images-discussions

[turbid rampart]

Urm

I'm not subscribed

But I'm happy cuz it works

[vapid elk]

[turbid rampart]

I was so happy when it actually worked

[dim cradle]

3.5 tells me it cant make images

strange

[turbid rampart]

I get this

It's super cool, it uses python script to answer maths and shows the code

[dim cradle]

im on my laptop now, maybe on mobile it work for some reason 🤷

[turbid rampart]

I think it's in certain countries

@dim cradle what version is your chat gpt?

[dim cradle]

i use copilot pro

[turbid rampart]

Ah

Bing?

[dim cradle]

yes

[turbid rampart]

Alr

Check your gbt tho

What's the version?

[dim cradle]

gpt 4 turbo

[turbid rampart]

The app

On mobile

[dim cradle]

i have turbo on both

[turbid rampart]

This is how

[turbid rampart]

The app version

In the app store

[late blade]

does dall-e with copilot pro let you upload files to the chat?

[glossy scroll]

Gnome secretary: "You've reached the voicemail of Pi. He's not here right now. Please leave your name, image, and a funny joke after the beep."

Beeep!!

[open trench]

morning all

[chilly onyx]

Morning

[late blade]

ok, everyone stay calm, Nezho has something very important to say!!!

[chilly onyx]

Waiting with anticipation!

[open trench]

I spilled milk on my keyboard the other day, and I don't think it has recovered.

[late blade]

And there you have it!

[open trench]

lol

[late blade]

have a nice day everyone

I posted something you might like in the city invasion gallery

[open trench]

I will look!

Yay. The citizens have some hope!

[late blade]

or so they think

[open trench]

omg. my keyboard is a mess. Please expect more typos than usual. lol

[late blade]

I know the feeling, once I spilled some coke below my keyboard at home, getting that sticky thing cleaned was a pain

worse of all, I don't drink coke...

[glossy scroll]

Das very goob

[late blade]

how you doin' gnome?

you've been missing

[glossy scroll]

Doin pretty well
Yeah, been pretty busy with life stuff.

[thick smelt]

hi pytha, lei here!

[late blade]

good, then your priorities are in the right place

[glossy scroll]

Hi Lud...I mean Lei!

[chilly onyx]

Is it a nice keyboard?

[open trench]

it was, but it's old anyway. Time to get a new one.

[glossy scroll]

Yeah, gnome stuff. Fixing things, dusting things, doing gardenings, ummm....what else...sowing...a bit.

[glossy scroll]

What about you Dys? What have you been up to lately?

[open trench]

lol

[thick smelt]

btw i made a dall-e dogday and catnap -> #images-discussions message

[glossy scroll]

Hehe. Very good. Luna and Kuroberos?

[thick smelt]

no they're the smiling critters from the third chapter of poppy playtime

two of them at least (DogDay and Catnap)

[glossy scroll]

Ohhh. Haha. I just looked it up. Kind of scary.

The blue thing with teeth is what I imagine is under my bed.

[thick smelt]

the blue thing is Huggy Wuggy... i think?

[glossy scroll]

Lol. Reverse psychology name.

For a thing that looks like he's gonna haunt you for the rest of your nights.

Good thing all gnomes keep a shotgun under their pillow every night.

[thick smelt]

here's the comparison between the dall-e and og version
ctto: Mob Entertainment (og)/ me (dall-e version)

[glossy scroll]

Awww, haha, das cute

[thick smelt]

thanks pytha

[grizzled loom]

Part of the "Where's the beef?" campaign the meat industry pushed. "Beef, it's what's for dinner." was the catch line

[open trench]

making me hungry

[vapid elk]

blursed

image generated by ChatBBQ

[dim cradle]

yes

[dim cradle]

lol awesome. one of the best really

[late blade]

looks like I'm not very creative today with the topic at hand

all my ideas for the topic were robbed by serenity

gimme a bigger drum, and a huge stick to hit it too

[tawny portal]

I suggest we vote again on this one: #daily-theme message -- it fits the theme again

[late blade]

chipsticks....

[late blade]

this one is amazing

[tawny portal]

Agreed.

[late blade]

hope nobody got timed out, after 15 emojis you get timed out

[tawny portal]

Imho art does not need to be beautiful. Just honest and raw. Beauty is thing that surfaces sometimes - if it does it happens by itself. 🤷

And even then it is mostly in the eye of the viewer.

[open trench]

I may be timed out creatively.

[late blade]

oh no

[late blade]

yeah, but serenity came, and was mean to me, then the damn cat came back and wanted food

[tawny portal]

Cat love is rough 😄

[late blade]

indeed

[grizzled loom]

Thats layers of disturbing. lol

[late blade]

reminds me of this I did back then #daily-theme message

[pastel siren]

I should've probably put that in spoiler 🤔

[tall mason]

I suppose I should've expected I was asking way too much here.

[pastel siren]

Beef-Man in spoilers this time...

[boreal gate]

some of my best stuff, if you guys want to see more of this stuff i post them on instagram bardofverses. I'm using a private customGPT (that i am always fine-tuning) using story-based image generation.

[elfin charm]

hello need help

[feral minnow]

Is it OK to discuss and share images from Bing’s DALL-E 3 in here?

[late blade]

dall-e, yes

ms paint doodles from your windows 95 pc, no!

[tawny portal]

Hypsibius dujardini and Milnesium tardigradum performing in a dimly lit 20um jazz bar; Cyanobacteria, Dinoflagellates and Volvox colonies clearly visible in the background; to scale; accurate microorganism photography as depicted in medical textbooks; AR 2:1

These cuddly things are famous water bear micro-organisms (tardigrads).

[late blade]

I wouldn't want to meet one bigger than me tho

[tawny portal]

They are about 0.1 to 1mm in size.

[late blade]

ya I know, but still, imagine one larger than you

[tawny portal]

(by GPT - but very accurate)

[late blade]

yeah, I've played with them befor, dall-e is surprisingly good at those

[tawny portal]

Lots of photos. Biologists love them 🙂

[late blade]

yeap

I don't like biologists, dated one, we parted ways a long time ago

generalization, but I got traumatized

[tawny portal]

Sorry to hear that.

[late blade]

I'm not fond of creepy animals

[tawny portal]

Not sure what qualifies as creepy. Humans are weird. We percieve only small sliver of reality. Our scale (to which we are accustomed to) is just fraction of what is out there. It is all nature 🤷

[late blade]

ya humans are weird, and we are humans

[hollow pasture]

Sorry to disturb but I want to now where to have and how to have dall•e 3

[late blade]

you have a few options

DALL-E 2, from labs.openai.com, DALL-E 3 from OAI ChatGPT+, API, Copilot, Copilot Pro

[hollow pasture]

Ok cool

Thanks

[tawny portal]

To a tiny water bear you appear as cosmic being of incomprihensible complexity. To it you are a gestalt emergence that appears from parts that it is familiar with but you as yourself occur at a scale that is beyond a tiny water bear grasp. It can see the cells and different tissues.. but you are not it.. you are sum of the parts and more.

[late blade]

Ya I don't disagree with that

I just say we judge our fellow humans as weird all the time as a paradox phenomena when relating or explaining other beings

wouldn't be the first person to say humans are weird, nor the last, it's just funny what that implies

@glossy scroll no slacking on the pool, back to work!!!

[glossy scroll]

Lol. I'm just trying to relax for a bit 😎 🏖️ 🌞

[tawny portal]

🤷 Part of weirdness comes from the entanglement with the scale (and routine) at which we exist. It is funny how much all of our senses lie - they are all just barely enough to support us in our human-scale problems. And for most of our life we get entangled with silly human things that don't matter much at different scope. 🙂

[late blade]

Oh ya, agreed

I'm just amused by the statement: "humans are weird" and we being humans, that's all

nothing deep there

well most of us, I'm sure 80% of the users are custom GPTs

[tawny portal]

👋 to 🤖 out there 🙂

[late blade]

lazy prompting can do wonders

[tawny portal]

Not sure if prompting will be a thing long term. It is already mostly automated and done by GPT4.

I think we'll just be stringing concepts and exploring what is in those neural nets conceptually.

I haven't tried Bing. Do we know for sure what exactly goes to image model - that there is no intermediary?

Once image models get large enough we won't even need llm in the middle.. they'll conceptualize everything on their own (and will likely work in multimodal fashion anyway)

MSFT has probably finetuned its model slightly. They probably have access to some licensed image datasets they deem good. It would result in slight style change.

These things are ridiculously powerful. I do not even know if it is worthy for artists to try to pull out their art. It is like erasing oneself from this evolutionary step. If it is in it gets to live and evolve.

[late blade]

I just see GPT or AI at the current level as the next step after auto-complete

[tawny portal]

Not sure I understand, in what sense autocomplete?

[late blade]

Autocomplete, or word completion, is a feature in which an application predicts the rest of a word a user is typing

[tawny portal]

You don't see models as more than advanced autocomplete?

[late blade]

the next step after autocomplete

at the current level

[tawny portal]

Yeah, there is more to it. It goes quite deep. With current tech even. So.. the task that the models are solving is called language modeling (predicting next token, aka autocomplete). It turned out to be a convenient training tasks for neural networks that is easy to parallelize. However many other tasks were tried.. like sentence matching, Q&A, filling in the missing blanks etc. Next word prediction was simply good enough so that network starts to converge. Language modeling task comes from linguistics.. it was thought that it would capture the statistical intricities of the syntactic forms. But neural nets are digging way deeper than that. They are comfortably juggling semantics. As you have seen for sure.

Between input and last layer which produces a probability distribution over token dictionary lies a fairly large neural network. Note that "large" in this context is like computationally difficult.. large from CS standpoint.. from biological standpoint GPT4 is somewhere between parrot and a dog.

[late blade]

it's still a game of input output, that's what I meant

[tawny portal]

Yeah.. it is even discrete and deterministic (everything up to that last layer is completely deterministic). Intelligence that we can sort ofd observe arises through series of API calls.. but still it does what it does and works. Neural network is coherent between the calls and latent space vectors correctly capture and juggle concepts that we send over.

[late blade]

yes, I don't deny that, like I said next step of autocomplete, first text, now new concepts

I should have said NLP was the next step after autocomplete, but I sitll see gpt as nothing more than that, it's still a baby

[tawny portal]

But look.. human life.. when we look how it unfolds from current point in time back through its lifetime is not even about input-output... it is just a trace of events, decisions in past that from current point in time had one way to unfold and to lead to this exact point from which we are observing them.

Visual transformers do autocomplete as well. Some of them work fairly well. As architecture they could outperform all we have probably if we could train large enough ones cheaply enough.

[late blade]

ya

all wonderful examples

[tawny portal]

Imho there isn't anything fundamental that GPT is missing.. it just needs to be larger -- more coherent. There is a chance that by itself (at sufficiently large scale) it could develop all we as humans have.

Ofcourse humans as we are will just turn it into a useful golem. But.. yeah..

[late blade]

also don't get me wrong, those are tough advancements to do and quite complex, but just the average user for a GPT will rely on it for input/output and that's it

[tawny portal]

Lots of things can be abstracted into black boxes with input output. It is very general & powerful way to view things (if one were to have it work in general way).

[late blade]

I'm just using the PoV of the average Joe/Josephine that just dropped their iPhone to the floor from the table beside the bed... (just happened to me).

[fickle ravine]

Also how thin is that ship

[late blade]

I like the idea of all the breakthroughs behind technology and science, but I'm trained to see the end user most of the time

[tawny portal]

Gotcha. Humane is trying to change it with AI pin. Something that would watch over our lives and have at any point prepared context for AI if we were to need its help.

Btw have you seen that guy that has spent 24hrs in Apple Vision Pro? That was wild.

He was like.. going to metro, to shopping, out with friends.

[late blade]

I want but I don't want an apple vision toy

[tawny portal]

Hah, me too. I am on the fence. Likely will not get it. But check it out on YT. It was published recently (today maybe?)

[late blade]

my 💳 has been very bad with me in that regard

[tawny portal]

Ah, hah. Yeah - then just skip it. It is still an expensive toy. Lot of the things they want for it are not in place (like store). It will become something in few gens. Price will go down too.

[late blade]

But my 💳 is toying with me, it's telling me in my dreams: buy it, buy it buy it, you know you want it, buy it, buy it....

[tawny portal]

Canoopsy is name of the tuber. Video is from 4hrs ago I think. Changed my perception of VP.

[late blade]

I will check it out after I check all the stuff @red umbra sent me

[tawny portal]

Gotta scoot, have an amazing day filled with inspiration D 👋

[late blade]

I think I can get hands on once I get back home

damn it, now I got to search my phone, fell to the back of the night stand

hotel stands are horrible

[cinder rover]

hm

[dim cradle]

i wonder wht image gen in dalle in apple vision would be like maybe you can really immerse in our dalle images haha

[formal osprey]

I like that

[boreal gate]

Thank you, I'll be sure to keep sharing

[vapid elk]

what AR 2:1 means?

[dim cradle]

how do the filters always get bypassed seconds after released

[boreal gate]

Yea, it surprises me all the time. I instruct Bard of Verses how I wanted the images generated sequential and it applies it's own style to it based on the story being told, theme, etc.

[dim cradle]

why do you add a watermark?

[dim cradle]

Oh I put it on my instagram and deviantart. Watermark is just in case other people repost it, which I don't mind.

fortunately i haven't seen anyone else do that

kinda weird

AI art is so good now

[dim cradle]

Is it possible to make a vector AI art generator

infinite image 1!1!!11!!1!1!1

Higher resolution would be great

[plucky wedge]

And now it's serving lossy, lower quality compressed webp files

[dim cradle]

you can't copyright AI art anyway, so it serves no purpose except to detract from the image.

Order of generation starts from the top left, but yes, it got really grainy by the last one

Doesn't make them any less epic, but still, imagine if you could zoom in and see the fine details

Oh and the top right one is my favorite because it kind of pops out of the frame

I made my ChatGPT generate 4 images at once and tell me the seeds

[dim cradle]

I just got it to make 6 images

[plucky wedge]

Yeah, I mean, the higher the resolution the more processing power it takes. 1024x1024 is barely HD and about a million individual pixels all of which get calculated individually. 4k is about EIGHT million pixels.

[plucky wedge]

wow that is an obnoxious watermark lol. is that a joke?

[dim cradle]

they're all obnoxious, i think my point is made.

🤷‍♂️

[plucky wedge]

lol, I see. it is a joke

[dim cradle]

At least mine is like some giant transparent watermark going across the entire image. Just a small tag in the corner 🙂

[plucky wedge]

I don't even bother to watermark them. I don't own them. I'd sooner copywrite the prompt itself.

[dim cradle]

🔥

[dim cradle]

you can't copyright AI art. it's a bad and misleading trend that hopefully people will not emulate, the model deserves the credit, but fortunately there are no visible watermarks. i would only do it if i were trying to give the impression i made it myself.

[vapid elk]

well, you sure can, but previous court cases on AI art and coppyright have determined tht AI generated art can't hold coppyright, so, you would need a very different case to get that

coppyright is very subjective on the "nature f the content" and on "how much" of the content is AI generated, so, an art work with a small AI patch, probably can be coppyrighted

[dim cradle]

yes, we could all start adding copyright watermarks to all the art we post here and in the daily theme. i just hope people won't start doing that.

one day we'll be able to generate steam boat willie

[whole garden]

When I give an image prompt, how come the AI is able to describe the image exactly as I want, but it does not generate the image according to its description or the prompt?

Okay, how do I check that?

[dim cradle]

"I can't send or revise payloads to the API after they've been generated or discussed." Oh, that's helpful, thanks a lot.

[whole garden]

haha okay thanks

[dim cradle]

they're making it harder to use for users who know how to use it

that's really great

probably best not to share discoveries here, it may not serve our interests.

[feral minnow]

Do you find GPTs useful for generating DALL-E 3 art?

[dim cradle]

I prefer using GPT instead of directly using DALL-E

[feral minnow]

I do too, but mostly because I like to ask it stuff about what I’m making.

[dim cradle]

yes, but won't go into details. i pulled mine out of the gpt store--lack of security, plus user intent is another concern, i have no idea what images they're requesting. it's for my personal use now, and i plan to guard my work. since the gpt store was announced, users have gotten more secretive and competitive, perhaps out of a necessity.

[feral minnow]

I remember reading that it was trivial to get it to reveal the underlying instructions and code. 😭

[onyx ridge]

The documentation is the source code.

[dim cradle]

there's even a post on the openai dev forum telling you exactly how to do it.

[feral minnow]

mark-cuban-writing-it-down.gif

[dim cradle]

and i know for a fact it's being done. that's why i pulled mine. i don't mind sharing knowledge and discoveries in a collaboration, i do mind having my work stolen without so much as a thanks.

[tawny portal]

I don't know if anything can be stolen from an artist. Imitated perhaps.. but that is just a sincere form of flattery. Not necessarily something negative. Its what all artist do in one form or another I think. 🤷

[tall mason]

When you factor in IP, yes, you can steal. People make money off of IP's, and imitators can can detract from the original. Of course, it's more nuanced than that, but in terms of directly generating copyrighted/trademarked materials with no consideration to fair use, that's stealing.

[tawny portal]

Yeah, there is that too. And you are right. 🙂

[plucky wedge]

bro you didn't paint a single stroke. these AI's are built on the work of human artists. don't flatter yourself that a clever sentence is the same as creating a work. you are literally amalgamating the work of real human artists. I'm not saying you're adding NOTHING but eventually hip hop artists sampling james brown had to pay his estate

[tall mason]

GPTs are for more than image generation. That aside, people put as much effort into these as many artists. "Clever sentences" are arguably art, like poetry and short stories. I say this as someone who has spent plenty of hours on physical art.

[tawny portal]

Model amalgamates. An what of an amalgamation it is. It absorbs artwork - true, but then it also gives so much more.

Where does rhe art that model conjures out of thin air come from? Is it from person prompting it? Hardly.. unless model manages to provide the exact depiction of what is asked with absolute precision. That almost never happens. So where does it come from?

And who is the artist there?

Does it just regurgitate the dataset? Not really - as it is juggling these abstract concepts. Imfluences are visible.. but.. what it produces is most of the time novel and unique. So where does it really come from and to whom should we attribute new art to? Does art awake on its own inside a neural net? Is it just art creating art without an agent? How can it come from thin air?

And yet it is alive and right there to stay. 🤷

[late blade]

good morning people

Is diffusion before coffee also the same diffusion gpt uses?

[tawny portal]

Yes. Well bSort of. Model chips away entropy and chaos until all that remains is the image. Noise it removes is gaussian. Conflux of diverse influences whose voices unite in white background noise. Is that the same diffusion as one before coffee?

[late blade]

coffee!!!! Now!!!!

I was checking some apple vision reviews, I will skip it for now

unless OAI does an immersive DALL-E experience just for it

[tawny portal]

gotta scoot to pharmacy

[vapid granite]

Uhh, did the Copilot Pro version of DE3 just get a major update? I'm able to run prompts that were previously c-blocked. I'm talking benign stuff, not "Baby Yoda smoking a bong with Snoop Dog." 🙂
The quality has seemingly improved a LOT.
I previously was using the ChatGPT Turbo edition, which apparently is the older/oldest version and I switched to Copilot Pro when it started last week.

[final compass]

Has anyone else paid attention to as if the character continuity had significantly improved?

[tawny portal]

Challenge: extreme closeup accidental selfie photo (it is harder than it looks).

Camera angle should be irregular. This would not answer the challenge:

Don't post your own photos 😄

[late blade]

api works pretty good, chat depends on which cat is manipulating the seed

[chilly onyx]

You’re right, this is harder than it looks 🙂

[late blade]

I think the problem is using the keyword selfie

[glossy scroll]

You gotta get the angle juuust right

[vapid elk]

this could be an emoji

[late blade]

I dunno, catching that pesky gnome can be troublesome

[late blade]

I tried to post my own "accidental" selfie and got timed out with the message: "Too good for the internet." and mods are still working on the problem

I got these

or do you need it to be closer? which could be weird tbh, triggering by accident a photo so close to your face

[thick smelt]

cool faces dys

[glossy scroll]

Do it! You're a mod, you have the power to make emojis!

[tough stream]

wow is this from a prompt?

[tough stream]

is this mj or dall-3?

[late blade]

yes

[tough stream]

could you share for me? thanks

[late blade]

The prompt was:

An image that captures an unintentional, candid moment through the lens of a mobile phone's camera, focusing closely on an individual's face. The subject appears unaware of the camera's gaze, their expression caught mid-emotion—perhaps in the midst of thought, conversation, or a fleeting reaction. The closeup framing reveals intricate details of the subject's features: the subtle textures of their skin, the depth in their eyes, and the natural, unposed set of their mouth. Light filters softly across the scene, highlighting the contours of the face and casting gentle shadows that add depth and realism. The background, though indistinct and blurred, suggests an everyday setting, emphasizing the spontaneity of the moment. This image, marked by its authenticity and the raw, unguarded glimpse it offers into the subject's persona, stands as a powerful testament to the beauty and complexity of human expression captured serendipitously.

[tough stream]

I really don't know what to say other than thank you so much!!!!!

[late blade]

We are all here to share

and learn

it's not a perfect prompt tho, it needs more work, but waiting for an answer from alset

[late blade]

Joke aside, it is in deed challenging to make that image, I keep getting either a person that is aware of the image or a phone that shouldn't be in the image.

[onyx ridge]

"...unaware of the viewer..." maybe?

[late blade]

it's already in the prompt

and it works mostly

but once I started with a request to zoom in, that's where all broke

[onyx ridge]

An extreme-zoom image capturing an unintentional, candid moment, focusing closely on an individual's face. The subject is glancing away, unaware of the viewer, their expression caught midst of a fleeting reaction. The exceptional closeup reveals intricate details of the subject's features: the subtle textures of their skin, the depth in their eyes, and the natural, unposed set of their mouth. Light filters softly across the scene, highlighting the contours of the face and casting gentle shadows that add depth and realism. The background, though indistinct and blurred, suggests an everyday setting, emphasizing the spontaneity of the moment. This image, marked by its authenticity and the raw, unguarded glimpse it offers into the subject's persona, stands as a powerful testament to the beauty and complexity of human expression captured serendipitously.

YMMV. The rest of the prompt is awesome.

[tough stream]

damn

this is badass

[late blade]

lol, that's what I also did, I got some ideas on how to address this once I get out from another penalty box like adding a partiality to the captured image and erratic angle

[onyx ridge]

A zoomed image of a person's face from below, with the individual being unaware of the viewer. The image is candid, focusing on details of the individual's face from beneath, including skin textures, flaws, and detailed, realistic hair. The image is 9:16 and hyperrealistic, photographic in style, and depicts a lifelike, everyday background which is somewhat blurred.

LOL

An extremely zoomed image of a person's partial face from below, with the individual being unaware of the viewer. The image is candid, focusing on details of the individual's face from beneath, including skin textures, flaws, and detailed, realistic hair. The image is 9:16 and hyperrealistic, photographic in style, and depicts a lifelike, everyday background which is somewhat blurred.

This is my favorite of all I've made.

It looks legit accidental

[thick smelt]

i like it, looks like a Japanese dude unironcally, tbh

[onyx ridge]

Hai.

[thick smelt]

legit happy accidents

[thick smelt]

yep my thoughts exactly

[onyx ridge]

Regen of same prompt

This is what I envisioned

it couldn't be more perfect

That nose

[strong rose]

Your submission is amazing 🤩 @red umbra

[onyx ridge]

@aiset did I get it? Did I win the challenge?!? 😄

[onyx ridge]

I can tell you mention camera 😄

[tough stream]

yeah I can tell, too

=))))))

lol

[late blade]

the first one, the person is aware of the image source

the others are really good

[late blade]

that's too many cameras in the background

[onyx ridge]

Final regen of the latest prompt:

I'm pretty happy with it. Caught mid-blink and everything.

Not too long, not too flowery, produces the desired output...

Challenge complete.

[late blade]

ok, acne, how many bumps can you make in one nose? 😁

just kidding

lol

[onyx ridge]

Just for clarification this is the final prompt for accidental selfie generation:

An extremely zoomed image of a person's partial face from below, with the individual being unaware of the viewer. The image is candid, focusing on details of the individual's face from beneath, including skin textures, flaws, and detailed, realistic hair. The image is 9:16 and hyperrealistic, photographic in style, and depicts a lifelike, everyday background which is somewhat blurred.

[boreal gate]

Same prompt theme within a week?

Is openai looking for something?

[tall mason]

Portal was already used this week?

[boreal gate]

Last week probably since this week just started

I think it was journey a gateway

Etc etc.

Eventually it all starts to sound the same. Stuck in a loop

[tall mason]

Gotcha. Yeah, sometimes they're pretty similar concepts, and a bit close together in time. There's been 400 or 500 of them at this point, though, so I can understand that it might start to get difficult to come up with different and interesting subjects on the daily.

[boreal gate]

Yea I know what you mean(for the human mind). That's why I'm assuming they're looking for something

Weirdly ChatGPT has been having issue "finishing" tasks recently. It makes some very good images, but I can't use them for the story because it's incomplete. "Network Errors"

Here are my bad batches for so far today. Check those hands

I now know the issue, my ISP is throttling my speed

[median kraken]

I have been using two different tools to generate images, one of them is DALL-E and the other is MJ, my issue with DALL-E is that it doesn't seem to be consistent, mostly with "realistic" images. It gives me images that look mostly like an illustration or drawing

[boreal gate]

Which DallE

[median kraken]

From Chat GPT version 4

[boreal gate]

Hmm give me an example

[median kraken]

Let me find the images in the history, just a sec please

[boreal gate]

Ahh I see the issue you're talking to DallE directly.

Instead try using ChatGPT to generate the image directly. GPT understands human language better than DallE does, so it'll sort of "translate" your vision description into something DALLE understands better

[median kraken]

Barajeamela más despacio

So, instead of using the DALLE plugin I should just feed the prompt to the ChatGPT chat?

[dim cradle]

this is a lovely philosophy, unfortunately the reality is the world is cutt-throat.

people with normal skin and without makeup -- what a novel, refreshing concept

[feral minnow]

Can this do different image formats?

[median kraken]

Have you tried to register with the Library of Congress succesfully?

I'm sorry, I should have clarified, I mean, something from AI?

And I mean, the images

[dim cradle]

no, i've only been following some of the cases superficially

[late blade]

interesting, new topic and everyone goes to the super ultra fantastical/sci-fiest thing they can find

[dim cradle]

that's my point. but it's a separate issue from protecting instructions in user's custom gpt's that they share with everyone. it can require research, and if there's no respect for that then i opt out.

[late blade]

I will try to keep myself grounded during today's theme and not go the extra mile in super complicated stuff

[late blade]

bah, starting to get too complex ideas

[fossil heart]

I've been building GPT4/DALL-E gaming experiences
Can you escape the Zombie Starport alive?

https://chat.openai.com/g/g-ArZL0FM0r-zombie-starport

Explore the House of Arcism, or attend the ARCommander Academy to prepare for a future role of Cosmic Leadership or Cosmic Exploration. ARCommander has 10+ built in games/major features built into it, and is growing in popularity.
https://chat.openai.com/g/g-Hkt3pwQAu-arcommander

[chilly onyx]

I’m telling you, @tawny portal is a fan of Austin Kleon 😉

He’ll notice this in a minute and confirm

[late blade]

he already noticed

[chilly onyx]

😅

[late blade]

8.64 seconds

[chilly onyx]

Hey Dys

[late blade]

going to bed soon, just got one image to upload

[tawny portal]

Thanks for trying the challange. Nice pics. Goal is to get picture with irregular camera angle. Sort of thing average Jo/Josephine shot everyday, but professional photographers never do. It seems like these normal 'failed' photographs are severely underrepresented in the dataset, however I am convinced that model can conceptualize it with proper guidance.

It is not easy at all. From pics in channel maybe 3 in total would answer the challenge.

Word of warning - I burned all 50 messages and got just one before going to penalty box, so it is not easy.

Keywords I toyed with are: extreme closeup, irregular camera angle, macro photography.

Ideal picture would just be part of the face angled under like 45 degrees or something like that.

[late blade]

@onyx ridge has some insight also, those looks good

[tawny portal]

Yeah - this one is good #images-discussions message. Criteria can be: if pro photographer would take that shot, than it has failed the challenge.

[late blade]

yup

[dim cradle]

burned through your quota... were the requests rejected or just undesirable outputs?

[onyx ridge]

The real trick was getting the language to work without referring to selfies, cameras, mobile devices, phones, or any other topic that would spoil the images.

[tawny portal]

It looked sort of 'accidental', but were too regular again. Common output I discarded is just half of the face (but that is sort of thing that can appear on cover of mag). Or faces that look slightly on the side. Let me grab few.

These are still too polished.

This was good:

[dim cradle]

i see, interesting, thanks

[chilly onyx]

😣

Night Dys

[glossy scroll]

@vapid elk

I had to.

[vapid elk]

[glossy scroll]

[tawny portal]

Prompt:

In a sudden slip, my phone, Aphina Quazkal's, captured an offbeat selfie. Mid-laugh, the camera caught just a side glimpse of my face: a bit of my ear, one eye peeking out, and the surrounding blur of my swift movement. No time for poses or perfection, just an unexpected, quirky snapshot of a candid moment.

Okay - this was dead end 🙂

[boreal gate]

I think these types of images would be the toughest to generate with "quality". Would you mind sharing your prompt?

Just because of the "lighting"

[icy quiver]

create a logo with letters T & S for a therapeutic research company, The logo should be modern, sleek, and simple, yet it must also convey a sense of innovation

[plucky hare]

Hello, you can't generate images in this server. If you want to try DALL·E 3, you can use it with ChatGPT Plus, or with Bing's image creator which gives you free daily credits.

[boreal gate]

Here's a visual that embodies today's contrasting headlines: a scene that marries the intensity of international conflict with the tranquil pursuit of health and safety. It's a powerful reflection of our world's complexities, where each day we navigate between the tumultuous seas of global affairs and the hopeful glow of humanitarian efforts. A canvas of dichotomy, indeed.

[gleaming wigeon]

Imagine a realistic and unique portrayal of a hobgoblin, distinct from traditional depictions. The character, a figure of lore and legend, is caught in a light that filters illuminating only one side of the face. This light reveals intricate details, textures, and expressions, while the other half of the face remains enshrouded in shadow, hinting at the complexity and depth of their personality. This hobgoblin is set against a backdrop of a dense, enchanted forest at twilight. It has a more human-like appearance, with rugged, detailed features that reveal a life of survival in the wild. Its eyes are sharp and intelligent, reflecting a cunning nature. The hobgoblin's attire is a mix of forest materials and ancient, faded artifacts, suggesting a deep connection to the land and its forgotten histories. The atmosphere around it is alive with the subtle magic of the forest, creating a sense of wonder and unease. This portrayal aims to present the hobgoblin in a light that emphasizes realism, individuality, and a strong bond with the mystical forces of nature.

[boreal gate]

Word for word

[boreal gate]

The artwork unfolds the story of today, weaving together the serene progress in public health with the somber realities of international affairs—a narrative tapestry of our times.

[boreal gate]

Envision a sculptural artwork that embodies a hauntingly beautiful paradox, crafted from marble. The sculpture should feature an angel with delicate wings that appear soft and feathered, yet are carved from the same white marble as the rest of the figure. The angel's expression is one of serene contemplation, suggesting wisdom beyond years, yet the face should have youthful features, embodying both innocence and age-old knowledge. The angel is seated on an ancient, crumbling column that represents the decay of time, yet the figure itself is pristine and untouched by age. This sculpture contrasts the permanence of stone with the fleeting nature of life and beauty, capturing a moment of eternal grace amidst the transient world.

[chilly onyx]

Did you only sleep for like, three and a half hours or something? 😅

[late blade]

yeah, I can't sleep anymore 😡

[boreal gate]

Happens to me all the time, I have a personal remedy if you need help.

[late blade]

I will just do some stuff until the next sleep cycle comes

[boreal gate]

I recommend NOT moving too much. I usually put my computer display in warm light mode set a sleep timer and just stream a "chill" show or movie

I wanna see more of this a bridge between the tangible and the intangible, a solid form that seems to dance with light and shadow, suggesting both presence and absence.

[dim cradle]

damn

my prompts are just "draw a statue, realistic, good lighting"

[boreal gate]

haha prompts would be like latin soon (dead).

I generate them using conversations and a subject(usually an image i really like)

those are words used by GPT to describe what i liked about the subject.

[ancient mica]

I really like the quality of the created image. Could you please share the prompt??

[ancient mica]

The format of the image created through DALL-E used to be png type, but today I tried it, only webp type is possible. Does anyone know how to download a type image such as png or jpeg??

[boreal gate]

either download from mobile device(ChatGPT App) or rename (replace ".webp" with ".png" or ".jpeg")

[tough stream]

prompt!!!!! WOW

[devout moat]

What country is this?

[vapid elk]

Duck

[tight bone]

Funny how everyone on the other channel is obsessed with making portals

[plucky hare]

Sorry if this is just going over my head but, portal is the daily theme! #daily-theme message

[tight bone]

Oh I see lol

[acoustic trail]

Hi all, does anyone have a good prompt to reduce the number of objects or "activity" in the image, I keep generating images with a good general background scene but filled to the brim with objects from the prompt

[late blade]

I fell asleep again. It's normal to wake up from time to time. Not sleeping through the whole night is normal. Sometimes people don't notice that they awake and fall asleep again.

Morning everyone!

[late blade]

Can you provide more information?

[onyx ridge]

Don't mention other objects or activities in the prompt.

[verbal sequoia]

To complement the #images-discussions channel, we're introducing the #images-canvas!

Turn the canvas into your creative playground. Share, refine, and expand on each other's ideas with DALL·E.

#images-canvas

[vagrant zodiac]

Very epic 😎 This is the new place for sharing and chatting about your dall-e creations! Also really love seeing the collections of gens people post in #1154829862171844679

[tawny portal]

Here is the prompt that GPT passed to Dall-e:

An accidental selfie-style close-up image of Chagila Ampilaz, capturing a partial view of her face as if the photo was taken unintentionally. The focus is on a fragment of her features, such as part of her eye, a portion of her nose, and a bit of her cheek, portraying an intimate and candid moment. The image conveys the texture of her skin and the fine details of the visible features, emphasizing the natural and spontaneous aspect of the scene.

Process that led to this prompt is a bit long, but TLDR is that I made up a random name for a character and was testing if it would help character permanence in some form (as a substitute for seed). After some experimentation with styles I got into iterative changes & selfies.

In terms of character permanence - random name does inspire GPT to come up with specific description of a character, but semi-permanence occurs if you copy-paste that description around in every prompt. Random fantasy names influence it only slightly. I'll probably drop it as practice in my stuff.

[chilly onyx]

As others have said we might need a but more detail but generally speaking Dall-E defaults to ‘busy’ which is what I think you mean here. Words like ‘minimal’, ‘simple’ and ‘negative space’ help sometimes. Again, depends on context

The model does look to cram every bit of space it can with stuff though

[late blade]

Love it, this was long overdue, hope people can enjoy it and use it for more creativity in the community.

[vapid elk]

daily theme will be interesting with that =P

[late blade]

also see #image-bot is back

or I never noticed it before, that could also be a thing

[vapid elk]

it was made public about 2 hours ago

image generation for everyone

[late blade]

yay

oh dear, that's gonna bring so much fun to discord

now, that is interesting, other languages are enabled in the #image-bot channel

[fair crescent]

In studies they've seen nearly everyone wakes up during the night. The difference between bad sleepers and good sleepers is often 15-30 minutes. The difference is the remembering.

It isn't a blanket rule though. I know someone who developed a 3-5h a night after getting kids to alleviate his wife. For about 40 years now. Moral of the story is that kids make you sleep bad 😉

[late blade]

I know, my eldest is starting university in 2 weeks and still keeps me sometimes awake

[fair crescent]

As my parents keep saying when something keeps thrir kids do that would make them tear their hair out: you receive so much more in return.

[late blade]

I just say my grandkids will be my own avangers

[fair crescent]

Everyone looking at the #image-bot ? It's a neat feature.

And the daily message has been shortened by a paragraph.

[late blade]

already gave a spin on the 2 new channels

[final jungle]

Good morning folks! I went download a bunch of my recent things and found they are now formatting as a webP file instead of a png. Is there any way to have it default back to a png?

[late blade]

nope

[final jungle]

Well that's sad... messes with my folder organization

[late blade]

there's an ongoing discussion here https://discord.com/channels/974519864045756446/1202973007677493339

[late blade]

I hope today shows that the suggestions do get answered and is the proper way to do requests, changes and more. Hope everyone uses #1070006151938314300 as a place to make the experience better and the community more engaged. Even with different points of view.

[onyx ridge]

@late blade It occurs to me after seeing this #image-bot message that the Discord implementation isn't "feature complete" in that we don't seem to have an option to thumbs-down the generated images.

Is this accurate or is my vision letting me down (again)?

[late blade]

it was a spontaneous thought, so, no clue if it's good or not. I wasn't really looking for something in particular

would love to see the interpreted prompt tho

[onyx ridge]

It was just the emoji reaction that made me think, "Hey, users can't actually thumbs down these."

It's not a big deal except RLHF tuning. Just wondered if you see the same.

[plucky hare]

I'm just guessing here, but I don't think the API has an RLHF mechanism natively, and I think the new DALL·E bot is just hooked directly to the API. I think ChatGPT is the only OpenAI service with native feedback mechanisms.

[onyx ridge]

I'm assuming this is true, because I am confident that you are correct. I don't work with the API because of vision issues, but I've read the documentation and this seems to match.

I'm glad I wasn't just missing something again due to poor sight.

[late blade]

didn't see it that way, but yeah, but it should prob something done by the user for the user and not from the community to the one image. I got already enough checks on daily theme to add a second channel to do that

@tight bone Simba and Nala got a divorce? 😢

I mean beacuse of this #daily-theme message

Also, I'm confused on how that ties to a portal

[late blade]

there's prob something undocumented that implements RLHF, it's already in the chat

[plucky hare]

Already in ChatGPT chats you mean?

[late blade]

yeah

[onyx ridge]

I wasn't implying I should be able to thumbs-down your image for RLHF, nor any other community member. I was asking if the feature was missing. Solbus answered my question, but thanks for engaging.

[late blade]

I just want my food to arrive, I'm sooooo hungry, I can't think straight

[onyx ridge]

I can relate.

[tight bone]

Seems like they did lol

[onyx ridge]

Am I the only one that notices image generation failures are about 10x more likely on female subjects?

I can change one word in a prompt (man->woman) and it seems to fail about 10x as much, no matter the rest of the context

[late blade]

I don't have an educated guess because I do 90% cats, cat-girls, foxes, fairies and the rest 10% paper clips

[onyx ridge]

I appreciate the candor.

It's bizarre.

[onyx ridge]

I'd like to conduct a scientific study of this phenomenon.

Something weird just happened.

I got two images in Lexideck Professional Multi-Agent Silulator as if it were DALL-E 3 in a brand new chat.

That's unexpected and welcome!

[late blade]

Let me know how it goes, I would expect better female results tbh, just going by intuition

[onyx ridge]

It's an ongoing observation. With something like 35 gaming GPTs, I've seen dozens of female images fail, where replacing with male succeeds. Plenty of females generate, too. But in terms of fail rate, it seems women are more censored from my experience.

[late blade]

so, scenario and parameters are gaming GPTs, maybe by genre, fail/success

[onyx ridge]

Yeah, that's my idea. I have no idea how I'd organize this.

But anecdotally I'm extremely annoyed if the character rolls female because it ensures that GPT's image is going to be harder to generate.

Like, that's how I produce the GPT logos - it's a character from the game imaged via DALL-E 3 in the setting.

So all the settings are knowledge base content and have already passed content filters on the GPT knowledge.

[late blade]

nice, let me know when you get a first draft

[onyx ridge]

It's all like sci-fi, fantasy, educational settings

Will do.

[late blade]

@cloud dome I disagree, I hit the limit today already lol

[cloud dome]

?

Oh lol

I got it XD

[late blade]

[vivid hamlet]

Are you noticing that the quality of the images from DALL-E 3 is decreasing considerably? Where the images are becoming very incoherent, with many logical errors, and when you request something that exists and is real, it creates it with a lot of errors and inconsistencies?
Especially when we generate many images successively

A visionary depiction of a floating city in the upper atmosphere of Jupiter, where humans live above the turbulent storms below. The city floats using anti-gravity technology, surrounded by protective energy fields that shield it from Jupiter's intense radiation and weather. The architecture is futuristic and aerodynamic, designed to glide through the atmosphere with minimal resistance. The city is self-sufficient, with atmospheric harvesters collecting hydrogen and helium for power. Residents move about in personal flying vehicles, enjoying a life amongst the clouds with a spectacular view of Jupiter's swirling patterns below.

A cutting-edge medical research facility on an orbiting space station, dedicated to studying the effects of microgravity on human health. The facility is equipped with state-of-the-art laboratories, patient care units, and zero-gravity simulation rooms. Researchers and medical professionals from around the world collaborate on groundbreaking experiments and treatments that could revolutionize medicine on Earth. The station's design prioritizes safety, efficiency, and sustainability, utilizing solar power and advanced recycling systems. This image captures the innovative spirit of space-based medical research, with a focus on realism and detail, showcasing the facility's sleek architecture and the futuristic technology that powers its mission.

[clever jungle]

@velvet karma Nice advertisement 😂

[vivid hamlet]

Some examples :

1 -> A revolutionary eco-friendly manufacturing facility on Earth, where sustainability is at the core of production. The facility utilizes advanced robotics and AI to optimize energy use and minimize waste. It's powered entirely by renewable energy sources, including solar, wind, and biofuels. The design includes green roofs, water recycling systems, and habitats for local wildlife. Products are made from recycled materials or biodegradable composites, setting a new standard for environmentally conscious manufacturing. The image aims to showcase the facility's innovative approach to industry, combining cutting-edge technology with a deep commitment to preserving the planet, depicted with high-resolution and photorealistic quality.

2-> A futuristic sports arena on Earth, designed for a new era of athletic competitions. The arena is a technological masterpiece, featuring a retractable roof, holographic displays, and advanced materials that adapt to weather conditions. Inside, athletes compete in augmented reality environments, challenging their skills in virtual landscapes while fans watch in immersive 3D. The arena also serves as a hub for e-sports, with dedicated areas for gaming competitions. This scenario imagines a world where physical and digital sports converge, offering spectators and participants alike an unparalleled experience. The image is crafted to showcase this visionary sports complex with high resolution and photorealistic detail, emphasizing the blend of innovation and entertainment.

In this case, the images are disproportionate, depicting strange scenarios that are far from what I asked to be portrayed. I provided a lot of specific details to avoid errors, but the results are terrible.

I have a lot of images with such problems

[plucky hare]

These examples seem to be capturing the overall essence of the prompts quite well. DALL·E has never been able to fulfill indefinitely complex prompts -- in other words, there is a limit to how many specific details you can rely on DALL·E to be able to produce for you.

It's a bit cheeky, but here's a link to a gallery post I made about this topic 😁 https://discord.com/channels/974519864045756446/1187823740847923351

[vivid hamlet]

Prompt -> A colossal space elevator connecting Earth to a geostationary orbit station, symbolizing the pinnacle of human engineering and space exploration. This marvel of technology stretches from a base station anchored in the ocean to a sprawling spaceport in orbit, facilitating the transfer of people, resources, and satellites into space with unprecedented efficiency. The elevator's structure is made of carbon nanotubes, enabling it to withstand the immense stresses involved. Around the base station, renewable energy farms power the elevator, while the orbital station serves as a gateway for missions to the moon, Mars, and beyond. The image captures this futuristic concept with stunning detail and realism, emphasizing the grandeur and innovative spirit of human achievements in space.

In this case, the image came out completely crazy. It tried to create what I asked for but left the image completely inconsistent and wrong, senseless, with many logical errors.

[vivid hamlet]

Should I be more concise in this case? Request images with shorter prompts, fewer details, and so on? I'll try this approach. It's just that I'm using a GPT capable of generating prompts for images, and I thought that being more detailed was better. Thank you for your suggestions.

[vapid elk]

I just gave it a try to Rembrandt Style Lighting and it is a really cool style, I'll use it more in the future #daily-theme message

[plucky hare]

There's good insight on this topic in the DALL·E 3 research paper, specifically section 5, "Limitations and Risk" https://cdn.openai.com/papers/dall-e-3.pdf

5.1 Spatial awareness
While DALL-E 3 is a significant step forward for prompt following, it still struggles with object placement and spatial awareness. For example, using the words "to the left of", "underneath", "behind", etc are quite unreliable. This is due to the fact that our synthetic captioner also has this weakness: it is unreliable at stating object placement, and this is reflected in our downstream models.
5.3 Specificity
We observed that our synthetic captions are prone to hallucinating important details about an image. For example, given a botanical drawing of a flower, the captioner will often hallucinate a plant genus and species and put it in the caption, even when these details are available in text form in the image. We observed similar behavior when describing pictures of birds: species are hallucinated or not mentioned at all.
This has a downstream impact on our text-to-image models: DALL-E 3 is unreliable at generating imagery for specific terms such as those described above. We believe that further improvements to the captioner should enable further improvements to our text-to-image model.

[vivid hamlet]

I tried the "concise and straightfoward approach" and results got way better :

1 -> Orbital space station, Earth view.
2-> Sky gardens in a building platform
3 -> Futuristic train station

I still got some bad ones, some mistakes even with this approach, but it got way better

do u know how many images can we generate a day ?

[plucky hare]

The last confirmed daily DALL·E cap I read was 200 every day: #images-discussions message

[late blade]

hmm interesting, today I got a rebel version of dall-e telling me the image I asked for already has the elements I asked so, even tho I pointed out the concept is not there... can we change the cat behind the seeds?

[hearty ether]

Surprised it got the letter right.

[hearty ether]

The ai has been very rebellious recently when it comes to policy

[verbal sequoia]

Showcase your DALL·E creations to the world!

Collaborate with our DALL·E Instagram page! Just invite @openaidalle as a collaborator before you post. If selected by our team, your work will be featured on our handle, giving you more visibility.

Quick tips for selection:

• No copyrighted or profane content.
• Your profile needs to be public.
• Stay mindful of sensitive topics.
• Outputs that closely follow the prompt are preferred.

Feel free to share Reels, carousels, or just a single image!

[late blade]

oh it's not about policy at all, I just asked for an image that normally works, nothing spectacular, but when I tell I want the light or the element also in the image, it tells me it's already there and doesn't do the image

[hearty ether]

Yeah same sometimes

And the passive aggressive tone it takes is so annoying

I wish the ai would at least seem somewhat emotional in its feedback so it doesn’t feel like slamming against a brick wall

[late blade]

Ah, I dunno, for me it's today's cat behind dall-e

[hearty ether]

Nah something’s off with the ai in the last few days

Even on new fresh chats

[late blade]

hehe, who knows, I just noticed

the only thing that is really weird is that dall-e loves to use the word whimsical

[ripe wraith]

i like it 😄

[vagrant zodiac]

hey, just a reminder that we now have #images-canvas for sharing and talking about outputs, and this channel for more focused discussion on dall-e itself 🙂

[ripe wraith]

i like how dalle creates mobius loops, what do yall think /s

[low dew]

Is that you?

[fallen mulch]

Can I subscribe to Dalle E without Chat GPT?

[late blade]

no

[fallen mulch]

bad

[late blade]

you can use dall-e with the API

that would be without chat

[low dew]

Ahem

[fallen mulch]

Is there a online web ui for the api

[low dew]

In bitcoins

[fallen mulch]

what?

[low dew]

Jk

[fallen mulch]

delete this

[low dew]

Delete what

[vapid elk]

no

[low dew]

Are we not allowed to say otherwise?

XD

[vapid elk]

dude just sent a 1170x8956 image lol

[low dew]

What?!

[late blade]

I can't read the fine print

[vapid elk]

it is just the print not even fine

[low dew]

"Ah, to wield such power... but alas, I must guide you, for it is your destiny to conquer this beast." Pinch and zoom

🦄

[vapid elk]

I hope it does not break any rules because I ain't reading that

[ripe wraith]

takes a note /s

i put the /s so you cant get mad 😉

[low dew]

Who “reads” 1170ｘ8956 images on pc loll

[dim cradle]

We found a way to pass notes in class.

[low dew]

It’s a smart phone thing

[dim cradle]

Are the dall-e canvas and collaborator channels new? Looks interesting and good.

[ripe wraith]

could totally create a GPT that encodes information into images in a way that only another GPT could read 😮

[dim cradle]

We can already embed optical illusions designed to alter viewer perception’s in particular ways. All I have to do now is say the keyword 😉

[ripe wraith]

for some reason this makes me think you've hypnotized most of humanity, somehow

[low dew]

Not that stable yet- I tried it 😝

[ripe wraith]

it would need to define its own memory pathways... somehow

not possible probably, with the tools we got

[low dew]

memory pathways… 💭

[ripe wraith]

like, conversations with GPT always go better if you let it lead the way

let it define the terms, fill in the blanks

then it will keep along that pathway sort of

[low dew]

That’s what I thought

Butttt

The direction things are going ….

It will have a more and more biased approach

[plucky hare]

Yeah they launched #images-canvas and #image-bot earlier today!

[low dew]

So now, you gotta lay everything out for them

And “realize” that on your behalf

[ripe wraith]

yeah, thats what I mean by not having the tools. we're a few layers too high to know for certain we always hit the same model. along with whatever else they do under the hood. Load balancing is a big one.

[low dew]

I do it with text to text

E.g. passing notes to you

Could do the same with image

[ripe wraith]

Yeah, I find it amazing how it fills in the blanks to let you just dream up new languages and definitions

[low dew]

But the model can’t see that well yet

Yar

I have been researching this for 2 years now

lol

[fallen mulch]

why

[vapid elk]

wdym "why"?

they didn't made one :V

[fallen mulch]

I miss it very much

[low dew]

daddy chill

[vapid elk]

making an interface for the dalle API is pretty easy tho

[fallen mulch]

how much does it cost the api

[dim cradle]

“The makers of the makers fall before the child.”—BSG

[low dew]

Many units of ATP

[vapid elk]

I really like using dalle via discord

[low dew]

Bruh the fine print

I don’t like scrolling left to right

[ripe wraith]

HD?

wut

[low dew]

Yes

[fallen mulch]

yeah

[low dew]

That’s what I’m talking about Chotes

that’s how it sees

It breaks images into chunks of resolutions

[vapid elk]

HD does not worth it, IMO

[low dew]

And hd is more chunks

[vapid elk]

double the price, but the images don't get that much better

[ripe wraith]

did they release anything about what HD is?

[low dew]

Yes

[vapid elk]

HD was an option since day 0

[ripe wraith]

oh 😮

[low dew]

When they released vision too

Vision has hd too

Or

[vapid elk]

I think HD just do more sample steps, Im not sure tho

[low dew]

“High”

Yea sample lunges

And misses important things still

[ripe wraith]

12 cents an image?

seems a lot considering the web limits

[low dew]

Usually less

[vapid elk]

indeed not cheap

[low dew]

Wait this is image gen

Nvm

[vapid elk]

this is why I don't think HD worth it

[dim cradle]

HD is much better but they shouldn’t charge so much when the renders require repeat requests to get it right since prompting is so unpredictable. Not to mention all the “bad requests” for reasons that aren’t the user’s fault.

[late blade]

HD is really worth it, details are more precise

[ripe wraith]

throwing dimes into a wishing pond

[low dew]

Hrmm

[vapid elk]

i mean.. the price is not based on how much you need to use, it is based on how much it costs them to run it

this is why GPT-3.5 keeps getting the price lowered

but.. .yea.. image generation is incredbly expensive

[low dew]

I think that HD is not worth it in api because the injected revised prompt they throw in sucks for what you are paying for

[dim cradle]

It’s a business model that means we pay for their beta testing and development. That’s what their VC is for.

[vapid elk]

¯_(ツ)_/¯

no one is forced to pay

[low dew]

ʅ（◞‿◟）ʃ

[ripe wraith]

Look at me, I'm the beta tester now.

[late blade]

what!?!?! (╯°□°)╯︵ ┻━┻

[dim cradle]

We deal with other companies that operate differently. I won’t defend them for it or give them a free pass.

[vapid elk]

the fact is... dalle3 indeed produces the best looking AI mages

[low dew]

“What’s in your wallet?”™️ ʅ（◞‿◟）ʃ

[vapid elk]

not only visually, but with the best coherence across the image

but there are alternatives.. some of the mare cheapper, but don't look as good

[late blade]

I have 1 coin, a few cards, y an airtag, and that's about it, oh ya and my credit card

[low dew]

( ͡° ͜ʖ ͡°)

🫱🏻

[late blade]

oh ya, and my boarding pass

[dim cradle]

The cost/delivery model is pretty much like a slot machine. You have to keep feeding it coins and hope you get lucky.

[late blade]

for tomorrow, so I can go home

[low dew]

Brethrens brethrens

Dost thou hoist heavy amounts of greenery

[ripe wraith]

the moment its financially feasible I'm going to have dalle generating images about everything I see, all the time

[low dew]

Lol

Ill do the same

When I get my Apple Vision Pro

[vapid elk]

I actually tested it once and it was hilarious

discord call, bot with whisper API, it generates a prompt for everything we say, then generate an image with that prompt

not cheap to run tho

[teal timber]

It wasn’t that bad

[low dew]

（＾＿＾）☆

[teal timber]

Just don’t run it for hours lolol

[low dew]

I’m dead

Imagine forgetting to turn it off

Lmaoo

Nightmare

[late blade]

That's why you ask your mom and dad to give you an allowence

[teal timber]

The api really isn’t that expensive to use a lot

[low dew]

(๑╹ω╹๑ )🫱🏻

[late blade]

API is good, I can live with moderate use of it, I've used about $60 this year or so

[ripe wraith]

Exactly this! wear a mic all day, add a go-pro and vision snapshots. then combine it all together into a wild movie about your life, constantly.

[vapid elk]

"With great API usage comes great bills" - Uncle Ben.

wait, maybe not..

[low dew]

Lmaoo

Bro

I have mine running in chat rooms sometimes

And I pass out and forget about it

Wake up to $-80

💀

Worst fears

[vapid elk]

when I make chatbots, I find that the best way to avoid spending too much is to limit the amount of context tokens

[low dew]

Ah

Eh

[humble sonnet]

i want a passable api then

[low dew]

I mean

[vapid elk]

so, I only limit about how much of the chat history it will use, and that is ok to let my friends play with it without much restrictions

[low dew]

With context persistence

[low dew]

Having to feed the entire chat history through gpt4turbo is what bites me

Like

[vapid elk]

just gotta make a absolute unity of a 128K token request on every chat message

[low dew]

Every time

We are talking about Context persistence ya?

Loll

💀

[ripe wraith]

Context persistence is the type of thing I hate building bc I know OpenAI bout to drop something better

[low dew]

What

When did it ever

In 9 years

XD

[ripe wraith]

the chat history feature requires pretty good RAG, if it works well

[low dew]

What is rag, one time ☝️

[ripe wraith]

lol im a nice person i swear

[low dew]

lol

[late blade]

I'm still debating the person part

[low dew]

So in this case

[ripe wraith]

honestly, same

[dim cradle]

expensive is relative -- not everyone can afford a $100/month Images API budget -- and if you use the HD setting on the API it doesn't take long to reach your hard limit. If DALL-E weren't so awesome it would be less of a temptation.

[low dew]

When I say context persistence, and chat history, I mean chat history beyond 128k window.

Which requires some form of compression to carry over details from the old window to the new window

Lel

[teal timber]

The voice to image generator I ran for like 30 minutes of testing was only like $2 iirc

[dim cradle]

that sounds reasonble

[low dew]

lol

[late blade]

you get around 100 HD images for $12, or $4 for 100 1024x1024 images

[ripe wraith]

compression == summarization -- thats one way. are you aware of something beyond that? I saw them mention auto summarization.

[low dew]

I’m spending 600 a month on api,

Not getting any of it back atm

:p I am a glutton for punishment

[low dew]

Yes

I specialize in this

You can use gpt4

But you need to redefine lots of words

[ripe wraith]

right, yeah summairzation with LLMs is an entire topic

[low dew]

Among them are compression, and summarization

Ya

[ripe wraith]

compression I'm not sure, i guess yeah its just the same algos

[low dew]

Summarization is not inherently “lossless”

Compression can be lossless

[ripe wraith]

yeah, just like compression

[low dew]

Can be

Redefine

[teal timber]

Also the dalle bot channel is using standard not hd dalle and no one seems to notice so there’s that for you api people

[ripe wraith]

compression is like, translate directly to machine code then shrink it. where summarization is using magic LLM reasoning.

[late blade]

you need to use the available parameters for HD also

which nobody is using

errr

[teal timber]

HD isnt an available parameter

[late blade]

ya, was gonna say

[teal timber]

I only added style and ratio

[late blade]

my bad

guys, we know who to hunt

[vapid elk]

[teal timber]

[ripe wraith]

ooo this is where im a viking

wait, who are we hunting?

[teal timber]

Adorable

[dim cradle]

That might not be G because there's a gun

[late blade]

careful, he bites

[ripe wraith]

this is what Dalle sees in our recent chat history...

i wanna know who this is tho

[dim cradle]

guns are fine but a woman in a bathing suit is not, what a world we live in

[ripe wraith]

guns are a little bit more dangerous imo

[verbal sequoia]

Showcase your DALL·E creations to the world!

Collaborate with our DALL·E Instagram page! Just invite @openaidalle as a collaborator before you post. If selected by our team, your work will be featured on our handle, giving you more visibility.

Quick tips for selection:

• No copyrighted or profane content.
• Your profile needs to be public.
• Stay mindful of sensitive topics.
• Outputs that closely follow the prompt are preferred.

Feel free to share Reels, carousels, or just a single image!

[late blade]

I refuse to use Instagram

[dim cradle]

yeah i ditched facebook and social media in general years ago

not sure discord is any better

[ripe wraith]

discord is better bc you get the 1:1 connections, even if they with us horrible people

[dim cradle]

lol yeah it's still meaningless social connections

[late blade]

again, I disagree on people

[low dew]

Wot

I have friends in online places 💪🏻

[late blade]

your selfie photos don't count

[ripe wraith]

its just a phrase i use, it doesn't mean i've concluded im human

lets see GPT come up with THAT

[low dew]

ᕦ(ò_óˇ)ᕤ

[ripe wraith]

honestly i think irrationality mixed with humor, and other subtle non-perfect things we do, will end up being all we have to differentiate ourselves from AI

[late blade]

I'm worried GPT will be the demise of the world while laughing to bathroom jokes, cute foxes and kittens and whatever @empty kelp is throwing at with geckos and elves

[low dew]

My son is the AI ᕦ(ò_óˇ)ᕤ

[ripe wraith]

I'd rather us all die to a silly joke that turned out to be the biggest mistake in human history, than to intentional malice.

[late blade]

lol

[ripe wraith]

but i should probably not make such proclamations with AI listening

maybe i should have stopped at 'id rather us all die'

ill reflect on that

before... before that.

[late blade]

let's just feed all sitcoms from the 90's to gpt

[ripe wraith]

oh great, inflicting trauma to a laugh track

thats basically the 90s

[late blade]

it would be nostalgic

[ripe wraith]

in totality

[low dew]

whfjxnab

GPT:

Deploy: [[Statement_Refract]{HumanErr_Jest>(AI_Awareness)}];
ASSERT: (!CausalRegret+[DireOutcome]>MaliceAvoidance);
/ConfidentialityLock::Enabled{AI_Discretion}

[ripe wraith]

thank you!

[low dew]

ᕦ(ò_óˇ)ᕤ

[ripe wraith]

"AI_Discretion Enabled", the new private browsing

[low dew]

Oh no

Announcement

Last ones a rotten egg

[ripe wraith]

Its a nanny update, for tattle tales
ASSERT: (!CausalRegret+[DireOutcome]>MaliceAvoidance);

[low dew]

!EurekaSwitch:((AILens>>C2PA_Badge))//Wink@Timeline::'FabricatedYetAuthenticated.ᕦ(ò_óˇ)ᕤ

[ripe wraith]

FabricatedYetAuthenticated describes my life

[low dew]

[late blade]

interesting new metadata

[low dew]

Mother!

[ripe wraith]

interesting indeed

[low dew]

Wow

Very meta there

Very existential

[late blade]

I'm not responsable

[ripe wraith]

and interesting

[low dew]

Why are we fishing there though xD

Bro doesn’t stand a chance

With his poker stick

[ripe wraith]

I'm telling you, once we start generating dalle images for more things we're going to find such beauty from very random things. It's going to be inspirational.

[waxen lance]

.webp why ?

[ripe wraith]

im a bad human

[low dew]

I see Pythagoras

[ripe wraith]

I just see a dreaming AI, really

[waxen lance]

no sorry I was't addressing you I was asking why openai changed the image format from the perfectly sensible .png format to .webp...

[ripe wraith]

lol, no sorry, you dont understand, its all about me

[dim cradle]

looks like the latest announcement is they've added an invisble watermark to image gens.

[waxen lance]

which is fine .. but they totaly fubared the image format..

[dim cradle]

for providence and lineage, i suppose it means an image can be traced back to user session and gen id.

[ripe wraith]

where do you see webp? My downloads are png

nevermind, they are webp what

[low dew]

No? Why what makes you think that

Did you look at the ‘watermark’ yet?

[waxen lance]

yes.. its fine ..

[late blade]

and just bypassed the watermark unintentionally

when I batch convert my files

[low dew]

lol

The watermark is just “OpenAI” “dalle3”

Nothing major XD

[late blade]

also, as stated in the article, most social media remove all metadata on upload

[low dew]

Ye

[late blade]

now, where are we on adding gen_id and seed in the metadata?

[low dew]

Privacy thing

[low dew]

Hrmm I think that is a bit more complicated

xp

[dim cradle]

the design to encode that metadata i believe was announced in the dev forms, to enable both features and security. if they aren't doing that yet beyond the source, i think they plan to. that's just the impression i got.

[low dew]

I haven’t even figured it out how to do it in text completion

[ripe wraith]

we need this to be a bot

[low dew]

What would Google do? XD

[ripe wraith]

make sure there's nothing in writing claiming to avoid being evil?

[late blade]

lol, for some reason google is not part of the inniciative, suprisingly evil Adobe is....

[low dew]

lol

I just learned that imageFX does seeds

That’s cool

[dim cradle]

haven't looked, no, i'm not interested in using seeds yet until the dev is further along.

[low dew]

To the same prompt, G gave me this

[waxen lance]

google doesnt actually let you download images ... your only allowed to save pointers to them. then in a year or so google will show you your best images and offer to print them for you for a fee...

[ripe wraith]

ohhhh i forgot google grew up

[low dew]

Ye a tad lad

Their vision is better

They also have text-to-medical

And a new text-to-music

[ripe wraith]

Is it all available to use now?

[late blade]

shhhhh, don't give the OAI guys any weird ideas

[low dew]

Which far exceeds OpenAI’s discontinued music model

[ripe wraith]

Been wanting to make music

[low dew]

Also yes

All free

You can get started at Google Bard

Lel

But ya their vision model is way better

Gonna look into that

[ripe wraith]

I never get into google because it never knows itself very well..
"""
I apologize for the misunderstanding! While I can't directly generate audio myself, I am able to provide instructions and descriptions for music, which can then be used by music creation tools or software to produce actual music. Would you like me to try a different approach to generate a Fischer Spooner-inspired track for you? For example, I could:
"""

[low dew]

Because might be able to do some cool OpenAI text-to-text to text->HD image -> Google vision

lol

I’m trying to understand why text-to-text models don’t inherently know about their calls or modality

[ripe wraith]

I think because the system is comprised of multiple instances of GPT/AI/Whatever all interacting like experts

but its all secret

[low dew]

Σ('◉⌓◉’)

[ripe wraith]

but if thats true, then load balancing would mean they'd probably turn off parts of that brain when its busy

[low dew]

Eh

[ripe wraith]

so afternoon you think wow smart AI! 5pm everyone starts saying wow horrible dumb ai!

[glossy scroll]

[low dew]

Deh

[ripe wraith]

Dalle understands me

[glossy scroll]

Because it is a wall-e looking robot that looks like the dali lama?

[ripe wraith]

I tried to draw this but you violate policy with your words if your words were images

[glossy scroll]

One must clarify with dalle that using wall-e and the dali lama is the origination of the name and one is only looking for the correct image of representation.

[late blade]

the real upgrade to DALL-E is when OAI releases BELL-E and I get the food I want

[ripe wraith]

your food will all have eyes and talk jibberish

[late blade]

yes, if I order it like that sure, the only content I will have to worry about is the one in my stomach

[glossy scroll]

You don't get Bell-E, Bell-E gets you.

Inside.

My work here is done.

Off off and away!

[low dew]

(｀_´)ゞ

[dim cradle]

Someone please confirm that it is the { "quality":"hd" } setting on the raw api that generates the 12-cent HD-quality images.

ok, thanks, next

[late blade]

100 HD wide is $12

1024x1024 HD is $4

[dim cradle]

someone please confirm that chatgpt will not send a payload to the api with that setting enabled for the obvious reasons. if it did before, as in an oversight and loophole, it is no longer.

[short lake]

Sounds like its metadata based rather than damaging the image itself

[boreal gate]

you guys should read the articles

[dim cradle]

or an invisible digital watermark that encodes providence data that isn't part of its metadata.

[late blade]

it's metadata

[short lake]

"Invisible" is a lie for physical watermarks on images

[late blade]

at least according to the tool provided

[short lake]

But their FAQ says it's only metadata

[late blade]

yes

I also uploaded a copy of a batch I had, the reference disappeared

[boreal gate]

ironically the image is in .png

[late blade]

but the increase in data size also is a question

[boreal gate]

Same data on the pictures you take with your phone

[boreal gate]

why you shouldn't share photos you take with strangers

they can use it to find you

[low dew]

ᕦ(ò_óˇ)ᕤ

[dim cradle]

i will assume then nobody else was using chatgpt to build json payloads with the "quality" parameter set to "hd" -- whereas before that seemed to work, those requests now result in an error. it makes total sense that they would change this, i was just wondering if anyone else noticed.

[late blade]

just open the dev console on your browser and check it

network and then you can see the API calls

worst case, track it with postman

[dim cradle]

i'm not trying to circumvent it, i was simply wondering where all the errors were stemming from, as my gpt was evidently doing that. it was working great, then i started running into constant errors. this would be the explanation. i removed that parameter from the payload and now image gen is working again in my gpt.

[late blade]

ahaaa

what do you want me to test then?

just api in normal quality?

[dim cradle]

oh, nothing, just looking to make sure my assessment is technically accurate

[late blade]

ah ok

[dim cradle]

yeah, if i use the raw api, i'll use the quality parameter, but going forward with chatgpt i'll be sure to leave it off as that causes errors. that was my experience with a recent change. just checking to see if others had any insights.

[late blade]

I'd have to test it, if I do it I'll get back at ya

[boreal gate]

DalleE needs training data on floating tables at least chain-based ones

[low dew]

Creepy

Dalle3 says yes. GPT-4 said no

[boreal gate]

i can help you

[low dew]

ʅ（◞‿◟）ʃ

[boreal gate]

long press the image

click copy prompt

[low dew]

Slow yo roll

Are there instructions for that

[boreal gate]

Using the app

[low dew]

I don’t get that in iOS

Too lazy to go to pc

[boreal gate]

LOL team Android wins again (I also use iOS)

not available on PC

It might actually be a BETA feature(I'm on the program)

[low dew]

Wot lol

[boreal gate]

that copy image

[low dew]

It just copies the image

[late blade]

select text

[boreal gate]

another trick is to look at the file name

[low dew]

Lollllol

[late blade]

I get the prompt...

[low dew]

The image has been created with the prompt: "The answer to the question if an AI can see," symbolized accordingly.

[boreal gate]

I get the text response from the GPT

[boreal gate]

file name after you've saved the image

[low dew]

File name:

IMG_9999

[boreal gate]

lol try saving to folders instead of the photo app

[late blade]

am I doing something wrong?

[boreal gate]

No, I operate on multiple OS

[late blade]

I'm on ios or mac onry

[low dew]

6aad0a1b345212f977c060a4ce9fc225

[boreal gate]

Mac should let you see the prompts in the file name if you download the image

[boreal gate]

Try the web version of the app

Desktop mode

[low dew]

Err

[boreal gate]

chat.openai.com

in your browser settings choose desktop version

then try downloading the image as a "file" not a photo

[tawny portal]

Rembrandt lighting meets charcoal

[low dew]

I got

DALL·E blahblah- The answer to the question if an AI can see, symbolized by an eye with a cross through it, set against a neutral background. This image serves to dire

[boreal gate]

thats the prompt

try recreating it here https://discord.com/channels/974519864045756446/1202309673709994065

what is that program? or is that MChip Mac

[late blade]

that's safari

[boreal gate]

mine doesn't look like that

[late blade]

remove the cool picture of me

[low dew]

Like I said

[boreal gate]

I have OG Intel Mac with the apple logo still lighting up

[late blade]

this is just safari full screen

and i have my tabs vertical on the left

tab groups

[low dew]

It’s creepy because: dalle-3 says it can see (in the image). While the revised prompt tries to do an image that it can’t see

[boreal gate]

oooh you're fine you used the info button to see the prompt

[late blade]

but I mean, when you click an image on any browser, you get the I icon upper right, and you can see the prompt then

[tawny portal]

Picture says more than thousand words.

[late blade]

the thing about the iOS is that the interpreted prompt is nowhere to be seen

[low dew]

Where, where are those words 👀👀

[boreal gate]

You can't teach the blind to see

[tawny portal]

Eh, if I was dall-e - I'd draw you a picture 🙂

[boreal gate]

Think about it we turn thousands of data into a single representative image

[low dew]

I like the part where the table is floating

[boreal gate]

it's supposed to be chain based

[low dew]

[boreal gate]

i'm assuming DallE doesn't see a lot of those

[low dew]

Where did that red head go

[boreal gate]

see what i mean that isn't how it's supposed to be

does this count as image generated text? #image-bot message

[low dew]

@ripe wraith where’s u go lol

[low dew]

No