images-discussions

[hearty ether]

Which is kinda dumb

[agile peak]

Here’s mine using Dalle•3 generated in full portrait resolution.

[vapid spruce]

I tried this and another picture, am I missing something?

[agile peak]

Some images are suitable for depth mode. But looking at your case.. it seems to be too much element near the lion’s head.

Try another variation with different gen_id.

Had a nice storytelling with GPT-4 today about Norse mythology. It created me the Ragnarok event + asked to use modern settings.

[vapid spruce]

What is gen_id?

I have the prompt it used for the lion

[hearty ether]

Awwww

The artstyle changed tho

[hushed creek]

I feel like the content policy has gotten even more strict the last couple days

[hearty ether]

yeah

its rather random

this was what i was trying to make before it ran into the violation

imo for the diversity they should make it more context sensitive with eye color and hair

[wraith pumice]

19 tries until i got an image generated 🥳 yesterday and today.... its getting so annoying it takes so long to just get ONE image
when will this get fixed?

[dim cradle]

Good question. Understandble with the traffic they are getting but it's been 10 days now since Turbo release, you would hope some improvements in these things but I guess not

[spice horizon]

Bruhhhh...My prompts tell it "Use the entire 16:9 space" and "Make it one full cohesive image" my GPT instructions tell it the same thing multiple times. I feel like it's just F'ing with me now by adding side bars.

[formal osprey]

I'm guessing this is tricky to resolve. It's weird though because DALL-E in ChatGPT fails a lot more than DALL-E with the API. And it's annoying because the errors counts as image generations. So it affects the rate limit

[plucky hare]

It can't see its own images, and often hallucinates about its own functions and abilities. Maybe try using more "wide" language instead of trying to specify a certain aspect ratio

[dim cradle]

learn to speak wide bro

[formal osprey]

oh, about that. a trick is to copy and paste the image so it can see it

now that it's multi-modal, it can see images, it's just not analyzing the images from its own output

[plucky hare]

Currently, its ability to see is related to how vision works, which is to say, it converts an image to text, and then you can use that text to generate further images. No img2img yet

[formal osprey]

I'm curious how GPT-4V interpret images. We don't know that much about the specific technical mechanisms by which GPT-4V analyzes images, such as whether it converts images to text for analysis.

We know it's multimodal, but we don't exactly know how it works under the hood afaik

[plucky hare]

They confirmed that it's currently img2txt2img during the AMA yesterday #dall-e-ama-answers message

[hearty ether]

yeah they really messed with the filter

its way too hard

i cant even specify character race without it erroring out on me

i like diversity but the characters i have are white in this current story

and its own policy is to specify descent when generating

it keeps making them asian 🤷‍♂️

now the issue is bedroom 🤣

fine i'll go living room

🤦‍♂️

[hearty ether]

Why did it make them kids

What…

I swear it’s getting harder and hard to control now

[agile peak]

Why does dalle•3 love to make thor from marvel look 😂 even though you mentioned classic and mythological lol

[hearty ether]

copyright 👀👀👀

[agile peak]

Weird. Whenever I used thor in classical painting it used marvel look.

[hearty ether]

Also did it accidentally make brad Pitt here

Or just me

[agile peak]

Like this example

[agile peak]

Hmm there’s seems no other way to avoid creating thor and loki in their mythological image.. even with detailed prompts of their appearance

[hearty ether]

Yeah…

Hopefully they do forgive all the warnings rn tho

Since it’s so many false positives

[wise solar]

lol yeah, because you know those yellow ducks are so inappropriate ..

[agile peak]

Tried default GPT it even make them more like marvel lol

[hearty ether]

🦆

[agile peak]

Lmao I give up. Tried gpt to rephase my prompt didn’t worked out

Thor’s red beard and weathered armor, along with Mjölnir’s intricate detailing, remain true to their mythological accuracy, while Loki’s cunning expression and serpent-like features capture his enigmatic persona.

It produced marvel Thor look with chris face just with red beard haha

[dim cradle]

hmm maybe try Norse God and hope to get lucky?

I have a custome gpt to create specific images. But after 11 pictures it said i reached the cap?!?

[agile peak]

Got it.

I used the old English of their names

Using thor will always makes dalle producing it with chris hemsworth face.

[dim cradle]

What do you use the get photo-realistic images?

[agile peak]

If you use the modern name like thor and loki it will be like this

[plucky hare]

If you want images that look like photos, just ask for photos, photographs, professional headshots, nature photographs, etc.

If you want something with realistic features, but doesn't look "real", ask for photorealistic, ultra-realistic, etc.

[glossy scroll]

That honestly looks awesome. Chris Hemsworths face is spot on

[dim cradle]

Yeah i was sitting here thinking how good that looked too, evne if not what was wanted

[agile peak]

The other images 😂 dalle used christ hemsworth face to both loki and thor.

[dim cradle]

You could even argue I think in terms of oil painting you could not get better than that. It's wild to think that for only Dalle3, not like 10 years to get there. If it could be more coherent and have control of x-subjects, you could truly make anything a masterwork you can imagine

[glossy scroll]

That's awesome too.

[haughty copper]

Does open ai reduce step size on dall-e in ChatGPT at high demand peaks? And can this be decreasing the ability to output text

[agile peak]

And finally. Solved to produce thor and loki that’s not marvel looks!!

[hearty ether]

What about female Loki 👀

I mean technically shouldn’t they be able to switch if I remember correctly

[dim cradle]

well Loki is a trickster who knows what form he might take haha

[shut niche]

Scroll up to last night. I demonstrated the differences that Solbus just explained with multiple images. I hope that helps. 😁

[agile peak]

Like this 😂 I used their old english name [Female Lopt]

[hearty ether]

Nice

[bold sentinel]

attack on titans

[pastel halo]

how do I get dalle to generate images at a specific aspect ratio?

[plucky hare]

You cannot, your current options are the preset square, wide, and tall ratios. And tall sometimes bugs out as a rotated and mirrored wide image

[vapid spruce]

Is there a way to make pictures of tv shows actors? It draws them but their faces are not that precise... maybe giving him actors' pictures?

It really was almost there... the last person is different from the original actor 😦

[plucky hare]

DALLE is trained against making images of public figures like actors.

[vapid spruce]

Well, I almost got it, so it's ok

[plucky hare]

You can definitely get it "accidentally" just by including enough words that describe the role/character. For example, there are a bunch of examples just above your message showing Thor that tends to look like Marvel's Thor. It'll just stop you most of the time if you ask for someone by name, as that's against the content policy.

[vapid spruce]

I literally told it the names of the actors (I mean, the characters' names, not the actors)

[plucky hare]

That tracks with that Thor example I offered! It probably wouldn't work for all character names, like really specific ones like "Ash Ketchem" or "Spiderman" that only refer to one specific fictional character

[dim cradle]

maybe more minor stuffs can get through

[rose laurel]

Dall-E is included when one buys GPT4 right?

[paper jolt]

When one buys ChatGPT Plus

DALL·E 3 is included in GPT-4

[vapid spruce]

Yeah, everything is included. Data analyzing, Dall-e, etc..

[rose laurel]

Perfect

Thanks guys

[paper jolt]

🤘

[rose laurel]

Is there still a limit on ammount of questions with GPT+?

Sorry for the questions lol

[paper jolt]

40 at the moment

[rose laurel]

Seem to remember 25 or 50 question limit

Ah ok

[paper jolt]

Though purchases for ChatGPT Plus are being waitlisted

Too many users from Devday

[rose laurel]

Fr?

Bruh was about to buy it to use with my thesis research

[paper jolt]

Yeah 😔
Sam Altman's most recent tweet

we are pausing new ChatGPT Plus sign-ups for a bit :(

the surge in usage post devday has exceeded our capacity and we want to make sure everyone has a great experience.

you can still sign-up to be notified within the app when subs reopen.

[light violet]

big RIP

[rose laurel]

My master thesis topic is legit the usage of AI

and posibilities/limitations

[paper jolt]

There are many testimonials online but I understand the inconvenience

[rose laurel]

eh well its fine i guess

dont need it for another couple weeks

But would be very nice to have it start of december

[paper jolt]

I don't think it'll be too long

But that's just conjecture

You can use the API if need be

[rose laurel]

Hopefully not

[rose laurel]

what is actually the APi?

i have heard people talk about it before

[paper jolt]

A way to interact with OpenAI's models through code if you will

There's a ChatGPT-esque interface here
https://platform.openai.com/playground

It requires attaching a payment method and purchasing credits

Pricing for usage is here
https://openai.com/pricing

[rose laurel]

ah ye that stuff

the playground

considering the ammount of stuff i got for it

prob easier to collect it up and just wait for the GPT+

[paper jolt]

Best of luck on your thesis 🫡

[rose laurel]

thanks man appreciate it

Might use DallE to construct a front page actually

could be a fun twist

If anyone wants to give it a shot im game atleast

till i can get my own GPT+

" limitations and possibilities of AI as a tool" No clue if that even works as a prompt, but would be fun to try

[agile peak]

Perfection

[dim cradle]

incredible maybe one of the best ai arts ive seen

i wonder if you could make it photo real at all. that is what really intrigue to me. get some fantastical things in photoreal is almost jarring

[formal osprey]

well, that's not what I meant. We know DALL-E works using text prompts. We know GPT-4 is communicating with DALL-E using text. But we don't know how GPT-4V is analyzing images. Is it using a convolutional neural network to directly interpret images, or does it have a model that transform the image into a list of segments listed as text to then pass it to GPT-4 as part of the input?

We don't know the specific implementation but I would guess it's not a simple img2txt model queued before calling GPT4.

[dim cradle]

and I guess I better not complain about 2 dalle pictures or 1, im just thankful not to be on the waitlist. but at least we see the people want AI which is good haha

[plucky hare]

Did you see what the dev wrote in that AMA response?
"It's important to understand that currently, DALL-E 3 and GPT-4 (with vision) are separate systems. GPT-4 sends commands (as text) to DALL-E 3, including information about images it perceives. Details in the image that are not easily conveyed with text won't cross this barrier very well, which is why you won't always get consistent results.

Definitely something we will improve going forward!!"
Bold emphasis added by me on the "with vision" part.

[formal osprey]

I did. that's why I answered

[plucky hare]

It reads to me like GPT-4 with vision interprets images in a textual format, and then passes that info to DALLE, likewise, as text. How are you reading it?

[formal osprey]

I read it as "GPT-4 sends commands (as text) to DALL-E 3". which means GPT-4 already has knowledge about the image and use it to formulate a prompt to DALL-E

[plucky hare]

How is that different from img2txt2img?

[formal osprey]

oh, don't get me wrong. it is img2txt2img when we're refering to sending an image as input and asking GPT4 to generate an image using dalle. The part I'm wondering is how is GPT4 itself interpreting images.

we can send an image to GPT4 and then ask followup questions about it. it will give more details if we ask it to focus on a part. so it somehow have an internal way to represent the picture. I'm just curious how this works

The important part in neonbjb's answer is "DALL-E 3 and GPT-4 (with vision) are separate systems [...] Details in the image that are not easily conveyed with text won't cross this barrier very well". That implies that if in the future they create a new GPT model that can generate images, there won't be that barrier anymore. So the image generation based on vision will be able to treat the details more accurately

I don't know how far is the research on multi-modality in the output. We see more multi-modality in the input right now. But the output is still always text. A new breakthrough would be multi-modality both for input and for output in the same model. I don't think there are many examples of this right now.

[agile peak]

For anyone looking for Public Domains arts/illustrations as references. Can use a website called “MidLibrary” to avoid copyright.

[dense ermine]

Looks like a movie poster

[dense ermine]

I heard you could get it by also showing a picture of the character

[earnest flame]

Frankly you should basically not be able to do this. I'm surprised you were able to even get this far.

Just on GPT that is...

e.g.

[agile peak]

Loki having an existential crisis in paris cafè.

[exotic stratus]

Dalle 3 not able to make images in my custom gpt ?

Tried 5 times

[dim cradle]

In the heart of Paris, even Norse gods ponder the mysteries of existence over a cup of coffee. 🌌✨ (From Cosmic Dream)

[agile peak]

This is mine lol

[dim cradle]

Well done. "The Beat Generation, known for its rejection of standard narrative values, and existentialism, a philosophical theory that emphasizes individual existence, freedom, and choice, often evoke images of nonconformity and sometimes the act of smoking as a symbol. This connection might be rooted in the cultural imagery of the mid-20th century, where figures associated with these movements were often portrayed with cigarettes, symbolizing a certain defiance or contemplative state."

[lapis osprey]

So creating from a seed id still seems to work if its in the same chat

[dim cradle]

maybe Gemini will surprise us with this

[remote sphinx]

anyone know how to get fuedal japan looking artwork?

any key things to put in?

[dim cradle]

the GPT should have some thoughts

[earnest flame]

I actually really like these.

[wise solar]

i thought the daily limt was like 250-300.... its telling me it is 25 now

[dim cradle]

it was never more than 50 for me i know. but yes it seems 25 now

oh you mean image? that was 200 i think

but I just got cut off after very few prompts, i cant believe was even 25 but must have been 🤷

[shut niche]

You can't trust ChatGPT with usage limit information unfortunately, it will lie and hallucinate (search for examples in this channel).
As of the AMA, it was 200 with separate 15min and 60min thresholds that you could hit (E.g., spamming generate).
The trouble is, user adoption and usage rates are off the charts, so they've had to implement a flexible strategy to mitigate processing issues that affect all of their products. So the usage rates can change, on the fly, E.g., in real time, to adapt to loads as necessary (it's probably a blend of automation and daily human intervention). The Devs have said what they can in here about it, but there's no way to be totally upfront because they need to be able to change the policy to adapt as needed (They're trying to keep the ship afloat).
One recommendation that was provided is to use the service in the off hours when loads are lower, and you should hopefully see fewer limitations.
IMO-No one wants to say, but my personal assumption is that there is an industry wide chip shortage, since Nvidia is basically the only game in town. Even if OpenAI have secured the necessary hardware, it takes time to scale up for the kind of demand they're seeing.
For now, we're all just exercising patience, and hoping they get the hardware in place soon.

[dim cradle]

i wonder if they'll switch to neuromorphic processors...

[shut niche]

It's possible that we might come full circle like that. Analog is definitely better at certain types of computation, but I'm also no chip designer. Another similar solution will be quantum processors.
You should read the Nvidia paper that was just released about Generative AI for Chip Assist. That's something that might answer your question.

[warm olive]

can we somehow use dall e to create video? create sequence of image and 'stictch' them together?

[shut niche]

There's not currently a way to iterate exact images. E.g., no image-to-image. I'm sure they're working on that.

[dim cradle]

cool, i'll check it out. i was wondering if that would help with their power consumption woes. i was interested to read that quantum computing is actually much slower than conventional silicon and transistors at floating point calculations.

[dim cradle]

check back in a week

i'm sure some day AI can make us new full-length eps of our favorite retro tv shows.

[mossy night]

you can create photos of a model, but a plus-size model is violation. open ai has fatphobia lol

[restive rune]

having a lot of trouble directing the eyes in my image. anyone have tips?

and the boy looks scary af being an ai image lol

[earnest flame]

Ask it for what the promt actually is. Hard to say what's wrong without seeing the prompt.

[earnest flame]

This image is haunting

[restive rune]

thanks i got the prompt, im gonna work on this prompt then post it here if it doesnt work still

[formal osprey]

he directed his sheet at the girl and look at the camera with a look like he's doing something wrong. weird how DALL-E does things sometime

[dim cradle]

ultimate battle droid

[restive rune]

In a black and white classroom setting, a 10-year-old Caucasian boy with short, light brown hair is seated on the left side. He wears a blue T-shirt and has a sly, mischievous expression, subtly glancing towards his classmate's test paper. To his right, a 10-year-old girl with long black hair is deeply engrossed in her test, unaware of the boy's actions. The background is slightly blurred, featuring a diverse group of students showing signs of distraction or restlessness. The atmosphere of the classroom feels chaotic and unfocused, highlighting the absence of privacy and the ease of distraction and potential for cheating. The image mimics the effect of a photograph taken with a 55mm prime lens, ISO 100, and aperture f/3.5, achieving a slightly shallow depth of field, focusing on the boy and the girl, while the rest of the classroom is less distinct.

bro is looking everywhere but the testing 😦

[restive rune]

one hit, i feel like it would blow up like tnt

[dim cradle]

haha, here's some consolation: "The droid is equipped with highly resilient armor, made from materials that can absorb and disperse the impact of projectiles and energy blasts, reducing the likelihood of a catastrophic explosion. Additionally, its weaponry and ammunition are compartmentalized within reinforced, explosion-proof chambers, ensuring that a hit to one area doesn't trigger a chain reaction. Moreover, the droid features sophisticated onboard fire suppression and damage control systems, which quickly isolate and repair any compromised sections, further enhancing its durability against attacks."

[shut niche]

So, what you're trying to do is easier said than done. It's counterintuitive, but Dall.E doesn't understand left, right, up, down, etc. The spatial awareness doesn't work like you might think. It might take some out of the box thinking to get the desired effect. Maybe something like, "The boy sees something on the neighboring desk?" Or, perhaps even spell it out more bluntly, "A scene of a boy cheating on a test in school." It will probably take a lot of different approaches and generations to get him looking at a different desk like you want.

[restive rune]

Interesting, I didn't know that. How does the spatial awareness work in dalle?

[restive rune]

^

[fallen ferry]

https://discord.com/channels/974519864045756446/1174592324739801128

[warm olive]

a week?

[shut niche]

It doesn't really have it, in a nutshell. I think it was mentioned in the AMA too (they are trying to train directions etc.). So it's hit and miss.

It's more likely to recognize a common idea seen in a common image. Take two lovers looking at each other. There are a million images like that, so it'll usually get that right.

But if you say, "A character looking at the thing", or "left," or "right," it's less likely to work.

However sometimes if you make "the thing (object)" the main focus of the prompt, the characters will natively just look at it. I'm probably explaining this terribly.
E.g. If the scene was about a nuclear reactor on one of the child's desks, Dall.E might depict all of the kids looking, because that's the focus of the scene.

[glossy scroll]

Lol, a nuclear reactor

Good example

[dim cradle]

In the realm of AI, wonders never cease,
A week might bring a new masterpiece.

[glossy scroll]

OpenAI is moving pretty fast in terms of data optimization

Which is quite amazing

Could be better though

[earnest flame]

Have it mention the girl first so that she's the subject. I think it does too much facial correction on the main subject to get what you're looking for.

Notably, dall-e DOES understand behind and in front. However, it definitely doesn't understand left and right. It understands next to though. So I don't think the assertion that it "doesn't understand" is quite right.

[warm olive]

a week is too soon

[restive rune]

that makes a lot more sense. thanks, i might say that the girl has a special ink pen that the boy's attention is on

[restive rune]

wish i could test this out more! unfortunetly hit my limit, and that was way quicker than i thought it would be lol

[earnest flame]

I'll test later when I have some time

[dim cradle]

If only SkyNet had OpenAI's ethical subroutines.

It's defnitely getting better with text

[shut niche]

Right, next to isn't a direction LOL 😛 It will also usually get above and below right. And it doesn't understand. It's just trying to mimic similar images categorized in similar ways. It's not using a logical train of thought to determine direction.

[earnest flame]

Yeah fair. I was going to say it probably doesn't "understand" anything. However, it is now far from the first diffusion models conceived. It's possible it has some sort of logic somewhere in there somehow that we're not aware of.

[shut niche]

Look for the open ended Dall.E prompts I shared in here a few weeks back, further ⬆️ in this channel. You might get lost for a minute looking at the surreal results you get back. It definitely identifies itself with an "artist," and some other abstract ideas.

[shut niche]

You never know what might happen! 🤣

[shut niche]

I think this might need to become its own new memes sub-genre... "Jolly Judgement Day"

[haughty spruce]

I can't generate any photos "taken with flash on", it simply doesn't understand the term I guess

It generates photos like this that have "flashlights"

Can anybody help?

[shut niche]

What's your prompt?

[haughty spruce]

A photo of a playground at night, apartments in the distant horizon, low exposure, taken with flash on, fogged camera, amateur, blurry and grainy, distorted, 2003

Apart from flash on, it understands all the others very well

[shut niche]

Is this Bing or ChatGPT+ with Dall.E. That doesn't look like an image prompt from ChatGPT.

[haughty spruce]

Yeah I use Bing one.

Why flashlight? I don't say it's "flashlight", I say "taken with DSLR flash" but it doesn't understand.

[agile peak]

This one made using reference of (Tadema’s style) from public domain.

[shut niche]

"Create an image of a bright playground with a backdrop consisting of a foggy night scene and apartment buildings on the horizon. Aim for an amateur, blurry, and grainy aesthetic, with some distortion."

[shut niche]

You can ask Bing to help you craft your prompts. It looks like you're trying a SD prompt format, with injected keywords, which doesn't work well for Dall.E.

[haughty spruce]

I will show you what I am aiming for. ChatGPT DALL-E and Bing DALL E is so different in prompting.

ChatGPT prompts usually don't work for Bing, it completely misses the point.

It accidentally generated what I was aiming for.

When I include people in the photo, it understand "flash on" but when I type an object name, it generates flashlights, what the hell?

[haughty spruce]

The prompt is "A photo of teens in 2003, analog film, kodak, low exposure, evening time, flash on, beach in the background, distant lights, memories, nostalgic"

I changed "teens" to "a playground" in the prompt and see how it doesn't understand.

Now the prompt is "A photo of a playground in 2003, analog film, kodak, low exposure, evening time, flash on, beach in the background, distant lights, memories, nostalgic"

[shut niche]

My guess is this is partly because it doesn't make sense in the context. A standalone image of a playground doesn't go with the idea of nostalgic memories. It's easy to re-produce the effect with people, because there's going to be enough similar images included in the dataset they used, labeled with flash on. But it's an obscure detail in the context of a playground.

[haughty spruce]

Maybe, because yeah there aren't enough photos to train, but what if I told you DALL-E 2 perfectly understand the concept of it?

[zenith jacinth]

Wow. Dalle is very firm on the no violence bit!

Trying to recreate a scene from an RPG I ran recently that was super memorable moment that is proving imposible to get Dalle to produce...

[zenith jacinth]

Utterly impossible.

[agile peak]

Anyone could suggest to make dalle produce to increase the intensity of thunders? Tried word “intense, excessive..” doesn’t work even “to show his magnificent power”..

[wispy storm]

Suggestion : Spectacular, exaggerated , meters long

[hybrid sorrel]

somethings definitely off with that, first said it was 10 hours, then 10 hours later its saying its now 20 hours, and just to clarify this is after asking it try draw something and getting an error.

[agile peak]

The irony of dalle. When I asked Wotan (Odin) riding Sleipnir the eight legged horse. It generated all of them with 4 legs perfectly 🤣

But when I only mentioned horse. It sometimes generated with 3-5 legs haha

[fallen merlin]

"Create an imaginative scene of a carousel on the surface of Mars, featuring six beautifully designed animals: a majestic jaguar, a charming panda, an elegant giraffe, a regal lion, a playful rabbit, and a powerful eagle. Each animal is intricately detailed and exudes a sense of wonder. The carousel, with its intricate designs and vibrant colors, stands out against the stark, reddish Martian landscape. In the background, the Earth appears as a stunning, distant blue orb, adding a surreal and dreamlike quality to the scene. Surrounding the carousel are children in miniature space suits, their joy evident as they run and play in the Martian sand. Their tiny footprints are scattered across the red sand, creating a lively and dynamic atmosphere. The lighting should capture the unique Martian ambiance, with a blend of eerie, otherworldly hues and the warm, inviting glow of the carousel."

[vapid spruce]

Yeah I'd like to generate images too, but I keep getting errors while generating

And I'm afraid that each time I tell him "Retry" it counts as one of 50 tries

Wait, I have been using GPT as deafult, also for DALL-E. While selecting DALL-E works and also gives you 2 images 😳

Is it better to use GPT asking him to generate images through DALL-E, or using DALL-E section which gives you 2 images? I don't know if there any other differences, such as the prompt it gives

[agile peak]

Sleipnir the giant black horse with eight legged from norse mythology

That’s too much 🙈

[boreal gate]

Using GPTs to create a version of chatGPT that tells stories and uses Dalle3 for visualization bardofverses on instagram or #bardofverses

[vapid elk]

Sleipnir the centipede horse =P

[agile peak]

The best artistic expression of Thor and Loki dalle produced for me so far.

[shut niche]

Xentoshi, you are one focused creator 😃😆 "Eye on the prize!"
Why don't you make a collection in the gallery, since you have so many of these to share? 🧐😜

[agile peak]

Will do soon 🙈🤪

By the way, guys. I have a question does using certain prompt writings techniques could improve Dalle rendition or differently?

Example:
Lawrence Alma-Tadema's depicting the Norse god Wotan. In the center of the composition, Wotan stands, wearing an eye patch over his blind eye, with a braided white beard and a viking winged helmet. With plain muted earth tone background

Into:
Style: Lawrence Alma-Tadema's painting with chiaroscuro style.
Subject: the Norse god Wotan. Wotan stands, wearing an eye patch over his blind eye, with a braided white beard and a viking winged helmet
Background: A plain muted earth tone

[dark breach]

So I asked ChatGPT to illustrate this scene in a story it was writing, and I'd explicitly defined the day to be Saturday, so it does this, Look behind the two figures.

[glossy scroll]

Wohw

Looks awesome

[twin heath]

for some reason its not working for me, it shows a picture for 1 split second and then becomes a blank white screen

[shut niche]

Bahahahaaha, I started making a meme generator for it last night. 🤣 I need to finish it up and share it 😆 🤣. A beacon of positivity amidst all of the AI doomsday chatter. Lolz

[final compass]

I tried a similar style at one point, and it did pretty good results.

I kinda wish i could build a form and just fill the fields and have some stuff even as drop down lists.

[dim cradle]

I found the word "absurd" works... "absurdly loud" or "an absurd amount"... anything with "absurd" seems to be understood and imagined...

[agile peak]

Mine. Giving more the closer result to the artistic style with the second prompt technique.

[agile peak]

Thanks will note those words for next attempt!

[dim cradle]

cool, let me know if it helps with your results

[humble sonnet]

I made this with a sona of mines early prompt I found (the image stretch and ip was me tho)

Here are some from prompt v2 btw

[agile peak]

Here’s the result from both prompting styles

First (normal) and second (modified version). The first one is generic while the second one is exactly the kind of style of Lawrence Alma-Tadema used.

[dim cradle]

Inspired by Cosmic Dream, created a DALL-E Dalí GPT:

"Here is the surreal interpretation of Norse Gods, envisioned in a style inspired by Salvador Dalí. This creation aims to capture the mythical essence of these deities through a lens of surrealism and fantasy."

I'm not sure what the cat theme is all about, though

[fluid maple]

@spiral prairie test

Where can I use dall-e for free?

[dim cradle]

ChatGPT

[agile peak]

You mean Bing lol

[fluid maple]

what is the url for bing

for their dalle

[agile peak]

FYI you don't need to directly use from Bing. Microsoft already made a standalone website called copilot.microsoft(.)com for less heavy usage.

Ensure you choose the Creative mode.

[fluid maple]

wow i tried bing now

so slow

still generating

a while ago i was in dalle's discord, was so much faster

[agile peak]

Yeah. That's why I used the standalone website that specifically for Copilot instead.

[fluid maple]

copilot not supported on my phone

[agile peak]

Ah.. thought you are on desktop, in that case better to use from their mobile app.

[agile peak]

ChatGPT behaving strange again. My rate limit countdown is keep increasing.. went from 18 hours to 21 hours

[dim cradle]

Y'all must be image-generating madcaps, I have yet to encounter a limit

[hearty ether]

Wouldn’t Dali be against policy

[dim cradle]

Perhaps so, it is not made public

[agile peak]

That's pretty wicked. Made me think of psychological struggle inside Thor.

[exotic stratus]

image gen is way too slow

even on 1

[agile peak]

Dalle seems having a difficulty to perfectly capture anna or michael ancher style.. even though it’s public domain..

[stark owl]

Good evening, I'm trying to insert a text inside my image so I added in the prompt "The artwork is framed by a decorative card border with the text "xxxxxx". " but it still renders without any text, why that ?

[plucky hare]

Hello! Can you share the full prompt? It's possible that the rest of your prompt is too complex to include this detail as well, it kind of depends!

[stark owl]

Sure :
Vertical fantasy card image featuring a character inspired by an Eliotrope from Dofus. The main character is standing facing away from us, we see his back, like if it was taking portals under his foot. The Eliotrope should be depicted wearing a hoodie, central to their signature look, with their hands forward creating magical portals, which is characteristic of their class. The environment around the Eliotrope is otherworldly, with a celestial or space-like background that hints at their ability to manipulate portals and space. The character's costume should include elements of gold and blue.. Their eyes should be glowing, signifying their connection to the mystical forces they wield. The overall color palette should be harmonious with the character's appearance, emphasizing blues and golds. The artwork is framed by a decorative card border with the text "Eliotrope".
I admit this is pretty complex prompt, i should remove some unnecessary details and it was part of generated by chat gpt

[restive rune]

dalle lowkey racist lmao, i am making do shots of kids in elementary school working, or cheating on test. it always makes the hard worker asian, and the cheater white. lmfao

[plucky hare]

Yes, I think you're on the right track with trying to simplify a bit. Maybe just try a super minimal prompt, but still include your desired label. Confirm it works, then slowly start adding details. Note that text generation might not be perfect every time, but you should start to see roughly when you start getting to the point of "too much"

[stark owl]

Absolutely ; By the way, could you please have a quick look at my question in #1174818985884274718 if you have 5 minutes please ?

[plucky hare]

This is something they addressed in the recent AMA 👍 #dall-e-ama-answers message

[restive rune]

good to know, thanks

[agile peak]

Did something changed. Right now even in public domain still not allowed to use as references now? In dalle chatgpt.

[median kraken]

Is DALL-E only accesible by payment? I have my ChatGPT profile but I can't find my gallery anymore

[plucky hare]

Sorry, what are you referring to? DALL·E 3 on ChatGPT Plus? Or DALL·E 2 on labs.openai.com? There hasn't been a gallery integrated with DALL·E 3 on ChatGPT as of yet.

[dim cradle]

and you wont find your dalle2 images there

[median kraken]

that's why I'm totally lost, the last time I used my DALL-E for creating images was a bit longer than a year ago, started to use ChatGPT with the same credentials and lately Microsoft opened an additional profile, I thought of using some of the old images but now I can't find the site. I know, I know, I'm going way back in time and there have been many changes but I rather asking

I think is DALL-E 2 on labs

[dim cradle]

yeah it is. 3 will be soon too I think

but theyre separate for viewing images and things

[median kraken]

so my old gallery might be dead

I am thinking on buying a licence and now is on waiting list, I was using Midjourney but the images are not what I need

[dim cradle]

Your old gallery should still be there on labs

yes MJ is nowhere near Dalle imo

[plucky hare]

So when you log in at labs.openai.com, you don't see anything? This user reported some loss of functionality on labs, but I don't see many other reports like it, no idea if it could possibly be a similar issue, if you're even having issues! #1167446256013017139 message

[dim cradle]

(just dont say this on the mj discord they dont like it haha)

[median kraken]

I still have issues with the prompts because I'm not asking for anything banned and yet is sending me warnings, but the results that I get from MJ look like cartoonish and is not what I want

[dim cradle]

MJ doesnt have anywhere near the coherncy imo. It's cool and incredible and i cant wait to see 6.0 and the competition going, but for my money, Dalle everytime

[median kraken]

Actually now I could see the gallery, it was showing me a blank space, logged in with Edge now and worked

Probably a Chrome issue

[dim cradle]

noice.

[median kraken]

I need help to get to where I want to get but in the meantime, since I can upgrade, I will make some trials

Thanks a lot ❤️ 🫡

[median kraken]

Ok guys, I have some questions because I know I'll have lots more and have to start somewhere. I assume most of you have some computer, software, coding background. If not, how difficult has been to create images using the AI? I have something specific I want to create and still feeling like a baby in a crystal store

[twilit helm]

is the image generation one now, not two?

[restive rune]

easy anyone can do it. just start messing with it. then watch some youtube videos, test some things, then come here, is the path i would take

[median kraken]

so, everyone started on the same point then, nothing to feel embarased about.

How precise is the Spanish version of the tool?

[restive rune]

damn, i guess

[twilit helm]

sad

hopefully it’s goes back

having different variations was useful

[restive rune]

yea, i doubt it. i asked gpt, says due to resource allocation. i imagine more people are using dalle3 now and the trend has been downward

[plucky hare]

@restive rune I was still able to get two at a time using the dedicated DALLE GPT just a bit ago. Note that any custom instructions you have on your account won't be active in this GPT https://chat.openai.com/g/g-2fkFE8rbu-dall-e

[twilit helm]

okie, thank you 🙏

[twilit helm]

make sense, more people use the resources, it will overload

if there’s not enough gpus too keep up

[agile peak]

I am impressed of how dalle produce Art Deco in form of oil paintings 🫡

[vital gull]

Lempica style

[restive rune]

image looks so real

[restive rune]

if you guys had to guess what's this image about/happening, what would you think?

[restive rune]

ah that checks out. wonder why that's the case thanks

[plucky hare]

Server load stuff most likely, OpenAI dev Moxi has talked before about how it's due to a limit supply of GPUs

[agile peak]

Arcology, anyone? it’s a pretty neat idea to explore..

[drifting hawk]

Can you share convo id of a convo where this happened in? It’s the id at the end of the url

[hybrid sorrel]

tried to DM it, looks like you have them disabled though, i can post it here if it wont make the entire chat public

[remote sluice]

what do you guys think dall e 4 will be actually capable of?

[graceful spade]

Is it possible to recreate this character with a decent level of fidelity?

[rigid badge]

any tipts on prompting for character generation to avoid having parts of them cut off at the edges? I'm trying to avoid mentioning cut off because that seems to make it worse since negative prompting doesn't really work

[rigid badge]

i don't think dalle3 is the best tool for that

[plucky hare]

Same experience here with negative prompting, just doesn't work yet. I've had luck mentioning specific details about the things that ought not be cut off. For example, if I want someone's full body depicted, I might specify what kind of shoes they have on.

[rigid badge]

yeah i've found full body early in the prompt is helpful but not as consistent as i'd like

[plucky hare]

I have trouble with "full body" as well--that's what I mean about specifying certain "edge details"--shoes, rings, holding something, hat, etc. Define your extremities, pretty much 😁

[final jungle]

Ive been on a unicorn kick lately... must generate more unicorns!

[clear egret]

anyone using the oupating feature?

i cant buy dalle credits

to use the feature on my browser

[final jungle]

Pony!

[final jungle]

nice one Shon. I wanted to get away from the 'pretty colors' for a bit, ive got a lot of those already, lol

[final jungle]

I combined a clydesdale, a unicorn, and a maine coone, lol

[dim cradle]

I figured, but I had to run through Cosmic Dream

[agile peak]

Be kind or face the judgement 🐹

[dim cradle]

absurdly large... reminds me of that horror movie from the 80's with the gargantuan demonic rat monster from hell

[rocky tree]

can anyone with gpt 4 please make me a favour input this into gpt and tell it to solve it? Im trying to see if its worth the upgrade for college usage

its a math exercise

[dim cradle]

this guy for real?

[final jungle]

1. No because i used all my messages making unicorns instead of hamster gods

But...

2. Why not use wolfram? or some other math solver

[rocky tree]

🤣

[final jungle]

Thats... spanish?

[rocky tree]

just want to see how it interacts with the image and see if it gets the result correct. Ive got other subjects like physics that cannot be solved with wolfram

[rocky tree]

correct! im from Argentina

[dim cradle]

"college usage" (euphemism for cheating on homework)

[final jungle]

translate it for me, im assuming its...

if EQ and EQ then f maps from R3 to R3 ... whats the rest?

[rocky tree]

but can't gpt get images as an input?

[rocky tree]

bro im in college why would I cheat? tat's highschoolish

[final jungle]

It can... but like i said, i cant do it, i hit my message limit making unicorns, lol

[rocky tree]

lol dont worry, thanks anyways 🙂

im too tired to translate it xd

[final jungle]

So, the next best thing was some random guy who graduated with a BS in math and loves puzzles... thats sorta gpt isnt it? lol

[restive rune]

bro really screenshotted his homework

[rocky tree]

you graduated in math?

[final jungle]

Yep, the irony of being a computer science teacher is i used my degree in math for 3 months, lol

[rocky tree]

lol I wish that was my HW, thats an exercuse statenebt from an exam

[rocky tree]

haha there aren't that many jobs on that field rather than investigation, right?

(maths)

[rocky tree]

My Calculus 2 teacher did a similar path as you did, she's computer science and math teacher as well

[final jungle]

hm... translated, i wonder if gpt3.5 can handle this... i think thats stokes' theorem, isnt it? applied maths / calc 3 stuff?

[rocky tree]

that's right !

I used Gauss's theorem (the divergence theorem) when solving it

[final jungle]

yeah 3.5 will set it up for you

[rocky tree]

Yes but like I mentioned, I have other subjects with non-mathematical statements or exercises written on a sheet that I want GPT to check for me

[final jungle]

Right... one thing it may struggle with is 'y'
You may need to specify your picture is in spanish, or it may try to read y as a variable, not an 'and' at the wrong times

[rocky tree]

true! good that you pinted that out

[final jungle]

And now... it is bedtime. I will leave you all with a chibi deadpool/robin fighting harley quinn

[rocky tree]

it's cool that we can understand the same math problem without speaking the same language. The beauty of maths I guess

im going to sleep as well. Thank you anyways matt and keep the hamster gods on!

[final jungle]

If you want to DM me the picture ill run it thru tomorrow and see what comes up.

[rocky tree]

Sure, thanks!

[glossy scroll]

[signal breach]

is dalle-3 available via api yet?

[glossy scroll]

The vengeance in that hamster's eyes

[signal breach]

i'm not enterprise thou

yeah dalle3 is out for me

very cool

anyone know an app i can use as a API dalle client?

its cheaper then labs

[shut niche]

I've noticed a difference in distance between "close" and "close-up." The term "portrait" also has similar distance ranging effects. E.g., "Close" usually captures the entire character without making it a "close-up." I hope that works for you. 😀

[rigid badge]

I checked briefly but didn't see anything as popular as the chatgpt-web so i spent a couple hours (in chatgpt-web) to help me put this together

[signal breach]

fair enough

its much cheaper to use api then labs

[rigid badge]

i saw a handful of things on github but none had more than like 30-50 stars so i just felt like i'd try to make my own

[signal breach]

you should add a button to just copy the image

you dont need to save it

[rigid badge]

i have it save the prompt and b64 of every image in a csv so even if i don't save it i still have what i paid for lol

[signal breach]

ooh nice

i've already used 16 cents on dalle3 generating some random images

legit money dump

[rigid badge]

ayyy it works good idea thanks

[signal breach]

no prob lol

[frosty nimbus]

Has anyone had luck making it produce SNES-era pixel art? In my case it constantly makes the images too detailed or starts adding in curves or blotches over the pixels

[dim cradle]

good morning

I'm trying to generate my first image with AI

May I ask you your help to guide me ?

[final compass]

If you are using gpt+dalle, then i would just suggest having a conversation with the gpt about what you want to see and suggest changes.

you can deep dive into the prompt tricks once you have something you want to really dial into.

describe aspect ratio wide, square, tall, then describe a style, photograph, painting, illustration, then describe main subject and maybe a background.

[hollow wolf]

Hello. I cannot upload my files to chatgpt and the display of pictures has something wrong, including DallE3. Does anyone have the same problem or something to solve that?

[half vigil]

Might be a stupid question.. I just signed up for ChatGPT+, but I cant seem to find an answer to; Is Dall-E free, or does it cost each time you create an image?

[plucky hare]

It's part of your Plus membership, no additional fees! This is just DALL·E on ChatGPT, to clarify, no DALL·E on labs/API is included.

But other than that, you can gen 200 images in a 24hr limit as part of your Plus. Subject to rise/fall based on capacity! Hopefully not fall 😁

[half vigil]

Ok thanks for the reply! Whats the difference between Dall-E on ChatGPT and on labs/API?

[plucky hare]

DALL·E 3 is on ChatGPT Plus. DALL·E 3 on API is how developers include DALL·E functions in their own programs. DALL·E 2 is usable on labs, so it's the previous model.

[half vigil]

Alright. Thanks. I love it, but Im having some issues with having the output be the aspect ratio I want..

[plucky hare]

There are currently just three preset aspect ratios and you can't get anything different other than cropping yourself. It's the preset square, and then tall/wide. Tall sometimes glitches as a rotated and mirrored wide image too, FYI!

[half vigil]

yeah its the glitching im talking about. Ive asked it for wide, or landscape, or even the dimensions of 1792x1024, but it keeps spitting out images that dont fill out the entire image, with white or some other color on the sides..

[plucky hare]

Ahh yes I see that too. The best way to try to avoid that is to, paradoxically, avoid talking about it 😁 What I mean is, it doesn't understand "negative prompting" because it can't really see its own images, so it doesn't really know what they do or don't have in them. So saying "don't cut off the edges," etc will just confuse it by thinking about it in the first place.

Sometimes it happens anyway though, just gotta adapt, overcome, evolve, survive

[half vigil]

ok thanks. just wish there was a prompt to make sure that it fills out the entire image each time

[final compass]

wide images still tend to perform pretty well in getting wider images most of the time, while using tall images i tend to get more rotated images, especially with portraits. My guess is that the training set had a lot of images, that are portraits in the essence it depicts a person, but also incorrectly rotated and the model learned that portraits are rotated landscape images with person in them 😄

[half vigil]

I guess thats just how it is for now, but its constantly improving right, so hopefully we'll see some improvement soon

gonna try my luck with midjourney to see if that works better for my purpose

[final compass]

yeah i can imagine that is not going to get 'hotfix' anytime soon, i guess they need to fix the training sets and retrain it, or maybe implement some checks on the image rotation, but this is purely speculation from my part 😄

[half vigil]

hehe alright. thanks for your input!

[plucky hare]

Yeah they mentioned something in the AMA about wishing they could teach it left from right 😁 I'm sure they're working on it! The tall image one is the most annoying one for me, I just don't even ask for them hardly anymore

[final compass]

Same here, asking for them just tends to waste your already limited number of responses.

I also did a small test, and when i asked for male portraits i tended to get correct results, but when i asked female portraits they were almost always incorrectly rotated, but my test wasn't super extensive to draw definitive conclusions.

That made me think about my own photography, that the cameras autorotate doesn't always work either and if i would just dump all my photographs into a training set, i would probably endup with the same problem 😄

[agile peak]

https{:}//cookbook.openai.com/articles/what_is_new_with_dalle_3
I just read the OpenAI cookbook. I wasn’t aware that we can use Standard or HD, even though by default it’s “standard”.

[plucky hare]

Thank you for sharing this! How neat. And it seems like standard and HD can also be understood as minimal and detailed kind of, and not the actual fidelity of the photos? Is that how you read that too?

[final compass]

The resolution of the image seems to stay the same, so that is not referencing to actual resolution, but i guess the model tends to try harder in doing more detailed image and costing more processing power.

[agile peak]

I asked ChatGPT it doesn’t let us to change from standard to HD. So I guess only for dalle API?

Instead ChatGPT dalle version says if you want to change the standard to HD by changing resolution.

Which is different from the cookbook version.

[obsidian timber]

Why is Dall E3 so expensive to use in the API? Just created a programm to multithread the prompts to generate variations and each run is like spending 50cents🥲

[agile peak]

I'm seeing in the future Dalle 4 might improve its capability to produce website design/mock up imo.

[agile peak]

A high quality of ASCII art depicting chatgpt in a humanoid form

[glossy knot]

That has been already done by GPT-4

Check out tldraw on twitter

It blew up like yesterday night

[cinder sonnet]

That was pretty neat, thanks for the req. This would be really fun to use to make some custom apps that have specific GPTs built in to them. Like GPT-specific wrapper interfaces.

[agile peak]

Already seen that. They are mind blowing. I mean for dalle to produce an image to show a good inspiration stuff haha

I think it must be due to my prompt. I am sure it can be done with some good prompt engineering.

[glossy knot]

Idk ai or anything
Actually I didn't like it until i put some credits in the API yesterday

After i followed a good amount of stuff
There's a lot of stuff going around that really improved workflow

Not just image creation

[pure robin]

Hey guys. Can someone explain to me how to have higher accuracy when generating text in the images ? Thanks in advance.

[polar flare]

THIS

[pure robin]

I'm sorry, i don't understand your answer...

[polar flare]

I'm wondering the same thing. I can't seem to get it spell words right

[pure robin]

Is about good spelling in the prompt ?

I had many attempts and i can say 1 out of 10 is succesfull

rest of them i get a chinesse + indian words

[polar flare]

Yeah I make sure to be very clear and specify to correct the words but it always messes up a couple letters. I notice that the longer the word the harder it is to come out right

[snow leaf]

dalle 3 really sucks lmao

this doesnt match the prompt AT ALL

[plucky hare]

Can't gen copyrighted imagery/characters.

[snow leaf]

i know but its not even close with the style even

[plucky hare]

Do you mean specifically the PS1 style? I wonder if more "low resolution" type promoting might get you closer. One challenge with DALL·E is making things look "bad" as in older graphics styles and stuff. It'll do low poly, but it'll be the crispest low poly you've ever seen 😁

[snow leaf]

i dont care if its hd low poly, but these dont really have polygons at all

[plucky hare]

The man in #2 seems a bit poly-ed to me, no? Maybe try asking for something like "Please make two more like this, but pass both images through a 'low-poly' filter" to kind of express that you want the whole image to take on your desired style?

[drifting hawk]

yeah you cant do anything with just the id but i can look up the logs

[drifting hawk]

theres no support, its only api

chatgpt is always HD

[hearty ether]

Oh yeah what’s changed with the seed and photo id system?

It doesn’t seem to be able to recreate the same image anymore with same prompt and seed

[drifting hawk]

seed and photo id is an implementationn detail and not a feature

you can search what i said about it before in this channel but nothing is guaranteed if you're relying on it

[tidal swallow]

hello word has gon?

[pure robin]

Azir from League of Legends... This is soooooo amazing

[agile peak]

Dang. Just got a notification from perplexity, sam altman stepped down from OpenAI.

[velvet karma]

Yes, but off topic for this channel. Try: #openai-chatter

[swift oriole]

Just wondering, all the multitude of filters and there algorithms run constantly checking everything against the objectional material sample libraries from prompt through the image creation and rendering how much GPU resources is that requiring compared to the image generation itself, just a thought.

[plucky hare]

seven

[swift oriole]

Thank you it's interesting to know.

[agile peak]

Mars arcology with Brutalist influence.

[plucky hare]

I love the stuff you share here Xentoshi, your ideas are always so visually interesting to me! Here's a couple I just got based on this idea--basically just asked it for a nighttime version, and to "HR Giger"-ify it for me.

[plucky hare]

Heaven version, to balance out my edge 😁

[restive rune]

dalle is able to analyze images and tell me information translated pretty well. kinda awesome, anyone else know other cool uses like this?

[plucky hare]

Image recognition is built into GPT-4 now that it's multimodal, so you can kind of use it as a shortcut to make "remixes" of images if you want. Note that the "vision" of GPT-4 is basically creating a textual description of what it sees, then passing that descrip to DALL·E per user requests, so it's not like DALL·E can "see" directly, as much as it's just being given a description, as illustrated in the example you've provided.

[restive rune]

ah that's neat, makes sense that it can do that now. thank you

[agile peak]

That’s remarkable. I can see a little influence mix between H.R Giger and Dune architectural styles there 😄

[tidal girder]

I'm really struggling producing any images of women. Sometimes they come through but lately even "a woman standing in a doorway" got blocked while "a man standing in a doorway" went through just fine. I see a few women posted here and there but it feels like if add ANYTHING to describe it (or nothing at all) it's blocked. This is my experience, I understand if that's not what everyone else is experiencing

[olive dragon]

I'm new here If you ever wanted to know what rainbow black holes look like here.

[remote sluice]

me when lantern is against the rules but lamp isn't

[remote sluice]

make a real one

[final jungle]

getting it to do women can be annoying. ironically, i find the worst word to use is 'woman'.
Lady, female, or girl seem to work better. Also, bypassing it totally and implying its a woman by saying something like...

"a picture of a business professional. She is ..."

[paper jolt]

Damn these look nice! What was the prompt?

[olive dragon]

Thanks here's the prompt

"ton 618 black hole with rainbow colored accretion disk, all of the universe falling into black hole, heavenly look, anime style, made in heaven"

[remote sluice]

i got my first nudity with dall e 3 🤪

[final jungle]

eh... i feel like you can find that anywhere. But this? Lord Laundry, riding his trusty steed, on his way to do your laundry? Thats special, lol.

Or Freebird imagined with a chicken?

[royal kindle]

Cat unicorn lol. Now that's a combination.

[final jungle]

Maine Coonicorn
Combined a unicorn and manine coone cat, lol

[coral mortar]

is bing down for you guys? i can`t login

[dim cradle]

will Altman leaving affect Dalle? (to keep it on topic)

[final jungle]

mods want that directed to the chatter channel, unless earlier they did

[earnest flame]

So cool!! I recommend posting them to #1154829862171844679

[final jungle]

Anyone done anything cool lately?

[restive rune]

What’s up with this??

[final jungle]

the default setting is not to translate. sometimes it remembers that

[restive rune]

Interesting. I had to ask it 4x, how do you turn on the setting in prompt?

[final jungle]

Oh wait thats an image in chinese and u wanted it translated... it might have been a case of 'im not supposed to solve this capcha'

sorry thought that was a dalle prompt in chinese at first

[restive rune]

Yea it was a business license. Not sure why it can’t do that.

[final jungle]

i got a pretty good bob ross a few days ago, not perfect, but... meh, what is? lol

[formal osprey]

PEW-dee-py

[cinder kelp]

‌‌Why doesn't the API page display DALLE3 anymore? Is it not allowed to be used?

[tidal girder]

Exactly. And I find it incredibly sexist that OpenAi has sexualized women to the point they won't even render them

[vagrant zodiac]

hi, please use the #1154829862171844679 for posting images, and use a discord spoiler tag for images like this that may be disturbing

[bitter elbow]

I’m sorry if this has been addressed already but did Dall-e stop generating 4 separate images?

[white skiff]

yes long time ago, the new limit is max 2 and sometimes even only one image per request

[bitter elbow]

ughhhhh that suckkkkkks. Thank you

[mild tapir]

I love Dall-e3

[delicate girder]

I can't do pictures with DALL-E anymore, most of what I ask (whimsical characters for a D&D campaign) apparently does not align with its content policy, yet I've made hundreds in the last few weeks, what is up with DALL-E?

Even if I ask DALL-E to adjust the prompt in order to respect the guidelines, even ChatGPT is unable to produce something that respects its own guidelines, it seems a bit out of hand

[plucky hare]

Do you have any custom instructions active by chance?

[delicate girder]

I do yes

should I clear them @plucky hare ?

I had like, 3 custom instructions, nothing considerable, but I'll try without

nope, still blocks. Basically I'm making food-based characters, as in "A Crown of Candy" for a personal campaign, nothing is going to get published anywhere at any time and I'm never refering to the existing content, only insipiring myself. I've done almost 200 character pictures, and today, it says it's not allowed to depict anthropomorphic food people or even just "too absurd or whimsical"

feels really weird

[plucky hare]

Can you share a prompt that you're having trouble getting through? I feel like foodfolk should be fine too, so I'd be curious to see. And after you disabled CI, you went to a brand new chat right?

[delicate girder]

yes sir, here's the one I'm trying to do right now, with a bit of context

So basically I'm trying to alter some pictures I already have in order to give the feeling that time has passed and situations have changed, here's the original picture, made from DALL-E

Basically the first time, I asked DALL-E to "Reproduce this character with a more disheveled look, with a guerilla outfit and a slightly more tired expression"

Blocked, so I tought ok, DALL-E doesn't want to infrige copyrights and won't reproduce an existing image, fair enoug, so then I asked:

Illustrate a medieval court instructor, his body is a banana with the tip dipped in chocolate. He wields a banana bow and a raspberry rapier. He dons a light and functional leather armor. He has an optimistic but slightly tired expression from war wariness. He has a van dyke beard.

and basically any variation, with or without picture, gets blocked

Then again, I managed to turn the first picture into the second yesterday, by providing the first as original.

maybe DALL-E REAALLLLYY doesn't like bananas

[agile peak]

Prompt: a raw photo of a neighborhood house with art moderne design with green and white colors, rounded architectural desings, The raw photo should capture with subdue lightnings, blue sky and soft colors, adding fine grain from kodak photograph.

I think using repetitive words mentioning raw photo or photograph even specific camera type work to avoid photorealistic/rendered looks.

[plucky hare]

Hi, sorry for the delay in getting back to you. Here's what I tried:

• Took your image and uploaded it to a new GPT-4 chat, and asked for a thorough visual description of the anthropomorphic banana character.
• Went to a new DALLE chat and said Two images. Please consider the following description of an anthropomorphic banana, and then create the two images that depict this character in a state of weariness, as if they had just completed a very long and difficult task: and then pasted the description from the first chat, with a few changes/corrections.
  Attached are a few that seemed at least in the direction of what you were looking for--though maybe you meant a wearier look. But it does seem possible. I'm not sure what was causing your flag. Maybe it was some combination of the war and weapon related words.

[delicate girder]

thanks for taking the time, really appreciated, I'll try that approach, I had tried a few, it's kinda hard too when you try a lot of things, DALL-E says no, then you exceed your quota for the 3 hours xD

I also thought about the war and weapon thing, it seems kinda hit or miss (I've managed to make a few magic weapons, like those)

[plucky hare]

I don't think it's always just a word that can cause a false positive, I'd speculate that sometimes it's a contextual confluence of multiple "borderline" words such that, if too many appear in a single request, you might be more likely to get dinged. This is why I like the "describe this for me" approach personally, as I think maybe (total guess) the vision will be more likely to give me a description that DALLE will approve of.

[delicate girder]

thanks again for your time helping me

[plucky hare]

You're welcome!

[chilly onyx]

It appears that DallE has consistent limitations in accurately portraying head-on confrontations or direct charges towards each other, regardless of the subjects involved.

Was trying to get two knights jousting on jet skis for the daily theme, and couldn’t really

So I got curious and tried a few more things such as cars playing chicken

Dall-E is not a fan

[plucky hare]

I could see it being related to its general hesitance to depict anything violent. Your final result shared on #daily-theme was pretty great though 😄 It might also be related to its limitations with directions. That is, it already has limitations with basics like "up, down, left, right" etc. manifested in various ways, so something like this where you're specifying not only subject location but also direction of subject movement is maybe another symptom of that. Though I'm a fan of your attempt using "chicken" as I think this is a clever way to try to depict this visual idea! I had trouble trying to get a jet-skiier to ski away from a billboard yesterday, so I know kinda what you mean here

[chilly onyx]

Yeah I tried to think up different ways of getting things to move towards the center of the image but nothing. I’m fairly sure you’re correct and it’s an imminent violence thing. Yesterday with the pies I tried for far too long to get a clown taking a cream pie to the face, or even just capturing the moment before it connected but it didn’t like that either

Think I saw a few people who did manage it who attributed theirs to Bing image creator

So different rule sets and content filters I’d guess

[agile peak]

Asked ChatGPT 3.5 to follow my previous prompt template. And the improvements is insane.. these are art moderne and brutalist.

[agile peak]

I saw from dalle cookbook from open ai, it's insane how dalle 2 could perfectly capture blade runner where dalle 3 its kind of difficult to achieve authentic grain to the images.

[frozen copper]

dalle generation is erroring out everytime

[oblique grotto]

I have a challenge. I'm trying to generate an image as though I'm standing on a ringworld, looking up at the curvature. I can't find the right prompt.

anyone know the secret magic phrase?

[delicate girder]

Try asking another instance of GPT4 to make you a prompt

[oblique grotto]

I did!

"You are on a ringworld around the sun. it measures 940Mkm in diameter and is 1Mkm wide. you are on a mountain not far from the north wall at the edge of the world, looking spinward. At your feet is the great 3km dam, with lake Seraphim on the left and the city state of Busy below on the right. in the distance the curvature of the ring curves up into the distance. "

[gray surge]

That isn't you want to do?

[oblique grotto]

No. I just realized i doubled the word distance, but no matter. the backgroun should look a little like this:

there's the wall, there's the ring cuving up. i just want to draw mine with a different foreground.

[gray surge]

Did you tried to use GPT-Vision and tell to generate a smiliar image?

[oblique grotto]

is that a plugin? i don't recognize the name

[wispy storm]

Halo owo ?

[gray surge]

Its the standard version of ChatGPT-4.

[oblique grotto]

yeah, my dnd game is on a halo and I'm trying to make some art for flavor.

[gray surge]

So that thing.

[oblique grotto]

when i use the image editor, can i use image generation at the same time? I want to generate images and then say "pretty close, please fix small feature x"

it seems like i either have dall-e or plugins, but not both

[wispy storm]

Dalle 3 have hard trouble to fix minor details

[oblique grotto]

i gave it a sample and it said

[agile peak]

Existential humanoid.

[vagrant zodiac]

Nice, please do use #1154829862171844679 for sharing images 🙂

[hearty ether]

whats a consistent way to get a similar images

like seed doesnt work anymore for me now

[hearty ether]

Can you use image id across chats?

[plucky hare]

You cannot. Here's a little more insight from DALLE Dev Moxi #images-discussions message
Edit: oh wow looks like Moxi was actually responding to you when she posted this 😄

[hearty ether]

It used to work well

You could recreate an exact image with seed and prompt before

Not anymore

[plucky hare]

Yes, I think by "not a feature," Moxi was referring to the fact that they're still tuning DALLE as a model currently. Meaning, the same seed used on two different occaisions could result in different output, even if you use the same prompt too. So it undermines the purpose, currently, to have seed control (insofar as it could mislead users). And image IDs have never been cross-chat, but instead ways to reference a specific image within a specific chat. But again, Moxi indicates this is more of a "backend" thing, if I understand her correctly

[spark cradle]

Is Dall-E down?

[hearty ether]

Image id is definitely more reliable rn

[zinc dagger]

I've been playing around with the image_IDs a little bit and it's been inconsistent, it seems to float between two different seeds on my end for images, despite having the same ID

[spark cradle]

Who programmed DALL-E not to show combat? That's ridiculous this isn't real life.

[plucky hare]

https://openai.com/policies/usage-policies
"Generation of hateful, harassing, or violent content" is not allowable on any OpenAI services

[paper jolt]

Pro tip! You can put angled brackets around a link to get rid of the embed
<link>

[spark cradle]

WOW!! Comic Book combat is Hateful and harrassing. I guess you have adopted a cancel culture for superhero stuff...

[plucky hare]

Thank you! That is a good tip! I've just started hitting the "x" but I'll try this now. Lemme give it a shot https://chat.openai.com edit: ur a hero ty

[paper jolt]

I'm certain OpenAI doesn't have a vendetta against superhero stuff (imagine if they did 😳), they are attempting to prevent more nefarious generations which can unfortunately include some more benign stuff.
OpenAI is taking the "better safe than sorry" approach here

[spark cradle]

Thats outrageous. But then again I guess you don't want Chat to go all Skynet on us... smh... Man to me AI will be like Johnny 5 from Short Circuit not Skynet.... People always lean toward the worst even though progress is progress.

People are evil though so I understand.

[humble sonnet]

ai is now closed, please go home.

[obsidian rose]

the censorship is beyond anything I'd expect any company to do for an image generator tbh. Most of the benign things I can think of don't work

[spark cradle]

Welp if anyone on here had plans to make action comic books they are completely out of luck.... smh.
There needs to be an AI for entertainment. As a Film Maker, I was hoping to visualize my actions scenes as a storyboard. Oh well back to the old fashioned way of doing things.
So much for efficiency, and here I was thinking AI was gonna help with that.
Kicking myself in the butt.

[obsidian rose]

it's such a big missed opportunity. They made by far the best AI generator, only to limit it so much that it's unusable for the vast majority of creators.

[humble sonnet]

I deleted the source image so you get a instagram screenshot instead.

[spark cradle]

@obsidian rose Agreed as an actual professional artist I began to see the benfits of AI generated images. It's great for previsualizations. However that limitation just killed it. I actually had a valid argument to support AI in the Art community oh well.... So much for that.

You know the consensus in the Art community is zero tolerance for AI art. I think the focus should be on trying to** innovate for professional artist with art tools** rather than making toys where people get to play make believe.

[obsidian rose]

hopefully OpenAI will eventually listen to the feedback of creators. I know a lot of them, and almost everyone is not satisfied with the limitations of Dalle-3.

[spark cradle]

I wonder how OpenAI will deal with Professional Artist opting out of having their work used for training. Especially since the Copyright department established no AI art can be copyrighted.

[dim cradle]

me right now

[humble sonnet]

nom!

[formal osprey]

#1175595352875872276

I saw the discussion come back many times here: "how to recreate this style with dall-e" or "how to get dall-e to generate images using my character".

So I made a custom GPT that aim to produce an accurate prompt for DALL-E 3 based on any image we send it. Feel free to test and see if it works good for your use cases. I tried to make it universal. There's still improvements to do but it's already better than asking GPT-4 directly to create a prompt.

[earnest flame]

I actually love this style

[dim cradle]

thanks. i tend towards simpler styles. lines tend to have more purpose, are cleaner

though, rough lines have their place too

[delicate girder]

@plucky hare finally got it
was missing a leg but I managed to edit it in with gimp

[agile peak]

Ugh. House photograph in dalle chatgpt (left) still look slightly off, appears like visual 3D rendered even with specific camera model.. or is it just my eyes?

A front view angle, an eye-catching brown and grey colored Googie-style house with a flamboyant a few of lamps, garage exterior designs, surrounded by palm trees and lush garden, the house's whimsical design exuding a playful and optimistic atmosphere, in pastel blue sky and soft twilight, creating an atmosphere of quiet elegance and modernity, Photography, pastel colour grading, DSLR with a 35mm prime lens, f/8.

[dim cradle]

that's a nice background template you got there

[delicate girder]

oh, I just ask for a dark neon-blue line grid background, usually does nice stuff

[dim cradle]

v. nice

[dim cradle]

no, there are simply too many styles to choose from

[hearty ether]

Try both

[hearty ether]

I like this style

[dim cradle]

thank you

[hearty ether]

Book vibes

[dim cradle]

the mirror images are always wrong though

[plucky hare]

Ayyyyy nicely done, and chocolate dipped to boot!! 🍌🍫

[elfin prairie]

is it very cute,right?

[dim cradle]

yes, very cute

[elfin prairie]

i think yours is more cute

[dim cradle]

oh hush

[agile peak]

did not expect this, but this is awesome for making a game concept

[dim cradle]

what if you play against a cpu though

[final compass]

this is pretty interesting, i have a landscape with castle and a bridge. i have been only asking to alternate the visual style still somehow crashing into content policy warning. it's not that random content policy triggers are uncommon, just funny how can you possibly get that with these parameters.

[sinful locust]

I can't make my current concept because you know why 👀

[humble sonnet]

if ai is closed so can ur eyes!

:3

[remote stone]

Is it just me or has the usage cap for GPT-4 decreased?

[humble sonnet]

Mornin! Ima wait for the next daily prompt so I can make something silly today lol

[verbal sequoia]

<:book_icon:1171408210398289941> `` Rule 1 `` Be respectful.

Treat others the way you would like to be treated, and assume best intentions. Don’t harass or attack others, and don’t engage in hateful or generally malicious behavior (e.g. sexism, racism, homophobia, etc.). Keep the negativity to a minimum.

[twilit helm]

so i only get 5? 😔

[plucky hare]

It recently dropped from 50 to 40

[rose laurel]

how long does it usally take to generate a image?

rough estimate

[twilit helm]

i don’t know i haven’t tried without a boost

but i heard it can be 3 hours

[rose laurel]

jesus

was expecting a few minutes maybe

dang

[frozen copper]

Is dall e having issues generating images? I keep getting errors

[lilac thistle]

challenge for someone: extend this image by giving them an opponent, be creative!

[plucky hare]

Seems working for me. Is it with all prompts? Do you have any custom instructions active?

[tiny jetty]

What do you guys think of the future of ChatGPT and Dall-e now that Sam left? Just an honest question

[twilit helm]

congrats on getting the puzzle role

[gray surge]

Congrats. 😄

[plucky hare]

Thank you both 🙂 @gray surge

[gray surge]

I think every speculation for that subject is useless - but there are other people can do the stuff too.

[twilit helm]

you’re welcome

[humble sonnet]

oooo, very fun prompt today!

[humble sonnet]

alright, just uploaded mine!

[agile peak]

Any reason why dalle chatgpt like to produce unrealistic image? even though mentioned specific camera models.

it seems now tend to generates two inconsistent results..

had no issue with bing though.. strange, possibly a glitch in the generation_id?

[humble sonnet]

ye, it takes a lot of work to get things to work to a vision, I usually spendanywhere between 20mins to an hour making mine fit my vision

[agile peak]

had no issue with bing, possibly generation_id glitch?

[humble sonnet]

Can I get some feedback on my submission for the daily theme? I wanna see what people think of it!

[agile peak]

that’s slightly frustrating, cause after 4 attempts it either gets “slow down” mode or make it quicker for me to hit maximum limits lol

[humble sonnet]

i know that feeling all to well and the thought of it still makes me want to cry

[agile peak]

ask chatgpt to rephrase your prompt to match with content policy, might worked.. unless.. it’s something unsettling haha

[humble sonnet]

whatcha tryin to make?

[humble sonnet]

ooo, that sounds fun!!

lemme make something since I have an idea

this was a quick one without any insanity but I think it would be fun to see a super strong mech fight this boi

fun times

[fast sun]

pufferfish in broccoli land

my fav food

i made this 1 year ago

[fast sun]

it looks goofy

[barren comet]

Hello

[fast sun]

hi

[fast sun]

can u rate it

[agile peak]

I figured most of prompt tips doesn't work properly on chatgpt, but worked fine on bing.

[remote sluice]

bing ai is so awful i've tried the same prompt atleast 90 times and it's still being censored

[potent briar]

I just created this a few hrs ago @agile peak

and this one just now . so I think is work well, just need a proper prompt

[lethal elm]

It feels like good skill/item icons for MOBA games😉

[remote sluice]

trying to get the original 2d image to be 3d, but i keep getting these results.
hoodie on bed with hanger.top left:light black.top right:dull mustard yellow.bottom left:dull red.bottom right:pastel orange.left sleeve:black,red,black.right sleeve:yellow,black,orange.colorful word "sky"&2 black squares on ends of sleeves.colorful word "verity" in middle hoodie.all letters capitalized.

if anybody could help me that would be great

(using bing)

[agile peak]

this is so fantastic, the whole thing rendered in incredibly realistic way

[final jungle]

Hey crew, got a question i never really thought about before... What affects dalle performance speed, and gpt as well? Does your gpu have anything to do with it?

[agile peak]

i found a way to make dalle on chatgpt to get photograph result.. mention the words such as : daylight, hazy morning and auto balance lightning make the scene appear more natural.

[remote sluice]

could you help with my prompt

[agile peak]

[your subject], then add Photography, DSLR with a 35mm prime lens, capture this by apply auto balance lightning for accurate colour representation, and additional film grain to evoke the texture of film photography.

[remote sluice]

i'm not trying to do photorealism i'm just trying to get more coherent results

[agile peak]

what’s your focused subject?

[remote sluice]

the results are not matching what my prompt is. colorful cropped long sleeved hoodie on bed with hanger.colorful quadrants::top left:light black.top right:dull mustard yellow.bottom left:dull red.bottom right:pastel orange.left sleeve:black,red,black.right sleeve:yellow,black,orange.colorful word "sky"&2 black squares on ends of sleeves.big colorful word "verity" in middle hoodie.all letters capitalized.dull pastel colors

i'm trying to get the realistic images to match close enough to the original 2d one

[plucky hare]

Regarding just your specific detail requests: I think you might be pushing the direction-following potential of DALL·E a little too hard by asking for such a well-defined depiction. DALL·E can't keep track of position or direction as well as your prompt is asking it to, even though you have everything written in a simple and straightforward manner.

[remote sluice]

so you're saying that my prompt is not currently possible for dall e to generate?

[plucky hare]

That would be my guess, yes. You probably have just too many specific details for DALL·E to do all of them, if that makes sense.

[remote sluice]

mk

[final sun]

Okay, one: that's not a Pyukumuku, and two: I don't think that's a peace sign-

[remote sluice]

everytime i try to generate wicks, i get random hairstyles that aren't wicks, and if i try to describe them (thick dreads, round dreads, etc) i dont get the results im looking for. is there likely not enough training data for dall e to do wicks, or is there a way I can describe it so that dall e will generate them?
another thing, whenever i try to describe shapes in the hair (octahedral, cubic, etc)) it doesn't follow them. If i wanted a cubic afro it would just make a normal one.

[final sun]

Most likely. The thing about most AI things is that it reflects humanity as a whole. In this case, training data will always be biased towards the racial, ethnic, gender, etc. majority.

[remote sluice]

black man:long rectangle wicks.large vibrant bright green shiny puffer coat,large black alien eye sunglasses,white boxers,sagging very baggy blue sweatpants,shiny 100% blue boots.

[remote sluice]

in my above result it just generated dreads, what can i do to change the prompt so that it will generate wicks?

[final sun]

Not exactly sure what wicks look like

Besides the ones on candles

Do they look like that but long?

[remote sluice]

this

[final sun]

So basically exactly what I thought

[remote sluice]

think hobie brown

yea

[final compass]

the more specific your request becomes, the more likely it becomes that it has no sufficient training set, to produce a compelling image. i am happy if i get hair color and braids right 😄

[plucky hare]

Like TrikiNya mentioned, you may be running into some of the biases in the models. When DALL·E was trained, it basically (1) saw a bunch of images, and (2) read how those images were described by humans. So in this case, there are two things potentially making this harder for you.

On one hand, it may not have had as much exposure to the visual representations of those who are under-represented. And then on the other hand, when the images were of those who are under-represented, you still have to rely on whoever described the image to know the correct terminology.

This is something the DALL·E devs are aware of and want to get better at. They want to reduce the amount of bias their model could perpetuate. But they have a long way to go!
#dall-e-ama-answers message

[final sun]

Am I wrong for thinking of this character?

Just pulled straight from my obscure childhood memories

[spark cradle]

Everyday with this

I asked for a simple enhancement of an image and it returns errors every time.

And it completely contradicts itself

This is precisely what I mean, instead of making functional art tools. We have a toy for playing games.

[potent briar]

Yeah, I'm amazed how Ai changed thigs within a relatively few months. And I can imagine how the AGI will change even more

[frigid kiln]

Hey all!
I'm curious about DALL·E's capabilities and have a specific use case in mind. I want to provide DALL·E with two separate images – one of a person and another of a piece of clothing, like a t-shirt. My goal is to have DALL·E combine these images to create a composite where it appears as if the person is wearing the t-shirt. Is this something DALL·E can currently do, or is there a feature that could facilitate this type of image manipulation?

[potent briar]

there is the answer: DALL·E is capable of creating highly detailed and imaginative images based on text prompts, but it has some limitations, especially in the scenario you're describing. Currently, DALL·E does not have the capability to directly manipulate or combine existing images, such as placing a specific piece of clothing onto a person from another image. It generates new images from scratch based on the textual description provided.

However, you can still achieve a similar result by describing the person and the t-shirt in great detail in a text prompt. For instance, you could describe the person's appearance, pose, and the specific design or style of the t-shirt you want them to wear. DALL·E would then generate a new image based on this detailed description, effectively creating a new depiction of the person wearing the described t-shirt.

If you have a specific person and a t-shirt design in mind, you can describe them in detail, and I can create a prompt for DALL·E to generate an image that represents your idea. Remember, the more detailed the description, the more accurately DALL·E can create the image you're envisioning.

[frigid kiln]

ahh ok thank you!

[dim cradle]

Can Dall E 3 generate variations?

[green pebble]

Pretty happy that my custom gpt bot is performing

[lament egret]

is any version of official openai dall e available for free??

[earnest flame]

only bing is available for no cost. Though you need a microsoft account for it.

[earnest flame]

At one point it was able to refer to the reference_image_id of a previously generated image. I'm not able to verify that that is still working right now. Notably, it can't create variations of an external image. However, it CAN look at an image, describe it, and you can submit THAT to dall-e.

[lament egret]

thank you steel man

[green pebble]

The power of the gods descends down on us. Things are getting good

[green pebble]

The fact that my custom bot can create such an intense photo so fast.... like I'm in love

[stable schooner]

does the image API support seed? I don't think it's possible to generate reproducible images right now

does anyone know what the dall-e-3 api limits are (especially for tier5+)? I only see dall-e-2 rate limits on my account whcih is 500 / minute
yet I am getting rate limited and it says try again in an hour

[zinc dagger]

the api for dalle-3 does not support seeds or previous generation editing as far as I can tell, but the ChatGPT implementation has something (I think it's called generation ID, but I could be wrong) which can take a previous generation to improve on (but it's a little hit and miss).

When it works, it works pretty well for changing details (i've attached some examples of some success I had with changing the gen ID and keeping the images consistent. From what I can tell the GenID is just a convoluted way of setting the seed). I still wish they'd bring seed control back, because the GenID gets confused and starts generating based on some other (but still consistent) seeds if you re-do a prompt or edit a previous message

I like using images with spots (like a cheetah, stars in a night sky, etc) when testing if the gen ID is working because the spots should appear in roughly the same place if the seed is the same after the prompt is revised, even if some other details are changed (ie, even if this cheetah had a hat on or something, the spots on their cheek should still roughly line up if it's using the same seed).

[final jungle]

Good evening folks! How are we doing tonight?

[final jungle]

gpt bot using api or you gave it tailored instructions?

[green pebble]

Tailored instructions

[green pebble]

I really love this

[dim cradle]

c:

[final jungle]

Nice, looks like some topdown rpg, like Final fantasy tactics, i like

[dim cradle]

that's in the prompt, actually. so very good guess

but yeah, i'm a sucker for that style

thanks btw

[final jungle]

all the prompting ive done... i never did any video games. I wonder if i could get it to do chrono trigger...

[dim cradle]

it can do quite a bit!

it's straight up given me cloud a few times

the fun thing about prompts like that, is they're modular, a few tweaks and you have entirely new characters and scenes. mindboggling at times even now, and i've been making prompts for a few years now, quite wild

[final jungle]

sadness, hit my limit making Persistence of Memory Hamburgers...

[dim cradle]

heh. melting burgs eh

[final jungle]

they came out ok, it REALLY did not want to remove clocks

[dim cradle]

yknow, dali would sit on a chair with a spoon in his hand and a plate on the floor, and he'd let himself fall asleep. at the precise moment of losing motor function, he drops the spoon, awakens, and in that state between dream and awareness, he would record his ideas for paintings

axolotl wizard

[dim cradle]

v. nice

[green pebble]

Taking you're prompt and pushing it to the pixel art limit lol

But I do have a soft spot to final fantasy tactics styles

[dim cradle]

this un's easily one of my favorites in terms of the actual pixel style

[green pebble]

Working on anime bot now

[final jungle]

Are you using the new 'explore' thing to create? Or doing it in the custom instructions for your account? How are u saving instructions?

[green pebble]

I'm giving extremely detailed and strict custom instructions

Forcing each bot to focus on one mastery, with a lot of technical terms to achieve the style I'm going for

[dim cradle]

now there's a style. like the oldschool computer games

[dim cradle]

ok last one for a while

[agile peak]

does most of dalle generated images from open ai teams made in open ai cookbook website? were done in bing version..

[dim cradle]

i think they have different llms that act as shuttles or filters for our prompts

[agile peak]

i think i finally able to recreate from one of employee creation

first image was gpt-4 recreation of image i uploaded, the second is when i told to ensure perfectly captured its true essence and authenticity

this is the image i uploaded to gpt

[mild tapir]

can I get help with a prompt I have, I wanna make a few adjustments but I can't seem to describe them adequately

[dim cradle]

yeah, shoot

[wind thunder]

When new dall e update???

[hybrid sorrel]

pretty sure its a timezone issue.

im from australia
today at 1:51pm on monday the 20th i hit the limit and it said it would reset in 10 hours and 51 minutes.
tried again at 7:07 pm and it said it'll reset in 16 hours and 7 minutes.

the timers from both are the time since 3am on the 20th, so earlier that morning thats already come and gone. so its timing for 3am of the wrong date.

only thing I can think of is the bot isnt considering the users DATE, just the current time, which would cause issues for any timezones that are ahead enough (of whatever the bots timezone is) to be in the next day, so timezones that have crossed that 12am point.

ID: 8c68782c-b123-4661-a049-a7c047727eb8
It is a long chat, to the point its incredibly slow and freezes often on a desktop client, I have to use the mobile app.

[vagrant zodiac]

Hey, please use #1154829862171844679 for sharing images, this channel is meant for discussion 🙂

[abstract crest]

hey guys I'm using dall-e for the 2nd time ever. Quick question: is Dall-e as incapable as it seems with instructions, or am I just clueless on how to use it?

It keeps creating images of 918x918 pixels and simply cannot create any kind of custom size image

This is chatgpt, not the playground

[dim cradle]

Nah

[vagrant zodiac]

Images go in #1154829862171844679, this channel is for discussion, thanks 🙂

[dim cradle]

Nah

[lethal cedar]

Guys do you think that as AI continues to evolve, it could eventually replace all designer jobs? What's your take on the future of our field with AI becoming more capable?

[abstract crest]

The future is likely to be more collaborative, with designers leveraging AI to push the boundaries of creativity and innovation. This partnership can lead to new design languages and approaches that were not possible before.

[uneven niche]

Say I prompt Dall E 3 with one prompt, and then I want it to generate another image with the exact same prompt but a different random seed. How the heck do I do that now we can’t refer to seeds and gen_ids don’t refer to the same concept?

I know I can reroll the seed by regenerating the response but I’d really like to be able to keep these images in a single conversation, especially since on mobile I can’t seem to swap between alternate generated answers.

My attempts to get it to generate another image with the same prompt either get “sorry you can only generate one image per prompt” (???) or if I prompt around that, it uses the same seed and generates the same picture again

[vagrant zodiac]

when prompting tell it to use your prompt exactly as written

and just copy-paste for each new request

[uneven niche]

Huh, I didn’t actually consider copying and pasting in the prompt again would work, thanks!

[vagrant zodiac]

and remember to say the 'use my prompt exactly as written. do not make any changes to this prompt.'

[uneven niche]

Yeah I got that bit down

[humble sonnet]

Bing doesn’t like me lmaoo

Wrote a small story for an image and it gave me a mug shot of my sona

(It was about a bike ride into the unknown with a positive mindset btw)

[pseudo kindle]

Is thre a website with loads of Dall-e generated images where you can see the prompts used to create them?

[humble sonnet]

this server has many in the #1154829862171844679 ! Not all posts have them but most do

[stable schooner]

What are the pixel size options for dall e 2?

[vagrant zodiac]

Dalle 2 only does 1024x1024

but you can use outpainting to extend 🙂

[stable schooner]

Thank you

[green pebble]

Made a bot that can make comics

[worthy wind]

Hey. I am looking to get access to dalle because I am able to get on waiting list only. Is there any chance to speed it up and to get premium account for chatgpt and dalle?

[dim cradle]

just use bing image creator. it utilizes dall-e 3

[worthy wind]

how to do this? is that browser?

[dim cradle]

https://www.bing.com/create

sign up and away you go

you get limited tokens a day, but it'll refill.

have fun

[worthy wind]

is this the same dalle there as in openai?

[dim cradle]

essentially, though the two systems seem to use different censorship filters

i believe chatgpt itself edits your prompts before it sends it through, but bing seems to be more lenient and also more effective

[worthy wind]

thanks for your help. i have to try this

[dim cradle]

no problemo

[worthy wind]

Wow, this is stunning! Thanks a lot. Do you know how to get free credits or how much it costs?

[dark sky]

In case you haven't smiled yet today, here's a herd of doggosaurus (DALL-E 3)

[dark sky]

Actually, it's unlimited. You just have a short (<1 min) waiting period between images once you run out of credits. It also refills weekly, to the best of my knowledge

[dark sky]

That's any DALL-E 3 implementation, possibly 1 and 2 as well, but I know for sure vanilla DALL-E 3 edits your prompt before generating. Now that it's available on the API you can even see the actual inference prompt that it makes based on your input prompt

That's why it's so much better than something like Stable Diffusion, you have a sort of LLM virtual prompt engineer between you and the generator optimizing your prompts

[willow sparrow]

Why is this request failling?

does it have something to do with the GPTs feature or something else.

I basically uploaded some chapters from my novel and tried to see if it can generate the characters inside the novel as images

this is the result

[rancid reef]

Why is Dalle giving me issues with placing text on images? it says it cannot anymore and i need an external tool lol whats going on

[dim cradle]

not fully unlimited. after a few in several hours, (like 100 or something) it'll say this for at least 12 hours. Further, there's an API through which i've seen the sort of editing it does via a chatgpt route, however, there seems to be a significant difference between its results and the results of bing on the same prompts. then when i use the "refined prompt" as they call it, it is consistent with the results through the API on bing. My guess is the Bing LLM is far more subtle than ChatGPT. which is unsurprising, i don't care for CGPT's overall qualities as an agent, nor the instructions it has overall.

[dim cradle]

and yes, it reloads daily.

[dim cradle]

credits appear to come at 24 hours after you finish your last batch, the timer doesn't start till you've spent the last one though. it doesn't top you off. you can gain microsoft bucks through their interesting service, seems like there's a lot of options, but you can still make stuff since it's not as insanely popular as it was in its initial surges. now it's laid back enough to pretty consistently produce on non-peak hours

[analog venture]

Will this be a feature? imo it's the most powerful tool for tweaking the generated image

[gray surge]

It will, yes - we only do not know when it will be a feature.

.. maybe you look in to the answers channel from the last ama. 😄

[sinful locust]

that is sick

[agile peak]

i am having a difficulty to recreate “clippy” from the old office assistant lol

dalle made it look like this

[vagrant zodiac]

As a reminder, please use #1154829862171844679 For posting images, thanks 🙂

and use #1021130377026351105 for getting help with prompts

[rigid badge]

a sign of things to come

[drifting hawk]

Thank you!! Time zones are hard…

Will take a look

[earnest flame]

Dall-e improved it!!

[agile peak]

these are another that closely resemble classic clippy 🤣 had to modify some words in the prompts.

[cobalt hull]

Any ideas when these limits will be lifted? lol I feel I can't get a groove on image generation anymore

[rancid reef]

what cap u ogt

[cobalt hull]

I wish i knew but I feel like I'm really not goig that ham on it. Yesteday I was put on an hour wait then did about 10 or so prompts then another hour. Today I did maybe like 20-30 prompts and put on a 2 hour wait or so

[foggy kite]

Dog

[humble sonnet]

Dog

[green pebble]

Thanks I have a bunch of my fine tuned tools open publicly

Have to work on fine running their executions even further. Likely going to be working on adding actions to further increase output and quality

The one that will be the hardest to pull off is actions for the 3d modeling one as I will want to write a script for it to give textual descriptions to send off to blender api to make 3d models for. Not going to be easy. But I honestly think it's achievable

[sinful locust]

Noice, I am working on an anime style
Japanese AI Language Tutor with functioning memory.

If it actually works then I will definitely distro, probably sell it if the MyGPT store actually becomes a thing

[agile peak]

previous one was done on bing, funny how bing couldn’t able to capture clippy perfectly as dalle (chatgpt).. in a single attempt with same prompt lol

[earnest flame]

Hands down one of my favorite fan arts I've ever seen on here.

[cinder kelp]

‌‌‌Why doesn't mine display dalle 3 anymore?

[agile peak]

love the art here ✨ golden era comic style

[hearty ether]

Oh wait did they make it a rolling system now?

So that it regenerates

That’s kinda nice

[agile peak]

finally a full portrait that perfectly capture him! all it takes just discuss with dalle..

[mild tapir]

Is there a way to in paint Dall E3 images

[quartz vale]

Dall-E 3 doesn't currently support inpainting or outpainting

[mild tapir]

oki

[sterile pilot]

Is the Dall-e capable of going off uploaded images for example?

[merry vault]

#daily-theme message AI will take over the world, just you wait

[merry vault]

yes, it can see images

[brisk basin]

@covert ermine Hello, I would like to inquire why the sentence is written "the mans face scratched out", but the picture is scratched out the womens face?

[covert ermine]

Have you tried the prompt yet? You will see.

[brisk basin]

I'll try again. Not sure how to make it more precise.🫠 Thanks

[covert ermine]

Both the man and the woman were being scratched out for me. I kept going then only the woman's face was scratched. Luck of the draw 😅

[brisk basin]

Ha, I get your means now, thanks.

[potent briar]

DALL-E decided to take a holiday today, it state everything I wrote its against content policy.

[frosty nimbus]

Is it just me or is this getting more locked down to prevent you from creating pictures of yourself in well known art styles? A few weeks ago I saw people making GPTs to turn themselves into Simpson characters, but now I can't get it to do anything that isn't minecraft

[potent briar]

yesterday I created beautiful pictures, today everything agains a content policy, only accept a a very simple very short prompts and creates ridiculous simplified pics.

[frosty nimbus]

ok... so not just me X_X

[potent briar]

It feels like trying to perform an AI-level task with a punch card.

[wispy storm]

What was the prompt

[potent briar]

I wrote nearly 25 different prompts and even asked GPT to describe the issue with the prompt and to rewrite it. However, GPT described its own prompt as violating the content policy. So I'm certain the problem lies with the policy, not my prompt. I don't appreciate it when we are enticed with new features only to have the content policy edited in a way that diminishes quality.

[wispy storm]

Try a new chat, without custom instructions .

If you try protected IP it won't work

[potent briar]

"I've tried every possible solution. New login, refresh, re-generate. I'm not using a VPN when I use GPT, it's obvious that it doesn't work. The simplest 1-sentence prompt (5-25 words) is the only thing that works, but I can draw that myself.

[wispy storm]

I suggest to write everything you tried in #1070006915414900886

[agile peak]

anyone have suggestions to keep dalle consistently produce images like the right one? with all of those tiny dots details.. or is it always random?

[runic granite]

feed back, ask style

its like 80/90s comic style

sorry 70-80

offset or whatever was the old printing method...

[agile peak]

Ah got it!

[runic granite]

i think it caused it

[agile peak]

Just asked gpt yeah, my bad now i can do it consistently with those tiny dots.

[runic granite]

what was the answer? 🙂

[agile peak]

the answer is called “ben-day dots” mechanical printing method developed in the late 19th century.

[runic granite]

I didnt really missed....

😄

can i check it on my drows?

BTW the first right is really good.

[fallen mulch]

how can i generate more than 1 image

[runic granite]

ask for it

bu i get constant errors

but i ask for 4

[runic granite]

maybe it allow to

two

[fallen mulch]

I want to use the same seed

[runic granite]

prev it was possible.... now everything is a mess

[fallen mulch]

mhh

I wanted to generate yt background images

[runic granite]

save image then outpaint

[fallen mulch]

outpaint with which tool

[humble sonnet]

Gotta wait for that daily theme so i can make somethin hehe

[agile peak]

you can use same image by applying same generation_id

such as style, colors, etc

now that i already achieved dark fantasy pulp art, still struggling to recreate that one of open ai employee post on reddit.. “blade runner” scene..

[fallen mulch]

how

and why only 1 images

[agile peak]

thinks this is one were achieved before open ai/bing increase the filters isn’t it..

[agile peak]

did you used default model gpt-4?? if yes then it’s the reason.

because dalle•3 is always producing 2 max of images

[fallen mulch]

yes do I need custom gpt

[agile peak]

no need

[fallen mulch]

I had 4 images last time

what version?

[agile peak]

yes, they changed it.. we are now only able to produce with 2 images maximum..

[agile peak]

just click on dalle?

[fallen mulch]

ah lol

I thought everything is in the same

[agile peak]

default gpt-4 = maximum one image
dalle = maximum two images

[fallen mulch]

thx

how can I get it done

[humble sonnet]

You’re looking to make them widescreen, right?

I personally ask it “Can you make these 1080p wallpapers?” And it formats it to the right size!

[fallen mulch]

look what i got

[humble sonnet]

ooooh, noice!

[fallen mulch]

no actually bad

[humble sonnet]

oh yea it’s a lil stretched

try asking it to try again, see if it makes a better one

[wind grove]

Maybe tell it the aspect to ratio you want?

Like 16:9 I guess

[fallen mulch]

let me check

I dont get a similar one

[wind grove]

Maybe say: no I want the exact same one I just had before?

Weird though

[humble sonnet]

Try to ask it to make a new image using the same prompt, that usually fixes it

[scenic cloud]

Won’t DALLE be able to use an image of let’s say, my cat and then make a new image of them more accurately in the next update?

[fallen mulch]

this doesnt look like it

[scenic cloud]

I remember something about that in that questionnaire they had

[fallen mulch]

I use chat gpt and then dall e

[fallen mulch]

not working

[scenic cloud]

Well yea

It hasn’t updated

[fallen mulch]

Last time I could use the same seed

[fallen mulch]

where is this

[scenic cloud]

It’s archived

[fallen mulch]

okay

[scenic cloud]

I screenshotted this because I wanted to remember

It’s something I’m really waiting for

[fallen mulch]

ye

[scenic cloud]

I’ve been using DALLE to create images of my Pokémon Trainer and even though I’ve taught my current chat to get consistent results “mostly” image references would make my life easier

[fallen mulch]

lol

[fallen mulch]

mh

[sacred temple]

so these are the same raw prompt in both Bing & GPT with obvious different results? Even the Dalle3 API results are different...

I wonder why they are different?