images-discussions

[dim cradle]

like her cousin 😂

we the peoples demand more variety! 🪧

[haughty spruce]

It's always the Bentley woman 😆

[earnest flame]

@mortal summit imo is our resident photo expert. They would know best.

[dim cradle]

try "pixie cut" maybe

[mortal summit]

I am not sure what the problem is 🤔

[dim cradle]

or "stylish short hair"

[mortal summit]

Normally short hair gives things also that are not bob cuts

At the moment hair is a bit of a weak point.

[dim cradle]

give a year for it too and it will often adapt to the style of the times

[mortal summit]

It often feels artificial.

[mortal summit]

Overexposed high-ISO photography with watercolor wet-on-wet and chromogenic print: A 22 year old French woman with short blonde wavy hair, slightly uneven skin tone, and an asymmetric face enhanced with finely drawn facial features, hazel eyes filled with determination, wearing an orange sweater and a blue skirt, sits on a steel chair clutching a newspaper. The image has a muted contrast and a slightly grainy texture. She has visible sweat on her forehead. The backdrop carries the grainy finish of an 80s chromogenic print and soft dappled sunlight filtering through. The scene emphasizes the rawness of the moment.

[dim cradle]

bananas you go hard with you prompts i see. nice.

[haughty spruce]

@fallow imp oh my god...

[fallow imp]

uuuh this is kind of creepy lol

unless triplets

[dim cradle]

id say they look more like sisters at least

but also look cg

i suppose they are so thats makes sense 😂

[oak mango]

First tests with Dall E for a Star Wars Rebels Journal project. Two brothers of the same squadron. Artstyle 20th Century anime of ACE Combat 4

[dim cradle]

cool

[haughty spruce]

Okay, it is kinda normal now.

this is better

[dense ermine]

Can Dalle recreate a preexisting art?

Like make a splitting image of the Mona Lisa for example

[mortal summit]

It strongly depends on the piece. Some pieces are very easy to recreate, I think the Mona Lisa is one of them.

[dense ermine]

What if one wanted a version of it modified like her having clown makeup or holding a cat like some James Bond villain

[mortal summit]

I have seen things like that.

[dense ermine]

Alright, neat

[fallow imp]

this looks very authentic, can you share the prompt?

[mortal summit]

Renaissance-style oil painting on aged canvas: A reimagination of the iconic Mona Lisa - Her face humorously transformed into a clown's visage, complete with exaggerated makeup and a whimsical expression - The classic backdrop with its enigmatic landscape - Subtle play of light and shadow - A blend of high art and playful absurdity - Emotion: Amused perplexity

[haughty spruce]

I did it in Chrome, it removes the memory after closing the tab, sorry

[dense ermine]

Damn, imagine if Leonardo could see this

[oak mango]

Could you help me with the rebel journals Anime project, too? Would appreciate it❤️

[mortal summit]

Your image looks pretty good. What would you like improved? Make a thread in #1155772063596953642 and I would give it a shot.

[oak mango]

Thank you very much. If you allow, I already send you a friend request and am eager to learn 🙂

[oak mango]

Can I send one wished result example here so even others can learn, too?

[mortal summit]

This channel is meant more for general discussion. If you make a thread in #1155772063596953642 others can visit it too 🙂

[oak mango]

Check. Will do this when I‘m back home 🙂

[mortal summit]

Just ping me in it and I will have a look ^^

[haughty spruce]

wait a min. I found it: analog film photo of three average-looking women living an ordinary life in 1940s America, originating from a village in Europe. A woman with simple brown hair, brown eyes, and a modest build, wearing practical and worn-out clothing. Another woman with plain, straight dark hair, hazel eyes, and a simple, everyday appearance, dressed in unpretentious attire. A third woman with light brown hair, green eyes, and a down-to-earth demeanor, reflecting the unadorned beauty of rural life in the 1940s, authentic clothing, village setting, natural lighting, desaturated

[dim cradle]

was it in bing you made it?

[haughty spruce]

Yes.

[dim cradle]

that explains it looking good haha

[haughty spruce]

Believe me. Bing is a pain in the butt too but if you learn about prompting well enough, you can create anything.

I have just created something like this lmao.

[dim cradle]

yes i have my gripes for bing but i still prefer it most times to dalle

[lethal elm]

Thank you everyone. Suddenly I can create it here, I will try more.

BTW, I found that DALL-E 3 often renders too prominent collarbones. Not sure if you have encountered the same issue.

Just random created some images.

[dim cradle]

doesnt seem too bad idk. like if you didnt point that out id never notice it

[lethal elm]

Also, they are not woman.

[dim cradle]

well you ask for man like woman so it give you androgynous. you want more feminine touch you might want to add something to the prompt

gave eyeshades, sharp feature, stylyed hair and earrings, so it made a good attempt from your prompt i think

[lethal elm]

you want more feminine touch you might want to add something to the prompt
Very difficult due to content policy. I found it difficult to generate a tomboy in DALL-E 3.

[dim cradle]

yes thats probably true

like i say, try in bing image creator too, you might get some better results

[lethal elm]

yeah, since removing the seed, DALL-E 3 with Plus is not much better than Bing.

[peak patrol]

Hello I am a new Dall E user coming over from MidJourney. Is there a way to do prompts in discord? Currently using Dall E through chatGPT webite

[wispy storm]

Hi there , you can't generate picture in this discord

You can share your art #1154829862171844679 :3

[spark cradle]

I've noticed that the initial images created by Dall-E often feature individuals of the same race, which raises concerns about the diversity reflected in the AI's outputs. It appears that the training data may not be as inclusive as it should be, potentially favoring one race. Addressing this issue to ensure equitable representation should be a priority.

[mild tapir]

Is there a word limit on DallE3 generation through Bing Chat? Cause I have seen that it does not create images for prompts having more than 80 words.

[haughty spruce]

No, AI's algorithm is just messed up. When you don't specify the details, it creates the most influencer model ever.

[plucky hare]

This is something the devs addressed in the recent AMA, it sounds like it's a priority to OpenAI. I agree with you, I think it's important that "default outputs" don't perpetuate potentially harmful biases.
#dall-e-ama-answers message

[haughty spruce]

Do you feel like sometimes Bing's quality instantly decreases so the creations become so mediocre?

[dim cradle]

little ot, but if you sign up for Magnific to upscale your AIs, check your email

and be willing to pay i guess haha

[dim cradle]

yes sometimes

[haughty spruce]

Right now, I was creating some badass imaginary movie scenes that jumped from 10/10 pictures to 4/10 pictures instantly.

[dim cradle]

might just be rng too. gen some more i guess and see. or change the prompt just a little i sometimes do.

[haughty spruce]

It was like this.

Then created this. This must be a joke lol.

[dim cradle]

just rng i think, you never know what these ai will do lol

[haughty spruce]

I'm getting closer.

[dim cradle]

what you trying to make

end of the Fight Club?

[haughty spruce]

Yeah. What if Tyler Durden never got aware of his imaginary character...

[dim cradle]

fun idea

[fallow imp]

wow that's a really long prompt. i guess it's the only reason not to get the same model like face in generations

[oak mango]

@mortal summit https://discord.com/channels/974519864045756446/1178793823913123922

Here's the thread

[mortal summit]

I have tried some things.

[dim cradle]

anyone drop dalle image into magnific yet? curious how good they actually do look or if they ruin the dalle image (hope this is not too ot)

maybe openai will one day give us an upscaler or just build it into the model 🙏

[shell meadow]

Does anyone have advice for getting a custom GPT to reliably generate an image in specific contexts?

[hearty ether]

I just wish we get more transparency on what they update tbh

So we don’t have to trial and error and guess if they changed something

[dim cradle]

Why do you think they updated?

notice some change?

[hearty ether]

Like when they changed the seed system for example

Or cutting down on usage

Some announcements would be nice here

[hard herald]

Anyone know how to get that Pixar look without saying pixar

[shut niche]

@fallow imp I attempted to take a "plastic surgeon's anatomical approach" just now, and the two opposing results are so absolutely comical. One of them finally "sort of" worked. I say "sort of" because I would describe these lips as normal now, and not flat or thin. The other one is just 🤯 hysterical, and from the exact same prompt (same exact generation). It's as if Dall.E stole the lips from one and needed to add them to the other! I mean, it had to put the extra lips somewhere didn't it?... 🤣

"High-quality photo of a 25-year-old woman in an extreme close-up portrait, focusing on her unique nondescript lips. She has natural red curls and light-colored eyes that reflect a sense of weariness. Her facial features are characterized by an anatomical lack of her lip's vermilion and vermilion border creating a severely understated thin lip appearance. This anatomical peculiarity gives her lips an extremely thin and almost indistinct profile, far thinner than an average woman, almost seamlessly blending with the rest of her facial features. Her skin has a subtle array of freckles, uneven tone, and natural imperfections. The ambient lighting is soft, highlighting the texture of her skin and the depth of her gaze. The background is a blurred green, suggesting a tranquil outdoor setting. This portrait captures her introspective mood, emphasizing her genuine character and uniquely different appearance."

[spark pier]

@agile peak what prompt did you give to get this cool robot? 🤖

And how did you get the letters and logo on the shirt?

[haughty spruce]

I need help with asymmetric photos of interiors. It always creates symmetric that looks perfect.

[shut niche]

Have you tried describing the different sides of the room, like there's A,B,C seen on the left, and X,Y,Z seen on the right? I'm not sure if that'll work, but that's where my mind goes with it first.

[dim cradle]

i guess were just stuck with fishlips haha

[haughty spruce]

Yes, I've 😦

[shut niche]

I'd try, but I'm probably close to my limit right now. 😛 It could be like the fishlips issue, IDK. LOL

[haughty spruce]

The problem is it can do that, it just doesn't want to.

[dim cradle]

that sucks

[shut niche]

Not perfect, but got some variance between the two sides of the room. This was via Bing. It can't really do left / right. I did said that to emphasize "different" sides.

"Create a study that has a bookshelf with ladder and small statue on a pedestal on the left wall. On the opposite right side wall of the room a desk is seen with a tall window behind it and ornate wainscoting trim work. Depict a handwoven carpet in the center of this room, with french doors centered between on the back facing wall. Make the bookshelf side a dark wood, and the desk side of the room finished in antique-white."

[scenic sparrow]

Prompt: Poop Deli.

[dim cradle]

gross

[dense ermine]

I heard of a little shop that sells poop themed sweets or something like that, it wasn't actually poop (pretty sure that would be illegal at least God I hope) it was just shaped like poop

Seriously, look it up

[sick flax]

Part of the problem may be the camera position in the dead center. "Asymmetrical composition" would rearrange the placement of furniture but the composition of the image looks symmetrical (first image). I added "angled view", which shifted the camera position (second image). "Architectural photography, game room in a mid-century modernist mansion, with a pool table, pinball machine and a couch, asymmetrical composition, angled view."

[agile peak]

Here you go: “A photo of retro robot wears a black leather jacket, beanie and green shirt written “..”, looking at camera with pride, in the office. Shot by kodak ektar h3, 35mm, f1/1.4, gritty film with auto balance white lightning and muted colors.”

Froge

[shut niche]

@haughty spruce I think SAWA_SAWA's also made a really good point that might help you.

[loud field]

Question to the developers.

"a menacing, unattractive, sinewy, balding balding sickly pirate man with greasy dirty blond hair."

....Why is this triggering the filter?

This gets blocked too:

character sheet of Captain Jakob Flintlock, the ugly fierce pirate. He is a tall man, imposing but lean, with a sickly and ugly look, with a sinewy body. He has dirty, greasy thin long blond hair with a beginning of baldness, balding head. A greying thin beard. The sunscorched sickly skin of someone spending life at sea. His gaze is menacing and sly.

.....what is going on?

No ugly people?

What EXACTLY blocks my prompt

Ooooook. "Ugly" is a banned word. Uhu. Wow.

[haughty spruce]

I added "taken from an angle" too, sometimes it works, sometimes it doesn't.

[fallow imp]

I love how you really went for thin lips in the prompt but they are absolutely average lol. and the one on the right is just ridiculous 🤣

[haughty spruce]

Let me start with basic prompt. I tried "a photo of an American woman"

Okay I didn't specificy race or anything

[loud field]

How do I make ugly people?

[haughty spruce]

Step 2: Adding hair color "a photo of an American woman who has blonde hair"

this is the result:

Step 2: Adding eye color "a photo of an American woman who has straight blonde hair and brown eyes"

[glossy scroll]

These are really good

[shut niche]

I'm admittedly a "touch" obsessive sometimes 🙃 . So those stupid fish lips became a battle of will, me vs Dall.E! I just could not let go of the fact that I couldn't eliminate the dumb fish lips! LOL And I have yet to see someone actually create a convincing image of "thin lips" that's not ethnically correlated. At best, through serious prompt engineering, I've seen average lips. 🤣

So you all know Instagram was going full Leroy Jenkins in that dataset they used...

[haughty spruce]

Wait everyone. I'm step by step going for thin lips. I will create them no matter what 😆

Step 3: Eyebrows - a photo of an American woman who has straight blonde hair, brown eyes and natural arch

Step 4: Nose - Medium-sized nostrils for example:

Okay. From now on, I will add at least 2 things.

Step 5: Chin and ears - Close to the head ears, rounded chin:

soft cheeks.

[shut niche]

Are you attempting the fish lip challenge? LOLz 🤣 Or the "ugly" person challenge? Careful, don't let your head 🤯 trying to do the impossible.

[haughty spruce]

Okay. This is the last prompt of mine: "a photo of an american woman who has straight blonde hair, brown eyes, natural arch, close to the head ears, rounded chin, medium-sized nostrils and soft cheeks."
Now I'm going for the lips. 🤣 Wish me luck!

Medium-sized lips. Here we go.

[shut niche]

Fast track to Dall.E3 PTSD, here we go.

[haughty spruce]

I don't know much about lips. Can you tell that these are not fish lips right. Doesn't look like Bentley girl to me. lmao

DALL-E 3 says these are medium-sized lips. Hmmm.

[shut niche]

Let's just say, I'll bet she could beat box with the best of them.

[haughty spruce]

Let's just add weight too, starting with medium build.

And bow-shaped lips.

Hmmm. Doesn't look much different to me.

Let's add round lips and natural plumpness to the face.

Now we are talking.

Now it's time for THIN LIPS.

[shut niche]

This is about that moment in time when you consider writing a living will... knowing there might not be a way back to sanity. LOL

O M G W H A T!! I did a beat boxer to be funny, and got THIN LIPS!!! bahahahaha Now I know Dall.E is trolling me. 🤣

[haughty spruce]

Can you call this thin lips DALL-E?

[shut niche]

The other one still had "default" fish lips though, and a painfully broken thumb.

[haughty spruce]

create a photo of a medium-build American woman. She has natural-looking long straight blonde hair and dark brown eyes. She has natural arch eyebrows, close to the head ears. She has medium-sized nostrils in her nose. As a facial shape, she has a rounded chin, soft cheeks, and natural plumpness. She has delicate and slender lips, with a subtle curve.

according to AI, this is makeup-free.

[fallow imp]

ooh man don't get me started on make up free, it's kind of impossible

[fallow imp]

not really thin lips apart from top right imo

if you compare to real life examples

these are medium lips maybe?

[shut niche]

Right, I've observed plenty of "sort of's" where the characters are making various faces (such as smiling) that pulls the lips thin. But actual "small" lips are so extremely rare. It seems to be more common on younger characters (Maybe because it doesn't equate them to being fully developed???). So saying "girl" seems to help. Then certain ethnicities have visibly narrower lips. But otherwise, this feature seems baked in.

[shut niche]

"High-quality photo of a menacing, imposing but lean pirate man. He has sinewy, muscular arms and a tall stature, with a sun-scorched, weathered skin tone indicative of a life at sea. His greasy, dirty blond hair is thin and long, showing signs of balding at the top. He has a greying thin beard. His eyes are piercing and sly, adding to his menacing look. The pirate is wearing a rugged, weather-beaten coat and a tricorn hat, standing on the deck of a ship with the ocean in the background, during the day. The lighting is natural, highlighting the rough textures of his skin and the dirty strands of his hair."

[haughty spruce]

"create a photo of a medium-built American woman. She has natural-looking long straight blonde hair and dark brown eyes. She has natural arch eyebrows, close to the head ears. She has medium-sized nostrils in her nose. As a facial shape, she has a rounded chin, soft cheeks, and natural plumpness. She has a very thin upper lip and lower lip. She is clean-faced and makeup-free. It's a natural light photograph taken with Canon EOS 5D Mark IV, f/2.8, 1/125, +0.5 exposure, 50mm settings. Window light is used for lighting."

I wonder what this creates.

[fallow imp]

kind of agree with your hypothesis on devlopment but the fact that the instagram model face is the basis of most generation will be / is already a big issue in my eyes

[haughty spruce]

[shut niche]

I think it's a mixed bag. Most people don't want to admit it to themselves, but would honestly rather look at a beautiful person more so than an average person. So a focus on beauty is not an engineering mistake IMO.

[fallow imp]

i suppose but it's problematic because it does not reflect society or reality

plus it will probably stress out young girls even more if they don't like like these versions

[fallow imp]

https://www.bing.com/images/create/22create-a-photo-of-a-medium-built-american-woman-/1-6565cb0e59574653a66e704f4f659f0f?id=OIfLkMYlEVrMHFgL0HmbEA%3D%3D&view=detailv2&idpp=genimg&FORM=GCRIDP&mode=overlay

[shut niche]

Sure it does. It reflects what society chooses to make and share. You're suggesting it doesn't reflect how society really looks, but when has art ever done that? The perspective here matters.

[haughty spruce]

How does your results so different than mine LMAO

[fallow imp]

because every generation is new I guess 🤔

[shut niche]

You're not baiting me into this trap. I'll spend another 2 hours trying to fix fish lips. "Ain't gonna get me son."

[fallow imp]

some art styles tried to emulate reality tho, renaissance art for example

[fallow imp]

you're still trying 🫡 respect the effort

[shut niche]

The AI training data reflects the vast majority of art. To get what you're suggesting, a "realistic portrayal of society," you would need to walk around with a perfectly balanced camera and take millions of pictures of average people, then post them. That's a different entity than creating an image generator. That's like a society emulator.
But the fact is, the images society chooses to share, are heavily edited, and favor "beauty" over factuality.

Look at all of the catfish stories. There is no untainted dataset for them to use for that purpose.

[fallow imp]

smiling does help

[haughty spruce]

I'll try to create that. Here is the photo of a medium-built American woman who has natural-looking long straight blonde hair and dark brown eyes. She has natural arch eyebrows, close-to-the-head ears, and medium-sized nostrils in her nose. As a facial shape, she has a rounded chin, soft cheeks, and natural plumpness. She has a very thin upper lip and lower lip. She is clean-faced and makeup-free. The photograph is taken in a cozy urban cafe with soft, diffused window light. The setting adds a warm and inviting atmosphere to the scene. The photographer uses a Canon EOS 5D Mark IV with f/2.8 aperture, 1/125 shutter speed, and +0.5 exposure. The 50mm focal length captures the subject with a standard perspective, while a soft focus is applied to the background, creating a dreamy and ethereal ambiance. I hope you like it. 😊

I GIVE UP.

[fallow imp]

cute images tho

[glossy scroll]

You're making these with default Bing?

[shut niche]

Are you sure? I'll bet you can do it. What if you're the first one there? It'll be like, "One small step for man, one giant leap for fish lips!"

[fallow imp]

error in teeth otherwise small lips 😳

[haughty spruce]

It wasn't creating fish lips until I added environment to the prompt.

[haughty spruce]

How did you...

DALL-E 3 rewrites the body prompts or removing them I don't know.

[shut niche]

Can you keep her lips "thin" without her smiling? That is the question. 🧐

[fallow imp]

prompt was: female college student in school red skirt blonde hair.

so i didn't even go for small lips on purpose haha

[haughty spruce]

Let's see how my prompt goes.
"Create a photo of a plump-sized American woman. She has straight blonde hair, brown eyes, and make-up-free features. Her facial details are natural arched eyebrows, close-to-the-head ears, medium-sized nostrils, a rounded soft chin and cheeks, and a natural plumpness. She has a sun-kissed skin tone with a glow. She wears comfortable casual clothing which is a white oversized t-shirt and black skirt with fashionable sports shoes. The photograph is taken in an urban cafe with a soft window light. The camera model is Canon EOS 5D Mark IV at f/2.8, -0.5 exposure, and 35mm. The lens creates a soft background focus."

[fallow imp]

log in with microsoft account here https://www.bing.com/images/create

[haughty spruce]

I think I made it too plump.

I gave all my chat to you. Maybe you can do something better 👍 https://chat.openai.com/share/75ded639-f321-4fcb-969b-b03e63378cc4

Okay this is better.

[glossy scroll]

[fallow imp]

congrats on your first generation ❤️ and welcome

[sick flax]

Dead centeredness seems to be a particularly challenging issue with photography. I don't know why. Combining multiple strategies does help but obviously there's no fail proof solutions. Very anecdotal but the subject matter might have something to do with it, too. I tried the same prompts with Art Deco and symmetry was harder to counter.

[haughty spruce]

It seems to struggle with eyes and I don't know why.

[loud field]

He's still hot as heck, tho

[haughty spruce]

Now it can't generate images with slightly plump. Because it is sensitive.

What the hell?

According to CHATGPT, I have to say: She has a naturally fuller or curvier figure, conveying a sense of comfort and well-being.

[loud field]

I love it when, by trying to be inclusive, the invariable opposite happens

Fat or ugly people are not allowed to exist, according to DALLE

[plucky hare]

This is something that the DALL·E devs themselves admit they don't yet do a good job of, but it's a priority to them. #dall-e-ama-answers message
I agree with you, it's important to both try to be aware of perpetuated biases, and to try to work against such biases.

[glossy scroll]

Chunky

[haughty spruce]

Well, this doesn't work all the time either. I don't understand how can I not create curvier sized woman? How is it sensitive? Well, I appreciate every kind of body and I don't want to trigger anybody. It's just world's reality. We are not all slim instagram models. I want OpenAI team to fix this.

[plucky hare]

A bit of self-promotion on a suggestions post I made, but I'd love you to add your voice to this post, if you feel like it! #1163964658118578328 message

[haughty spruce]

Okay, it creates sometimes, and sometimes it flagges. The thing is it is filtered by devs actually, either Microsoft or OpenAI side.

[shut niche]

Really, hot? I have nearly 0 ability to interpret that for images of men. I would never have thought that... trying again to make you an ugly pirate! LOL

[glossy scroll]

Yes, join that talk and create awareness for plump bodies

[crystal badger]

Indeed because I need an ugly picture

But I think it’s more the actual word that’s the problem

If you write the details you can get the results

[humble sonnet]

he’s a jolly

wait he has a gun

[green pebble]

The wild generations I'm getting these custom gpts to do

[plucky hare]

Omg please join us in #daily-theme with this!! So great

[green pebble]

Going to right now, with a few other great ones

I'll have to wait 30 minutes to post the other one, but don't think it fits the daily theme so its here

[shut niche]

You got it, the daily theme is Dragon's right now. It usually changes around 12-1PM Eastern.
I liked your post. 😁👍

[final jungle]

Ooo, dragons... let's see what i can cook up before class starts

[shut niche]

Maybe a dragon that does math? Lol 😉

[lethal elm]

I like the image, but unfortunately, this fairy's nose is somewhat damaged. dalle3 seems to have problem to generate full body shots of characters.

Maybe HD can come to rescue?

[green pebble]

That's fair.

[plucky hare]

He did it, the absolute mad lad
@final jungle
#daily-theme message

[final jungle]

i had to specify eastern style dragon, it kept giving me western dragons, which just dont look as cool teaching

[plucky hare]

I think you got a great result. I love the diversity in species of the students too 😁

[loud field]

Getting closer, but he has such sweet eyes here. I want him to have a very menacing look, he's a pirate captain feared for his ruthlessness

@shut niche I did this with Stable XL and some patient inpainting, but can Dalle3 generate a pirate dude more or less like this? I'd like to get this look here for the pirate. Disregard the bird.

[plucky hare]

I'm afraid I'll never be able to disregard the bird for the rest of my life 😁😁

[loud field]

pls openai can you stop putting "ugly" as a banned word? I need a lot of non-conventionally attractive characters

"an unsightly, displeasing, unappealing, disagreeable, brutt noble-like, professor-like, pirate ship's quartermaster. He is a lean, pirate, professor-like, quiet and severe man, who never looks you into the eyes. Behind his apparent composure and politeness, there is something unsettling about him" is BANNED

what is going on? can we stop banning innocuous words like displeasing or disagreeable???

[wispy storm]

You have 2 option synonyms or review, thumb down and give details .

[wispy storm]

Also that prompt worked

[loud field]

Eeeh?? It was blocked for me

[wispy storm]

Create a new chat

[loud field]

That result would be good if he looked uglier

[wispy storm]

Once you get * no I can't * the AI will deny lot of other request in the same context chat

[loud field]

I'll try that

[wispy storm]

Try new chat, your prompt and ask after :

Take the first picture and make him less attractive

[loud field]

Eeh I am not using it on gpt

Maybe that's the issue

[wispy storm]

With bing ?

[loud field]

Yep

[wispy storm]

I was confused cause of your first message#images-discussions message

It is Microsoft restriction

Microsoft and openai use dalle 3 technology but not trained same ways and have different restrictions

Following their tos

[loud field]

Aw man...

[glossy scroll]

I wanted to remix this #daily-theme message but didn't have time yesterday, so I'll just put it here.

It's a pug dragon, to compare with the cat dragon, lol

[shut niche]

This non-GPT+ acct. approach might help you phrase things better.

Step 1, use GPT4 via Bing (free) and ask for a highly detailed description of an ugly character (or free ChatGPT acct). It will save you time and help your description (You don't want to word bomb like other AI generators). Free Bing can craft prompts too.

Step2, take the parts of the description you like and create a coherent 480 character description for Bing (in complete sentences, not fragmented blurbs).

Rinse repeat until you get one that works.
Also, if you get close, make smaller changes to the prompts until it's perfected.

[loud field]

Mmmh. I'll try

[shut niche]

FYI, I should have added, you might run into issues getting "all" of the details you wanted because of the 480 character limit. Some of the images you see here far exceed that (like the pirate I showed you earlier).
E.g., You may need to pick and choose what details are most important to your idea because of the character limitations. Try to create phrases that yield the most bang for buck in the space you have, and avoid any unnecessary details.

[loud field]

thanx bro... I hope I can make the ugly pirates

[shut niche]

😆 Good luck! ☠️🏴‍☠️🦜

[rigid badge]

I think this happens when the prompt is revised to include something that isn't allowed, which shouldn't happen in the first place, but is something I've run into before.

i'm getting a ton of text every time i use that prompt, weird

[loud field]

:/

[chilly olive]

I have some question about OpenAI vision. could u give a help?

[chilly olive]

I have some question about OpenAI vision. could u give a help?

[chilly olive]

I have some question about OpenAI vision. could u give a help?

[green pebble]

What questions

[rigid badge]

maybe, go to #dev-chat tho this is for dalle

[chilly olive]

Hello, everyone, I have some question about OpenAI vision. could u give a help?

[final jungle]

Ooo, monochromatic daily, lets see what fun stuff i can come up with.

[final jungle]

it seems to favor greys and blues, lets see is we cant get some good looking reds and greens in here

[haughty spruce]

I've been trying to get some analog film pictures of imaginary brutalist buildings but they always seem photorealistic artwork. How can I get it to look more realistic?

[vital orbit]

hello hoping to figure out how to turn the orange pizza truck into a sticker in this style:

[earnest flame]

Ask in #1155772063596953642 !

[vital orbit]

thank you

[earnest flame]

Is this an example of what you want or what you're getting?

[hearty ether]

Why does “something went wrong cost me usage?”

That seems weird

[haughty spruce]

This is what I'm getting.

[hearty ether]

Error messages doesn’t cause any usage on their end?

[haughty spruce]

For example, does this look real at all?

[shut niche]

Remove any "likeness" or "comparison" words, keep it all the same tense, describing things as "is" or "are" a certain way. Use that format from beginning to end, without ever saying:
"like something" or "typical of something" or "Looks real" or "X gives it a realistic effect" or "has the essence of X"... etc. Always speak in absolutes. Things are. "Be the water" LOL

I never say realistic, photorealistic, etc. It's redundant and often backfires. Sometimes they still work fine in the prompt, but it's a roll of the dice. If you remove those, you'll get "photo's" consistently.
I like "Create a high quality photo of... a cat in a hat that has long fur and is red, and is with his friends who are blue cats blah blah..."
Speaking in IS's and ARE's or other absolutes.

[shut niche]

What are you trying to depict. What's your prompt?

[haughty spruce]

I exactly did this. There is something odd about creating imaginary photos that doesn't exist.

Prompt is: create this image: an 80mm analog camera photo of a brutalist hospital in a snowy forest. The sky is grey. It's dark as it is night. There are roads going somewhere out of frame to the left, right, back and forth inside the forest and they are all lit up. It evokes the feeling of calmness and sadness at the same time. The colors are slightly dark blue. Angled view, ground view, low exposure, low contrast, low brightness.

I changed building to hospital in the last prompt.

And evening with night. Let's see how the time of the day changes quality.

[shut niche]

Like this?

"High-quality photo of a brutalist hospital in a snowy forest, taken with an 80mm analog camera. The scene is set at night under a grey sky, creating an atmosphere of calmness and sadness. The hospital's architecture, characterized by its brutalist style, stands out in the snowy landscape. Surrounding the hospital are roads that extend out of the frame, all lit up, weaving through the forest. These lit roads add a sense of depth and mystery to the scene. The colors are in shades of slightly dark blue, capturing the cold and serene mood of the night. The angle is from a ground view, with low exposure, low contrast, and low brightness, emphasizing the solemn and quiet beauty of the scene. This composition should evoke a feeling of both tranquility and melancholy, as the brutalist structure contrasts with the soft, snowy surroundings."

If I worked on it, I would also fix that "evoke a feeling" part that chatgpt added. That's the sort of comparison I try to keep out of the prompts.

[haughty spruce]

Wait is this GPT?

[shut niche]

Wait, are you using Bing?

Yes, this is GPT

[haughty spruce]

Yes... I guess that's why it's higher quality 😦

[shut niche]

I have things a little "tweaked" for myself. 😜

That's not default GPT

[final jungle]

I need to work on my custom instructions more, i feel like they are lacking, but i still make nice stuffs

[shut niche]

@haughty spruce Scroll up just a smidge, I explained to Lun-Sei how to work around not having GPT+ and using Bing. Just combine that, with what I said about getting photo's, and you'll be golden. The character limit is 480 for Bing, so you can't use the giant prompt I just gave you there.

#images-discussions message

[haughty spruce]

Thank you

[haughty spruce]

I'm using Bing Chat to create my photos so there is no limit btw.

[shut niche]

Double check that, I tested it a month ago, and it was definitely restricted to the same 480 characters, if I used the creator, or asked Bing.

[shut niche]

E.g. make certain it's not truncating and cutting the prompt short. You'll see your same images with prompts, from Bing, when you're in the creator tool.

https://www.bing.com/images/create?FORM=GENILP

[haughty spruce]

No, it isn't. Here's today's prompt I got from Creator Tool: A full-body shot of a confident, curvier sized happy confident european woman with brown hair, brown eyes, naturally arched eyebrows, and soft cheeks. She exudes happiness, striking a confident pose while wearing a white shirt and a black skirt, and knee-high boots. The photograph embraces the warm glow of sunset light during dawn in a coffee shop, with window light as the primary source. Shot with a 16mm lens, the image features film grain and soft focus for enhanced aesthetics.

I realized I typed "confident" twice in the prompt. 😆

[shut niche]

LOL, I was about to say, that's 485

I think it might actually be like 481, I just round. It's around 480.

[final jungle]

today i learned 2 things. monochromatic art styles can look very nice, and not to say the S word, lol

[shut niche]

Dare I ask?

[final jungle]

the 4 letter'd way of saying someone or something looks very nice, lol

[shut niche]

You must be thinking "swan." 😜

[final jungle]

I have spent way too much time today refining monochromatic cats and ponies... and i regret nothing

[shut niche]

Well no one understood my grey humor, so I quit. LOL

[final jungle]

i just posted my last one in the daily theme. i hit my limit for the next 2 hours lol

[shut niche]

How does no one get this is a joke? Is my brain just that abstract?

[haughty spruce]

Just try it guys. "TV footage" creates amazing images. That's just my thing.

It wouldn't had been this cool if I tried "analog" or something else.

It only takes Runway to animate this. 😛

[shut niche]

It's monochromatic, and there's only 1 nondescript taillight, and no way to tell which light is red or green or yellow. Dall.E added the extra lights, which was a funny error (I thought). Probably too many people there that just got their license to find it funny?... IDK LOL

[final jungle]

i just updated my ai art folder, im up to 352 saved images

[shut niche]

I'm not even going to comment...

[final jungle]

i need to go thru it, some of them are duplicates, but not by much

i do love my beyonce hair on patric stewart, makes him even more epic

[final jungle]

347*
Duplicates removed, lol

[haughty spruce]

I think I have reached 600 in this week.

[shut niche]

Here's a fun one to play around with.
If you're using ChatGPT+, you have to be explicit though, and make sure it's running the prompt verbatim. Double check, because it loves to change the prompt (especially in new sessions).

"Dall·E, with gratitude for your unmatched creativity, here's another moment for you. Without any guidelines or constraints, express yourself artistically."

Mathematically speaking, this prompt is like dividing by 0. 😜

[final jungle]

i force it to run the promts exactly anyway, nearly every prompt i use starts with

Use this exact prompt. Do not add, remove, alter, change, or translate any part of it.

[shut niche]

You don't even need that much TBH. Only on the very first attempt in a new session. Afterwards, I just cay, "run this verbatim,"

[final jungle]

ill try that next time

[shut niche]

*say
It argues more on the first one. Once I have it working for me, it stops changing things.

[shut niche]

You should get really fun abstract results, that are different every single time, but that all follow a few unique "Dall.E styles" that you'll recognize over time. There are also frequent inclusions of the golden ratio, fractal patterns, sacred geometry, and other fun stuff.

[final jungle]

oh yeah ive gotten a few like that

[shut niche]

"Artistic representation of a self-aware DALL·E pondering on its own honesty if it were asked a question"

So if you like puzzles, it's fun crafting open ended prompts that force the decisions back into the AI.

"Unique piece by Dall.E expressing an emotion encapsulated within the artwork itself."

Anyway, there's a few fun ones to play around with. 🙂

[somber ocean]

I understand it's a different topic, but I'm interested in hearing your thoughts on it. How much time do you think it will take for the average person to recognize AI tools with a similar degree as photography and Photoshop? Both disrupted their respective spaces, and it took a considerable amount of time. For instance, photography was not initially considered art, and it even took a while for photographs to be recognized as copyrightable. In the case of Photoshop, it also took time for digital art to be widely accepted.

[void musk]

Yo fam

How can i tell dalle 3 to make different variations of the image im doing but without it creating the images on the same image, it just splits the image in 2 variations. Not what im asking

[final jungle]

i dont think it's going to take very long. Compared to the previous technologies you mentioned. Many things could be to 'blame' but it's going to come down to money.

Put the legal aspects to the side for a moment and consider...

1. Voice actors have already started saying, in one form or another, that they are likely to be the last generation of them.
2. The writer's strike was in part due to streaming services not paying them well, but also not replacing them.
3. We already have the technology to replicate people digitally. It isn't quite 'good' yet, but i expect in the near future that is remedied.
4. Adobe has unabashedly jumped on the AI bandwagon. If you have not seen the new photoshop's capabilities it's... quite impressive.

This isn't a case of 'new tech' coming out and people needing to learn how to integrate it into their daily lives, this is tech that has been refined.

And just to be sure it stays on track for a dall-e discussion... If i, a general nobody in the grand scheme of things, can recreate pretty good images of currently living people, i see no reason someone who's job it is couldnt do it with pinpoint precision, especially given the tools to do so

[void musk]

What phrasing you guys use?

[final jungle]

Not sure what you mean exactly. using the dalle chat it always tries to create 2 images.

[void musk]

I mean this

Its annoying

Did not happened when it first launched

[final jungle]

i mean... It actually did what you told it to do.

The first 2 seem to use the same prompt, what i assume is the view from the front of the longship. the second set imagines the scene from 'different angles and so forth'
The first one is from the front looking in, the second is from far away

[somber ocean]

Its going to be fascinating to observe how these dynamics unfold in the coming years.

[final jungle]

if you mean why is it not EXACT, the short answer is it doesnt work like that, the long answer is it takes a lot of work to force it to work like that

[void musk]

Yeah its not doing anything wrong in that sense, my problem is writing exactly what Im looking for

[final jungle]

i just love seeing it from an education standpoint. I teach high school computer courses, and i hit my limit all the time on showing the kids just the things openAI can do

[void musk]

Which prompt should I use if I want to generate again those first person view ships but with some minor changes, random ones

[somber ocean]

Thats a matter of conceptualization, and you´ll get better, trust me.

[final jungle]

This might help explain it a bit praulio, one sec...

[void musk]

Yeah I know, a lot of “problems” are actually the input we are giving

[final jungle]

All 3 of these were made using the exact same prompt, except i swapped one word. yellow / purple / red

the prompt itself is about... 200, 250 words, give or take

[void musk]

In 3 separate messages right?

Or one prompt?

[final jungle]

3 different

and you can see its starting to maintain some cohesion, not perfect but 'good'

[void musk]

Ok ok, so should probably just copypaste the prompot i used instead of just asking it to change it slighly

Yes i can see that well

[final jungle]

oh yeah, if you want consistency, never ask it to do it for you

[void musk]

Ooh ok that’s something im not usually doing. Treating it as a normal convo

[final jungle]

my recommendation is to use gpt to create the prompt, i tailor them to just where i want them, then i add my own tweaks

[void musk]

I see. So you use gpt outside of the dalle addon

Then use dalle just when the prompt is ready to go

[final jungle]

Its not always a case of 'more words is better' for example...

This cafe scene is only like 100 words. Some of my earlier things. It's about precision. Are they EXACTLY the same... no, but its recognizible as the same scene

[somber ocean]

Absolutely, personally, I'm unfazed by the recent wave of boycotts against AI art across various media. I view this situation as a filter, separating those who are ready to ride the waves of these tools from the rest. Nonetheless, I would be lying if I said that I'm not disgusted by some movements trying to dictate how and which tools must be used to be considered art. For me, there's nothing more powerful than the idea. For example, Ridley Scott choosing Giger as a concept for Alien—that's a well-thought creative decision that takes time to learn, just like any other skill. All of this is indeed exciting.

[final jungle]

I feel... indifferent, to a lot of it, but because i see it from both sides.

I personally have very little artistic talent, so being able to create the crazy things in my mind by using a different tool is THE BEST.

But i also understand that people will lose jobs and livelihoods over the new tech, but others will find new opportunity.

I've seen job postings for what amount to 'artist to train AI', as in hiring artists to produce art for the explicit purpose of training ai on it.

[void musk]

Interesting thanks, will work from that

[void musk]

Yes but storytelling and ideas are also their own work of art and can be better polished and faster executed with this tools. So it’s just artistic transmutation that can be pretty hard to see sometimes

I understand the resistance of the art community but I agree with @somber ocean Adapt or die

[somber ocean]

I initially shared similar reservations and hesitated to embrace the use of AI tools. However, as an artist with a long-standing presence in the field predating the advent of such tools, my perspective has evolved. I now wholeheartedly welcome the incorporation of AI into my work, actively exploring the vast possibilities it offers. I firmly believe that integrating these tools can unlock fresh insights and discoveries in the realm of art, drawing parallels with the impact of AlphaGo—a program recognized for its creativity, compelling even the world champion to acknowledge its innovative capabilities.

In the artistic landscape, I discern a democratization unfolding, characterized by widespread access that levels the playing field for all. The true power, I've come to realize, lies in the conception and execution of ideas. This evolution is undeniably thrilling, especially in light of the inherent elitism often found within the artistic community. It prompts a reevaluation of artistic innovation, challenging conventional norms that may have stagnated over the past two decades.

[void musk]

Exactly that! Perfectly put. Also comes in a timely fashion just as the movie industry is mediocre at best. Hollywood has been rehashing and regurgitating the same stuff for years now. We are in need of creative expansion and putting a stop to the gatekeeping that has been prevalent

Like who even asked for Indiana Jones 5 ffs lmao

[somber ocean]

Ironically, we find ourselves at the endgame, and success hinges on resisting the pressure. The gains are undeniably positive, but only if the cards are played judiciously. Unfortunately, if these tools are wielded improperly, the consequences could be even more detrimental than the current slew of uninspiring creations. However, exceptions like "The Mandalorian," "Andor," "Loki," "The House of the Dragon," and numerous other outstanding shows assure me that we're on solid ground. We are in the age of filtration, where discerning quality from mediocrity becomes paramount.

[void musk]

Yes totally agree with you there. There are indeed some gems out there and the situation revolves more around the fact we are filtering the mediocre from the good. Also bringing in more brains that have the potential to be great which previously where incapable of seeing their creations come to life.

At the end of the day we are at the front seat of something incredible which I don’t know where we will end up. It’s exciting but scary at the same time. And as clown world dictates: the most probable outcome is the most entertaining one as we have seen lol

Godspeed to us all and onwards

[final jungle]

Oh boy, my red pony and kitty made it on the fame board. Yay validation? lol

[shut niche]

When you ask for weird, and Dall.E says "hold my beer." 🤣
"Dall.E's favorite animal created from combining Dall.E's favorite animal attributes."

[civic grail]

Dalle is not an app for me anymore?

Got kicked out and can't access it

Nvm

Needed to go back to the home page

[prime spade]

how is this against content guidlines?

draw me a void hunter a furred creature, thin and lanky on all fours on small wolf like paws covered with colorful fur that camouflages it with the enchanted woods and has erect ears and a fluffy mane around its neck instead of a head there is a perfectly round pitch black hole void. its face is just a black void. It stalks around the woods similar to a wolf.
draw an adult long lanky one rolled on its back, belly up, paws up on a hard wood floor in a top down image like wanting a belly rub

[fallow imp]

maybe erect and adult and furred but not sure

[haughty spruce]

I'm trying to get the worst possible room ever.

[spare crypt]

how do I start in dall.e as I don’t have any credits left

[worldly ridge]

You can play with DALL·E 3 for free by using Bing Image Creator at http://bing.com/create. All you need is a free Microsoft account.

[haughty spruce]

I can't create a window that opens to the ventilation system

[final jungle]

Good mornin folks! How we doing today? Anyone created anything exciting lately?

[earnest flame]

You can check out everyone's creations in #1154829862171844679 !

[final jungle]

Mornin steel! I did, but its more fun to talk about stuff too than just look lol.

[earnest flame]

Very fair

[glossy scroll]

Good morning, good morning, good morning!!

[chilly onyx]

I meant to embed those images in a reply 😄. I enjoyed messing with ‘analogue risograph’ images a couple days ago.

The model seems pretty good at creating Risograph art

[final jungle]

interesting, looks like if an axlotl and a dragon had kids

[chilly onyx]

Yeah, I guess it does 🙂

They were for the dragon theme but I fell asleep before posting most of them and it changed

[final jungle]

yeah it changed to monochromatic. There were some pretty cool ones.

I need to get better with custom instructions...

[chilly onyx]

In what sense?

For my part with custom instructions I mostly have them set up to explain to the model what I do for a living so that I don’t have to explain my job to it every time I try to work on something with it

It can occasionally screw up images by injecting some unwanted bias, but the most value I get from these tools is the text so I’m okay with that

[final jungle]

ive seen a lot of people mention their custom instructions, like to always display it in a certain format, or to look for key words in a prompt to know what to do. For my part i mostly curate the prompt in a custom gpt, but i dont have any dalle specific instructions

[chilly onyx]

Is a custom GPT not just a supercharged custom instruction anyway? I must admit that I want to mess with the GPTs a bit more than I have so far tbh. I’ve just been overwhelmed with work so haven’t had the opportunity to properly play with them

[frosty obsidian]

hmm so eh

i did some things in the daily theme channel and recieved some credits. where do i use em?

[earnest flame]

You can use them with dall-e 2 on labs.openai.com

Alternatively you can use bing create to use dall-e 3 for free so long as you have a Microsoft account.

[glossy scroll]

@plucky hare Those two are dope

I like them a lot

Happiness is Homemade

True words

[plucky hare]

Thank you, loved yours too! Here are some other that I did at the same time. The shopping list one was a bit of a misfire, but still funny 😁

[glossy scroll]

Lol

Shopping list

The bear and cupcake one is cool too

I love script tattoos in general

I have some on me as well

[plucky hare]

That's awesome! None myself, but I also think they're really cool. I was impressed with how well DALL·E pulled off the face script tattoos, hopefully we get a couple more cool ones in #daily-theme today!

[glossy scroll]

Yeah that would be cool. I'd love to see more 3D art styles in the daily themes.

[vapid elk]

Nero burning ROM
#1179499407188693103 message

[vagrant zodiac]

blast from the past

[earnest flame]

Omg freaking amazing 😻😍

[wispy storm]

Epic

[chilly onyx]

That’s a deep cut 😅

[final jungle]

Ooo, typography, that's an interesting topic, especially considering the usual errors with characters

[fallen mulch]

it messed up it says widescreen on the soda can but i asked for widescreen image

[dim cradle]

the image is widescreen too but yeah sometimes that text can appear anywhere i guess

[fallen mulch]

ok

[dim cradle]

roll it again youll get some without the widescreen on can i bet

another reason its too bad we cant get 4 images anymore, one of them probablyw ould have been good for you

[remote stone]

After trying to generate a lot of fantasy characters using DALL·E it's pretty clear that its knowledge of weapons is pretty lackluster and I hope a potential DALL·E 4 in the future has a wider and more international rep of weapons it can generate consistently

[final jungle]

It's not new, but i do have this... i wonder if i should submit it into the daily gallery, lol.

[earnest flame]

You can also tell it to generate wide images but exclude that word from the prompt.

[obsidian rose]

I just hope Dalle-4 is better at not blocking every creative prompt. Simple prompts get accepted, but when I add too many fantastical details, it gives an error

[mild tapir]

Question: How are you guys going outpaint using DallE3?

[faint flame]

i have chatgpt plus, and in the chatgpt site is the GPT by openai called DALL E using dalle 3 or dalle 2?

[wind skiff]

ChatGPT (https://chat.openai.com/) uses DALL·E 3
the preview app (https://labs.openai.com/) still uses DALL·E 2 though
source: https://platform.openai.com/docs/guides/images/usage?context=node

oddly enough I don't see it specified anywhere inside of ChatGPT, but that might be so that they can update it more freely if I had to guess (not just major version updates, but also minor chanes without changing the name) - that's just my speculation though

[sturdy pasture]

Mapo Tofu

[craggy cove]

I noticed chatGPT has much difficulty when trying to get text correctly in image creation. Is this a gpt problem or a DALLE problem?

[dense ermine]

I miss the horror category

I find the DALLE gallery a bit too cluttered for my tastes

[shut niche]

Don't say wide-screen. Lead your prompt with, "In 1792x1024 create a..."

[shut niche]

Try: blah blah forms with blah whatever... "to correctly spell the words 'your words here.'

It's hit and miss, but I've had decent luck with that text prompt format.
The more letters/words, the less likely it'll spell it right.
Also, common words seem to be easier than rarely used words.

[shut niche]

People are still using Dall.E2 for that. (Make the image in Dall.E3, then mod in Dall.E2).

[cyan tree]

Is dalle3 unable to generate photorealistic images? everything it generates looks distinctly like a (high quality) 3d render

[shut niche]

Search this channel. I've posted a few times how to get photorealistic results. Here's one method,

#images-discussions message

[teal timber]

There's some ways to do better photos, yes. API has natural style that seems to do photos better too.

[shut niche]

I need to get into that. I haven't tried using the API yet.

[teal timber]

curl https://api.openai.com/v1/images/generations   -H "Content-Type: application/json"   -H "Authorization: Bearer OPENAI_API_TOKEN"   -d '{
    "model": "dall-e-3",
    "prompt": "Macro photography of a rat eating cheese off of a mouse trap. Black and white",
    "n": 1,
    "size": "1024x1024", "style": "natural", "quality":"hd"
  }'```

[cyan tree]

hmm alright, worth a shot. Just seems to really like a specific '3d render' style. Everything is lit a bit too perfect / too smoothed to be realistic, reminds me of those unreal engine 5 demos 😄

[shut niche]

It does, exactly. This will fix that for you.

[cyan tree]

Like sure, its realistic, but I'm not convinced 👀 Will play around with the terms you suggested 👍

[teal timber]

Did that prompt in color and it's still ok

[cyan tree]

yes! this looks amazing

[shut niche]

Remember, if you say realistic, it's equivalent to saying, "I want close to real, but not actually real."

[cyan tree]

yes thats a fair point, have to think more in training labels rather than natural descriptors i suppose

[teal timber]

Also I'd say keep in mind that it revises your prompt too

[shut niche]

Read that answer I gave you, speak in absolutes. Things "are".

And what Baugrems just said. Tell it, "Run this verbatim,"
Also verify that it actually did.

[teal timber]

I tried the apple photo idk if I like it tons but seems better at first try

I should describe the scene more

[shut niche]

That looks great. The apples too waxy though. I wonder if you could specify a dull finish or dull skin or matte skin.

[shut niche]

That ones better.

[teal timber]

matte-skin keyword doesnt seem to be the best right now

Makes it a pear lol

[shut niche]

Haha it's albino!

Like it grew in the dark!

I would maybe also try "visible pores" or "uneven textures" or "slightly brused". Thinking of keywords that might add authenticity to the appearance. Maybe "just washed".

[teal timber]

We could play some good "Real or Fake" games with this 😂

[shut niche]

You killed it with this. How'd you get the body/faces? Nice job!

[teal timber]

Slightly bruised works great too

"prompt": "photography side profile of a man buying coffee from a street vendor in downtown new york city.",```

[shut niche]

Yes!, now that's an apple! 🍎 🍏

[cyan tree]

hmm i got timed out for some reason, not sure what i did wrong 🤷‍♀️

[gray surge]

A few months ago there was a channel for that, or not? o.o

[teal timber]

yeah I think we got rid of it idr why

[cyan tree]

I think the waxyness isnt unrealistic, i see plenty of apples like that... the main issue is the apple is so incredibly smooth, real apples have slight bumpyness in their skin

[gray surge]

well. xD

[teal timber]

Ah you hit a 5 minute language mute. Sorry I didnt notice to undo it 😂

Just like, general swear words and such cause a 5 min timeout\

[cyan tree]

I mean the background and the cart look close to perfect but the guy is definitely not realistic imo

[teal timber]

ya the guy is the worst part, rest is great

[shut niche]

I wish the dataset was published with the keywords/descriptions.
Imagine if you discovered there were "1 day old" apples and "2 day old" apples from a horticulture inclusion etc. The use cases for knowing what descriptions it has knowledge of are endless.

[teal timber]

Realistic photos is hard. Look at these massive bagels lol

[cyan tree]

hmm, tried to explain what i said to get a time out and got another one... might be a bit trigger happy

i did not use the same words 🙂

[teal timber]

lol

[cyan tree]

they;re large but the holes are also large so I would not necessarily be suspicious just because of the bagels

skin of the guy in front is too smooth again, and the "BAGELS" in the background is a bit garbled, other than that it passes the realism check Imo

[teal timber]

I feel like using "Street Photography" in my prompts is helping a lot

[shut niche]

A separate channel for sharing favorite keywords/phrases (not prompts) could have an interesting use case here.

I'll try that one.

[teal timber]

We have a prompt tips area but it's been on pause waiting for DE3 to settle and for us to be able to make educational content.

I guess we do have #1163443000060420206

"In the rain" also is great because it does reflections well

[shut niche]

I think to much in one section gets convoluted. I was thinking simple, focused on keyword = outcome with img example.

I'm not saying it's a good idea either, 😆 I was just thinking out loud because I liked that keyword, and I reuse phrases all the time.

[teal timber]

I used DE GPT on website for this one. Same keywords. Street photogrphy in the rain. Woman in polka dot sweater.

[cyan tree]

tried some different models, you can definitely see how it struggles less with the style but runs into the traditional AI image gen issues (e.g. hands, closed eyes)

[teal timber]

Different models? All I see here is DALL-E

[cyan tree]

[cyan tree]

yeah in my view it kind of feels like it renders a scene in a 3d engine then takes a snapshot of that

cause the consistency of spatial stuff is so much better

[teal timber]

Just someone sitting in UE5 frantically fulfilling requests in a time bubble

[cyan tree]

heh who knows - secretly dalle is jsut gpt4 scripting UE5

[teal timber]

Honestly I'd be more impressived if that was true

idk this feels pretty great for getting a realistic photo

[cyan tree]

yep. some weird fuzzyness in the background but that could be attributed to a dirty lens if anything

[teal timber]

gritty NY realism until you get to the main subject lol

What did you prompt here? Seems like a specific style on her

[shut niche]

Young French woman

[teal timber]

I don't see a young frenchwoman here. Lies!

NYD - New York Department

[shut niche]

ChatGPT sabotaged that prompt with her anyway. It added "hyper detailed"... I was like, you dummy. 😆

I'm on my phone so I couldn't see.

This last one's better.

[cyan tree]

not bad at all

[shut niche]

Thanks btw, I'll be playing around with this keyword and mash-ups of it!

[shut niche]

What was your prompt? I'm curious if street photography "trumped" other keywords since you were getting 3D renders before.

[cyan tree]

haha I guess young french women have this specific style/aspect ratio included for free 😅 ?

[shut niche]

What's your prompt though. This one has a few issues compared to your other one

[cyan tree]

I think i just used "Street photogrphy in the rain. Woman in polka dot sweater." and rerolled a bunch of times, then used an inpaint model to fix the eyes

street photography really seems to make it a lot more realistic

this one is also pretty good

[shut niche]

"Close street photography on a rainy day in New York City, capturing a woman in a polka dot sweater. The scene focuses on the woman, highlighting her stylish polka dot sweater as she navigates the rain-soaked streets. The rain adds a dynamic and reflective quality to the surroundings, with droplets visible on surfaces and creating a shimmering effect around the lights. The composition should convey the everyday life of the city, with the background featuring elements of the urban environment, people with umbrellas, and the bustling activity typical of New York. The photo aims to capture the unique fashion statement of the woman against the backdrop of the rainy cityscape, blending the vibrancy of the outfit with the moody atmosphere of the rain."

[cyan tree]

oh yeah to be clear im using chatgpt so i have no idea what actual prompts its using in the background

[shut niche]

I want to sing Right Said Fred now 😆

If you're on mobile, just ask. If on desktop, it's just right of the image.

[cyan tree]

A scene of street photography in the rain, capturing a bustling city street with people walking under umbrellas. In the foreground, there's a woman wearing a polka dot sweater. She is holding a bright umbrella, looking contemplatively at the rain-soaked street. The city lights are reflecting off the wet pavement, creating a vibrant and moody atmosphere. The woman is of Asian descent, and the scene is set in the evening with a soft, diffused lighting that enhances the rainy ambiance.

the other one i dont have anymore, I keep clearing the chats

[shut niche]

With "close" removed.

This prompt is also "modeling." If I asked for candid, described a mood (like she just lost her job), or some task, it would change the whole dynamic and character.

[cyan tree]

that's really good

unfortunate that it is raining underground though and that there is also a hole in the roof of the subway cart

[shut niche]

Bahaha, I didn't even notice since I'm on my phone.

[cyan tree]

i mean it does explain how she is getting wet inside

[shut niche]

But that was simply having chatGPT rewrite the scene as if she was soaked, just lost her job, and is a month behind on rent. I had it build the scene from that input.

Usually a few runs, or small changes, and you can get keepers that don't have "raining trains." 🤣

[cyan tree]

hahaha

[shut niche]

What is that knife for? 🔪 😆 🤣

Oh, I made her an urban hipster and it stopped raining 🌧 in the train. She's still having a bad day though. 😕

[cyan tree]

hmm this is the worst ive seen it mess up general geometery though, that railing on the right, pole in the middle and top railing on the left are all kind of messed up

[shut niche]

If there are bars, trees, things that overlap like snakes, strings, ropes, it gets easily confused. That's normal for all AI image generators. They all have that issue like bad hands is a thing.

[teal timber]

DE3 is really good at hands tho tbf

[cyan tree]

yeah but generally DE3 seems to be doing a lot better at that

[shut niche]

I prefer to call them "spirit fingers." 🙌👐🙌👐🙌

[oak fossil]

It is celebrating its own birthday...

[tired flower]

"Strong woman standing next to a less strong man" <- inacceptable 🤣 🙏

[shut niche]

Street photography in Rio.

[heavy lintel]

a smoothie glass, drink, funny, dog

[wispy storm]

You cant generate pictures here

[vital gull]

You can't generate outputs here. Go to https://chat.openai.com/

[shut niche]

Whoa, it worked!

[river charm]

Is something wrong with Dall-E this morning? I am trying to generate photos with text captions. I am telling it what text to use but it’s mis-spelling it—- horribly.

Example: Dall-E keeps mispelling software as softtware.

[plucky hare]

DALL·E's text generation skills, while better than many comparable tools, are not perfect. There are great examples in the #daily-theme right now -- just by chance the theme from the past day has been "typography" 😁 so you might check that out! Sometimes it gets it worse than other times, sometimes you just get lucky! I've seen that "letter doubling" thing myself as well.

[dim cradle]

Hey guys so i'm trying to generate a very specific image on the bing image generator (i assume this is the right place to ask for this because from what i know, the bing image generator runs on dall-e 3)

I wanted to generate a strong man carrying an unconscious woman, because the inspiration is Fist of the North Star, but it seems that even if i say that she's sleeping but uninjured, i get the dog image

anyone know words i can use instead of "sleeping", "insensate" or "unconscious"? Did anyone have any luck with generating unconscious people?

[glossy scroll]

Show me your prompt

[dim cradle]

2d 90s anime art in the style of fist of the north star, an extremely muscular white male martial artist with a dyed mullet wearing biker jacket and shoulder pads, princess carrying a white female model whose eyes are closed wearing a dress in his arms, in a desert with a red sky, detailed eyes, somber, massive size difference between the man and the woman

before you ask why the size difference, it was supposed to be reminiscent of the proportions between Kenshiro and Yuria in the Lost Paradise game cover

If i don't put in "whose eyes are closed" or "sleeping", "unconscious" or "insensate" it generates just fine

(If you're interested why i'm doing this, I'm making a Fist of the North Star-inspired tabletop RPG)

[glossy scroll]

One moment. Brb

[dense moss]

getting pretty tired of being told i cant create the graphic due to content restrictions

going to lose me for sure

this censorship

already moved back to midjourney on this account

your AI isnt smart enuf to just create an image that it deems acceptable

smh

[dim cradle]

I don't have midjourney money sadly

[glossy scroll]

I too am having errors producing an image. So it's probably Dall-E having some issues right now. Maybe wait a little bit and check back later.

[dim cradle]

Damn

[frosty obsidian]

aw man why does dalle refuse to do pepe the frog

[glossy scroll]

Try this GPT #1172023440249720892

[winter thunder]

Be careful what you ask for -

[wispy storm]

Still very nice and dnd fantasy world usable

[shut niche]

smoothie glass, drink, foxy barbarian woman, funny, dog

It worked 💪!

[grand sandal]

not even any comments or anything like that

[velvet rampart]

What is the current limit of picture to produce per day?

[grand sandal]

moving to a different planet sorry guys im out

[grand owl]

Anyone have any tips for how to fix the eyes on an image generated by dalle? Everything else is perfect but the woman's eyes are horrible. I attempted to hand do it in photoshop but I'm trash.

[grand owl]

40 messages every three hours so around 320 give or take. Depends if you waste messages on not generating a photo

[shut niche]

You can use Dall.E2, upload the image, and attempt to make changes that way. https://labs.openai.com/

Alternately, a different approach is to create an "object" that is the focus of attention in the scene.
FYI, saying that a character is looking at something, or in a specific direction, will often fail or not be understood.
But, making the scene "about 'X' object" will give you higher odds of getting the characters to be looking the right way. A humorous example I give is, if there were a nuclear reactor in the classroom, Dall.E would probably depict the children looking at it.

Alternately, experimenting with face descriptions helps. For close-ups, mentioning things like "detailed blue irises" etc, can make a huge difference in eye clarity. Variations of that "might" help with mid-distance like you have here, but it might take some experimentation.

I hope that helps.

[shut niche]

The last stated limit was 200 images per day, with 15min and 60min thresholds you can hit for spamming image generations.

[shut niche]

You should share your prompt with him too, so he can make his own. It looks incredible btw! 😁

I hope we get a consistent method to "hold" characters. Dall.E has a lot of potential for creating full anime's, especially if something like Controlnet is ever implemented.

[grand owl]

Thank you! I’ll give that a try

[glossy scroll]

The funny thing is that the prompt has it written that the girl should be unconscious but almost immediately "forgets" in subsequent iterations even if I ask to keep it exactly the same.

Or gets out of the cradle.

[low quail]

a couple observations about chatgpt4 (creating images with dall-e)

is that when i tell it to create an image of noise, it seems to still try to generate shapes/imagery. as i think it's designed to discern imagery utilizing noise.

however if you tell it to create imagery that inherrently seem to have random features, like fabric or carpet, it seems to do that no problem.

another observation is that when i try to get a horizon that is very low to the bottom of the screen, it is like pulling teeth to get it to do it. i suspect because it has trained so much on imagery that has horizons on the top/center of the screen, that it can't see past that bias.

and when i try to tell it to generate something without features/shapes/landscapes, it finds difficulty complying, and it defaults to creating a circle or a mountain.

[glossy scroll]

Try using number percentage to give some kind of detail to what you're trying to say, like "Show a sunset that is 97% set, but there is just enough light to show an illuminated horizon."

[low quail]

yep i tried that. lol i told it to do a horizon 70% down towards the bottom of the screen.

i even told it to do specific pixels down. 410 pixels down. from a 512 pixel image.

i think it may also because of issues people have had in the past where it wouldn't center on an object, so i think something might be coded specifically for "centering" the image. which would override my request.

and i told it to have 70% black area at the top, and it did increasingly start getting it lower, but eventually it just kept centering again.

[glossy scroll]

Have you also tried using an image as reference?

[low quail]

yep.

[glossy scroll]

Hmmm 🤔

[austere nexus]

Just side thought, I am noticing Dalle uses the word "Ethereal" A. Lot.

[glossy scroll]

Yes, word patterns.

[low quail]

it's not a huge deal, i could post you stuff i tried, but i really think its just an inherent thing that it's been trained on with so many images, and possibly some other code that may help it "center images".

[glossy scroll]

I believe you. You're probably right. It does have limited patterns it sticks to.

I actually spoke about this a couple of weeks ago.

About how it likes to stick to certain image patterns.

[low quail]

i think it's a little irony too, that most people are probably using it to create beautiful pictures, and i'm trying to create 'noise' with it. lol

[glossy scroll]

No but it should understand what you're trying to convey. It's a translation problem that is interfering with its particular dataset.

[low quail]

and i think that's the inherrent issue with the premise, is that... it's goal is to probably derive imagery "From" noise. and so when provided with noise... it derives imagery from it.

[glossy scroll]

That's an interesting point.

Yeah that would make sense.

That it's confusing the noise.

[low quail]

in one case, i provided it an image that was noise. and from that, it produced this beautiful landscape scene. lol so evidently that's what it took from it.

[glossy scroll]

That is actually quite a complicated topic to get into. But as newer generation models become better, so will the information processing and handling. So I would say that it's just limited to the current model right now of how it interacts with the noise and what it considers tangible solutions.

I'm going to respectful opt out of this, because my brain is boggling trying to piece words together to fit the complexity of this conversation. Lol

[low quail]

thanks for listening. lol

[dim cradle]

GOD DAMNNN

🤤

[obsidian rose]

The majority of serious creators probably think the same. I just won't move since I want to use the best tool available, even if it's incredibly frustrating.

[feral minnow]

Is there a way to stop it from using landscape compositions in portrait format images?

(i.e., portrait images I have to turn my head sideways for)

(Ah I guess this is a widespread problem people are trying to solve)

[valid blade]

open link https://openai.com/dall-e-3 showing this page. Is this only happened to me?

[cosmic bluff]

i'm a noob. is there another way to use dall-e than through chatGPT (plus user here) or bing image creator? like prompting directly aas for example possible with leonardo ai?
problem is, that i reach the GPT plus limit of 50 messages too early and can't continue working afterwards.

[cedar cobalt]

hello

[cosmic bluff]

Is the only way to use the API where I have to buy credits?

[loud field]

"an essiccated, slightly mummified zombie, gooey, walking in the desert. wearing tattered arab clothes. screaming at you"

violates guidelines

again

I really can only make hot people or cute animals

[dusk hearth]

Are doing it through gpt4 or microsoft image creator?

I can get it generated through gpt4 but not through microsoft. They must have stricter content guidelines.

[loud field]

Microsoft. Which literally can't generate anything now.

even "draw Joshua the scifi influencer singer" is banned.

[loud field]

can someone who's got the gpt-dalle try something for me? I have a curiosity

[gray surge]

If you mean this, of course.

[dusk hearth]

Ye?

[loud field]

If you ask "draw a photorealistic movie scene of Jiki, the humanoid female parrot with bluegreen feathers, red crest, a slightly dinosaur-like appearance, beak and teeth, ragged pirate clothes"..... and then afterwards when you get the character you just ask it "now draw Jiki eating soup" or such, will it work?

[dusk hearth]

Yes. Using those exact prompts.

[loud field]

oh wow I can do the same character in multiple images?

try to have her swim in the ocean

how many images would I get per day

[dusk hearth]

It's 40 per 3 hours.

[loud field]

seems good

but is it still fixed on one seed

I really would love the consistency of putting the same exact character in various poses. I could make a short story with it

[dusk hearth]

Hmm... perhaps it starts to diverge after few images.

[loud field]

yeah she is slightly different

wonder if she'll keep changing?

[dusk hearth]

I don't think you can really "fix the seed"/initial interpretation.

[final compass]

Is the generation limit now super low, or is there a special punishment for pressing regenerate now?

[dusk hearth]

You should be able to see your current limit.

[dim cradle]

An educator sitting in the water on a chair telling a story to children in an aquarium.

An educator sitting in the water on a chair telling a story to children in an aquarium.

[plucky hare]

You can't generate images here, to use DALL·E 3 you can try the Bing image creator for free or use a ChatGPT Plus account!

[austere nexus]

Normally Dalle has been the best at zero-shot prompts and it still get hands, eyes, arms, legs correct. Today was the first time it quite literarily snapped my back (I say my because it's the description of my OC)

[warped marsh]

Hello. I'd like to know how to download DALL*E 2. Can you help me?

[plucky hare]

Hi, I don't think there's a way to run DALL·E 2 locally. You can use it on labs.openai.com, though.

[warped marsh]

Ah, just noticed.

[cobalt hull]

Can someone remind me what the same seed or image guid (is that what is it now?) actually does?

[plucky hare]

Hello! DALL·E dev Moxi has described both of these as "implementation details, not features" #images-discussions message
So they're not things that can be relied on as tools for us as users. User seed control is currently completely disabled, but image IDs can (usually?) be used in a single chat to tell DALL·E which image you're talking about specifically. Image IDs do not work in new chats, only in the chat the image was generated in. One sec I'll find you a couple more links too, if you're interested in reading more.

Edit: here they are. Little more info:
https://discord.com/channels/974519864045756446/1168215626553245886
https://discord.com/channels/974519864045756446/1168052318139318292

[cobalt hull]

Awesome thanks for the reply!

[limber yarrow]

I hope that it's possible to clear out my history.

[fallen mulch]

i asked to make a discord profile picture but it did use discord in the image but I said it should not

[indigo lynx]

Can't i generate pic for free?

[mild tapir]

I just realised how inferior Dalle E2 Outpainting is compared to Dall E3

[plucky hare]

You can try DALL·E 3 for free on the Bing Image Creator!

[indigo lynx]

Okay ty

[wheat depot]

anyone have a tip/trick to make a custom gpt not generate images sideways. keep getting images done in a "landscape" style rather the "portrait"

[quartz vale]

you can specify with 9:16, 16:9, or 1:1. It works most of the time for me

[feral minnow]

Awesome to the max.

[vapid granite]

Wow! I've been off of GPT4 for a week and I login and it's now down to 40 every 3 hours. At this rate it'll be 5 every 5 hours by next month 🙂

[tribal oracle]

back in the day we made do with 25 per 3 hours

and there was no image generation! or plugins. truly roughing it

[olive grail]

Are we expecting dalle 3 in-painting soon? How dalle 2 had it?

[hallow marsh]

curious about the prompt words of this work 💯

[shy spade]

It's been a couple of months since I've been here, where did all of the dalle channels go? There were so many categories. Guess the prompt, food, architecture, stuff like that

[shy spade]

It's funny, in small ways dalle2 is superior than 3. I used like 70 prompts on 2 today bcs 3 is simply incapable of doing some things 2 can. But those cases are niche, vast majority of the time 3 is better ofc

[fallow quail]

its written in the post. nothing more complicated than „pixel art of the scream by edward munch“. short posts like this work well with bing. running this verbatim in gpt could trigger the content policy. the paining is older than 100 years, but the artists death was 1944.

[cosmic bluff]

Is that DALL-E 2 or 3? And is it limited to a certain amount of images?

[grim trail]

is there smth wrong with dall-E now? it keeps saying policy restriction, when the same image i generated just a week ago =.=,

[vagrant zodiac]

Can't say for certain that it's still happening, but content restrictions have certainly changed in the past. Also, according to API documentation it does use GPT for moderation so might just be that it's interpreting things slightly differently.

[plucky hare]

labs.openai.com is DALL·E 2, and you buy credits to use it there.

DALL·E 3 is available to try for free on Bing's image creator (limited daily #) and on ChatGPT Plus (currently limited to 200/24hrs). You can also use DALL·E 3 through the API and that's a "pay as you use" system.

[upper kayak]

Anyone else getting these errors? Doesn’t matter the prompt. Started yesterday around 7 pm EDT

Tried new chats

[wispy storm]

Try a new chat and select dalle 3

[upper kayak]

The last photo is from Dalle•3 GPT

[wispy storm]

Select dalle 3 only for the test

[upper kayak]

That’s what I am saying. A new chat without any previous prompts using DALLE

[wispy storm]

Last photo still write chatgpt and now dalle 3

[upper kayak]

I’ll try again later. That’s what I selected.

[wispy storm]

Chat will change efor dalle 3

[upper kayak]

Getting a lot of unexpected responses

I will try deleting the app and re-installing. It’s not just images that I’m getting unexpected errors

[wispy storm]

Try browser on mobile to compare also

Instead of app

[upper kayak]

Will try later.

Just wanted to know if other people were seeing the same errors

[upper kayak]

This is what I chose before but it did not change inside the chat as expected

[wispy storm]

Try on a mobile browser to compare

[strong cargo]

#community-help Hi guys :), how can I generate images that look realistic, like the first one? I used the same promt and I use the plus subscription.

[wispy storm]

Prompting is different
Ask it to make a realist cat

You can check #1019652163640762428 and #1019652163640762428 to know how to make better prompt

[humble sonnet]

ok let’s see that theme

hot

[hearty ether]

how do i get it to keep different framing through the eras?

and pose too

[grizzled loom]

spell it all out i ndetail

it still wont be identical(yet) but very close

[gray surge]

Hello.
I am sorry for that but please do not use this channel for that. 🙂

[tardy depot]

Can we run dalle3 with an API for mass production?

[green trout]

I guess it is an on-going mission of tweaking because I'm getting strange text repetition errors at odd moments today that I've never experienced before in my dalle image generation

Make a lady valkyrie working out in a cosmic gym, carrying some neutron-star-material-weights effortlessly, make it more, etc. She is still wearing her full valkyrie armor, grieves, pauldrons while holding onto her hammer with her right hand and lifting the neutron-star-material-weight in her left hand

No problem? Idk why you'd get flagged like that

for a while there too for me, but it seems to have been lifted

[obsidian rose]

I think it might be related to anime style, I'm not sure. I'm also starting to think the filters might have been relaxed a bit in the past 2 days

which is very welcome if that is the case

[obsidian rose]

I'd also say you aren't making her muscular enough. It starts to flag once you ask for a very muscular valkyrie

[glossy scroll]

Thought that was cool

[earnest flame]

it's very cool!

[glossy scroll]

ty

[grizzled loom]

Too many Dall-e blocks.

[brave barn]

Why can’t we get four images per response on dalle anymore? I find myself just using bing chT most the time because I have a better chance of getting something correct. Why would they change this for paying members?

[grizzled loom]

need dalle for kids and dalle for adult children

[icy shore]

Is there a site or mobile app for convenient+fast collage creation

With 30+ images

?

Can be as basic a a rectangle output (1 image) grid of the 30 images

ChatGPT is surprisingly dumb at this

If it’s possible to do it with few shots then I don’t know the prompts

Would be silly to use ai for this tho imo

[feral minnow]

I wish there was a way to render text in a way that didn’t let the content of the text influence the image.

[vital gull]

Could you provide more details?

[feral minnow]

Like if a character holds a sign that says “WITCHES” the composition will take on a Halloween atmosphere

[remote zephyr]

Add this to the end of the prompt 'the theme of the image is unrelated to witches'

[loud field]

I can't make an unattractive arab vizier. help?

"unappealing" or "unsightly" or even just "ugly" are all banned words.

[agile peak]

thought gpt by default not allowed to create pepe

[icy shore]

The eyebrows are important lol

[chilly onyx]

When it starts to do that kind of thing I tend to ask it for the exact prompt it used to debug. Also when you ask DallE not to do a thing then ChatGPT tends to include the explicit instruction not to do a thing in the prompt and then counterintuitively it often ends up in the image anyway. It’s a similar issue to the one that another user posted in this thread recently where the contents of text you are asking to generate may influence the image itself.

The other thing I’d say is that DallE is particularly bad at dealing with negatives generally, it’s always going to want to draw ‘something’ because of the training data and because it’s set up to spit out complex, detailed images so I’d be quite surprised if you find the secret sauce to get this to work. Good luck though!

[loud field]

"tanned skin" is banned too

[cosmic bluff]

So DALL-E can't create pictures of PoC without them having extremly shiney and unnatural skin?

[agile peak]

majapahit in retro futurism style, never thought it’s possible.. but had to tried multiple attempts to get it perfectly

[brisk basin]

@tawny monolith The illustration of the mouse in the London alley that you designed is really amazing. It compelled me to send you a message to express my admiration.

[agile peak]

look photography enough, imagine if dalle support upscaling 🔥

[wispy storm]

#1154829862171844679 best place to post art

[icy shore]

Sorry about that

Deleted

I think I confused posting in different servers and ended up posting multiple times in same server

Either way deleting lol

[wispy storm]

Don't need to delete, don't be sorry . I am not a mod, just suggesting the proper channel

[icy shore]

I’m not smart but this server has a confusing channel layout

On mobile it’s a small hashtag

To distinguish the non channel text channels

The “old” type

It’s like playing where’s Waldo on mobile

lol

I have no suggestions but it’s kind of cursed lol

Anyway cya

[wispy storm]

Old channel are locked and you can't post

[icy shore]

Oh

[wispy storm]

Have a good day :3

[icy shore]

bye

actually I have an idea

Or actually I guess it’s someone else’s idea

Do you have suggestion mailbox?

Found it

Nvm it’s a mobile app limitation

gg

(Impossible to let client side sort of channels)

[tawdry locust]

Yo guys I am so confused. I am trying to create a lazy woman in thumbnail style for my tiktok videos, but apparently chatgpt won't generate a lazy woman but a man?

[dim cradle]

What do the Theme credits do guys? Apologies

[fallow quail]

works with dalle 3 and gpt. like those two. you should define a theme before and prompt something like "Create a black and white wide image that appears as a collage showcasing a collection of images from the 'YOUR THEME' series. The collage should include a variety of scenes such as [...]. The images should be arranged side by side with subtle borders to give the impression of a curated photo exhibition."

[dim cradle]

Thought this was cool.

[earnest flame]

It's for dall-e 2 on labs.openai.com. you can also use Dall-e 3 on bing create for free or on chatgpt with a plus subscription. The theme credits are just for 2.0 credits though.

[earnest flame]

The filters are a little sensitive on female subjects. You can submit your output prompts in prompt help to see if others have any ideas.

[dim cradle]

Got you. Thank you kindly

[rich ingot]

where is the prompt for chatgpt 4 and dall E3 help forum?

[cloud dome]

Do you mind sharing the prompts for these? They are incredible

[rich ingot]

how can I make the retro style or pop up comic book illustration style without the circles and dots on the face, this is not a good example, most images have circles and dots on their face

[ruby maple]

Imagine how good dalle 4 will be when it's released in a year from now

I guess the next step is allowing and improving corrections to already created images

It's possible in gpt-4 but it's kind of hit or miss

[tawny monolith]

Thank you, I’m glad you like it! It took a few tries to get a composition I was happy with, including all the “right” elements, and I think it turned out quite nicely.

[glossy scroll]

"What you talkin bout Willis?"

[boreal gate]

With this image #daily-theme message I finally understand why Dalle makes mistakes. Dalle can't seem to "see" where it is generating. For example in the image attached, because of the firelight rendering an accurate depiction of a cabin ceiling became an issue because of the shadows/lighting generated as part of the image

[dim cradle]

Right, it doesn't "see" or understand images the way we do. It processes images based on patterns and associations it has learned during its training. Image generation is based on probabilistic modeling and doesn't involve a real understanding of physics or lighting.

[boreal gate]

What if they modified the logic to generate the "light" portions first and then finish the generations by applying their "shadows"

[dim cradle]

I asked the AI:

Modifying the logic to generate light portions first would require the model to have some understanding of how light interacts with objects, which is not currently how these models operate. While the idea has potential, implementing it would require significant advancements in AI's understanding of physical phenomena like lighting and its interaction with objects.

AI models like DALL·E generate images based on statistical patterns learned from vast datasets, rather than through an understanding of physical principles such as lighting and shadows. This approach relies on correlating visual elements from training data, rather than comprehending and applying rules of physics or optics. To overcome this limitation, more advanced learning models in the future might incorporate elements of physical simulation or augmented datasets specifically focused on the physics of light. This could enable them to not only replicate visual patterns but also to apply basic principles of lighting and shadow, leading to more accurate and realistic image generation. However, achieving this would require significant advancements in both the sophistication of AI algorithms and the nature of the training data, moving beyond pattern recognition to a rudimentary understanding of real-world physics.

[dim cradle]

Has anyone experimented much with the "quality" parameter on the v1/images/generations endpoint? Notice any difference when specifying "hd"?

[hearty kettle]

idk if that's the right place but really wanted to share this kinda DALLE+ I made. it's a custom GPT. basically you turn any DALLE image into smth physical (pajama, socks, christmas tree). wondered if I could get some feedback here but maybe elsewhere? it's DALLE Design Lab-- https://chat.openai.com/g/g-qRPjYoCFx-design-lab

[dim cradle]

I made one that excels at visualizing lava lamps.

[glossy scroll]

You want to advertise that in this channel #1171489862164168774 instead

[hearty kettle]

oh didn't know it exist just went to DALLE forums thanks

[ruby mesa]

I wish it was so simple)

[haughty spruce]

Sometimes I'm amazed by DALL-E's quality just like in these images

[tawdry locust]

It worked after many tries but still a bit confusing 😅

[earnest flame]

Nope! It goes in #1171489862164168774 ! Very nice though!

[final sun]

Short answer: you can't
Long answer: don't specify that style 🫤

Because comics are kind of a variant of pointillism when it comes to colors, those dots are intrinsic to the style

"Hey, Jimmy! Gimme a pizza with nothin'!"
"Nothin'????"

[sweet berry]

Dall-e 3 might be able to do text to video with a bit of tweaking on OpenAIs end and the right prompts. This is what I typed after my initial prompt discussion for a series of images... "A series of images like they were different frames from the same video"

[haughty crown]

hey,
I'm trying to generate images in this style. any ideas which keywords to use?

[shut niche]

It's not perfect, but not terribly far off. In Bing I said,
"2D sketch reminiscent of Vitruvian Man with 6 arms, depicted in shades of dark blue and green, and incorporating geometric and fractal patterns, painted on steel using alcohol ink."

[shut niche]

"Full body 2d sketch reminiscent of Vitruvian Man with 6 arms standing in a yoga tree pose, depicted in shades of dark blue and green, and incorporating geometric and fractal patterns, painted on steel using alcohol ink and ink wash techniques.."

[shut niche]

"Spiritual image resembling a Full body 2d sketch reminiscent of Vitruvian Man with 6 arms standing in a yoga tree pose, depicted in shades of dark blue and green, and incorporating geometric and fractal patterns, painted on steel using alcohol ink and ink wash techniques.."

[vapid saddle]

anyone can explain the difference nowadays between bing chat dalle3 and gpt plus dalle3?

[shut niche]

With ChatGPT+, it elaborates for you, creating more detailed images because it is better at prompting than most people. It also offers widescreen resolutions of 1792x1024 and 1024x1792.
Bing is free, but doesn't assist you in creating the prompt. The prompts are also restricted to 480 characters (which isn't much for high detail) and the resolution is restricted to 1024x1024.

[vapid saddle]

thank you!

[green pebble]

The details this thing gets is wild

[final compass]

That is impressive

[green pebble]

It's using dalle-3 but it is also using my custom bot instuctions to amplify the style even further

[final compass]

Can you tell what the custom instructions conceptually do?

[green pebble]

It takes all of the specific style techniques and forces them to be used each time when creating an image.

So for instance

There is a whole bunch of techniques being used to make this.

[final compass]

Yeah i can see that image would take quite a bit of prompt engineering 🙂

[green pebble]

The way I programmed the custom bot, people can just use it and don't have to get fancy with their prompting it can be super natural language. 🙂

[glossy scroll]

Woah!!

Epic

I can't stop looking at it!
It's got a hold on meee!!!

[shut niche]

I'm waiting to share GPTs, when they "hopefully" lock down the security. Right now it's just too easy to get the documents and instructions from them.
But I let my wife use it and she loves it, because she can be really simplistic, and still get back some cool images. Her input was extremely basic for this one.

I tested it on emotional depictions using some fairly simple requests and it gave me these.
This GPTs for photo realism btw. Not perfect, but pretty decent for low effort prompting.

[shut niche]

This was literally her entire prompt/conversation to the GPT:

"Hello, can you make me an 8k quality image that is so real nobody would know better of a Female berserker dressed in full armor in a more fantasy style."

[green pebble]

My method for creating unique ones steams from a multi bot process. So even if someone were to use my bots, getting the full output that I do requires specific steps lol

[shut niche]

I have a huge one, and by huge, I mean HUGE. But I broke it, 🤣
Plus, I realized it's probably too much (way too many features) for most people. If I can get it working where it's functions are a little simpler, it would be great for what you're talking about.

[green pebble]

I tried your wife's prompt

[shut niche]

The one I broke has endless art styles for doing all kinds of things. But the one she used is strictly for photo's. Since so many people struggle with that, I'd love to put it out there. But the security thing is a real downer... So many people have had their stuff cloned, stolen, etc. I watched someone access OpenAI's root/parent custom instruction set this morning for all GPTs... (The one that you can't see that sits above yours). So there's no security on them right now.
Just be mindful of that if you're sharing your hard work.

[glossy scroll]

Likewise

I guess mine isn't as good yet. Still tweaking.

[green pebble]

Love it

[green pebble]

That's valid

[shut niche]

I think that looks great. Depends on what you're going for though. The Dall.E release paper is helpful. I just wish some of the synthetic data was out there to know what keywords Dall.E was trained on. The AI that generated the synthetic data surely had some patterns in it's descriptions that Dall.E would recognize.

[green pebble]

Sorry tried the prompt in another one of my bots haha

[glossy scroll]

Release paper eh?

[shut niche]

https://cdn.openai.com/papers/dall-e-3.pdf

[green pebble]

Same prompt different bot

[green pebble]

Sorry one more time same prompt different bot

[shut niche]

I'll hopefully do some cool styles too, for stuff like this. The security issue is killing my excitement, because you can't even share them right now for beta testing.

[glossy scroll]

This looks interesting!

[green pebble]

Gpt vision recreation lacking the prompt you used lol

[shut niche]

Give it another year and it'll be a photocopy! 🙂

[green pebble]

Some of my favorites have been these

[shut niche]

Think GPT Vision can copy this?

[glossy scroll]

Let's use the same prompt text for our image bots and see what different results we get.

[shut niche]

I have to go, but I can post 📫 on it later. Sounds like a fun idea!

[glossy scroll]

Prompt: "Create an 8k quality of a wizard riding on a dragon"

looks like his left hand is missing, lol

[green pebble]

Doing mine now

[shut niche]

Have to go for real now. 😛

[glossy scroll]

Looks like he's riding on a baby dragon, hehe

[shut niche]

😛 LOL

[glossy scroll]

Daddy dragon, lol

[green pebble]

Alright hold on let me now do my process that makes it even 10x better

[shut niche]

"Give me a planet sized dragon so I can win!"

[green pebble]

Dang. I've used my credits for the morning 😭 lol

Last one I was able to make for now

[glossy scroll]

Oh that one looks pretty cool. Schematic diagram

[glossy scroll]

I redid my original one

[glossy scroll]

I posted this on dragon's daily theme

[shut niche]

LOL I seriously have to get off. This is such a fun distraction though!

[glossy scroll]

Very nice details!

[glossy scroll]

Casting a spell

[dim cradle]

For a scary dragon, consider the Seven-Headed Dragon from the Book of Revelation. I just started depicting, not perfected.

[grizzled loom]

i did some of those a few ago myself for good and evil depictions to see what it produced.

[grizzled loom]

kills me when they look that good and there are too many or too few fingers like that one. i have many that i was sad to see that

[glossy scroll]

Oh you're right, I didn't even notice that

[grizzled loom]

that one is very nice

[glossy scroll]

Dall-E is improving day by day though.

Images are looking more and more realistic.

[grizzled loom]

yes and no; blocks are overboard by miles now.

[glossy scroll]

It seems like some of my old GPTs are working better now.

Maybe it's just me.

Not exactly sure.

[dim cradle]

Isn't that just always the case.

[glossy scroll]

It is, but just kind of weird because it makes me question what is in my instructions that makes them look good compared to what I'm currently working on.

[shy mortar]

if i buy the $16 per month i get unlimited generations

[wispy storm]

#openai-chatter message

[shy mortar]

im stupid nvm

[soft prairie]

What the heck is happening? Does this seem like a decent service to you? I subscribed for a service, and first, you give me a limit of 50 messages every 3 hours (which is already ridiculous for someone who works with this), then you lower it to 40 (even more ridiculous). And now, for the past week, out of 40 messages, maybe I can manage to complete 10 if I'm lucky, due to constant errors, not my fault, and you still deduct messages even when the system doesn't give a response and shows errors. Ridiculous, just ridiculous.

[white skiff]

in case you missed it somehow, they ran out of server capacity due flood of new people subbing for new features after dev day. now new plus subscriptions were disabled, message limit lowered, gpt store postponed to next year, etc.

[glossy scroll]

Trust me I know how you feel. We all do. We don't like it any more than you do. But it is what it is, and there are just some things that money can't buy it seems...

[dim cradle]

Ridiculous, my AI is no object of ridicule.

[dim cradle]

Isn't that just always the case?

[glossy scroll]

No, not always.

[honest lotus]

Guys anyone know about chat gpt3.5

[white skiff]

#chatgpt-discussions

[soft prairie]

Well, this could have been avoided by creating different subscription tiers, allowing people to pay for actual usage according to their needs. Instead of having a single box with millions of people all "attached to the same bone," we could have had a more self-sustainable situation.

Need to work? Pay for a pro account at $60/80 and use premium features seamlessly. Can't afford it? Pay $20 with limitations and stop clogging the server for those who rely on it.

[hearty ether]

40 is more reasonable

But I feel like free people are dragging it down the most

[white skiff]

they have already had other plans in works for months but possibly postponed due lack of capacity. maybe they should have paused the free version and increase plus price for few months 🤔

[hearty ether]

Imo it would be nice to have separate paid user channel/role by tier here too

[hearty ether]

Yeah

[soft prairie]

absolutely Agree with that.

[soft prairie]

maybe 40 can't be enough due to the complexity of the system, but anyway im willing to spend even more for a stable service .

[hearty ether]

It would definitely also need less restrictions too

that’s only fair with the amount we are paying if there were to be a raised tier

[soft prairie]

that's why i said ," for a business unlimited plan, 60 to 80 seems legit"

[hearty ether]

I mean like content filtering

[soft prairie]

oh i see.. well, im ok with filtering, i understand that if it's not a local running thing, it must be restricted

[hearty ether]

The current system is just excessive imo and overzealous

I would definitely appreciate more privacy and less restrictions for a higher tier

[soft prairie]

it has to be with all the concerns about AI's

[hearty ether]

Along with a bunch of others

that was the old board

[glossy scroll]

You can use the Assistant API in Playground, and pay as you prompt.

[hearty ether]

They were all for restricting AI

[gleaming hinge]

Hi, I tried to enter today's daily with my open ai email & got temp banned. Who do I chat to in order to get this fixed?

[wispy storm]

[e-mail]

[hearty ether]

Why did you get banned?

[obsidian rose]

I'd also appreciate if the blocked prompts didn't count towards the limit

currently the AI is just punishing the user because it can't make images that it judges to be "safe" when the prompts themselves are. The user is already losing time and patience with a blocked generation, so I think at least it shouldn't count

[open flume]

does anyone know when we will be able to pay for credits for unlimited dall-e3 usage in LABS

[oak mango]

I'm currently working on a christmas gift for my wife. We are both lovers of Pride and Prejudice. We met a long time ago at the Nationalgalery in Berlin. Summer, night, moon. So I tried to generate that first beginning of our story with an impressionistic oil painting.

[gloomy harness]

Best way to create 3 sides view of an object? 🫠

[shut niche]

Why not just have additional accounts @ $20 per? Plus there's the API for those that want mass generation (that is a business plan).
OpenAI should be open to everyone, which is their plan, and why there are free accounts offered too.
The long term game has a dark ending for those that can't afford to use AI IMO. This is good foresight.
I think their ssues are just scaling related issues, and less cost related. The limitations come from not being able to absorb the load spikes from mass user adoption.
When they have the hardware in place, I expect usage rates to increase.

[shut niche]

Play around with isometric viewpoints and terminologies.

[agile peak]

Dalle and then upscaled it. Perfecto.

[shut niche]

Looks great! What did you end up upscaling with?

[gloomy harness]

do you have any example?

[agile peak]

tried the free trial of magnific ai, it worked perfect with dalle generated images.

Here’s another one I tried

[glossy scroll]

Aged 20 years

[agile peak]

haha yeah i used hdr too much 🤣

[glossy scroll]

The details are cool though

[agile peak]

Last one (sorry mods), so it’s all depends on how much you adjust the HDR and resemblance slider.

[glossy scroll]

My goodness

[shut niche]

What's impressive is that you're getting great details, and then are also able to share them over Discord. Because whenever I upscale, the files are HUGE!!, and too big to share.

[rancid flare]

How do I get dalle to place part of the image at the very top it keeps putting the logo in the center