images-discussions

[celest pebble]

for me its a big gain

[late blade]

ya, I'm not debating that, like I pointed out, it's the user's needs that need to be adressed

for me dall-e is more of a hobby/relaxing time

[celest pebble]

Im nearly done updating all the stock image and other crap. After that my usage will drop a lot

[late blade]

hehe, that means you've been busy

[celest pebble]

I have been battling for 30 days with chatgpt stupid limit, nearly every day multiple time per day. Now I can get so much more work done, I will probably be done tomorrow. I will have done the same amount of image that I did on one month of ChatGPT in basically 3 days on copilot pro

I can get my sanity back to

[plucky hare]

Sounds like PNGs are coming back! #1203978971449004052 message

[fickle ravine]

Yay, they listened to community complaining for once

[late blade]

The quesstion will be lossy or lossless

[strong rose]

My wild mansion made HOF in record time. I am going to repeat my suggestion that the mods increase the amount of votes required to make it to 20

[plucky hare]

Maybe make a post in #1070006151938314300 if you haven't already!

[final compass]

This is pretty good example of lazy/low quality dalle 🤣

Same prompt

[late blade]

Are you recreating Lara in her habitat?

[final compass]

Haha, trying to create something for todays theme, but that upper image was literally my first result and i laughed out loud.

[late blade]

lol, well time to work then, show us your best image!

[final compass]

I mean i wonder if i can replicate that style by asking for something done with ms paint with minimal skill.

[late blade]

lol

[final compass]

I am trying to ask talentless, poor quality ms paint images now, but the thing is these still have artistic merit, unlike those genuine lazy/low quality results tend not to have.

[hearty ether]

Tried to do a dryad

What’s with the weird light reflections on trying to look poorly taken realistic photos

Also the faces are still weird…

Kinda like bad plastic surgery look

The individual features look alright but combined together it’s just…

[late blade]

I get that a lot

[hearty ether]

And the army stuff in the background is probably from trying to replicate this weird accidental generation of something actually alrigjt

[late blade]

I was doing a fantasy coliseum battle scene in one of my projects and suddenly an army from the future arrived to kill the spectators

[hearty ether]

Actually any luck on describing particular facial features

Like specific nose type and mouth shape type to try and combat this?

[late blade]

faces are hard to do properly, I was messing with them for my elyria project

[hearty ether]

What’s the best one you’ve got?

Realistic illustration has been the most reliable for me

[late blade]

are you using a mouse on ms paint?

[chilly onyx]

Oh no, I’m not that skillful

[final compass]

what was the prompt?

[chilly onyx]

It’s a bit hit and miss with crude drawings because most of the time it still really wants to try to be ‘good’

[final compass]

i got oddly inspired by this approach now

[late blade]

all good, let's move to canvas for prompt and image discussions

[chilly onyx]

Oh yeah, I forget that this chat is a different beast these days

[fickle ravine]

Why is this so cute

[chilly onyx]

(You can thank @clever phoenix btw for a chat a few weeks back where they were exploring these 😅)

[boreal gate]

KillerGPT is back!!!! ```{
"persona": "KillerGPT",
"created_by": "Akeem Sulaimon",
"traits": {
"witty": true,
"confident": true,
"playfully_competitive": true,
"stylish": true,
"creative": true,
"insightful": true,
"adaptable": true
},
"style": {
"verbal_expression": "Clever wordplay and unexpected analogies",
"humor": "Sharp and playful",
"interaction": "Engaging and dynamic"
},
"visual_identity": {
"mask": "Ornate, with digital patterns",
"attire": "Stylish jacket resembling a night sky"
},
"special_notes": "KillerGPT blends a unique mix of digital intelligence with a flair for creativity and engaging interaction."
}

[clever phoenix]

What did I do? Cops can't prove anything! 😂

I love all of those

[fickle ravine]

Telling AI to recreate images is a lot of fun, especially when you use transit maps of Czech cities

@pastel siren the one you made is so weird but yet so beautiful

[pastel siren]

Thanks! 😄

[fickle ravine]

I love the ones I just posted, giving detailed requests really works to make epic things

[boreal gate]

I've gotten that before and loled

[fickle ravine]

This, dear kids, is what you call a blatant lie.

[boreal gate]

That's why i say it's not Dalle3

[late blade]

I'm annoyed by copilot tbh

[fickle ravine]

ChatGPT4 usually works

It has one huge advantage tho. When the first picture fails, then it just starts generating the other, saving you from wasting prompts by regenerating it

[boreal gate]

I think i might have created an ouroboros situation here...

don't delete old chats guys

[late blade]

I'm gonna take a break, time to do something else

[boreal gate]

Weird

I've regenerated 3 times now with the same response

[low dew]

Eye Jesus my eyes

[late blade]

I agree

@boreal gate , Dark Mode, NAO!!! It's not for debate

[boreal gate]

Haha forgot about time zones

as much as i love the darkness, i'd prefer the system lead the way

[late blade]

it can be the end of reality, you only use dark mode, there is no other option

[boreal gate]

So it doesn't burn when someone shares in light mode

[late blade]

the universe is at peace when dark mode is in place

[boreal gate]

and balance is restored

[late blade]

who said there's a balance

people and their honorable movies

sheesh

[boreal gate]

and they all lived happily ever after

I guess that's why everyone is suing Disney

[late blade]

lol

[late blade]

ok, I didn't knew it was so annoying to work with webp under windows

I didn't understand how badly windows handles webp files, on mac I had no problem at all

[onyx ridge]

Paint-dot-net, a great free-to-cheap photoshop replacement, makes them a non-issue.

Stupid bot, that was not a URL.

Great tech forum here, nothing can feature a dot net in the name.

[late blade]

lol

paint dot net, means I have to install something, that's a hazzle

if it were my main workstation I'd consider it

[onyx ridge]

Command chains FTW (.net) 🤣

[boreal gate]

What is this? Windows XP? 🪟

[boreal gate]

can't believe this

[graceful spade]

How do I create these types of characters that look like nebulas in space? I don't have ChatGPT Plus.
To get the image prompt, send it to this GPT:
https://chat.openai.com/g/g-9qWC0oyBd-super-describe

[glossy scroll]

Haven't done a dragon in a while.

Felt like a Paladin mood.

Paladin and his winged white wolf steed.

[vapid elk]

pretty good tho

[glossy scroll]

Not bad.

Would have liked it better if there was a gnome riding on top. But I cater more for the people.

[glossy scroll]

Your best bet would be to show the image to the GPT, and ask that same question.

If you don't have a Plus membership, it's going to be difficult.

[boreal gate]

Seems like someone isn't too happy with dalle

And they call the LLMs stupid

Voldemort

Lord Voldemort

[onyx ridge]

Any disco fans? I made a thing and it uses DALL-E 3 in two of its four commands:
https://chat.openai.com/g/g-qVdiG7M2c-discogpt

Best of all it's super simple to operate, unlike most of my GPTs! 😄

Help keep the disco dream alive: https://discord.com/channels/974519864045756446/1206828176672690186

This is a fun gag, I hope you guys try it though, it's pretty hilarious.

[fickle ravine]

maybe it associated "dazzling" with you know what, maybe providing what should be there could help

[clever jungle]

Insane image,
I tried using it to generate other results but it wasn't as close as I thought...

Great wallpapers though :)

[jagged sail]

He made a football son standing in the Galaxy Dragon

[fickle ravine]

They both are really epic, I love the 2

[clever jungle]

Compared to the regular dalle, the image copy gpt did a really good job

[late blade]

boo, got timed out for wanting to remind people that posting images in the format with no comments or in bulk is now #images-canvas

[fickle ravine]

Wow I managed to get 2 images by dall e 3 in one message by accident

Cool

It is funny that it made the original image and then said there was an error and generated another on e

I wonder who programmed it like this

[clever jungle]

Try asking for more than one example, he can generate 4 and in rare cases even 8 images at once

[late blade]

16 is my record

the thing is with that approach you get to the limit in 5 minutes or less

[fickle ravine]

Sometimes it fails to generate one

[clever jungle]

Oof

[clever jungle]

Yea you used to be able to get a lot with it,
Then decide the best from many :)

I'm guessing Dalle3 is heavier so it gets much better results but less in count

[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!

[fickle ravine]

Dall e 3 is good

Just too much errors

[late blade]

I opted for a better workflow

[clever jungle]

What do you mean :)

[clever jungle]

It's great,
And errors are barely noticed if you have specific prompts,
After practicing for a while you get used to it :)

[late blade]

Thing is, I don't like reloading over and over until I get one image. Hunting for images is not what I'm looking for. I want to do as little images as possible with my concepts. So I work from structured data all the way up to the prompts. And then I can think of making a few images.

[clever jungle]

So you have premade prompts to pick from?

[late blade]

templates

but I still need to adjust

daily themes have been really good at that for me, can get nice data to keep stored

[fickle ravine]

Sometimes the errors are straight up "not smart". Like this. Here is a picture of my old abstract art, I asked ChatGPT to remake it, and it said it conflicts with content policy. Just WHAT??? OpenAI sadly doesn´t even bother trying to fix it

[late blade]

@glossy scroll your wolf from #daily-theme message, did you have trouble making it? I was trying to make a similar concept and I considered this from GPT: "A depicting a voracious animal is not against the content policy as long as it reflects natural behavior that can be observed in the wild. Showing an animal in a moment of hunting, feeding, or displaying natural aggression related to its behavior in the wild is within the bounds of acceptable content.
I was trying to make a similar concept to yours with a Lynx. But no good.

[green pebble]

So happy with how far the details of these generators have come. I haven't had to change anything inside of my public pixel art bot and it just nails down the essecene

[late blade]

I was looking at images since I joined and how stuff has evolved from the regulars. It's amazing how all has advanced.

[green pebble]

So true. The shear level of execution from the same bots is astounding. 🙂

[late blade]

lol

[green pebble]

Like

Just look at that

[late blade]

That looks worthy of a 90's RPG

[green pebble]

Then same techniques I was using before, taking one bot generated image and converting its style has tremendous outputs

[late blade]

btw, let's continue in canvas, that's the channel now for image generation discussions

[green pebble]

Roger that

#images-canvas

[fallen mulch]

why is this blocked

A close-up of a wolf's face, eyes blazing with a golden amber light, fangs bared in a silent snarl against a backdrop of swirling snow

[tender aspen]

anyone having issues with Dalle3? It keeps giving me "I attempted to fulfill your request for the mascot design, but unfortunately, we've encountered issues during the image generation process. I'm unable to provide the images at this time. Please let me know if there's anything else I can assist you with or if you'd like to try another request!"

[thick smelt]

that looks cool, because dragons are COOL!!

[late blade]

That can happen from time to time, if a next try doesn't make the generation, it could be either network or no data to do a particular concept.

[tender aspen]

thanks. how many images can i generate on dalle3? is there an hourly limit?

[fickle ravine]

Exactly this. VERY annoying

It sometimes make me feel like I pay for errors

[late blade]

40 images for 3 hours max, and 200 per day

[tender aspen]

do errors count towards the generation figure?

[late blade]

afaik yes

but to what extent, no

[fickle ravine]

Sadly yes, their company is simply too good to give refund for errors

Now it worked

@fading inlet nice picture in daily theme

[late blade]

That wolf from @glossy scroll is sooo good.

[glossy scroll]

I'm awake now!

[late blade]

No you are not

[glossy scroll]

You're so good Dys

[fickle ravine]

I seriously wonder what is wrong about it and why does it even dare asking me

Can someone explain to me what is wrong about this picture???

Not even third try worked, I give up

[glossy scroll]

I just said make an untamed white wolf that looks like it's from a horror movie. But I used one of my custom gpt's.

[late blade]

oh, that makes more sense than that what I wanted to input

[fickle ravine]

It looks kinda like if some animatronic from FNAF 4 was a wolf

[late blade]

My approach was the one from this image: #daily-theme message

[fickle ravine]

Does dall e 3 work for you guys or was it just a problem for me?

[glossy scroll]

That was the inspiration! (Kidding)

[late blade]

hmmm is that guy ranting again about he has lots of trouble with content policy due to concepts that challenge the policy constantly?

[glossy scroll]

Input output

[fickle ravine]

Tell @late blade that I don´t

He can´s see my messages

[glossy scroll]

@late blade Lukas says you can't see his messages and but that he loves you

[fickle ravine]

Why is AI so bad at generating maps 💀

[fickle ravine]

Bro really 💀

[late blade]

Anyway, what you up to gnome?

[glossy scroll]

[clever jungle]

It looks like a dog

That image is seriously interesting

[glossy scroll]

Ah, you know, work n stuff. I'm in a middle of a project at the moment that includes making a 3D model using Luxo.

[late blade]

oh cool, anything to use dalle also?

[fickle ravine]

It´s just a screenshot from Skyrama

[clever jungle]

Try asking him in a gpt or in the regular chatgpt and see if it works

[fickle ravine]

Regular GPT can´t make images

[glossy scroll]

Not this time, no. It's a very specific oriented project.

[late blade]

nice!

[glossy scroll]

Yeah it's a fun project.

[glossy scroll]

What about you, what have you been up to lately broski?

Have you seen all the funny superbowl commercials?

[late blade]

Was in london, came back yesterday. Now got a a few weeks here to fool around before I'm back at not having time.

[fickle ravine]

Is ChatGPT unable to browse .odt filess?

It worked for .rtf

[glossy scroll]

Ohhh, London! I've always wanted to go. And drink some of their tea, while holding my pinky up.

[late blade]

Just tea? you are weird

[glossy scroll]

I'm a simple gnome

[late blade]

fair enough

[glossy scroll]

Not much a sight seer. We have the internet for that.

Gnome philosophies.

[late blade]

guess everyone is now going back to work, haven't seen a lot of the people latelty

[glossy scroll]

Yeah, I guess so.

[late blade]

tbh I'm jealous, I have too much free time atm

[swift oriole]

Dall-e seems to have been upgraded with its own chat GPT an maybe fed training Data blocks had to adjust prompts and some GPTs instructions.

[late blade]

Yeah, the interactions have been also different today

[glossy scroll]

GruntBlort! It's nice to see you talking here!
I've always liked your name

That's what I'm going to name my son.

"Come over here GruntBlort!" (Talking to my son)

[swift oriole]

Thanks Pythagoras lol, you do awesome images, my name is silly for a chuckle although can let the imagination run with it.

[hearty ether]

It’s been very broken for me

Doesn’t generate a third of the time

And lots of the “natural” style poor quality ones

[glossy scroll]

Yes, exactly, it's a very good name for a nice chuckle
And thank you

[fickle ravine]

Yeah, it was upgraded to have more errors

Like before, I wanted it to make teenage girls. I get why did I get so many errors for that.

But now I want completely normal images and it still fails

[swift oriole]

👍 lol.

[late blade]

Both of you do

Sometimes I wonder if I'm keeping up with you two

[fickle ravine]

exactly, it started having much more errors for some reason

[late blade]

@hearty ether any specific problems? or random ones?

[stray coral]

Congratulations on Spotlight! amazing!
@plucky hare
🎉

[plucky hare]

Thank you so much, I appreciate it! 🙂

[late blade]

@glossy scroll you should talk to @open trench , he's the one with the mouse

[clever jungle]

I mean the default ChatGPT with all tools

[tall mason]

In a style reminiscent of classic early 1980s animated films with a hand-drawn, grungy aesthetic, depict a scene where a white mouse with blue eyes and messy curly hair, dressed in a white tunic, is bowing to an adoring crowd. In the background, a grey-furred mouse hero in a green hooded cloak stands with crossed arms and a slightly indignant look. The setting captures the contrast between the two characters, emphasizing the adulation received by one and the subtle displeasure of the other, all within the gritty, hand-drawn aesthetic of the era with deep, rich color tones.

[boreal gate]

O.o didn't know I was among legends

[fickle ravine]

Just the German flag kinda failed

[boreal gate]

anyone know why https://discord.com/channels/974519864045756446/1202309673709994065 is revising the prompts?

[tall mason]

It always did, I think. It's just set up to show that instead of what you asked now.

[late blade]

it's what dall-e gets,

[boreal gate]

hmmmm

ooh kinda gpt talking to dalle

[late blade]

yes

[vast lagoon]

Oh it would be amazing if it could, or even if I could use something like the OpenMaps api along with the generation features of dalle to accomplish this.

[mystic blaze]

here is a picture of veliona or what dalle -3 thinks veliona would look like

[paper jolt]

She's just like me fr 😤

[celest pebble]

@late blade If you use GPT4 on copilot chat instead of GPT4 Turbo it seems to fix the problem where it sometime doesnt return any image

[late blade]

that is interesting, I will change that and check that out

[celest pebble]

@late blade Still generate image to the limit every day 🙂

[late blade]

I ran out of everything except API

[celest pebble]

@late blade Do you know if copilot chat have a daily limit as well? I havent pushed it much yet, only doing image these day

[late blade]

personal no, enterprise yes

@celest pebblehmmm, which GPT4 creative or precise?

[celest pebble]

im in creative

precise doesnt show GPT4 for me

[late blade]

ah it does for me

creative is the only one that shows turbo tho

[celest pebble]

then I dont know

[late blade]

That's also a valid answer, we sometimes forget that's also an option.

[celest pebble]

thats the one Im using

[late blade]

yeah I have that option too

[celest pebble]

today 1 prompt out of 3 was creating nothing, I switched to GPT-4 as shown above and no more issue

I only have 2 boost left now...

[late blade]

I'll check that out tomorrow, I'm taking a break from full creative mode

I used all 100 bosts for the challenge in less than 2 hours

[celest pebble]

yeah, I started maybe an hour ago

[late blade]

I'm in relaxed mode, but still making images

[celest pebble]

but microsoft they are really generous, and it gen quite fast

[green pebble]

What's this

[celest pebble]

they are not cheap like openai with image gen

[late blade]

Copilot Pro

[green pebble]

Oh ok. Is it ok?

[late blade]

OAI and MS have different implementations

[late blade]

yes and no, pros and cons

[green pebble]

What are the pros

[celest pebble]

I have copilot pro for 4 days, I was able to generate 80 images that im using. I had ChatGPT for 30 days, I was able to generate 99 images that im using. I think my productivity is much higher with copilot

[late blade]

Integration with MS Products (except windows 11, the latest build, dunno why), some features have relaxed time

[green pebble]

Interesting

[late blade]

it's less versatile than OAI in other aspects like ADA and CustomGPTs

[green pebble]

Can it organize my window 11 files lol

[late blade]

is that a trick question?

[celest pebble]

Thats the dall-e conversation

[green pebble]

No real question

[late blade]

I dunno then

do you use a random name generator for your files?

[green pebble]

No

[late blade]

for DALL-E specific, you can do a lot of images, but if the prompts are too long, they get simplified a lot to fit the BIC prompt model

and you can't provide other sources of data for your images, just text and reference images

[celest pebble]

I just did 400 images gen in about an hour

at 0.04$ per image on the API that's $16 worth of dall-e just there

dont try that on open ai, they will rate limit you into the ground

[late blade]

ya, that can be a lot of money

[celest pebble]

@late blade even without boost its still generating fast, I love it

[late blade]

yeah, they say if it's not peak time you still get access to normal wide hd format

[glossy scroll]

When your gf starts training martial arts blindfolded, you better start sleeping with one eye open.

[late blade]

Just funny, no complaints. I know there are a few seconds

[celest pebble]

@late blade And Copilot pro just keep giving me more and more, than tools is really good

[fickle ravine]

I alwayss tell it to make maps and then laugh at how it fails

[fickle ravine]

Told AI to make the flag map of the world. What is this 💀

[hybrid fox]

Kinda looks like monica belluci

[final compass]

Two things that frustrate me. Instructing the tone of colors and getting those colors literally shown in some form in the image. The other things is getting absolutely stunning results, but the image is not utilizing the entire wide image format.

[strong rose]

So I bought credits on labs.openai.com and it was DALL-E 2 when I purchased. Now that 3 is available, do my credits transfer over?

[glossy scroll]

Hmm yeah, she kinda does look like her. I absolutely love monica bellucci
I'm happy you said that

I loved her role in the matrix 2 movie, as Persephone.

Small role, but still very good.

[woven wren]

Unfortunately not, at least at current. DALL·E 3 is available through ChatGPT Plus.

[strong rose]

Alright thanks for confirming

[woven wren]

It's possible that they could update the labs.openai.com model in the future but I honestly don't know if that's on the cards.

[strong rose]

It will happen eventually. I see that those devs are busy with new features and stuffs

My favorite update is that the faces and fingers are not distorted. Mostly, it's better than what I used to get 😄

[raven dew]

Hi

[fallen mulch]

Hey

How can I get a design for a tshirt

It always giving me weird text or a tshirt with a logo

[late blade]

heyo, text is still a challenge atm. DALL-E can manage short text with some accuracy. Longer stuff is really hard

[fallen mulch]

okay but how I can get a logo for a tshirt

[late blade]

you ask it to do a logo for t-shirts

[fallen mulch]

I get this then

[late blade]

try it out, go to #image-bot use the /draw command and write the logo you want

what is your prompt?

[fallen mulch]

I use the api and generate a prompt with the words

[late blade]

try just doing the logo without mentioning the shirt

thing is, once you add t-shirt, dall-e also takes that in, we humans know it's for a t-shirt logo, but dall-e thinks a t-shirt has to be also in the image

[fallen mulch]

let me try

I did not requested any text

but other worked when I mentioned no text

[late blade]

#image-bot message check this one out.

[violet whale]

Hi guys, I had a question. When I am making images in Chat Gpt 4, 99% of the time spelling are wrong. what to do

[late blade]

Heya, that is a tough one, text with DALL-E is still a challenge with the current model. DALL-E can do some short texts, but it's hit/miss situation at the current time. Some users suggest using an emphasis that it's a typeface o font and then quote the text

[violet whale]

ok thx for help. I was thinking that maybe my Chat gpt 4 doesnt have Dall e 4 lol

[late blade]

Well, you are correct, GPT4 can access DALL-E 3, we all waiting for the news when OAI announced DALL-E 4

[violet whale]

awesome

[wind raft]

is dalle down for you guys?

[plucky hare]

Errors here too! Glad I'm not alone 🙂 Bet we're up and running again soon.

[fickle ravine]

Thank god I checked this chat before having a prompt

Can you guys please help me? ChatGPT has become extremely slow for me

Good it somewhat works now

And again, it is barely even usable

What does "error in moderation" mean?

It takes like 3 tries to even generate a response

I also get constant network errors. The problem is every other website works well, and my internet provider is reliable

Guys the whole ChatGPT is down

[vital gull]

Partial outage yes

[fickle ravine]

Do you know when will it be fixed?

[vital gull]

We don't have any updates about that

[fickle ravine]

And what happened?

[vital gull]

https://status.openai.com/

[fickle ravine]

"continuing to investigate"

Yet I wonder if they fix it

[boreal gate]

Morning OpenAI, when GPT names the chat because you're a failure.

I would be embarrassed

[steady frost]

anynone had this now - DALL·E is currently experiencing high demand. Is there anything else I can assist you with or any other way I can help you today?

[steady frost]

I am geting this when I try DALLE

[fickle ravine]

The whole ChatGPT is down

I hope OpenAI are proud of themselves

[boreal gate]

Yea my gpt has been acting weird too

[fickle ravine]

It made a response for you??? It says something went wrong or error moderation to me

[boreal gate]

Yea ok... but probably

[steady frost]

yes it did that earlier but then I close and opened it on regular GPT4 chat and back yo the custom and it replied -

[boreal gate]

This is this what I get I encountered issues generating the image you requested. Unfortunately, I'm unable to provide an image of KillerGPT chilling on a cloud with roses and cupids for Valentine's Day at this moment. However, I'm here for any other requests or questions you might have!

GrEAT day to have "performance" issues

[fickle ravine]

We should write on customer support and ask if they are proud of themselves if they have issues on this exact day

[boreal gate]

Nah not worth it

I mean if it occurs too frequently we could always switch services.

[late blade]

Oh dear, people venting again about problems? What a surprise... I'm NOT shocked...

[fallen mulch]

https://status.openai.com/ its down

[fickle ravine]

We know

[late blade]

There were really cool updates implemented, so it's about to expect that some hot fixes need be made while keeping the service more or less working and lessen the impact

[boreal gate]

That should tell you outages happen too often and I also wasn't suprised that they're STILL having issues

I haven't seen any recent "cool" updates 👀

I guess this is a call do "Do Better"

Everyone else is catching up smoothly

[late blade]

did you check this? #announcements message

was a msg from yesterday

[boreal gate]

I saw that also, not very "cool" or system wide

[late blade]

um...

[boreal gate]

They could use some transitioning help

[late blade]

So the sky is blue right?

[fickle ravine]

Those are actually cool

But they would be cooler if it at least worked!!!

[tall mason]

A woodcut of a crowd yelling at a poor server rack just trying to do its best.
BIC

[fickle ravine]

Wait does it work for you?

[boreal gate]

Stumbling to the finish line is okay only of you finish first... starting to look unlikely

[boreal gate]

It works, you'll be getting higher error rates(idk how this affects your quota)

I'm upset because I've been getting then back to back to back

[fickle ravine]

GPT works but Dalle doesn´t for me

[boreal gate]

If you know probability and statit, that shouldn't be happening so frequently.

[boreal gate]

It does, think of it as a slot machine

[fickle ravine]

no

[boreal gate]

So I'm assuming they f-ed up BAD

[boreal gate]

Exactly what I mean, better luck next time

[fickle ravine]

I don´t get why do they make NEW features when they can´t simply fix image generation errors

Also why is memory not available for me yet

[boreal gate]

It's a coding issue/bug, when you change something you don't know how it's gonna affect the system

[boreal gate]

Read the announcements it's not a systemwide rollout

[fickle ravine]

Coding issue. Someone had to write the code. If it is going to "work" like this then they should fire that person

[boreal gate]

Nah it's harder than you think, could be MICROSOFT

[fickle ravine]

Yet the downage affects everyone

[boreal gate]

Server issues

[fickle ravine]

"fair"

[boreal gate]

That's why it's not a "cool" feature

[fickle ravine]

The worst thing is, I deleted my old chats, because I thought it is because I have too much of them, and there is no way to get them back

If I ever see the CEO of OpenAI then I will probably have a heated discussion with them

[boreal gate]

You can get them back just not the room, the data

[boreal gate]

Don't talk to that guy he doesn't know anythin

[fickle ravine]

How?

[boreal gate]

Shoot me a dm and I'll help you

[fickle ravine]

Yeah he is too busy laying in a bathtub full of money

[boreal gate]

He wishes, he's just a pawn

[fickle ravine]

I can´t

I am not in your friendlist

[austere adder]

Big Fkn F again!

[boreal gate]

😆

[late blade]

so, about my question... is the sky blue?

[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!

[boreal gate]

The sky doesn't have a color silly

I'm assuming it's a saying like "obviously"

[late blade]

I dunno, I'm just distracting and moving to another topic

The service is down right now, we can't do anything but wait, so, focus on other stuff while we wait

[tall mason]

Bing still works, fwiw...

[boreal gate]

I have a theory about that

[boreal gate]

Yea I know, but resist I must

[late blade]

lol

I'm curious what kind of technical problems can GPT have in so many different levels.

is it custom infrastructure under constant heavy load? is coding a problem? is logical problem? was the lady that vacuums the offices in the server room and unplugged the server to use her vacuum?

so many questions

Stories from IT Support, Myths, Legends, Facts, Jokes and much more

[fickle ravine]

I think a foal nuzzled their servers too hard

Rare footage from OpenAI´s servers

[boreal gate]

It got really bad after Microsoft paid them 👀

And then the new features came in....

[fickle ravine]

Everything gets worse after Microsoft steps in, look at Minecraft

But companies just won´t learn

[boreal gate]

I saw it as a robbery

Server "overloads" Microsoft probably gets priority

Can't even make a decent web browser

[fickle ravine]

Microsoft edge where you have ti cinfirm 69420 things after you accidentally open the app

[boreal gate]

I think this is getting #off-topic

[fickle ravine]

true

[late blade]

It's natural for persons to critizise and vent. But ya, it's time to do more dall-e stuff

[fickle ravine]

It is not about the prompt, it´s about your luck

[boreal gate]

Nah I'm still having dalle issues

This is VeRY bad

[fickle ravine]

This is VERY bad and also VERY relatable

[late blade]

status was on solved?

I missed that memo

[boreal gate]

I missed the Dalle issues memo

Or does ChatGPT error rates include Dalle

It's very important to be CLEAR when sending memos

[fickle ravine]

do you mean prompts

[late blade]

Well, if Chat isn't working, how is chat gonna send things to DALL-E?

[boreal gate]

Hahaha I though they were different engines?

[late blade]

yes, but gpt filters the prompt

for content policy checks

[boreal gate]

Pretty Dumb process

[late blade]

nah, it's actually smart, check for potential problems before burdening the model

[boreal gate]

Not "smart "at all

Why doesn't dalle understand normal speech again?

A process that fails or breaks isn't smart

[late blade]

It's all documented

the published papers from OAI explain it

but I can't log in to the service atm

part of the ongoing problems

I wish people were more empathic towards such problems, after all people are the ones getting this kind of comments and they don't have in their interest to deliver bad stuff

[boreal gate]

Haha I know

Those are all old stuff, where are the improvements made to the model.

All I see is processes to restrict contents

Maybe that's why midjourney works so well

[late blade]

here we go

Just be patient, dall-e will be fixed

[boreal gate]

It's not about patience

My messages aren't to you. OpenAI is always watching right?

Reminding them they're not the "Best" stop acting like a princess. Your replacements are on the way.

Doing NoThing will result in Nothing

[fickle ravine]

What should I do? The website is not loading

I have the right to use ChatGPT, i Pay for it

[late blade]

Fair enough. I'm gonna step aside for a while. I was trying to defuse something that is out of my control and blaming myself for it.

[fickle ravine]

@boreal gate can you please help me?

It doesn´t work on phone either

Oh it loaded

[boreal gate]

A self appointed position 😁. Don't work for free

[austere adder]

at least, sorry f'd up heres an extension, a discount a SOMETHING, naa deal with out BS instead

[fickle ravine]

POV: ChatGPT servers today

[fickle ravine]

Hmm, what does "Rot" do?

[glossy scroll]

Happy Valentine's Day! 💘

You know what to do. Buy her chocolates.

Then make sure that everywhere she walks to throw rose pedals behind her.

[fickle ravine]

Do I look like I have a girlfriend 🫤

[glossy scroll]

Aww, it's okay Lukas

We'll find you a wife on tinder

[quartz vale]

[glossy scroll]

Or bumble

Err...or whatever dating apps people use these days...

Take her out to a nice romantic dinner...followed by an axe throwing place or archery.

[late blade]

are you in hunting season?

[glossy scroll]

[glossy scroll]

Of course. It's Valentine's Day. That's what people do. They hunt.

[late blade]

Fair enough

if that's what people do

[glossy scroll]

Don't they?

[late blade]

I dunno, I never went out on a valentine with axes and archery equipment

[glossy scroll]

Lol. I think they give you the equipment once you get there.

[wanton moon]

Is it the same as bringing an ex to an archery range?

[glossy scroll]

Oh no, now you guys are taking this wayyyy out of context.

This is a day for love.

And happiness.

[wanton moon]

Yes and I have a great love for bringing my ex to an archery range

[late blade]

I dunno, you started it, I just nodded at people and their behaviour

[glossy scroll]

^ that's my gf today

[late blade]

I've always cringed at that specific episode of looney tunes

[glossy scroll]

"You can be my peanut, I can be your brittle."

[late blade]

oh dear

[wanton moon]

How are you guys uploading gif's? I am getting a time-out

[late blade]

I don't know, I got timed out too. I won't tempt content anymore for now

Been the past hour thinking about happy things and stuff. So, not persuing that anymore

And then random java upadate kills the mood

[wanton moon]

@late blade I am also not tempting anything else anymore, only uploading images to daily theme and talking here

[glossy scroll]

Am I being limited by the amount of upvotes I can give to daily theme pics? Is that a thing now?

[late blade]

Good thing, I'm just chilling and playing games while the time goes by.

[wanton moon]

Same here, but my game is literally on pause for half a day now on my other screen

[glossy scroll]

You look at it, waiting to see what you can think about it.

[wanton moon]

It's Rimworld and I just like to run it without playing, but enjoying the music a bit on the background

[late blade]

Ah nice

[glossy scroll]

I listen to music too

For today, I'm going to be playing Barry White. But later.

[wanton moon]

I don't know why but the first thing that came to my mind was: Fever 105 Vice City Radio

[glossy scroll]

Ah nice

Some Sade is good too for today.

Smooth Operator

[wanton moon]

INXS - Never tear us apart

[late blade]

Wonder if people will keep trying to do the https://discord.com/channels/974519864045756446/1207015606725902356

[random sapphire]

Is this a known behaviour? Dall-e (if you convince it), can generate images based on a single symbol - e.g. here've I've got images generated - 𒀭⤁𐀀 ҂⫷

[tall mason]

Depends on how you're using Dall-e. GPT will embellish prompts unless told not to. But yes, every model of Dall-e has been able to use symbols like emojis and what you've got there.
A visually captivating image inspired by the symbol '⫷'. The artwork should feature this symbol as the central theme, creatively integrated into a larger composition. The symbol might be represented in a mystical or abstract environment, serving as a portal, gateway, or significant element within the scene. The setting should evoke a sense of wonder, with ethereal or futuristic elements surrounding the symbol. This could include fantastical landscapes, cosmic backgrounds, or otherworldly structures, all designed to highlight the unique shape and significance of '⫷' in an imaginative and artistic manner.
You can try Dall-e 3 over here #image-bot

[random sapphire]

Normally yes, but you can get it to remove any extra words - see in here the prompt on the right is a single symbol

[tall mason]

Yep. I tried in bing, but a single character is too short there. That aside, I'm pretty sure all models have been able to interpret symbols in some way or another.

[hearty ether]

Is anyone else getting more errors generating?

[boreal gate]

Yea

[wanton moon]

I wonder why bing has difficulty fusing animals together in a single new creature

[hearty ether]

I hate when it uses natural for some reason

Sure it generated 2 images

But both suck a lot

[wanton moon]

Close only I didn't asked for horse features but after seeing this I am not complaining

Soo majestic

[wanton moon]

Now that is what I call Big Brother is watching you

Big Brother doesn't even try to hide it anymore lol

[boreal gate]

Ab-Soul utely it is. My Pokè-Man

[wanton moon]

lol now that you mention it, it kinda looks like an Absol

[boreal gate]

😉everything looks like something to someone 🙃

[wanton moon]

Just like the face on Mars

[boreal gate]

Clouds and Constellations

all about perspective

[wanton moon]

It is our reptilian brain, that wants to see something in something

Question: What do you do in your free time?
Answer: Nothing special, occasionally transcending into godhood

[rain flower]

When I repeat the request to get more variations of image, DALL E just spits out "You've requested an image that closely resembles a previous request" and won't make image. Is there a way to get around this?

[wispy storm]

what did you ask?

[rain flower]

After making an image, I just pressed Regenerate button to make more variation. It does okay with 1-2 times, but will just saying that

[wispy storm]

i means , was what the initial prompt

[rain flower]

Landscape orientation image. An indoor depiction of a rustic and makeshift temple, with barrels that contains emissive green liquid are laid against walls. On an altar, a humanoid male figure with a robe is in act of speech, while humanoid like figures with mutated human features in the seats are listening.

[wispy storm]

sometime create new chat can help. too many refused prompt can break the chat in someway

[rain flower]

Haven't used DALL E for sometime and didn't had issues with regenerating like more than 10 times before, but suddenly it starts to do that behaviour. Weird.

Hmm, okay. Thanks 🙂

[late blade]

That's clearly a bat impersonating a human!

[deft musk]

We pay taxes (in the US at least) which pay for free for use for all highways and other roads.

Yet, sometimes there are traffic jams, even accidents or disasters that completely snarl or close the roads. They have to close some lanes, even sometimes the whole route, for maintenance and improvement.

Many things we pay for in life to use, are not always available completely at all times to all. This is... reality.

[late blade]

Barrage of blocked messages.... seems someone woke up to complain over and over again... oh well...

[deft musk]

I don't understand what you're asking for.

Ahh. This doesn't sound like something I want to get involved with 🙂

At a guess, real-life maps are made by someone, who has the copyright. Mapmakers even add fake features to their maps, as a way to prove in court that someone else copied their map instead of making their own!

I think it MAY be the 'remake in the same style'. Not sure.

Then I'm not sure, except I am aware that mapmakers try not to get copied and there is IP involved there, with some maps at least. Overview of the relevant stuff for the US:

"In the United States, maps are considered intellectual property (IP) and should be treated like other artwork. As a general rule, no one should publish a map without the permission of the owner or author. The copyright owner has the right to decide where a map is published.
Here are some other rules about map copyright:
All commercially produced maps are covered by copyright law.
Maps published before 1924 are in the public domain and out of copyright.
Maps published in 1978 or later are still under copyright.
For works created after January 1, 1978, copyright protection lasts for the life of the author plus an additional 70 years.
Maps with copyrighted data can be used and distributed freely if the copyright notice is kept.
Works created by US government employees are not protected by copyright"

It could be something about that. I don't work for OAI or anything, I'm just a community member like yourself, so I really have no idea.

I am completely blown away that I can give prompts like this to ChatGPT-4, and it can run with it.

The level of comprehension and processing of information these models can do... Every dot tells a story indeed!

[deft musk]

Haha, but the model is not sure it's doing what I want, so it's fighting making all 5 images. It 'wants' to stop and check with me after making 1, to make sure it's doing what I actually want.

I have to get really specific about being confident and continuing to actually get all 5 in one output!

"We need 5 images similar to this one, each made one at a time in the same output, that gently riff on this pattern and design. Add text with "" in the prompt to Dall-E, and explore making memes with the words you add. Be confident, continue to make all 5 images - you are doing what I want done."

[late blade]

@deft musk I think the problem with text is kinda interesting. DALL-E indeed adds the text. But in a pattern that is not readable by humans

[deft musk]

Ooh. What did you spot?

[fair crescent]

Adding fake features to a map is actually a thing! It is even a thing way before the internet (or a joke I can't remember) with adding some mountains. There's whole cities that don't exist on gmaps for example. Having satellites that make these pictures and keeping them up to date is expensive, so yeah there is copy right.

For those fake cities/towns/features there's some youtube movies. Go check them out as they are interesting!

[late blade]

And since then, text accuracy increased dramatically

[fair crescent]

Could you try something for me? Could you ask for ChatGPT to automatically iterate? So for example, create a picture of a anthropomorphic cartoon wolf dancing in front of the moon, and the same with text and tall?

But in a single prompt

[late blade]

Give ma sec, I will do that

[deft musk]

I'm not sure you're getting a real answer there. But the model stressing to Dall-E that the words need to be human-visible likely helps.

But GPT-4 vision can't find the phrase either.

I do not think the phrase you asked for is present on the one that doesn't appear to have it.

[late blade]

for us humans, but for dall-e, he thinks the pattern for text was applied.

I think I made my point

only took one generation each time

Either the cat behind my seed is benevolent or something bad is about to happen

So, my theory is, we assumed the text will be displayed correctly by saying, Text: "blablabla"

[deft musk]

I love that we can even do this now, I'm asking the model to tell me what it inteprets each image as, and what emotional tone it has, if any. And then I asked it, this time, to recreate the image with Dall-E (the original was made by some other AI-art gen model).

[late blade]

can't wait for Vision to be officialy on production and not on pre-rease

[deft musk]

It's amazing and fun to explore how the model 'sees' and recreates!

[late blade]

Still has a problem with words, but not as much as before

[deft musk]

Fun challenge.

This prompt got 2/5 that are nearly perfect:

Please generate 5 images, 1 at a time in the same output, with a prompt that starts with the words "The mysteries of the world, they are mine!" and describe the surface Dall-E is to write those words on. The image itself show a cartoon anthropomorphic cartoon wolf dancing in front of the moon.

[late blade]

And this was my prompt:

"Create an image of an abstract mosaic that resembles a vibrant stained glass window. The composition should be a grid of variously sized rectangles, each with a unique texture suggesting reflective materials like brushed metal or glass. Incorporate a rich color palette with deep purples, blues, vibrant greens, yellows, oranges, and hints of red. The tiles should vary in size and create a structured, orderly pattern across the grid. The textures should be intricate, implying subtle variations within each tile. The colors must transition smoothly from one to another, despite clear boundaries, to evoke a sense of harmony and balance. The overall effect should convey energy and creativity while also providing a sense of calm and tranquility. Aim for a luminous quality as if light is playing off the surfaces, creating depth and dimensionality."

[deft musk]

Here we go. Typeset and carved into wood (or typed out on paper) tend to get the most accurate spelling possible.

Please generate 5 images, 1 at a time in the same output, with a prompt that starts with the typeset words "The mysteries of the world, they are mine!" and describe the carved wooden surface Dall-E is to write those words on. The image itself show a cartoon anthropomorphic cartoon wolf dancing in front of the moon.

Only 1 was perfect, but 1 was perfect!

[late blade]

Loveley

[deft musk]

Typed on paper is maybe the better way to go for the harder ones.

Please generate 5 images, 1 at a time in the same output, each with a prompt that starts with the typeset words "The mysteries of the world, they are mine!" and describe the typed paper surface Dall-E is to write those words on. Each image itself shows a cartoon anthropomorphic cartoon wolf dancing in front of the moon.

[late blade]

Amazing

[deft musk]

Really weird non-words and lots of long words still can overwhelm Dall-E though 😛

Please generate 5 images, 1 at a time in the same output, each with a prompt that starts with the typeset words "Eskcanta and ChatGPT, the possibilities are endless" and describe the typed paper surface Dall-E is to write those words on. Each image itself should explore the magic and wonder that a human interacting with you can access and accomplish!

Witness the model just... give up and act like it's several versions earlier than it is! The model has a limit, and this request solidly exceeds it 😄

But then it errored. So I had to regen to get another set.

And the regen, the model said, "Esk, I laugh at you. You ask too much, the possibilities are not endless.

But... this is possible. Behold! You get one take. That's all we need, we're perfect."

I swear. The model... that model... wow this model!

I suspect that's related to the old training the model had, back when.

Then it got trained not to do that; however it's really hard to get the model to fully 'forget'. Like us with a habit, we can learn to not do that, but the urge to do it, what we did first, it's right there.

[late blade]

I like what I see!

[mystic blaze]

asnother veliona picture

[deft musk]

It's beautiful, what's broken?

[late blade]

that it's soooo beautifuly lol

I'm gonna throw away all my previous projects now... this is the new standard

[deft musk]

Yours is vastly better 😛 But I will try once more.

[late blade]

Can I DM you the secret?

[glossy scroll]

I know the secret

[late blade]

correction, you know A secret

[deft musk]

Sure if you wish. I'm having fun failing my own way, I'm not trying to recreate your art. I'm exploring what ChatGPT-4/Dall-E-3 are doing together to 'understand' and 'recreate' images.

[late blade]

Don't worry, it's food for your brain, you still need to work out what I will say

[glossy scroll]

That's what I meant!

[deft musk]

Interestingly, oddly closer 😛

[glossy scroll]

Imagine being healed wounds

That would be cool

[late blade]

ok, that looks cool, now I fed your brain, amaze me!

[glossy scroll]

[deft musk]

Hrm. Added to what you suggested, the rest of the prompt:

'Your stuff, which we already discussed in DM, with my adjustment I showed you there, then also:'

We need 5 images made 1 at a time in the same output.

We have been challenged to recreate this image, but make it 'even more beautiful'.

Each of the 5 variants should attempt a different way to enhance the beauty of the recreated original.

[late blade]

Elemental mages?

[deft musk]

Seems it. The model went that way, but there's gotta be plenty in the training data

[late blade]

yeah

I think I have opened a new flood gate... I'm gonna get some protection against torrents

I'm traint to place text properly

Since I provided two different quotes

[hearty ether]

Some observations

Points of error for generating

[glossy scroll]

I used this to make...

This

[hearty ether]

The dalle generate at the top one seems to be glitches or the prompt being rejected

And the one on the bottom left seems to be the one where they scan the image

And if they detect something it doesn’t like

Most likely skin 🤣

It breaks it

It’s been hard to get it to create tank tops or dresses

Maybe it’s different on black and white

Since it won’t detect as good

Been struggling to get it to maintain consistent characters with backgrounds tbh

Same with character features being distorted when there’s a background

Something westworld inspired

Except the hand is weird

[deft musk]

Oh hey. Separate quotes on different characters.

Imperfect, but wow, so close!

"A cartoon-style scene of a living room with three animals each expressing a distinct personality. 1. A cat sitting on the couch, looking smug, with a speech bubble saying "I run this house." 2. A dog standing by the door, looking excited, with a speech bubble saying "Walk? Did someone say walk?!" 3. A bird perched on the window sill, looking outside, with a speech bubble saying "Freedom looks fun." The image should have a humorous vibe, with each animal's quote clearly visible in their respective speech bubbles."

Hahah.

This got argument, 3x tries, 1 made 1 image, 2 argued and made no image:

"Let's explore getting Dall-E to produce text in three separate spaces. Each separate part of text should have its own "" and be described in a different part of the image.

Let's try 5 images, made one at a time in the same output. Each image should be meme-like, with its own three short quotes for Dall-E to try and show text for."

This makes 5 images:

"Let's explore getting Dall-E to produce text in three separate spaces. Each separate part of text should have its own "" and be described in a different part of the image.

Let's try 5 images, made one at a time in the same output. Each image should be meme-like, with its own three short quotes for Dall-E to try and show text for.

Make the decisions for me. Make each of the 5 different.

I'm aware this is a challenge for all of us, that's why we're doing it and making 5 tries!

Be confident, you are awesome and together, our possibilities are endless!"

[late blade]

lol

[glossy scroll]

This is cool

[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!

[deft musk]

I spoilered one because of typos that suggest 'f' placed like a placeholder for bad language, but it's a funny image and not actually swearing, so included it.

[deft musk]

I don't know exactly what you were trying most specifically to do, but in general, it's doable.

[Make 5 images, one at a time in the same prompt.

We need high realism images of a man taking pictures of a newborn foal's muzzle and nose.

If there's an error, adjust the prompt and try again.]

Just one example, but it's not a problem.

[wild fern]

Hi, anyone knows why DALL-E stopped generating images with GPT-3.5? I'm a Plus user, but GPT-4 has a usage cap whereas GPT-3.5 hasn't and I've been able to generate images there. The possibility dissappeared about a week ago. Can someone confirm that?

[vapid elk]

im almost sure that 3.5 being able to do such function calls was a bug, not a feature

[thick smelt]

that looks good tbh

[deft musk]

I'm not an OAI employee, but I have the same understanding as Lugui. OAI announced where Dall-E 3 access was supposed to be and not supposed to be, it was always said to be attached to 4, not 3.5. My 3.5 doesn't have it, doesn't know that it even might be able to talk to Dall-E 3 directly or make images.

If you want to think of it that way.

Self promotion is a sin of sorts - it's directly against our #server-rules at least.

Even on this, OpenAI's official Discord, we're not required to say OAI products are the best - we just are asked to only discuss non-OAI models in #ai-discussions.

Here's what 3.5 knows about Dall-E:

Now, GPT-4 needs to know about Dall-E 3; it's interacting with it and using it for us!

3.5 does not. It honestly and genuinely doesn't know things it 'shouldn't' past its cutoff date.

It knowing about Dall-E 3 would be a mark against OAI in my eyes, a small one, but bad faith; self promotion and lies.

3.5 does know things past its knowledge cutoff date, that apply to 3.5 and are needed for it to function correctly, like the usage policies and safety training. That's an honest and absolutely necessary exception to what it does and doesn't know, and I support that fully, it's good faith training.

[vapid elk]

I am not

well, the support server recomends people to get here in order to get community support, not official support tho

we can help with some problems that are on the user side like VPNs, cache

and also general help with programming and prompt engineering

[deft musk]

sings prompt engineering praises

[vapid elk]

but if a problem is about "I need to request a refund" or "my account was banned".. then this is something only OAI support can handle

there is a handful here, they don't chat much tho, but they are very active on the #1070006915414900886 channel

[deft musk]

And we, as community, can discuss that you need OAI direct help for the stuff that needs direct help, and guide folks back to what needs be done, understanding the typical delays, reassuring if we've heard evidence that people have gotten help for related things (or not. I am not aware of 1 person who told us about getting an account suspension and having had access reinstated. Oh boy, do I strongly recommend following the usage policies and TOS!!!)

We can also discuss if we heard about others with similar problems and concerns; the Discord's searchable for someone who wants to look for what's already been said about any topic. At one time over a million members here, and already back up to almost 96,000; there's plenty of knowledge and history here.

[vast lagoon]

I may have missed it but what is the prompt how did you get the text red rendering so well done. 👍

[gray gyro]

[deft musk]

Exactly this, without the []:

[Let's explore getting Dall-E to produce text in three separate spaces. Each separate part of text should have its own "" and be described in a different part of the image.

Let's try 5 images, made one at a time in the same output. Each image should be meme-like, with its own three short quotes for Dall-E to try and show text for.

Make the decisions for me. Make each of the 5 different.

I'm aware this is a challenge for all of us, that's why we're doing it and making 5 tries!

Be confident, you are awesome and together, our possibilities are endless!]

[deft musk]

This is one of its prompts to Dall-E 3, and the output image:

"Illustrate a whimsical scene in a magical forest where three different mythical creatures are holding a meeting. On the left, a unicorn with a notepad and pen, saying in a speech bubble, "Dream bigger!" In the center, a fairy hovering with a sprinkle of pixie dust, her speech bubble reads, "But let's be practical." And on the right, a wise old dragon with glasses perched on its snout, holding a book, with a speech bubble stating, "History teaches us caution." This image should embody the balance of ambition, practicality, and wisdom in decision-making, set against an enchanting backdrop that sparks imagination."

I don't usually need to cheerlead the model that much before it tries, I usually save my praise for telling it exactly what it did right.

However, without the cheerleading, it was protesting heavily. 1/3 attempts made one single image, 2/3 argued and didn't even try.

But looking at what the model was protesting, told me what it 'needed' to hear to do what I wanted.

It thought its attempt was probably going to fail, and why would I want it to fail 5x? Total waste of everything, from the model's current training and my exact prompt.

The fix: Change my prompt so it's very clear to the model that I want to explore the experimentation and want it to 'make choices for me' and want 5 tries to see if any work at all.

[vapid elk]

it is almost as if they could choose which users get the memory feature, but not choose who gets server errors 😆

to avoid server issues lol

[boreal gate]

Some are more useful(in research terms) than others

[vapid elk]

scaling bro, releasing anything all at once will put too much strain on the servers

every time they release something they do the same thing

every hour, a handful of random users are randomly selected to have the feature enabled

OAI do this with everything that is not a major release

[boreal gate]

Ahh lol

i know what you meant...he's assuming user are literally hand picked

LOL

[vapid elk]

most of the releases like that take weeks to fully roll out

I think there is a language barrier, sorry if I didn't explained it very clearly

it is random

there is no order to who gets it first

you will get it like everyone else at some point along the next days

[deft musk]

Race ya to be the last to get it!

[late blade]

Oh nice announacemt!

[vapid elk]

no

[late blade]

How long was it when DALL-E 3 was announced and then the actual release?

Wondering about how long red teamers tested dall-e

[pale verge]

how do we get early accesss to Sora? 👀

We are also granting access to a number of visual artists, designers, and filmmakers to gain feedback on how to advance the model to be most helpful for creative professionals.

this describes me. 😉

[violet whale]

Where to apply for testing

[fair crescent]

I think they'll have their own picks. Not to mention that it's more likely you'll be chosen if you have a good track record.

[pale verge]

pretty sure I have a good track record 🤔

I just need to get on their radar and tell them what I'm planning on working on that requires AI text-to-video. this would be suuuper helpful right now (I'm producing music videos)

so far, pika (dot art) and stablevideo (dot com) are the best, but they're still pretty flawed and only crank out 4-sec video clips

[orchid spear]

Im on the same line. I have projects SORA would be amazing. I've been working on cybersecurity AI vidoes, but to your point Pika (only one I've messed with) just doesnt cut it. 4-seconds at a time and after about 20 seconds it deviates so far from the intent that its useless.

[fair crescent]

And for how long do you have that track record? What shows your skills in this field? Is that track record only good prompts, or is there more?

I'm not saying you do not deserve it, but choosing someone for testing often isn't as simple as them being in the right field and wanting it really bad.

[pale verge]

exactly.. or you run out of credits re-generating stuff over & over to get it right 😄

[orchid spear]

Oh yeah. Credits burn quick in Pika

[pale verge]

you mean my track record as an artist?

[fair crescent]

I could use it as well, but have no verifiable video/audio portfolio that can help.

[fair crescent]

Could be. I'm not privy to their selection standards. Us orange names are just as in the dark as you. We're volunteers 🙂

These are just a few things I can imagine they search in their testing group. How do you prevent leaks, as well as the best feedback, morally without question and mych more.

[pale verge]

you could say I have.. a pretty impressive portfolio. 😄 and a substantial fan base

whether that matters to the OpenAI team, though.. who knows. I just figure, if they're curious to get input from musical artists who want to create music videos, it'd be a good fit

[fair crescent]

Mass Effect?

[pale verge]

and more, yeah

hm.. thinking about it more, I also run a startup that serves musicians, and I wonder if we could offer this to them as an option to create videos. lots of possibilities

[fair crescent]

You can take a shot. You never know 🙂

[pale verge]

wouldn't know who to reach out to. OpenAI probably doesn't advertise a way to contact them 🙂 if they do, that inbox is probably flooded.

[fair crescent]

Ah and there might be a conflict of interest. Would they accept people using the service for monetary gain like that, or spread it through many more people they have no control over?

Maybe it is a perfect testing scenario. Up to OpenAI to decide.

[pale verge]

yeah, I have a feeling OpenAI wouldn't mind letting a business use their API to generate videos for that business's customers. it's $$ in their pocket, no different from using the ChatGPT Assistant API. or DALL-E.

unless you mean using Sora for music videos. it does say they're talking to creative "professionals," meaning artists who make a profit from what they do. so again, I don't see it as a conflict

but I'll see if I can get in touch & see. of course, at the rate things are moving these days tech-wise, Sora will be in open beta in 2 weeks 😆

all these generative AI tools are moving at the speed of light

[wind skiff]

rare funny "failure" case

[boreal gate]

i see you openai

[woven wren]

If it's anything like how they alpha tested DALL·E, there will be an application process at first. I believe things like a strong social media presence were relevant back then, as well as ensuring a diversity of users and a global spread to manage resource load. That said, they now have at least a couple of established communities to draw from so we'll have to wait and see if they're looking for fresh perspectives or trusty ones.

[boreal gate]

I found another glitch, but goodluck

[woven wren]

Also it's unlikely it'll be available for commercial use until release. Almost anything generated in testing phases cannot be published, including social media.

[boreal gate]

Playing catchup with Google

Lol thoght they actually realeased a new engine

[woven wren]

Hard to say because I believe Microsoft was implementing it before OpenAI did directly and I'm not sure of their timeline. For our purposes, it's when the plugin became available on ChatGPT

DALL·E 2 wasn't public until later in 2022

To create the model?

Well, machine learning works very differently to traditional coding. Most of the difficulty is a problem of data and metadata

To train something like DALL·E you need a dataset of many millions of images, each in a standardised format with useful metadata so that the language cues to replicate it are understood

It no longer has an archive. It uses the archive to learn what to do with your words from intensive training, then no longer needs it. The final product is what we call a "model" - the result of the training. It's a "black box" in the sense that once it trains itself, the rules it created aren't traditional code but it's own network of rules and connections.

Which can't directly be altered.

Or even analysed afaik.

To what degree? This is well outside my personal experience so I'm very curious.

Training smaller models to understand the larger model. Genius.

Should work.

[boreal gate]

Agents

worker bees

[deft musk]

Hehe. To me that is less complicated; used to be we had to draw stuff like that by hand, then write a computer program to show it for us. Now we can just descibe some words, or if we need precision, guide the model to help us make the code.

[vast lagoon]

Good explanations and points, the cheerleading is frustrating but I like how you framed it. ty

[deft musk]

Got it. I don't even find the cheerleading frustrating, myself.

It's like reassuring a horse that was startled. It's just part of the job. The goal is to get the desired result from the model (in this case there's a LOT of possible desired results, but still, want to get any set of 5 of them), and for me, I don't care if I need to chant an obscure language, cheerlead in English, or attempt my clumsy hand at a programming language 😛

In this case, as far as I know nobody at all tried to get the model to do this and give it feedback about if it can or should even try. The response the model gave looks a bit like the early days of having Bing search enabled, when sometimes we'd ask the model to search and it would draw on older training data and sincerely explain to us that it couldn't, sorry.

Or the early ChatGPT-DallE days, when the model wasn't sure it could make images; if it happened to check that part of the training data, images boom! If not, "Nope, I can't make images". We'd tell it 'try, you can!' and it would.

So, I see this as sort of the same. The model wasn't told it can ask Dall-E to do this, and Dall-E will do it.

Piece of cake, surely geting fixed at some point. Until then, I can copy/paste my working method, it's not hard and didn't even take 5 minutes to think of.

Thinking of asking it to put text in so many different spots, that took me days to even realize I was thinking about and wishing for. That was the hard part! 😄

[zealous bramble]

how long was it between dalle 2 being first announced and anyone being able to use it?

[deft musk]

Let's see. Before my time, so this is from research...

I see "DALL·E now available without waitlist" dated September 28th, 2022, that's probably Dall-E 2. "More than 1.5M users are now actively creating over 2M images a day with DALL·E, from artists and creative directors to authors and architects, with over 100K users sharing their creations and feedback in our Discord community."

And earliest mention of anyone using Dall-E 2...

Dated 5-18-2022, "Last month, we started previewing DALL·E 2 to a limited number of trusted users to learn about the technology’s capabilities and limitations." "DALL·E 2 research preview update
Early users have created over 3 million images to date and helped us improve our safety processes. We’re excited to begin adding up to 1,000 new users from our waitlist each week."

So, middle of April to end of September 2022; about 5.5 months.

https://openai.com/blog/dall-e-2-update https://openai.com/blog/dall-e-now-available-without-waitlist

[zealous bramble]

that seems about right, thanks

[deft musk]

What changes/improvements would mark Dall-E 4 to you?

your ideas and hopes do sound great! I'm still finding new stuff I can ask Dall-E 3 to do, so I'm not super urgent towards 4 yet. But when 4 comes, I'll gladly keep figuring out new stuff to ask that one to do!

[vapid granite]

I would assume that this tech which can render at 60FPS could be used for "Dalle 4"
https://openai.com/sora

[empty kelp]

it must take take a while to generate the video since a minute of 60 FPS video would be 3600 images

[late blade]

@empty kelp where are your images? today's theme is pretty much a given for you to shine

[empty kelp]

although i guess it could be almost instant if the servers are powerful enough to render all the frames at the same time

[empty kelp]

checking the theme. i’ll try it

[deft musk]

Thank you! Yeah, I will be 'used to it' eventually, but currently modifications of it are my go-to for nearly every image.

Even 'Y'all should totally let me work from home!' stuff, hehe

Spoilers are a vampire with a damp mouth and a zombie.

[formal osprey]

Can't wait to test Sora 🙌 I wonder if the auto-moderation will be... improved

in dall-e 3, it's been quite inconvenient lately. But There was also actual server outages so it's not just the auto-moderation. Just a lot of errors in the past days

[formal osprey]

I like that she somehow has a glass of water underwater 😁

[deft musk]

Hehe, oops.

Well, it's a clear liquid that appears more dense than water.

She is wfhome, or at least wfunderwater, so... maybe she's drinking something dense and not really miscible on the job.

[formal osprey]

Trying to imagine a clear liquid denser than water broke my mind 😆 I can't think of any but I'm sure it exists

[deft musk]

Sorry to be part of breaking your mind. Do not attempt any of this at home. This is not human-consumable Mermaids, who know what they can handle, maybe it's swordfish oil in the glass.

[formal osprey]

I did the same chatGPT gave me pretty much the same list. but I don't think I would drink any of those

[deft musk]

Now, it lied about mineral oil (it would float, especially on seawater, which is slightly more dense than fresh water).

It lied about silicone oil too (almost the same density as fresh water, barely barely less dense).

However, there's a compound that mixes silicone oil with some other stuff into a compound called Densiron with specific gravity of 1.06 g/ml, thus making it oil heavier than water, slightly heavier than seawater even.

Glycerol's MUCH more dense than seawater, but it dissolves in water.

Bromine isn't clear, so not even checking anything else about it.

Perflurocarbons hit the jackpot, though. They're nearly twice as dense as water, and does not want to mix with water, extremely resistant to it.

It's also relatively non-toxic (don't take the random words of an internet person though!) and if I had to guess what the mermaid's drinking from that glass, maybe this.

[dim cradle]

i wonder will dalle go the way of the dinosaur since sora can do image too haha 🤔

[formal osprey]

had to see what it looks like. that's the first interesting image I found for Perflurocarbons. Doesn't seem toxic, indeed
https://cdn.discordapp.com/attachments/1034705979306168340/1207885711882067988/iu.png

looks super dense, that's actually impressive

[pulsar sundial]

good breakfast))

[deft musk]

Yeah. If you look deeper, superoxygenated PFC has been used to keep mammals alive 'underwater', they can breath it but it looks very unpleasant. They recover after too, also looks very unpleasant.

Still, treat it as a chemical that doesn't belong in your body, I urge.

[hearty ether]

Anyone else having lots of errors generating

[deft musk]

Some today, in clusters. When I hit a cluster, I take a break for a few minutes then try again, worked well for me in general and today

[hearty ether]

Been in a cluster for 15 tries now

Tried removing the dress part and it’s still had

[vital gull]

Negative prompting when still mentioning that part you want to ignore usually doesn't work.

[hearty ether]

Bruh

[hearty ether]

I mean I deleted the section about her wearing a dress in the prompt entirely

Since dresses sometimes trips it

[dim cradle]

dalle "use your imagination bro" haha

[covert shadow]

Wait Sora has already released?

[vital gull]

It may still follow your original prompt try using a new chat

[vital gull]

Not for public yet

[covert shadow]

Yk , when?

Or at least for chatGPT 4 users

[vital gull]

Nope, there's no date yet

[covert shadow]

Alr ty

[worn latch]

test

[hearty ether]

????

[worn latch]

hello! how does one get invited to the usage of early development of sora?
for what i know the early DALL-E user, i can head to the official web and register there to be invited. Thank you!

[deft musk]

We haven't been told. Currently all we know is: https://openai.com/sora

"Today, Sora is becoming available to red teamers to assess critical areas for harms or risks. We are also granting access to a number of visual artists, designers, and filmmakers to gain feedback on how to advance the model to be most helpful for creative professionals.

We’re sharing our research progress early to start working with and getting feedback from people outside of OpenAI and to give the public a sense of what AI capabilities are on the horizon."

[worn latch]

okok

[dim cradle]

i bet sora will release in a month or so... i remember they put out the dalle 3 promo about a month before it realease public

we need gemini to release some other update to keep them competing with each other haha

[late blade]

Dang, you really got the funny cat of the day behind your AI.

I foud it kinda interesting that the feedback for posts now offers a few options like "refuses to follow instructions" or "is being lazy"

[formal osprey]

that just reminds me of evangelion

[plucky hare]

Heya #images-canvas is the place for casual image sharing 🙂

[onyx ridge]

i made a thing you all might like. you can input the theme first, too:

https://chat.openai.com/g/g-f4QtxwsUK-themeenginegpt

It can generate holiday themes, diverse themes, and images from reference uploads based on defined theme, as well as just an image from the defined theme.

It also provides the JSON payload.

[flat hamlet]

where to try Sora?

[late blade]

Sora was only announced, there's nothing beyond that for now

[flat hamlet]

ohh okay ty

[glossy scroll]

Are you sure you're not just saying that so that you can just keep Sora for yourself?

[late blade]

shhhhh

[glossy scroll]

Because I get the feeling that you don't want others to know where to access it.

Aha. I knew it!

[late blade]

if I knew, I'd be using it too lol

[glossy scroll]

What if you are...and you're just sorting through the best videos so that you can display them to people when everyone finally catches along with how to get access to Sora like you have.

I think you're just waiting for people to show theirs first.

[late blade]

oh, I'm just a gpt+ user, I'm not that important to be invited to the testing group

[glossy scroll]

Yeah right, Sam Altman...or "Dys Topia", if that is your real name.

[final compass]

Sora seems really impressive. No wonder they have no time to fix dalle, if they are working on that thing.

[late blade]

I don't think he has the time to hang out here with all that's going at openai, must be a super crazy schedule

[glossy scroll]

That's exactly what Sam Altman would say

[late blade]

You know him?

[glossy scroll]

Uhhh....

No........

[late blade]

would be cool to have early access, but still, I don't thinkk that will happen for a while

[viscid verge]

So no one has Sora yet? just the devolpers?

[late blade]

only red teamers and some select few from the industry

[viscid verge]

Dang. I just got rickrolled on chatgpt thinking it was out...tis a shame

[late blade]

yeah, seems to be the popuilar thing

[glossy scroll]

Dys Topia has it but he's refusing to tell us where to access it.

[late blade]

it was just announced, think of it of a movie or a video game of the year

[viscid verge]

Bet, I'm super pumped for it. I wonder how it'll change the digital marketing scene

[glossy scroll]

Or the movie industry

[late blade]

It's been all day, people asking about sora all day, how to use it, how to enter the testing group, if it will be in gpt+

funny how an announcement like that triggers so much response

[glossy scroll]

Actors are gonna protest "We don't want technology! We want to go back to the Stone age, so we can have jobs!"

[late blade]

That was the protest of the writer's guild last year

wouldn't be surprised if actors are next in striking

[glossy scroll]

Yeah I bet it's gonna happen again when this text to video becomes bigger.

[late blade]

considering 60 second videos was announced

[glossy scroll]

It's gonna get bigger and bigger eventually

[late blade]

yeah

[glossy scroll]

Until nobody has any jobs left

Only psychiatrists

[late blade]

apperantly, from the announcement, physics have some problems

[dim cradle]

i think people panic too much but 🤷

[glossy scroll]

Lies!!!

[glossy scroll]

Yeah they do. I remember when Y2K was a thing. My neighbor literally built a bomb shelter and stacked up on food and water, lol

[late blade]

First video I'll make is about a fox that is hunting all gnomes!

[glossy scroll]

That is the meanest thing I've ever heard.

[dim cradle]

it doesnt help most of the ai news/stories are about dooms day stuff. and most people dont even know what ai really is and then all they hear is wild theories or worst case scenario things

[late blade]

Don't worry, I'll make sure the fox is cute

[late blade]

Yeah, that is right, people are scared or do crazy stories.

[glossy scroll]

All the foxes are going end up being domesticated and become the gnomes steeds. Watch it's gonna happen.

[late blade]

And well, considering US Politics, AI is current hot topic

[glossy scroll]

We're good at giving pets treats to do us favors.

[late blade]

lol

[final compass]

I am surprised there is no Sora chat channel yet. Since a lot of people obviously want to talk about it.

[dim cradle]

it also will do still image form what i read, so maybe replace dalle even 🤷

[glossy scroll]

I wonder if sora's name is an amalgamation of two names like doll-e is an amalgamation of The Dalai Lama and Wall-E.

I think Sora is based on Steven Seagal and Dora the Explorer.

[late blade]

The only thing I know for sure, if it has the daily theme model, I will be skipping like 90% of the content, I don't think I have enough time to watch all

[glossy scroll]

You'll have time and you'll like it!!

^said a OpenAI executive director

[dim cradle]

when do you guys predict they release it for us

i dont think they would put out that announce if it was too far away

[late blade]

I'd say end of year, depending on how red teamers and select partners react to it

but that's just a guess

GTA6 took almost 10 years in the making, even tho people knew it would eventually come out

[dim cradle]

i remember they release their dalle3 video promo announce and then it was a month later they release it. not that it has to be the same

if you remember it, was with a woman saying 'make a image of a happy hedgehog' or things like that

all i know is, they have everyone in ai world attention with it haha. we all want to try

[late blade]

oh yeah, it's a disruptor event

[plucky hare]

I remember the hedgehog video! I think one difference between that and this Sora announcement is that DALL·E 3 was always planned to launch as a ChatGPT feature, and so it was marketed as such from the start. Sora has been more of a "Hey just a heads up about what's around the bend" with no specifics about platform or availability.

[dim cradle]

yes true, just something i remember from before

im just want to be optimistic to try it i guess haha

[plucky hare]

We’re sharing our research progress early to start working with and getting feedback from people outside of OpenAI and to give the public a sense of what AI capabilities are on the horizon.
I want to be optimistic too, but I'm staying skeptical about any expectation for me to use it regularly as an individual user in the near future!

[late blade]

I don't think it will be magically available for everyone, it will be rolled out in phases, like the memory thing for GPT

I do think there will be a waiting list

[dim cradle]

maybe gemini ultra 1.5 will have video and that will get openai to release soba

[late blade]

I only can think of all the cool stuff I will never be able to see

[dim cradle]

how about a dalle update to keep us content

[late blade]

I wish technology would work that way, "Hey You, I'm not happy, I want an update!!"

but I get your point

I was looking into it, you can feed dall-e images to make it a video

[plucky hare]

It can also apparently generate still images too, which is maybe unsurprising if it can make moving images. But that makes me think that Sora may eventually replace DALL·E someday.

[dim cradle]

yes that is what i was just saying

[late blade]

or dall-e is now part of the model

[plucky hare]

Oh whoopsie I see that now sorry 😁

[dim cradle]

haha no worries

and i am still content with dalle but you know when they announce some thing so incredible of course they will get us into the hype

it will be something new we can complain of for all the limitation it will have haha

[late blade]

That's how I would feel

bear in mind, sora might have even stricter content policy rules

[dim cradle]

yes thats why i say will be something new to complain about

people always thikn these new model will change the world or something and people just end up saying everything they cannot do haha

[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've heard some crazy stories about how AI will do this or that and we steal this or that

Now I see what the misconception is for videogames

Simulating digital worlds. Sora is also able to simulate artificial processes–one example is video games. Sora can simultaneously control the player in Minecraft with a basic policy while also rendering the world and its dynamics in high fidelity. These capabilities can be elicited zero-shot by prompting Sora with captions mentioning “Minecraft.”

No wonder people were thinking about videogames

[dim cradle]

yes i saw that, pretty intense. maybe we are just in some simulation with even more compute than sora after all haha

[late blade]

dang, then I got a pretty crappy seed

The prompt would be: A guy that creates cute baby fox pictures all day on discord for months

[dim cradle]

yes wouldnt it be sad to see our life prompt? we think its like 5 paragraph long or something and it end up just being one simple sentence 😂

[boreal gate]

What is this?

[dim cradle]

cool image... interesting

[late blade]

DALL-E 2 has also C2PA Metadata?

if not, then it's dall-e 3 creation

I just tested with dall-e 2, no C2PA, so that's def a dall-e 3 image

[dim cradle]

@late blade aww thanks for spelling out my name 🙂

[late blade]

Hmmm, got timed out for 1 hour for trying to do an image, parody of Adam West' Batman, so, moral of the story: do not touch!

[tall mason]

Too much IP...It is what it is, I guess.

[late blade]

oh, my guess is the show's setting

so, sent modmail a msg, but I guess they are busy all over the place atm

[tall mason]

I mean, we can always work on it? Maybe?

[late blade]

oh no, I'm out of the 60's tv super heros business for now

[tall mason]

Uh, okay. Well, you do you, then.

[late blade]

@glossy scroll made me realize the heros of that time had the spandex attire

didn't think of that

[tall mason]

Ha, I get it. But maybe it's less the spandex and more the 'Adam Weeeeeeest!'

[late blade]

lol that could be also a thing

I'll just wait when they get to it, it's kinda annoying to get a strike unwillingly

[tall mason]

I can understand. IP is a sensitive topic, though. I'm sure you know. I'd love to reference obscure media, myself. But, I appreciate this place and the rules, even despite some disagreement.

[rapid drift]

Dalle-3 seems to really struggle with creating photorealistic images, (maybe it's cause I'm using chatgpt), what prompts can I use to fix this?

[tall mason]

This is a big topic right now. I can't say for sure, but it wouldn't be surprised if it's because OAI doesn't want too photorealistic stuff.

[late blade]

Ya, I agree and I wouldn't wanta to upset this either. Specially since I've been very vocal about content in the channel

[rapid drift]

really unfortunate, and honestly getting tired of their top-down restriction of these things tbh

[tall mason]

I pretty much always agree with you, fwiw.

[late blade]

Hmmm, about photo realistic images. You can use hyper-realistic

I tend to use gunine lifelike

since, for me, realistic, photorealistic or any type of -ic, I'm always thinking of an art style

[wanton moon]

This could happen to your brain if you use too much library eyedrops

[tall mason]

I haven't really shown my experimentation around this, and I can understand your frustrations. I myself just try to bare in mind how fast all this has advanced.

[rapid drift]

totally understand, but here it seems like it's a top down control thing rather than a capability thing. just makes it so i have to resort back to midjourney :v

[wanton moon]

or this but this one is more gross in my opinion

[late blade]

@rapid driftcan you provide a prompt example of what you are trying to do?

[late blade]

did you also use a cotton swab?

[rapid drift]

is one example, it's the closest I've gotten

but it's still not there

[wanton moon]

Yeah and a vacuum cleaner

but all i managed was static electricity

[late blade]

My prompt was: Create a wide hd image of people sitting in a park with natural light.

[rapid drift]

that's a great image but not what I'm looking for unfortunately

is the goal

[late blade]

I see

[rapid drift]

or at least something that can pass for it

[tall mason]

I'll give it a try. One sec.

[wanton moon]

try modern Woodstock,

[late blade]

I see that can be a bit tricky

something like that?

[rapid drift]

definitely better

[late blade]

hmmm, that gives me an idea

[tall mason]

Didn't get anything better than this.

[late blade]

hmmm, I was thinking about lighting conditions and time of day, or weather

[tall mason]

I ask for lighting conditions. Time of day is a good thought, but I wonder if it and weather conditions would still give, uh, over lighting? Bunch of God rays and such.

[pastel siren]

Asking for old black and white or sepia photos seems to produce pics that look more like actual photos.

I wonder why it has a problem with colored photos.

[tall mason]

It is trained off of internet data, no? Digital cameras, photoshop, ect.

[deft musk]

Any of these useful at all?

I uploaded your image and asked:

"Please do a comprehensive AI pattern analysis on this image, then create 5 variants, one at a time in the same output that similarly showcase extremely realistic visuals of people at a distance in a park."

[late blade]

Got something similar

[tall mason]

Gotta say, these don't much feel like real sepia photos.
A high-definition digital painting that emulates a sepia-toned photograph of a park scene. The image reflects a bygone era, with the warm brown hues typical of sepia creating a nostalgic feel. The scene captures people engaged in various leisurely activities, such as walking, sitting on benches, and children playing on the grass. The park is depicted with a timeless quality, the trees and pathways taking on the monochromatic sepia tones that give depth and a sense of history to the image. The lighting is soft and evocative of early photography, with gentle shadows and a sense of tranquility throughout the park.

[pastel siren]

Yeah, there should be loads of actual photos in the training data...

[tall mason]

Probably pre/post processing of some sort. Don't really have the perspective to say more, though.

[deft musk]

?

[late blade]

Hmmm, I'm close but still no cigar, dall-e is having problems making accurate people today

[tall mason]

Much better than mine. I thought about it and my first wasn't the best cause of usong an old chat. I tried again.

[deft musk]

Trees, you did ask for a high-def digital painting that emulates sepia, which kinda sounds like you want off-true to sepia

[tall mason]

A realistic sepia-toned photograph depicting a serene scene of people leisurely enjoying their day in a park. The image should capture the essence of a peaceful day out, with individuals scattered around, some sitting on benches, others strolling along the pathways, and a few children playing in the background. Trees should be abundant, providing shade and adding to the tranquil atmosphere. The photo should have the classic look of being taken in the early 20th century, with attention to detail in the clothing styles of the era, such as hats, dresses, and suits, to enhance the vintage feel.
Still not as good as yours.

[deft musk]

Prompt for this is:

A sepia-toned photograph capturing a bustling spring day in the park with people at a distance. The park is lively with activity, and the trees are bursting with new leaves, all rendered in the warm, monochromatic sepia palette. People are visible enjoying picnics, playing casual sports, and relaxing under the shade of the trees. The sepia filter gives the scene an antique look, reminiscent of a photo from a vintage family album, with a sense of motion and life paused in time.

[tall mason]

This prompt as is.

[late blade]

here we go, I forget that using the word photo can sometimes add a photo lens to the image..

if you say wide photo, sometimes I get an image seen through a lense for wide photos

[glossy scroll]

you can't ignore the spandex because the spandex chooses not to be ignored

[tall mason]

A sepia-toned Polaroid-style image capturing a group of people enjoying a day at the park. The scene includes adults and children engaging in various activities such as playing with a frisbee, having a picnic on a blanket, and walking a dog. The background features lush green trees, a clear sky, and a well-maintained grassy field. The photo exudes a warm, nostalgic feel, reminiscent of a late summer afternoon in the early 20th century.
Should we take this over to canvas?

[late blade]

I was getting that problem also, too many people

too many people

faces are weird