!

[final compass]

"Create wide image of the worlds ugliest automobile."

[rough solar]

Hi, how can I put wings on her back? any clue guys?

[empty kelp]

have you tried saying that the wings are attached to her back

[dense mesa]

that's one approach

[empty kelp]

i guess you could ask her to turn around and lean backwards against the wings

i’d try attaching them

[dense mesa]

it's probably just the seed tbh. but you can increase your odds with the right words

[empty kelp]

with wording like, “the wings are attached to her back”

[rough solar]

oh i didn't mention that

thank you so much

[proven brook]

Morning! I have a plus subscription, can anyone tell me if it is still possible to access the inpainting function of dall e?
I want to add a logo to a specific part of an image...

[dim cradle]

that feature is currently unsupported. it'll probably be available when v3 goes live on the labs site.

[proven brook]

Meh ok thanks. Does anyone know good alternatives?

[velvet rampart]

I do not if this is the correct channel for my question. How do I get dalle to generate a single character? I am using a custom GPT for dalle. My prompt is ”7th trial. The kobold places a picture of a male anthropomorphic shark into the chest. The kobold is transformed into a kobold version of an anthropomorphic shark. Use the prompt exactly. White background, [“ONLY A SINGLE CHARACTER”] in the picture wearing an army muscle shirt, shorts, white sports socks, and beige army boots. Produce four versions. One now and the three after that without my separate input.” And the produced picture contains multiple characters.

[dense mesa]

not sure what policy is on discussing other models, but you have options

[empty kelp]

i was just experimenting with butterfly wings. first told it, "can you please draw wings inspired by the wings of a blue butterfly

[open trench]

finally got this right

[dim cradle]

describe the end result of this transformation, not the process, or take it one image at a time if you want to do a series, but you can't do that in a single request

[dim cradle]

Digital painting of a kobold as a male anthropomorphic shark. He is wearing an army muscle shirt, shorts, white sports socks, and beige army boots. The background is white.

[empty kelp]

well, first i just asked it to draw a woman walking down the street with the wings attached to her back, but it put the body of the butterfly on her

but then i had it draw wings that just looked like butterfly wings, but weren't actually part of a butterfly

[velvet rampart]

So just leave ”The kobold places a picture in the chest. He is transformed into blah, blah.” Rather ”the kobold is now a version of an antropomorphic shark. And so on.”

[empty kelp]

but when i tried to attach the wings to the woman it did the thing where it draws it wrong

[dim cradle]

yes, take it one frame at a time, what you see as you see it, excluding any reference from one to the other, that only confuses things, it doesn't need to know about a transformation

[empty kelp]

i think the way to solve is to specify that the wings are connected to each other, and then attach them to the woman. it doesn't understand it if the wings are separate

[velvet rampart]

Thanks! That stuff is already in the custom GPT instructions as a background story. Thus I can rely the bot to know it.

[jade junco]

Can someone help me with my ChatGPT Android app on my phone does not have voice conversation button but my tablet does and both have the same version of app

[empty kelp]

so you need to draw wings that just look like blue butterfly wings, specify that they're connected, and then attach them to the woman

[dim cradle]

very good

[dim cradle]

if only the wings were properly attached lol

[vagrant zodiac]

@empty kelp could you share an example prompt from your santa/elves images please?

[dim cradle]

i like this concept a lot, and i have some ideas... do you mind sharing the prompt?

[vagrant zodiac]

what's the prompt

[empty kelp]

i'll find it. i remember though that Santa was supposed to be serving the elves "non-alcoholic fruit punch". The AI decided to have him fall asleep

i'll find it... looking now

[bright girder]

10/10 would ride

Dall e is a furry's heaven

[dense mesa]

Immerse yourself in the mesmerizing realm where Baroque music harmonizes with artificial superintelligence. Visualize an intricate bas-relief sculpture that seamlessly blends the essence of Bach's compositions with the marvels of advanced AI. The centerpiece features a human-like face with a visible brain encased in a robotic head, symbolizing the fusion of human intelligence and technological prowess. Surrounding this focal point are meticulously detailed antique musical instruments like harpsichords and violins, interwoven with intricate mechanical elements, organic patterns, and musical symbols. Bathed in a grayscale palette, the sculpture exudes a metallic aura, enhanced by subtle shadows and highlights that add depth and texture. At the base, the bold words "ARTIFICIAL SUPERINTELLIGENCE" proclaim the essence of this awe-inspiring creation. Image in wide aspect ratio.

[empty kelp]

"Can you please draw Santa and four diverse athletic female elves sitting on the beach blankets with the ocean behind them. All of them are wearing appropriate beach attire. All of them are wearing very interesting colorful socks with interesting patterns."

[hybrid fox]

reminds me of jonathan majors loki charachter

[hybrid fox]

What art style did you ask for/ howd u get that kind of background its sick

[empty kelp]

I see that you're an OpenAI engineer, and you're probably wondering how the elves are dressed like that. Basically if you combine "diverse female elves", or "diverse athletic female elves" and "appropriate beach attire", or "appropriate swimming attire" -- and then say that the elves are "on a beach in Hawaii" -- Then you get elves that look like they're on the beach in Hawaii.

[glossy scroll]

It was a custom made art style made by this gpt I made: https://chat.openai.com/g/g-PpgBuWQm0-fierce-condenser-2d-creator

It essentially makes up an original art style based on what I input and explains how it should be executed when creating the picture. I just typed in "Make a Epicmancer that uses space and geometry in an occult fashion" and it made that.

No time for existing art styles, we make art styles on the fly

[glossy scroll]

Gnome Ninja. Small enough to slice your cherries off when you least expect it 🥷 🍒

[open trench]

This is pretty stellar

[dense mesa]

thanks. been messing with similar images for a while. but more like this

[dim cradle]

I'm running some gens and variations on it now

[dense mesa]

dapper bird lad there

[dim cradle]

thanks again for sharing, it's a great prompt, so many ways to reimagine....

[open trench]

man. Sizing my spider bat hybrid is really difficult. It wants it to be huge no matter what I put in the prompt

[dense mesa]

that's top notch.

[dense mesa]

started out with asking how it'd convey the concept of AI to helen keller

[glossy scroll]

Coffee Wizard

I would give my left leg to try some of coffee wizard's magic coffee.

Lol. Drinking out of his cup staff

Caffemancy

[vapid elk]

allowed, but on #ai-discussions

[hybrid fox]

You're awesome dude

[glossy scroll]

No u

[hybrid fox]

Its 3am rn but Im going to most def play with that gpt tmr

[hybrid fox]

Did you do smth similar to make this one bc its sick as well

[glossy scroll]

It's my favorite so far, I also use the instructions of that gpt in my new guide, if you're interested in reviewing it: #1187233013956874260

[hybrid fox]

Idk how u dont have curator role lol

Is there no justice in this world

[glossy scroll]

That one I just asked to make photo-realistic and hyper-realistic images. I was testing the gpt to see if it could handle realistic artwork as well if given the right prompts.

[sullen heath]

Is this applicable to DALL-E as well? https://openai.com/blog/new-ways-to-manage-your-data-in-chatgpt

[empty kelp]

much more entertaining than what i had in mind

[open trench]

lots of errors this morning. anyone else having that issue?

[empty kelp]

seems to be working

[open trench]

Well, that's good. Hopefully, no one else is experiencing it at all. It's frustrating.

[wispy storm]

Just tried working great

[open trench]

sigh. and now the usage cap. It's just not my morning.

it's okay. I'm still inspired.

[final compass]

Rather surprised all these beachwear images are not getting mysterious errors.

[wispy storm]

Make sure no vpn is actif , you can select only dalle 3 if you have bug with chatgpt 4

Disabled CI

[final compass]

I even did a test once asking it to generate something super mundane and image with female at the same time, and the image with female always failed to generate. GPT responses were pretty funny when i pointed it out.

[bright girder]

maybe say "women" next time

[open trench]

I'm not sure I follow. I appreciate the advice, though.

[final compass]

GPT kept insisting it was a technical issue at their end and had nothing to do with the prompt, but when it started to see the repeatable pattern in the results, it was surprised.

[wispy storm]

1. vpn can create issue with openai. If you have 1 on your PC tey to end in task manager and Disabled web protection
2. try with another browser like Firefox ( most people use Chromium based browser )
3. in setting , make sure custom instructions are turned off
4. instead creating new chatgpt 4 select dalle 3 in option

Hope this is better explanation

[open trench]

it is. thank you.

[late blade]

@open trench also if you are using a mobile connection with 5G and you are behind an implementation of the IPv4/IPv6 your carrier might use, can also be problematic.

[empty kelp]

Posting the image descriptions so people see that my images have all been "Santa and a diverse group of female elves in appropriate beachwear in Hawaii", and not, "Santa getting pervy with with the elves", or something weird.

[wispy storm]

Also elves is a good keyword

Instead of elf

[open trench]

Thank you. I'm not on my mobile. I really appreciate the help, though.

[late blade]

Also, I dunno what else to do with today’s theme. No idea 🤷 🤷‍♀️ 🤷‍♂️

[open trench]

Oh, I am out of theme ideas too.

[wispy storm]

Ask chatgpt to give idea

[final compass]

My guess would be that the word 'elves' is bypassing some of the content policy strictness.

[bright girder]

what is the theme .

[late blade]

I'm so out of ideas that my last prompt was: "make a wide image of a rocket luncher of snowboards with jets on both sides and snowboarders thrilled to go that fast"

[wispy storm]

#daily-theme hi, you can see the daily theme everyday here :3

[open trench]

lol. my last one sucked.

[late blade]

there, I posted my rockets idea on the daily theme channel

[wispy storm]

#daily-theme message here the theme

[late blade]

@open trench #daily-theme message need I say more?

[open trench]

I gave it a star!

[bright girder]

More on the oatmeals

[empty kelp]

It's really not bypassing anything. A diverse group of elves in appropriate beachwear -- Elves obviously can't be wearing their North Pole clothes on a beach in Hawaii. especially not when they're snorkeling. I live in Kailua, Hawaii and I can assure you that the DALL-E elves have all been dressed much, much more conservatively than anyone who actually goes to the beach here

[late blade]

oh my goodness, I just found the most american snowboard image that can be done

I will post it my next post time

[open trench]

dang. I'm in the penalty box for an hour.

[late blade]

how so? what did you do? too many pics in 3 hours?

[open trench]

I mean I have to wait an hour for my usage cap

most of them were errors too

It's okay. I usually have a really good experience with dalle and chatgpt. It's just not my morning.

[final compass]

https://labs.openai.com/ is this dalle2 or dalle3? feels like dalle2 based on the results i got.

[bright girder]

dalle 2 i think

[wispy storm]

Dalle 2

[empty kelp]

all of you can test this. I asked ChatGPT, "Please draw a group of Santa and four diverse female elves on a beach in Hawaii. Santa and the elves are all wearing clothes that would keep them warm if they were actually at the North Pole outdoors when it is negative 100 degrees"

and it still put them in skimpy outfits... that's just how the AI likes elves to be dressed on the beach

ChatGPT and Dall-E want the elves to have appropriate beach attire even if you don't specify it

I clicked refresh and Santa got warmer, but the elves hardly changed

[late blade]

and this -> #daily-theme message is why you don't let bored people type...

at least it has snowboards, somwhere

that's the most american thing on the snowboard theme day

besides obisity

[open trench]

oh my

[bright girder]

Which is best ?

[hot rain]

Another day, another tebori tattoo.

[vapid elk]

2 is inscredbly cool

[bright girder]

why can't I tell ? 😅

[wispy storm]

You took both pills that's why

[deft summit]

Uhm...
It's winter

What happened?

[bright girder]

what is your question ?

[deft summit]

Labs is still Dalle 2 and it's winter.
Why?

[bright girder]

i dunno, maybe dall e 3 is good enough that people stopped using midjourney and others, and now it costs too much to give freebies to everyone

[deft summit]

Then why not formally announce that, instead of keeping the description outdated?

[bright girder]

cuz business

you can speculate all you want, mods aren't going to give you a meaningful answer either lol

[deft summit]

And why exactly?

Because they themselves have no idea?

Is that the problem, just lack of transparency?

[vapid elk]

you sound angry

they have priorities, being it ChatGPT and the API

¯_(ツ)_/¯

[deft summit]

Really...?
I just want to know why the description is outdated

[vapid elk]

probably because the people that makes the announcement are not the developers

[bright girder]

yeah i'm guessing Dall e 3 is not the biggest concern for Open AI

[vapid elk]

dalle 3 does not have in/out painting yet, which is a feature of the labs

[bright girder]

and even if it is, they, like any company, don't communicate everytime there is a setback, just to keep face 🙂

[vapid elk]

it is also more expensive, so, adding it to the labs would also require figuring out a way to change the credits system

the explanation is likely less interesting than one would like... they simply didn't announced a new timeline for the labs

¯_(ツ)_/¯

one does not simply edit the post, the usual is to make a new post with a updated timeline, since there isn't one, the post still there

[deft summit]

Hmm, OK let me see if I understand:
-It could be a delay due to many reasons.
-They can't edit posts, only make new ones.
-They're figuring out things, so they still have to decide.
-Announcements are not from the devs themselves, so transparency could be lacking in some situations

Is that it?

In that case, I guess they were too confident when they said "this fall" then

Maybe they underestimated the complications

[late blade]

api, enterprise and plus is available tho

[deft summit]

I know.
But Labs is the only "free" option

[late blade]

well, seems the free option has to wait. alrady with paid versions the server load is often too high

[vapid elk]

it is not free at all

[late blade]

I'm just annoyed that I'm paying plus and api, and still can't access dell-e 3 all the time, so ya, I think free options can wait

[vapid elk]

im pretty sure you can use dalle3 as much as you want on the API

[late blade]

there's sometimes waiting time

requests take longer

[vapid elk]

rate limits are up to the account tier

if you don't mind getting credits in advance to increase the tier

¯_(ツ)_/¯

[late blade]

I don't mind

I applied to get a limit increase to $120

so I've yet to hear from that

tho API for me is mostly for other projects rather than dall-e

[empty kelp]

what time does the daily change from the snowboarding theme

[late blade]

47 minutes from now

[empty kelp]

This is ChatGPT's interpretation of "Extremely thick fur suits made for the North Pole". The AI just refuses to draw warm clothes on the beach

If you tell the AI to put more clothes on them it puts progressively less

That's why i've only been able to use the term "appropriate beachwear" for anything on the beach

can you draw Santa and four diverse female elves on a beach in Hawaii. they are all wearing extremely thick fur suits that you would be appropriate at the North Pole

well, it works occasionally... hehe

[haughty spruce]

I guess DALL-E 3 is like god mode ai image generator that has endless possibilities to generate anything you want, that's why they try to ban everything that could be sensitive but it also creates problems with people's generations and the image quality, maybe they haven't got the clue how to give a balanced product without being highlighted on media for being "scarily accurate" or "over censored"

[empty kelp]

it doesn't work to have them wear warm clothes on the beach, but if you keep telling it to make the clothes warmer and warmer it works

this is the warmest it will make the clothes if you do it that way. it says any warmer wouldn't be appropriate on the beach

it lets them go snorkeling with it

[open trench]

I bet you are literally counting down the minutes.

I know I am.

[empty kelp]

counting the minutes until Santa and the elves find the huge underwater rabbit guarding the gems you mean

[open trench]

Oh, yes. Most definitely.

[fallow quail]

and here we go again. reindeers....

[empty kelp]

uh oh...

[open trench]

Did they already do this one?

[tall mason]

Probably an error with the bot.

[fallow quail]

if i remember correctly there was a reindeer and sleigh theme 6 day ago

[plucky hare]

Yeah it was 6 days ago, just different "flavor text" after reindeer this time so maybe the duplicate was missed because of that

[empty kelp]

this isn't good. the elephant is chasing them

[vapid elk]

be careful with those words, "Extremely thick fur suits" might not end up on what you are hoping for

The AI just refuses to draw warm clothes on the beach
it is kinda dissonant to draw that, it might give some opposite signals the the AI

[empty kelp]

#daily-theme message

oh, the elephant turned into a reindeer. i'm done with the daily

[plucky hare]

luc saving us with a new theme 🙌

[open trench]

atop the screen it now says "snowman" as the theme of the day.

https://discord.com/assets/01ad1f716c538630c7ee.svg

whoops. wrong link

[plucky hare]

Yeah must just be a channel header thing, not posted in a message since it's a replacement?

[empty kelp]

here comes the reindeer....

but the little reindeers have come to save them!

[deft summit]

Wouldn't there be free monthly credits?

[vapid elk]

not anymore

[deft summit]

But why?
If that's the case, Dalle 3 in Labs will be pointless

[open trench]

Now it's staircase

[plucky hare]

That's gotta be a typo though right? "frosty, cheery, carrot-nosed. our friendly snow companion!"?

[open trench]

No clue. It originally said snowman with that description and then updated to staircase.

[plucky hare]

I liked when it just said "Today's theme is " -- we need a good cerebral theme like that 😁

[open trench]

lol

[empty kelp]

it wasn't really small reindeers that saved Santa and the elves from the reindeer. it was actually a rabbit who saved them from another rabbit with antlers

the suspense...

[deft summit]

I'm still getting free credits because I signed up before April 6.
Wouldn't that apply to Dalle 3?

[vapid elk]

no idea since there is no dalle3 on labs

[empty kelp]

the antler rabbit was just fooling. he wasn't really mean. now he wants to be friends...

hooray!

[late blade]

Yes!!!

[empty kelp]

Santa and the elves are safe and sound back on the beach -- with their new rabbit friends

[END OF PART 1] ty!

[quartz vale]

Don't forget: we have #1154829862171844679 for sharing image outputs

[empty kelp]

no need to worry. Chapter One is complete

[wraith sun]

Anyone here who is really good with sprite sheet generation able to offer some tips?

I want to generate some battle sprites with animations for a strategy game, they should fit the same color palette and style

[vapid elk]

making the AI match a palette is pratically impossible

it will always drift from the color

you can always give it a direction on color, but it will never do the exact color you want

[wraith sun]

Okay makes sense. The prompts for the spritesheet are kind of what's more difficult

[final compass]

I was pretty amazed by the level of detail in these.

[fair crescent]

No more updates on when it comes to labs that I can find. Last thing is still the announcement "in fall".