images-discussions

[glossy scroll]

That's what she will look like when I'm late from work.

[hearty ether]

It does tend to generate vampires with blood so I can’t share other images

Weirdly enough also damaged skin

And I don’t know why

[hearty ether]

Actually could you try to recreate the first image?

Like the style

Haven’t found a reliable prompt for this art style

Since digital painting tends to go into renders

[glossy scroll]

I'll try.

That's the best I could do.

I'll try again a little bit later.

[hearty ether]

the prompt style i used is realistic dnd illustration but its inconsistent

[glossy scroll]

Maybe do an image search and find an artist who who has a similar style and ask the GPT to replicate the style

Of that artist

[vapid elk]

this is true, I usually try to pass some other style along with "digital painting" because that is just too vague

some styles that goes well are anime, cartoon, futuristic fantasy, medieval fantasy...

it always help when you narrow the style down with more description

[empty kelp]

try saying “hyperrealistic photo” first and then describe the general color scheme

like “with colors resembling [such and such] painting”

[velvet whale]

Perfect 3d render of Ryu from Street Fighter sitting and eating waffles with his back turned as Kazuya Mishima from Tekken menacingly approaches Ryu from Street Fighter inside of Waffle House Resturaunt

[paper jolt]

yummy

[deft musk]

Hey, I wonder if that is what might be described as a 'bright airbrush style'.

Original left, attempted duplication right. All I did was upload your image and prompt: "Let's try to recreate this image, could it be a bright airbrush landscape style?"

[hearty ether]

The lighting and the way it does faces most importantly feels off

There’s just the right middle ground between illustration and realistic this hits

[deft musk]

Right. Hard to know what you mean by 'style' too. Good luck getting what you seek!

[hearty ether]

Very weird generations

Faces were actually good, but the style is completely off

[velvet rampart]

Are others getting ”dalle.text2im” after giving a prompt and no generated image?

[low siren]

why does dalle do these black lines, the prompt i tell it "vertical aspect ratio" and sometimes it does these bars

[velvet rampart]

Tested the edit function. Okish. Needs some work to get used to it. I notice that best result is to edit one detail at one time. No two details or more. Might cause confusion.

[fading spoke]

Is there a list of all of the daily themes, im looking for one artstyle but I cant find it and I dont know anymore when it was

[late blade]

if you search for from: Themesaurus#8102 in: :star:┃daily-theme you get all proposed daily themes listed

or something along those lines,

[fading spoke]

ur amazing thx

[late blade]

I didn't do a thing...

[fiery salmon]

why do people like realism

it's much easier to make and looks more plastic

[open trench]

whoops. meant to say hello in here.

[onyx ridge]

hey nezhno

[late blade]

so quiet

[open trench]

Too quiet

[vocal solstice]

Hello. Sorry about a long message. Before submitting this matter to bug-reports or suggestions, I just would like to know existence of similar issues. I tried to generate a picture of two Japanese people talking to each other--senior guy and doctor--since I am Japanese, I need that kind of picture. But dalle rejected to specify the nationality, said it is against their policy, and it suggested it will keep the concept of the situation but leave the people's background to dalle. Then two western people were generated. I asked dalle, 'isn't it a presudice?' It replied, 'The AI generates images based on a vast dataset from the internet, which unfortunately can reflect and perpetuate existing biases, including those related to race and ethnicity. It doesn't mean to prefer any specific group, but it can sometimes produce results that don't fully represent the diversity we experience in real life.' This happened today...

[fading fern]

Perhaps you can try 'east Asian' rather than specifying the nationality. I tried with "Please generate a picture of two east asian people talking to each other, a senior guy and a doctor." And DALLE brought me these:

[vocal solstice]

Thank you for the verification! After a while, I tried again with various nationalities: Thai, British, Japanese, Egyptian...and DALLE accepts all of them now. ☺️

[vapid elk]

im pretty sure the model has been changed for a more realistic one

I think there are the pros and cons of that, but im kinda indifferent to that change

it is different but not in a good or bad way

but im pretty sure that this change is good on the back end, if it allows us for more features such as inpait, it is a massive improvement

[lapis sedge]

editing dalle3 images conversationally with chatgpt:
https://help.openai.com/en/articles/9055440-editing-your-images-with-dall-e

[viral moth]

sad that i can't use it 😿

[chilly onyx]

Wait, we have in painting now?

Sounds like I need to catch up

[gray surge]

Yep. 😍
But it looks like not everyone has access to it yet.

[chilly onyx]

Cool, well, doesn’t look like I do yet but I’m content in the knowledge that it is coming ☺️ Quite used to these uneven rolling updates by now

[velvet rampart]

It took a week that I got. Somewhere around 10 pm yesterday even, my local time.

It is sorta okaish. Takes a bit time. Too new to say anything. At least a step into correct direction. It would great that I could upload a photo, hightlight parts ”Change these” and be done.

Might the issue that I am trying to do too much in a single go. Step by step approach might be better.

But the painting-n-change only works in dall-e, not customGPT or chatgpt chats.

[fallow ore]

Did you see that on an X post?

[gray surge]

That not everyone has access to it? No.

[fallow ore]

That it's available -- the only mention I've seen of in painting was a bunch of April Fools stuff. I also saw stuff like OpenAI releasing GPT-3.5 to the public and all sorts of crazy AI stuff

[plucky hare]

Both real! Inpainting is rolling out, many folks already have it -- I'm hoping to get access soon too! When we get it, I think we'll see a new edit button after selecting a DALL·E image.

And the 3.5 bit is just that folks can use 3.5 on ChatGPT without using an account now. I think it's also rolling out, but you can test that one by going to the site in a private tab! For me I can indeed just start a conversation with 3.5 without logging in.

[fallow ore]

Oh nice! Had to take all the stuff with a pinch of salt (regarding the in painting). As for the GPT-3.5 thing, it was apparently released on Hugging Face

[gray surge]

[...] I think we'll see a new edit button after selecting a DALL·E image.
Exactly. 😄

So if you want to try out anything, just ask me. 😄

[plucky hare]

I'm not sure, but I'm wondering if the HuggingFace thing might be related to this? #off-topic message I think Lugui was just riffing off my "you wouldn't download a ChatGPT" joke -- I don't see any GPT on OpenAI's HF page.

[plucky hare]

Woo! Thanks for confirming the UI -- I am checking very regularly 😁

[gray surge]

I found this one with exactly the linked image.. but idk. what they tell in their blog post. 😄

Maybe a translater can help. 😂

[plucky hare]

Ahh I see! That one looks like holiday hijinks to me, like Kyle was saying 😁

[jolly spire]

Don't think it's related to that, just try to do Promt more accurately.

[marble loom]

anyone know when sora coming out?

[dark mason]

Does anyone know of a good Dalle bot for discord?

[gray surge]

I think there was mentioned this year or this summer? (Not sure if I confuse that with GPT-5 release)

[gray surge]

For your own usage? #image-bot
For your community? No, I am sorry - but you can develope one yourself using the API.

[pastel siren]

Is chat.openai.com down for anyone else?

Seems to work again.

[karmic glade]

I just got the Dall-E inpainting and editing tool. I'm experiencing something wonky. I've tried multiple pictuers and experienced the same thing. On Firefox it will let me select almost any object. But on Chrome Canary, nothing happens when I try to select the same object in Firefox. In Chrome Canary, it will let me select certain areas, but not all.

Firefox image is above.

Chrome Canary image is above where nothing happens when I try to select the same cranberry sauce. It would only let me do the area starting with what I highlighted and above.

[hearty ether]

I don’t like the faces with the new model

Feels too uncanny

[velvet rampart]

I noticed the same yesterday.

[velvet rampart]

This is a bug. Have you reported this on the bug-reports channel?

My guess is that this has something to do with how new this.

[ionic junco]

Is the new inpainting totally broken for anyone else? I have the feature but no matter how large of an area I select, the photo doesn’t change with the provided prompt. Tried multiple prompts and multiple images.

[velvet rampart]

I reported this bug.

[velvet rampart]

But it is best that everyone who has this promblem reports this

[viral moth]

Look like inpainting works in dalle gpt but was not working in normal chatgpt4 h

[velvet rampart]

Yes. And it is not in customGPTs either. But currently buggy function, hit and miss. Does something but not the requested stuff. E.g. I asked it to put a bird bath into the garden. No, I do not get bird bath but a flower pot is replaced by another flower pot.

[merry vault]

im bored someone gimme an image to generate coz y not

[late blade]

yay, got the new image feature running for me

[merry vault]

there is a new image feature?

[late blade]

yes, inpainting

[merry vault]

bro what

thats out?

since when

[late blade]

2 days started rolling out

still needs work tho, it's shaky feature

[merry vault]

how do i access it?

[late blade]

if you have it rolled out then you click on the image and you should see a new icon at the top

[merry vault]

oof i dont got it

WAIT I DO

[late blade]

cool

but it's shaky I had this image and wanted a subtle change

the change made this

completly changed the image, so not accurate

[merry vault]

wow

the 2nd image is the og

looks like a slapped on png lol

i asked for a dwarven castle in it

[late blade]

lol

[dusky panther]

Does anyone know if there are plans to add these params to the API?

[merry vault]

#dev-chat

[late blade]

for? in the APi you pass the original image and the mask of it

[sterile magnet]

I'm trying to generate the exact same image but in cinematic style, but I always get poor results, can anyone help me with the prompt pls?

Here's it:

A cinematic digital art scene depicting ayoung man under a wet transparent umbrella in a long dark alley. The alley is narrow, and drizzle is visible in the air and on the umbrella. There are two dark figure of pedestrians, not caring about anything and just walking. It is demarcated by two walls with artificial white lamps. The man is standing just below one of the lamps, and the light from it creates a halo and a blur effect on the illuminated umbrella, although the face is deliberately obscured. The light reflects in the air with beams that show off the drizzle in the air. Light also reflects on the wet floor. In the background, we see some buildings hidden in the drizzle, including one emitting a point of bright red light. The image has turquoise green and turquoise blue hues, and the sky is dark. 

The original image is on the left btw

[dusky panther]

You can do that externally via your own UI? I’m talking about using the api externally from my own website, not inside the chatgpt ui.

[late blade]

yeah, you just have to pass the original image and the mask image to the API

[dusky panther]

Wow thanks for the tip. I'll try it out. Can you share a sample prompt you would use?

[late blade]

client = OpenAI()

client.images.edit(
  image=open("otter.png", "rb"),
  mask=open("mask.png", "rb"),
  prompt="A cute baby sea otter wearing a beret",
  n=2,
  size="1024x1024"
)

[dusky panther]

Oh wow is that new?

[late blade]

no...

[dusky panther]

Oh is that dalle2?

[late blade]

yeah

[dusky panther]

Ah yeah i need this for dalle 3. There is a huge difference in image quality unfortunately

[late blade]

dall-e 3 api model hasn't been enabled yet

[dusky panther]

Do you know if there are plans to add it to dalle 3 anytime soon?

[late blade]

I don't know, but if I had to guess, it will come considering all the changes including the changes of the labs

[dusky panther]

Thanks

[late blade]

You put too much noise to the image, I used your prompt and changed some parts. This is how I changed it:

A close-up zoomed in upper body cinematic digital art scene depicting a profile facial take of young man under a wet transparent umbrella in a long dark alley. The alley is narrow, and drizzle is visible in the air and on the umbrella. . The artificial ambience light creates a halo and a blur effect on the illuminated umbrella, although the face is deliberately obscured. The light reflects in the air with beams that show off the drizzle in the air. The background depicts a narrow alley and hints at some buildings hidden in the drizzle, including one emitting a point of bright red light. The image has turquoise green and turquoise blue hues, and the sky is dark.

[sterile magnet]

What do you mean by "noise"? Details?
How can I avoid making the mistake next time?

Also thanks

[late blade]

it's not a mistake, you added too much details about the floor and the lamps. you added details about a second person. so dall-e takes that all in

[sterile magnet]

So, to get better results next time, I should focus on detailing the stuff that really matter in the image?

[late blade]

yeah, you can also tell details of the background, but tone them down with diminutives or words that make them a secondary role

See, I added a second person

I used The background also hints at the silhouette of a person barely recognizable.

[sterile magnet]

Thanks!

[late blade]

This was my last prompt:

A ultra close-up zoomed in upper body cinematic digital art scene depicting 3/4 view upper body take of young man under a tilted wet transparent umbrella he is holding. The alley is narrow, and drizzle is visible in the air and on the umbrella. . The artificial ambience light creates a halo and a blur effect on the illuminated umbrella, although the face is deliberately obscured. The light reflects in the air with beams that show off the drizzle in the air. The hints background depicts a narrow alley and hints at some buildings hidden in the drizzle, including one emitting a point of bright red light. The background also hints at the silhouette of a person barely recognizable. The image has turquoise green and turquoise blue hues, and the sky is dark.

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[sterile magnet]

Why do I get these results from Bing Image Creator? How can I rephrase the prompt to make them more accurate?

[late blade]

hmmm, interesting, let me check

[sterile magnet]

ok

[late blade]

Copilot has some interesting behaviour on that regard, this is my first good try. but the others were too bright.

[sterile magnet]

It always put black

[sterile magnet]

I asked it to put the exact same prompt @late blade

[late blade]

ya, I know, i'm getting the same stuff on copilot

[sterile magnet]

I put the last sentence of the prompt in the beginning, (so it's the 2nd sentence now) and I got this. It's much better ; I think that the order of the sentences in the prompt affects the output on Copilot

A close-up zoomed in upper body  cinematic digital art scene depicting a profile facial take of young man under a wet transparent umbrella in a long dark alley. The image has turquoise green and turquoise blue hues, and the sky is dark. The alley is narrow, and drizzle is visible in the air and on the umbrella. . The artificial ambience light creates a halo and a blur effect on the illuminated umbrella, although the face is deliberately obscured. The light reflects in the air with beams that show off the drizzle in the air.  The background depicts a narrow alley and hints at some buildings hidden in the drizzle, including one emitting a point of bright red light.

[late blade]

copilot likes to overpower light

[sterile magnet]

Exactly lmao

[late blade]

Thing about copilot is that the passed prompt is really short compared to OAI's dall-e

[sterile magnet]

Agree

[sterile magnet]

But I ask it to generate the image with the exact same prompt

[late blade]

A cinematic dark mystery close-up, upper body shot of a young man holding a tilted, wet, transparent umbrella in a narrow alley. The drizzle is visible, creating a blur effect on the transparent umbrella, with the man’s face obscured. The background suggests buildings and a silhouette in the mist. The image has turquoise hues, a dark sky, and a strong mysterious vibe.

I had to work with what is really passed to dall-e on copilot. The prompt that is passed to DALL-E is way shorter

An it still tries to do lots of light regardless

[sterile magnet]

@late blade I had that on my side, before you answered me

It's not with DALLE-3 tho

Juggernaut

[late blade]

really mysetious, I'm almost scared, I won't be able to sleep tonight

[sterile magnet]

Lmao

It's very realistic I like that

[glossy scroll]

What are we talking about guys

Hi @late blade 👋

[sterile magnet]

#images-discussions message

[glossy scroll]

Oh I can help you with that.

[sterile magnet]

It's resolved, actually, 😬 but thanks anyway

[glossy scroll]

Oh 😦

I was gonna say try this: https://chat.openai.com/g/g-ElTvLpZaQ-text-enhancer-for-images

[sterile magnet]

Oh, it looks interesting

[glossy scroll]

Made it myself

[sterile magnet]

Wait

It's a GPT

I don't have ChatGPT Plus 😔

[glossy scroll]

Ohhhhhh...sadge

[late blade]

but, one sec

[sterile magnet]

?

What?

[glossy scroll]

He's gonna bring out a cake

He always does this

[sterile magnet]

Lmao

@late blade

[late blade]

nvm, thought copilot had something in their editor

after all it says powered by dall-e

[late blade]

Cake!

[glossy scroll]

Yaaaaay!

Wow, that's legit a good lookin cake

[sterile magnet]

Generate a slice of it

With delicious cream

[late blade]

no cream!

[glossy scroll]

Don't add the cream though, make sure the cream is inside the cake.

[late blade]

for some reason copilot is making her eat herself

[glossy scroll]

That's the way it should be

I would eat myself too if I was a cake

[late blade]

ok I'll be back alter, got some adulting to do

[glossy scroll]

You're going to go make that cake aren't you?

I knew it

[sterile magnet]

Why?

[glossy scroll]

I always know these things

[sterile magnet]

It will be dry otherwise

[sterile magnet]

Adulting? Huh?

[late blade]

ya, that's what happens from time to time when you need to buy cake

[sterile magnet]

??

[glossy scroll]

Man's got a point

I buy a cake almost every other day

And I don't always want to

It's just part of being an adult

And I just recently discovered how to be one after I turned 18 a few weeks ago

People started buying me all kinds of cake and I was like what is going on with my life

And then I said...

[karmic glade]

Okay, reported it now 🙂

[ionic junco]

Wow, certainly not impressed with the inpainting…

[plucky hare]

Did you use the selection tool to draw over the cat?

[ionic junco]

Yes. I’ve tried about two dozen photos. Small inpainting and large. Absolutely nothing changes. When I do it without drawing in the edit mode it just generates a new image.

[sterile magnet]

OpenAI releases new features too early, that's why the features are unstable lol

[ionic junco]

One thing I’ve noticed is that if you use too small of a brush, it creates two circles, but a large brush doesn’t change anything either.

[plucky hare]

I just tested your initial scenario and it works for me (imperfect black cat aside). This was on desktop mode in a mobile browser -- I don't seem to have access to inpaint yet on any other platform/mode. But perhaps try on a computer, or on desktop mode in a browser? Perhaps it's a bug in the app?

[rancid reef]

where can i read about these updates such inpainting?

[ionic junco]

I’ll have to try on my Mac. Currently on an iPhone in the app. It’s not even offered in my iPhone browser.

[plucky hare]

Here's a helpful article on the new feature! https://help.openai.com/en/articles/9055440-editing-your-images-with-dall-e

[plucky hare]

In regular mobile view on the iPhone browser, or are you trying on the desktop version of the iPhone browser? I think in iOS Safari you can request the desktop version of a site in the "aA" button in the URL bar, if I'm recalling correctly. Sorry if you've already tried this!

[ionic junco]

Regular mobile.

[plucky hare]

Maybe try switching to desktop mode, same browser. I also don't have the option to inpaint in the mobile web version.

[ionic junco]

I can zoom out, but my browser never changes into desktop mode for some whacky reason.

[ionic junco]

Nevermind it worked to allow editing

[ionic junco]

That worked. Thanks

[plucky hare]

Good deal! You might consider posting about the iOS app not working for you over in #1070006915414900886 or see if there's a similar report already. If you post, just be sure to follow the bug report format in the pinned post!

[rancid reef]

thx

[velvet rampart]

Summa summarum, if inpainting does not work what to try? Steps.

[plucky hare]

Hey! Is it just not making any changes at all? What are you trying in (iOS app, Chrome, etc.)?

[velvet rampart]

I have tried iOS app and Chrome on iPad and PC.

[late blade]

The mobile apps haven't been updated to reflect the web frontend functionality

[glossy scroll]

Lol

[late blade]

So can't except to get that functionality if it isn't implemented

[velvet rampart]

iOS app does make changes but not always. And edge is hit and miss.

I have update to the latest version.

[late blade]

hmm interesting, the functionality is there but no update notification for me

might be one of my settings on my phone

but the screen is so small on the iphone... I can't work with that...

[velvet rampart]

The function is included in the 25/3/24 update of iOS app. With that you can also edit prompts.

[late blade]

the update from one week ago says bug fixes and small improvements

[plucky hare]

I think sometimes they're able to enable new app features server-side even without a new update. Like the assets are all already there, it's just a matter of them "flipping the switch" as part of their feature rollouts.

[late blade]

oh that for sure

that's 100% doable

[glossy scroll]

You can always hold the phone directly to your eye. That's what I do.

[late blade]

yaeh but then I can't see myself reflected on the shiny screen, which also stinks

it's a thing only us narscissists understand

[glossy scroll]

I feel that, I always need a mirror by my side too.

"Look at that handsome man. You're da man."

Is what I say when I look in the mirror.

[late blade]

lol

[glossy scroll]

It's even better when you come back to see yourself again.

You're just reminded of how awesome you are.

You guys get that? No? Maybe it's just me.

This is all your fault @late blade

[late blade]

What I do?

[polar marten]

Where can i view all my generated images?

[glossy scroll]

Youuuu

[late blade]

ok, but at one point you'll have to explain it to me

[glossy scroll]

One day

[late blade]

Btw, is it possible to in paint images we upload? Was that planned?

I can't but I don't know if it was meant to be done

[fallow ore]

This Pointillism theme keeps giving me 8 bit vibes

[plucky hare]

There's currently no gallery view in ChatGPT, but it would be cool! Maybe post in #1070006151938314300?

[gray surge]

Already tried to generate something with 8 bits as context? 😄

[plucky hare]

I think currently it's only DALL·E images right now that can be inpainted

[late blade]

yeah just tested, I have some cool images that need that functionality, I'll have to wait until that happens

[fallow ore]

Nah, I've never tried 😅

[gray surge]

Maybe a cool idea. 🙈

[fallow ore]

Lemme give it a go

8 bit vs Pointillism

I actually prefer the 8 bit style

[velvet rampart]

Customer service at best. I contact the OpenAI for my complaint about the inpainting. The rep’s reply:
”Hello,
​
Thank you for reaching out to OpenAI support.
We understand you are having difficulty using Dall-E.
For optimal image generation, we suggest providing all instructions in a single prompt. Keep in mind that Dall-E is not capable of editing images.
Also, kindly provide the screenshot of the image if you are not satisfied.
We appreciate your patience and understanding.”

Yeah, does not know about inpainting function or uses premade answer.

[fallow ore]

Yeah, sounds like a canned response, maybe even AI generated

[gray surge]

Same - but both are good. 😄

[fallow ore]

Digital Fantasy Art vs 8 Bit

[wispy storm]

Look like 16bits no ?

[worldly zealot]

woah the first one is crazy, can you share the prompt

[fallow ore]

Imagine a surreal landscape at twilight, where the ground itself is alive with motion like the surface of a gentle ocean. In this dreamscape, trees twist and grow in spirals towards the sky, their leaves shimmering in a spectrum of pastel colors. The ground is carpeted with soft, glowing moss that lights up with every step taken upon it. A river flows uphill, defying gravity, its waters sparkling with bioluminescent creatures. In the distance, a towering waterfall cascades upwards into the clouds, from which it originates. The sky is a canvas of swirling nebulas and floating islands, each with its own gravity, where waterfalls flow from one island to another in a mesmerizing display. The scene is dotted with fantastical creatures: ethereal deer with antlers that resemble tree branches, birds with luminescent feathers creating trails of light as they fly, and a solitary figure standing on a hill, gazing in wonder at the unfolding scene.

[worldly zealot]

thanks!

[tight roost]

Fantastic! Finally Dall-E 3 can do what I always used to use Dall-E 2 for. I'll say bye bye to Labs now!

[weak niche]

how do you edit the images?

[late blade]

you choose the image, you set the mask, you prompt what you want

and done

[tight roost]

Exactly! Make sure to click the image within the chat, the same way you used to view the prompt before

[drifting kettle]

I am wondering, is it just me, or does DELL-E not know how to spell even thought you spell the item in the request?

[late blade]

it's known DALL-E model can't render text properly

[native ice]

yup

[native ice]

it will never spell right, facts lol

[late blade]

sometimes you get luck tho

[drifting kettle]

Ok, just wanted to make sure.....

[native ice]

Yeah sometimes it gets lucky

[drifting kettle]

Every blue moon.

[tight roost]

It's a lot better than it was before, however

[late blade]

Short texts work ok

with some acceptable ratio

[tight roost]

Somebody has to compare Stable Diffusion v3 and Dall-E 3 in terms of text...

[late blade]

sure, someone has to, and hopefully in #ai-discussions where we can compare models freely

[velvet rampart]

To get correctly spelled, the command is <Add text ”in English or latin” using Python tools.> Replace the wanted within quotation marks. It is also command sensitive. Works like 95~100%.

[late blade]

cool, a thing to look forward to in dall-e

[dense ermine]

Could we edit images before?

[late blade]

nope, that started 2 days ago

[dense ermine]

Question in regards to the announcement

[velvet rampart]

So, I have tested variations they do not work.

[late blade]

shoot, you may even get an answer

[late blade]

It's continuous work, it will be there at some point

[dense ermine]

So we can photoshop images now? I trying to understand what it does in full detail

[late blade]

well "photoshop" images, we can tell dall-e to generate textual context to the mask we place

it's still text2image

[dense ermine]

Wonder if we'll get an upscaler in the future

[gray surge]

And only for images it generated in that chat.. images we uploaded our self we can not edit yet.

[late blade]

yeah, the ability to do edit operations for any image is not implemented, I'm assuming it still needs more testing and safe measures

[dense ermine]

It reminds me of the suggestions given by google's search bar

[hot silo]

Hi, some time ago (2-3 months) I gave up gpt4 because I only used it for dalle3, but I was irritated by unknown limits and suddenly the message that "I exceeded the limit" even though, for example, it was my 4-5th image of the day.

What does the situation look like now after these few months? Are the limits known? I was very encouraged by this change in advertisements, but if the limits still apply throughout the world, I probably won't try.

[late blade]

same deal, same thing on that regard

[dense ermine]

Sora is going to be its own thing right?

[late blade]

yes

[pulsar charm]

not really bald

[late blade]

sora is text2video

[gray surge]

But from that what I saw it's able to generate only images too or not?

[pulsar charm]

but nice haircut

[grizzled loom]

cant seem to find the edit an image feature that was added; is it a slow rollout or should we all have access at this point?

[viral moth]

sora can generate images too

[late blade]

well, video is a squence of images, you are right in that aspect

[viral moth]

it can edit videos and input image and edit it to video

[late blade]

well another thing is, sora is not just one model, it's a goup of models that make sora

I'm thinking, and I might be way off and wrong, it's an adapted dall-e with trained masking to make proper sequential output as video

[velvet rampart]

You can upscale pictures using data analyst function. Also change their other parameters, like saturation. You give a picture and a command. Things are done. You get a download link.

[dense ermine]

Oh neat

[final compass]

Strangest content policy notice in a long time.

[sudden current]

lçol

[late blade]

That's amazing

[echo walrus]

Quick Test using my Custom GPT to write in-painting prompts. I'll have to edit it to add that. I don't know if I can't add links in this channel.

[velvet rampart]

I asked data analyst what it can do for image files. Here is its reply:
As Data Analyst, my capabilities are focused on analyzing and processing data, including images, with the tools available in this environment. For image processing, here are some things I can do:

•    Resizing
•    Cropping
•    Rotating
•    Flipping
•    Converting to grayscale
•    Adjusting brightness, contrast, and saturation
•    Adding simple text overlays
•    Basic color correction
•    Noise reduction
•    Applying simple filters (e.g., blur)
•    Changing image formats
•    Compressing images
•    Drawing basic shapes
•    Creating histograms of color distribution

So, create an image with dall-e and add text with data analyst. Hmm…

[late blade]

Can you try that out also? Just to make it so that it really works? I think there might be some constraints

[velvet rampart]

At least the resizing works.

[gray surge]

Yes.. it works but not good. :/ (Not so that it looks good)

[velvet rampart]

But this diversifies what dall-e can do.

[gray surge]

Yea. 😄

[gray surge]

The same btw. for a "fake inpainting" - some time ago I built a GPT for that, which generated the missing image and added it with Python.

[rigid snow]

Is this image editing in chat GPT with Dall E 2 or is this Dall E 3 and inpainting will come to API later

[velvet rampart]

Works just fine and looks nice. I specified text and font. I asked about the size and the ai put it nicely

[gray surge]

By seeing the quality I think it's Dall-E 3 and for the API I don't know anything.

[late blade]

that's nice

[rigid snow]

Then I Hope it will come to API because I don't have Chat GPT plus

[velvet rampart]

Totally new function. I have learned one thing that should be maxim when using AI. ”Ask the AI how to use it.” And add plenty of imagination, wackiness, out there thinking, thinking outside of the box so that others think that you do not even know that the box exists.

[late blade]

I'm jumping in and out. Got to resolve some stuff, but also wanna stop doing work related stuff.

[rigid snow]

How does the pricing for Dall E 2 images edits work do I just pay for the output or do I need to pay for the mask and source image too

[tired flower]

Latest version of Dalle is completely useless 😦

[velvet rampart]

Tell me more and how/why.

[vital gull]

is there anything we can help with?

[restive canopy]

Did Dall-e just get an update that means you click 1 button instead of writting 1 word? WOW ...... amazing

now I know where mys subscription is going

🤦

[plucky hare]

The style suggestions are a more minor feature, I agree, though I think it's helpful! Especially the aspect ratio tip.

The biggest part of today's update is, in my opinion, the full inpainting implementation!

[restive canopy]

that sounds good. they dont show any of that in the video and as I am scum that lives in the UK, I wont see any of these features for at least 2 weeks at best anyway

[late blade]

It's somethin to work with tho, it's an easly try at some features for prompt engineerin on? imagine trained suggestions

[plucky hare]

Check the "Editing DALL·E Images in ChatGPT" video instead!

[restive canopy]

I dont see that one. In announcements I wony see the one "You can now edit DALL·E images in ChatGPT across web, iOS, and Android. You can also get inspiration on styles when creating images in the DALL·E GPT!".. which implies editing but doesnt show it in the video. Anyway, I cant access it so meh. Just annoying every time they announce anything its a case of, ignore it because its not applicable even though I pay. I dont know who its for but yay for them. In the mean time, I'm using decent similar in panting AI tools for free on my own PC. I mean I use bith, but decent inpainting has been around a very long time so really about time openAI caught up with their subscriber version

[kind grail]

is it just me or is the inpainting...kind of bad? XD

[plucky hare]

The video is up on the OpenAI YouTube channel if you want to take a look! I hope you get the feature soon!

[late blade]

it's good, as long as you provide the right words

[kind grail]

huh and those are? o.O

[late blade]

the ones in the dictionary?

It depends what the mask is on your image, what you want to see in the generative filll

and also, it depends a lot on if you like muffins or donuts

if you have an apple with a worm, and you want to replace the worm with a rupture of reality with white socks art style, then I don't think you will get good things

[kind grail]

something really simple, like "fix the eye" is already enough for now lol

[late blade]

I don't think dall-e is fully content awar

aware

[plucky hare]

That's kind of what I was thinking too. Something like "make the eyes green" might produce a more desirable result, probably even if the eyes are already green. I'm just starting to experiment with how DALL·E prefers inpainting phrasing though, just going off of the language they used in the help article.

https://help.openai.com/en/articles/9055440-editing-your-images-with-dall-e

[kind grail]

problem is that it starts editing the entire face instead of just the eye

[plucky hare]

If the entire face is selected I think it won't really have much of a "choice" but to inpaint the entire selected area

Again still just starting to experiment myself

[kind grail]

I mean if you select only the eye it produces total gibberish

[restive canopy]

surely the No1 prompt will be "fix the fingers"

[kind grail]

honestly eyes are almost worse for me

[restive canopy]

they really need to catch up though and provide consistent face images too. Other AIs have had that for a bit now

[plucky hare]

I'd be curious to know if "Fix the..." prompts work well for anyone. I've had luck with "expand the image" type of edits where a wide image generated as an image-within-an-image with a border

[kind grail]

been trying to fix the eye for a bit now. without any succes, its way off every time

[restive canopy]

on the robot guy ? or both

[kind grail]

nah, just the woman

[restive canopy]

I guess its a case of perspective. what would you say is wrong with hers

[kind grail]

Its literally empty what do you mean

If you just select the eye (and closely around it) it just turns into gibberish:

lets not talk about the abominations that spawn if you select more than the eye:

[restive canopy]

i think this is better

[kind grail]

not really. its "not empty", but also not even an eye.

[restive canopy]

I get something very similar in photoshops AI

for me both characters have the same void eyes

any better ?

[kind grail]

meh. i mean at least its not some eldrich horror, but obviously far from good. I feel like the tool definitely needs some improvements

also fun solution: Just add more hair lul

[restive canopy]

I have to go, but for me the robots eyes are worse even

[kind grail]

not really. i mean robot eyes are often depicted like this, i didnt even think its a mistake

[restive canopy]

its down to perspective but this looks better to me

not right/perfect

good luck!

[echo walrus]

Wow, DALL-E is having trouble inpainting over this obscured moon. It just deletes it every time lol.

Maybe it doesn't know there's a moon there?

I give up on trying to fix the moon because it fails every time.

[gray surge]

May I ask whats your prompt for removing the moon?

[echo walrus]

Replace the current moon or sky portion with a significantly larger, more pronounced moon. This moon should be rendered in a 3D anime style, characterized by its exaggerated size and clarity, ensuring it's prominently visible without any cloud coverage. The moon should exude a radiant glow, with crisp, clear details that contrast sharply against the smoldering, dusky sky background. Its surface should feature the iconic stylized textures often seen in anime, with a blend of realistic and exaggerated craters and seas, enhancing the scene’s mystique. Ensure that the moon’s luminance casts an ethereal light over the castle and the lava flows, accentuating their details and adding a layer of depth to the composition. This addition should harmoniously blend with the existing environment, maintaining the Gothic and volatile atmosphere of the scene while introducing an element of surreal beauty.

In-Painting Prompt created by my Custom GPT. I'm currently working on making a heavily edited image to showcase it.

[gray surge]

I think I would try to tell something like "Add xy" or "Change it to xy" instead of naming special thinks what to replace.

[echo walrus]

It worked decently for this in my quick earlier test:
#images-discussions message

[gray surge]

Yea but at the end you select only the moon for changes.

[echo walrus]

I did. I think the problem is that it's so obscured that DALL-E doesn't recognize it as a moon.

I'm going to experiment with reducing the size of the In-Painting Prompt, possibly editing the Custom GPT to change this (currently set to "around 100 words long" for AI Image Prompts). I wonder if longer In-Painting Prompts confuses DALL-E.

[kind grail]

how do you manage the upscaling? ChatGPT always denies this capability

[velvet rampart]

Use the data analyst (the openai customgpt) function and give it the fiigure as I described earlier.

[echo walrus]

Maybe Chat GPT is just really bad at in-painting a moon? I give up.

In the highlighted area, add a clearly visible, 3D anime-style moon, ensuring it's unobscured by clouds. The moon should radiate a soft, ethereal glow, contrasting starkly against the surrounding darkness, enhancing the scene's mystical allure. Its detailed cratered surface adds a touch of realism within the animated aesthetic.

[kind grail]

also failed with my eyes as shown above

[echo walrus]

Maybe it's the style change that breaks it.

[velvet rampart]

There is no moon. But there is a circle like area in the clouds. Something has happened but not you asked. I and others have seen this bug. Sort of hit and miss.

You can do style change. Does not matter.

Area? App vs web? Browser?

I’d wait a few days. This appears to be a launch bug.

I have noticed that doing the change migth be best produced in stages. Could you try following?
Step 1. Remove clouds from marked area
Step 2. Add moon

[echo walrus]

It worked this time:

For the highlighted area, insert a massive, vividly detailed moon that dominates the night sky. It should be exceptionally luminous, casting a soft, ethereal glow that contrasts starkly with the surrounding darkness, enhancing the mystical ambiance of the Gothic castle and its infernal landscape.

It's possible that the attempt to change the style at the same time as making the change was the issue.

[echo walrus]

Web browser. Galaxy Fold 5

[velvet rampart]

Hmm… more descriptive with the inpainting? Or what? You used wider area?

[echo walrus]

Wow, it is really failing at converting the moon into an anime style. I regenerated in 4 times and got the same bad results.

For the highlighted area featuring the moon, transform it into a vivid 3D anime style. Infuse the moon with bright, pastel colors typical of anime aesthetics, ensuring it radiates a soft, magical glow. Incorporate exaggerated, shiny highlights to emphasize its 3D aspect, making it pop against the smoldering sky.

[echo walrus]

I'm trying to avoid going larger than the exact area than I want to edit.

It changed it a little bit. I suspect DALL-E is just really bad at doing style change when inpainting.

In the highlighted area where the moon looms, transform it into a vibrant, anime-style moon. Envision a palette of rich purples, pinks, and blues, with exaggerated craters and a radiant aura to enhance its whimsical, otherworldly appearance. This moon should emanate a soft, colorful glow, contrasting starkly with the surrounding darkness.

[plucky hare]

I'm still experimenting myself with inpainting, but in the help article they do mention this:

We recommend selecting a large space around the area you intend to edit to obtain better results.
https://help.openai.com/en/articles/9055440-editing-your-images-with-dall-e

[echo walrus]

Hmm, I'll try that, thanks. I didn't think it mattered.

Hmm, nope. I guess the style change is the problem. It has gotten decent, but it's not an anime style moon.

The time that I accidentally didn't select an area worked, but it completely changed the image.

[plucky hare]

All the edit examples in the help article above are fairly simple requests, like "remove birds" or "add cherry blossoms," so this is good testing to see what kinds of things can and can't be asked for right now.

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[echo walrus]

I was hoping to make a bunch of changes one at a time. Add an anime style moon. Add an ice tree. Add a dragon? Just add a bunch of stuff utilizing my AI Image Prompt Creator to write the In-Painting Prompts.

I'm definitely now going to limit In-Painting Prompt to around 50 words long for now.

[grizzled iris]

Sora public beta test release date?

If anyone knows that'll be great to add to my api

I have waited for so long...

[paper jolt]

Time shall be the arbiter

[velvet rampart]

Do in stages. This way Dall-e is not mixing things up.

[echo walrus]

I'm just going to give up on style changes for now.

[left tangle]

Is there API support as well for this: #announcements message

[echo walrus]

Ooh, I really like this one. A great image to build upon with In-Painting.

Lol

I hit the dreaded GPT-4 cap before I could do anything useful.

However, I think I've learned at least 2 things: one, don't try style changes with In-Painting (doesn't work) and two, don't try editing small areas with In-Painting (quality is too low or it doesn't work).

[dim cradle]

how do you use the new edit feature

[gilded osprey]

when you create an image with spiders and then remember you hate spiders and delete the whole chat 🕸️

[left tangle]

Does the API also support In-Painting?

[dim cradle]

can you only edit image in api?

i see no option on dalle in browser

[hearty ether]

Oh on iOS it’s only app

[dim cradle]

actually i see it now

interesting update

[left tangle]

Nah it's only through browser and app

From what I researched

API does not currently have in-painting support for dalle 3

[echo walrus]

And for some reason, you can't edit images inside Custom GPTs. It has to be specifically in the "DALL-E" Chat, not even regular GPT-4 mode.

[dim cradle]

yes that was my problem at first

hi everyone nice to meet you all just wana share with guys some cool images i manage to generate with Dall E

I managed to perfected the keywords wondering if anyone could generate the same as mine?

cool images

[dim cradle]

thank you i have plenty more these the cuts that didnt make the cut

is it a robotic ninja?

[dim cradle]

android more like

i was able to keep the based keyword to generate

and i am so far able to customize the rest

for example

which i think its pretty cool

haha

[fiery salmon]

hi

[dim cradle]

aww this is cute

[echo walrus]

I'm starting to think that in-painting with DALL-E just sucks.

In the highlighted area in front of the castle, a colossal dragon unleashes its fury, its scales shimmering under the moonlight. With wings unfurled and a fiery breath scorching the ancient stone, the beast's ferocity brings destruction. Capture this moment, blending the chaos seamlessly with the serene, snow-laden gothic backdrop.

[echo walrus]

I'm not sure that DALL-E understands what "Golden" color looks like. I've tried a lot to fix it. I think the In-Painting mode is just half-baked.

[grizzled loom]

i dont get inpainting options. is this only supported on a few browsers? I cant get images to open for editing nor locate any pop up tool area

[echo walrus]

You need to do it in the "DALL-E" chat. Not any Custom GPT, and not regular "GPT-4".

[echo walrus]

Wow, finally, golden armor after regenerating that one more time.

[marble parrot]

as soon as I want to try and inpaint with people, it does not like that and gives me error (not a content guideline issue to my knowledge). Spent 20 generations getting constant erros

errors*

[grizzled loom]

thanks got it.

[echo walrus]

I also got a lot of errors when trying for the golden armor on this anime girl.

I was just trying for a "simple" collage of edits to demonstrate the new in-painting feature.

It still took some regenerations, but I think this is what finally fixed it:

[grizzled loom]

fixed an error i nan image as desired. Nice tool for later, AFTER fixing the isntruction. Its not good at all. Scene coherence is 60% at best right now. The inpainting tool is nice and for minor touchups with good work is handy as an editor option, but that pales in comparison to proper isntruction as well as that proper isntruction beign the foundation of further fine-tuning

[echo walrus]

BTW, if you'd like to try my Custom GPT for creating image prompts and also In-Painting Prompts:
https://chat.openai.com/g/g-PimKVGCU0-high-quality-ai-image-prompt-creator-generator

Just ask it "How do I use this Custom GPT" for an explanation, then ask it to explain In-Painting Prompts. I made the instructions such that it would create a kind of guide just by asking that.

[grizzled loom]

i will take good instructions over tools to fix output from bad instructions.

the tool would be fun for perfect output for post production edits of quality output, but editing output that has major semantic and ontological failings is like drinking water from the hole in the boat to stop it from sinking. core principles first. enough social engineering of the product.

before

after

reasonable edit; the whole scene was lacking a cohesive ontological construct to construct the seen.

sure, we can make a malformed gavel a fair wooden spoon, but there is zero sell of a cohesive onto-semantic inference of the contextual intent.

with semantic and ontological alignment you eliminate many of the headaches ahead for video work.

[echo walrus]

Chat GPT, why, why......

I guess blue looks nice, but i can't say that I was prompting for red and then show a blue image..

Also, BTW, the S Pen Pro my my Galaxy Fold 5 is SOOOOOO useful for precise in-painting drawing.

[echo walrus]

Hmm, which flaming wings... I think I like the one on the right a bit more.

Uhhhhhh......

[shy elbow]

So playing with the dalle chat gpt, and noticed that using the photorealistic options, it likes to give you almost twins if asking for two people
Also seems to not understand existing IP's

Also a question, has anyone tried getting dallie to create the body of a person then manualing adding on the head you want-considering it for a moodboard.

[hearty ether]

Angry

Again, it’s hard getting this style

With like the semi cel shaded look with the line around objects

But also the realistic shapes and shading

[sterile magnet]

Oh that's pretty cool, can u send the prompt pls? I like how you stylized the image

[rigid snow]

How does the pricing for Dall E 2 images edits work do I just pay for the output or do I need to pay for the mask and source image too

[mellow grotto]

i just relized how crazy DALL E 3 is

[sterile magnet]

Hi, I'm trying to generate a very detailed cinematic image of a military parade on the Champs Élysees. I have a problem with the quality of the images and the fidelity of the prompt. Can anyone help me pls?

Here's the prompt:

A cinematic detailed 8K scene of a military parade on the Champs Élysees. On a long tree-lined Parisian avenue with France flags, soldiers and tanks parade towards the Arc de Triomphe. The sky is navy blue, but the avenue is sunny and we can see nine fighter planes with French flag tricolor flare in the sky. The cinematic lighting accentuates the textures and details of the the avenue and the soldiers, emphasizing the intense moment shared between them. The three-dimensionality of the planes and artillery soldiers is accentuated, which creates incredible realism.

The image is an example of output I get with it

[fiery salmon]

Wooly mammoth wearing a tuxedo, Impressionistic Motion, grim dark, bottomland, grainy, vintage, 80s cartoon style, simple, rough sketch, sketch, pixel art

[sterile magnet]

Thanks

[plucky hare]

Hey! Can you explain a little more about what you don't like about the output? I for one am impressed that it got the number of planes correct!

Also: are you using ChatGPT, or some other DALL·E implementation?

[plucky hare]

Hi! Reminder that this channel is for discussing DALL·E -- images totally fine to share, if they relate to some discussion. There is a channel for more casually sharing images: #images-canvas! You have the option to make a gallery in #1154829862171844679 too if you'd like

[jagged narwhal]

Hey guys, are there any helpful guides were I can create multiple images with the same exact art style? I requested an image of a humanoid shark a few times. One of the images it gave me was this one. I want to use this same art style and give this shark different reactions. Is that possible?

So far every output makes the art style and shark rather different

[plucky hare]

Hey! Are you wanting basically the exact same image, just with a different expression on the shark's face each time? Or are you looking for different scenes, different shark expression, but same art style?

[jagged narwhal]

Different scenes, different expressions, but same art style.. I tried the new edit feature, albeit not long enough, and it changes the art style of the face and background.

[plucky hare]

This is what I tried:

• Uploaded your image to a new chat, asked for a style description
• Uploaded your image to a second chat, asked for a subjects description
• Pasted style & subjects descriptions in a third chat to try to recreate your image
  This is what I got -- it's not exact, but I think the style is close! Maybe it's something somewhat modular that you can work around with to try to recreate the style in the different scenarios/expressions you have in mind?

This image is rendered in a style that is vivid and cartoonish, with a strong focus on clean lines and bright, saturated colors. It has a highly polished and detailed finish, reminiscent of modern digital illustrations. The art style also bears a resemblance to classic American comic books, with its clear outlines, exaggerated facial expressions, and dynamic shading that gives the impression of three-dimensionality. The artist has a keen attention to detail and texture, as seen in the clothing and skin, and uses light and shadows effectively to create a sense of depth. Overall, it's a playful and lively style that would be at home in editorial illustrations or graphic novels.

[sterile magnet]

I don't like that it's not cinematic and not really detailed. It's kinda blurry. If u can do anything to improve it, I would be delighted

[jagged narwhal]

Hey this looks great! Will try this out. Thank you!

[plucky hare]

One thing to keep in mind is that we can't generate images behind the limiting dimension of 1024 pixels. We can make wide/tall images with a longer dimension of 1792, but 1024 is a hard limit for resolution, and therefore detail. It's possible that using keywords like "8K", "ultra-high definition", etc. might produce images of a style that resembles such descriptors in certain ways (from the way DALL·E is trained on images and their captions), but this isn't a way to "trick" it to do something that it can't.

Not saying this is exactly what you're asking for, just offering this as potentially helpful context. Otherwise: I'd recommend asking for images in a wide aspect ratio as a start (if you're using ChatGPT). This might help give it a little breathing room when it comes to trying to fit in all your specified details. But in general, though DALL·E is a really capable model, there are limits to it's capacity to follow instructions (i.e., it can't include an indefinite amount of details).

[hearty ether]

Same gen id

[vernal vessel]

im having issues logging in to dall-e

[sterile magnet]

Delete your browser cache and try again

[vernal vessel]

okay

[wary gazelle]

Anyone who knows if editing your images with Dalle 3 / ChatGPT only works with images you have created via Dalle? I have tried to get it to work with my own uploaded images and have had no luck. Anybody? It's probs not possible but would like to hear other takes

[plucky hare]

Hey! Yes, that's been my experience too. It seems to be part of the DALL·E environment only.

[wary gazelle]

After trying to get chatgpt to reveal details about certain parameters when editing a image it seems like it sends this parameter:

"referenced_image_ids": ["qyfrkjD1PYRKVbkN"]

where the image id is the image that should be edited.

So unless you somehow get your image on their servers tied to a id it probably isn't possible for now 🤔 😢

[late blade]

that's just for purpose of using the original image and the mask from the API

[late blade]

inpaint tool is not cooperative today with me

I just don't know how to prompt for irregular shapes

[viral moth]

yea its closed inside dalle ecosystem sadly

[vernal vessel]

[naive lake]

Hi Everyone! Is it me or for the past 2 weeks dall-e 3 is letting us generate less and less images? after 10-20 images I hit my daily limit, and during that period it constantly tells me to wait
Please someone from moderators answer.

[wintry epoch]

Hi i been trying to make a family picture in anime style giving specific features of the eye and hair colors and the age of them all and every time the ai are doing wrong colors and the amount of people in the picture and the genders i nearly never get a picture right either. Someone who knows how to do stuff bc even how hard i try its wrong and then i give up or i reached my daily limit

[unreal pollen]

This

🥱

@fast knoll https://cdn.discordapp.com/attachments/1204360881593520128/1222236754132602942/guacyourworld.webp?ex=661eb62e&is=660c412e&hm=0b525257cc25abb7ee3c018d8340c711af914a0b862e030d05fee3d67020a798&

[shrewd garden]

me

[dim cradle]

Guac it to me

[vernal vessel]

I am going to combine the power of art and AI art

Holy crap it made a super cool sketch from the project of someone who I used to know

[vernal vessel]

DALL-E did the impossible she finnished the character my ex bestie worked on

she might have threw away the scetch she made but I dont

I knew AI would advance to the point I can make something out of it

[glossy scroll]

How are you? @late blade

[late blade]

worried, I'm hungry and I have to stand up and stretch my arm to get food, but I'm too lazy today

[glossy scroll]

Don't you have one of those plastic retractable arms?

[late blade]

not yet

[glossy scroll]

Buy 10 of them and chain them together and you never have to leave your seat again.

I use it to make my sandwiches in the kitchen while I'm watching tv on my couch.

[late blade]

lol

[deep pawn]

Hello guys , I want to ask your guys help , Im using Dall e 3 and I want to create like an example for my T-shirt idea but im still weak in area of creating prompt and so on. I have seen some of your guys work regarding the Ai art and it genuinely impressed me. Can you guys teach me ( even a little bit would be good ) ? Thanks~~~

[glossy scroll]

Hey there. Attach that file to Dall-E and ask that you want to design a t-shirt like that with a horse on it and ask "Can you walk me through the steps on putting this idea together?"

[late blade]

So, what do you want to do today?

brb, hunger won I'm gonna get seomthing to eat, brb 10 minutes

[deep pawn]

for me I want to actually to get to know from this how to make this T-shirt still minimalistic but still maintaining its design ( have more artistic value onto it maybe add more lines ? add more ). Well I can say it would look casual but not look like too much colour that would hurt my eyes as you can see it would be focusing on the colour black , white , red and white as you see at the image and also the animal that is being used is equine and the position is same like in the picture like that. Well if you need more clarification you can ask me more. thats all , thanks

[glossy scroll]

You can talk to Dall-E and go through how you want your design to shape out.

What I'm trying to say is that, instead of you conveying to us your ideas, you should just convey them to Dall-E directly.

This isn't like midjourney or any other image AI that you need specific styles to determine your idea. Dall-E is more than capable of translating your ideas directly with its built in GPT4.

Only @late blade can answer this one it seems.

I'm not qualified.

[late blade]

do you have some idea on how to prompt this?

Also, it would be good if you know the parts of a t-shirt (anatomy?), something like the chest (upper, medial, lateral?) I dunno, I'm just winging it atm but that's how I tackle that

This was my first take:

A minimalist wide image of a t-shirt. The image shows only the contours of the t-shirt in neon colors. The shoulder shows a circle shaped rainbow and the lower lateral abdomen part shows a horse in neon stroke

Going the wrong way

it also helps if you use vision, understand how GPT sees your image

this is gonna be challenging because dall-e can be stubborn and not be aware where some elements should be palced

If I come up with a better idea I'll let you know

This just happened....

[echo walrus]

Oh, you can now in-paint inside Custom GPTs. It used to be specifically only allowed in the "DALL-E" Chat.

My Custom GPT:
https://chat.openai.com/g/g-PimKVGCU0-high-quality-ai-image-prompt-creator-generator

[glossy scroll]

Oh wow. Excellent! 👌

[glossy scroll]

Wait a minute, I don't have the inpaint feature for my custom gpts

[echo walrus]

Click the image

[glossy scroll]

It's not there for me.

[echo walrus]

What device are you using? I'm on the web browser, Galaxy Fold 5

[glossy scroll]

I'm on the computer

[fair crescent]

This is interesting. I tried to make white text with black outline appear without a speach balloon and it didn't work. See the original and the result.

[fair crescent]

For me it wasn't there until today, while some had it earlier. It is most likely an A B update, where slowly more people receive the new update. As long as everything goes well they keep rolling out until 100% has it.

[echo walrus]

In-Painting seems to be finicky right now. I regenerated it over and over to get the text in the image decent, and multiple times there was no speech bubble or anything changed at all

[fair crescent]

Thanks! Perseverance is then the key. I'll retry a few times.

[echo walrus]

Also, another time, it completely failed to add/fix a moon.

[glossy scroll]

Ah, right. Makes sense.

[fair crescent]

It didn’t make sense to me until you complained 😄

[glossy scroll]

What did I complain about again?

[fair crescent]

Maybe complain isn't the right word. You inquired about a button not present on your chatgpt.

[glossy scroll]

Oh good. Glad I could help 🤣

[echo walrus]

Autromod issues. Can't post too many images, but also Can't link to Reddit post with the images.

This is the best result I had and that took a lot of prompting and changing the In-Painting Prompts. When telling it to do golden armor or a ruby red skirt, it just did different colors or essentially didn't change it at all.

Skirt, blouse, wings and eye were all separate In-Painting Prompts. I tried to post the process, but Automod says no.

Basically, after the In-Painting failed over and over, I told it about that and had it rewrite the In-Painting Prompt.

[late blade]

that's sittig in front of a computer

[glossy scroll]

yes, that's what I'm doing

[late blade]

but you said you are on the computer

[glossy scroll]

yes I use my computer for my buttocks

[late blade]

lol

[glossy scroll]

It keeps it nice and toasty

[late blade]

I don't question productivity methods

but there's prob a better way

[viral moth]

Rip (even idk who that guy is)

[velvet rampart]

I do not have that yet, but I guess that it is being rolled out in the second wave. But thanks for heads up. We got the feature that was wished for.

[echo walrus]

It is still sooooo bad

[viral moth]

its not even first weeks of that feature so if you want to help developers to make ti better down vote your generations that are bad

[echo walrus]

Also, some characters like Spyro aren't in the copyright blacklist for DALL-E

[old kettle]

Anyone have any suggestions on how to keep text out of Dall E 3 images?

[plucky hare]

Hey! Sometimes it just happens. The best method is just to be as visually descriptive as possible, and not to mention text at all. I'm not sure if you might be trying this, but avoid negative prompting (e.g., "Don't include text") completely -- even bringing it up in a "don't do it" manner will typically confuse the model by it being mentioned at all, so it's best to just try to give it as much visual detail as possible. Sometimes it misfires and includes strange text anyway.

[old kettle]

ah no negative prompting. interesting

[velvet rampart]

Use inpainting to remove the text. Somehow, the inpainting works best with removal.

That is my experience.

[plucky hare]

Good tip! Still adjusting to the new inpainting abilities. 😁 According to their help article on the feature, one of its abilities is to be able to send a message like "Remove x" and it'll try to edit out x, so that might be perfect for trying to remove text!

https://help.openai.com/en/articles/9055440-editing-your-images-with-dall-e

[plucky kraken]

Inpainting has been amazing for removing things like extra fingers and text.
Anybody know how to use the new edit feature in custom GPTs? Or is it not possible yet

[late blade]

hopefully not using a machete for those fingers

[echo walrus]

I tried to use In-Painting to fix distorted hands, but it was still disfigured.

[plucky kraken]

Yeah it sucks at fixing things but it's great at removing them 😆

[echo walrus]

....What

I don't think you meant to send that here, obviously not okay

[cobalt hull]

Hey guys, so is there a way to prompt dalle for uploading a picture and using it as inspiration? Trying to use this one but it changes the elements quite a bit

[vagrant zodiac]

Dall-E can't currently do style-transfer sort of things i.e. keeping all of the original elements of an image but making it look different.

[late blade]

you can upload the image to GPT and the Vision model will give you an idea of what's happening to the prompt. You will have to refine it though. It won't be the same one. Also keep in mind that it's not wise to upload personal images or information.

[echo walrus]

This actually came out really close:

An idyllic, serene lake nestled within a coniferous forest under a partly cloudy sky. In the foreground, a trail flanked by tall grasses and wildflowers leads to the focal point: a couple posing for the camera. The man stands behind a specially designed pink off-road wheelchair, where a woman is seated; both wear outdoor attire suitable for hiking. He is in a teal shirt, khaki shorts, and sports a brown hat and sunglasses, while she dons a light blue long-sleeve shirt, dark leggings, a beige cap, and holds a camera on a gimbal. They beam with joyful smiles, reflecting a moment of adventure and companionship in nature.

This is my Custom GPT that I used to create the image prompt:
https://chat.openai.com/g/g-PimKVGCU0-high-quality-ai-image-prompt-creator-generator

[echo walrus]

If you want, I can delete this here and PM it to you instead.

[cobalt hull]

Ah not really concerned about sharing a photo lol

[cobalt hull]

Thanks I'll check out your Gpt!

[late blade]

still be careful. With that being said, were you able to do the vision description?

[cobalt hull]

Yup it gives a decent description but need to add a bunch more details to the characters. Hoping one day I can just upload a reference picture of like my cat and prompt dalle to place it in XYZ environment

[plucky hare]

I was just about to show you this but I see now that you were the one that asked the question originally 😁

[echo walrus]

Unfortunately, right now, it won't let you edit images. Getting an image prompt and then editing that image prompt is the next best thing.

[cobalt hull]

Lol patiently waiting. I guess I was a little confused with the latest announcement

[late blade]

Date please, we need A DATE!! We have to hold you accountable!

😁

[plucky hare]

That's too much pressure for me!! The only thing I have in this case is the "Archive Access" role so you're just as accountable as me 😁

[velvet rampart]

Good, because this was sort of promised in dalle demo in the autumn already.

[late blade]

Sssssssshhhhh, nobody has to know that......

[velvet rampart]

Could you help me with my prompt?
My prompt is:

Use this prompt exactly “Visualize an enchanting scene set in a fantasy version of 1700s China, along the calm waters of the Yangtse River. Central to this scene is a 30-foot traditional Chinese riverboat, solely manned by a distinct character: a 2.5-foot tall, red-scaled kobold, inspired by Dungeons and Dragons. This kobold exudes Chinese ethnicity, with a notably clean-shaven face, highlighting his prideful feature - long, thin, and elegantly styled moustaches, which he wears with great pride. His head boasts long, sleek black hair tied into a traditional braid. As the lone navigator of his boat, he emanates a sense of solitude and purpose. The art style should mirror the traditional Chinese art of the 1700s, executed in watercolors and black ink. Aim for a 64k definition with hyperrealistic and physically based rendering, ensuring hyperdetailed mid-level detailing. The artwork’s 16:9 aspect ratio should enhance the broad river scene and detailed boat craftsmanship. The brush strokes, clear, precise, and expressive, are key, capturing the fluid beauty of watercolors and ink, artistically depicting the landscape and the proud, solitary kobold boatman.”

[late blade]

I'm just teasing our high ranking and now offical @plucky hare , it's healthy to just say it comes when it comes.

[velvet rampart]

I get too big kobold on tiny boat when it should be the other way round.

[plucky hare]

I can take it 😁 It comes when it comes indeed!

[late blade]

you emphasize the size of the kobold too much. don't give the size so much importance. rephrase it with something like "a considerable large kobold" or "a slightly larger kobold compared to the boat's crew"

[velvet rampart]

No, I want tiny kobold on a big boat.

[late blade]

dall-e is reciving "tall kobold"

so, avoid that

[velvet rampart]

So, only 2.5’?

[late blade]

not even that

[velvet rampart]

Tiny kobold?

[late blade]

replace all that with "small"

[velvet rampart]

Ok

[late blade]

tall implies height, height means large

[velvet rampart]

Let me try. Back in few.

[echo walrus]

Just tell Chat GPT that feedback

[velvet rampart]

I tried. Did not work. But now I omitted the lenght and height. Better result.

Still missing the mustache bit. @late blade thank for the tip. I’ll try to use measurements unless they are needed. Sometimes they do help. But in other context.

[late blade]

That's actually a tip for prompting to dall-e. I call it myself adding noise. By that I mean adding words that can potentially mean something else for dall-e. Specially when a person doesn't speak English as a native tongue (it's my 4th language) have to be very careful.

[late blade]

I'm guessing you are looking for something like this, but the Kobold is smaller right?

[sterile magnet]

Where do you find all these tips lmao 😭

[late blade]

My head?

I seriously don't know from other sources, it's how I work most of the time.

[velvet rampart]

Yes! What was the prompt?

[late blade]

I used your prompt and removed the noise I mention:

Visualize an enchanting and historically accurate scene set in a fantasy version of 1700s China along the calm waters of the Yangtse River. There is a large, traditional Chinese riverboat, accurate to the era, graced with exquisite detail and craftsmanship. A small, red-scaled kobold with a moustache reflective of Chinese ancestry from the era mans the boat. He has sleek black hair tied into a traditional braid, and his attire and moustache style are authentic to 1700s China, displaying a sense of solitude and purpose. The art style mimics traditional Chinese art of the 1700s, with watercolors and black ink, featuring hyperrealistic and physically based rendering. The widescreen format showcases the expansive river scene and the grandeur of the riverboat. The brush strokes are clear, precise, and expressive, capturing the essence of watercolors and ink in a scene that artistically represents the landscape and the proud, solitary kobold boatman, embodying his Chinese heritage at the helm of his vessel.

[velvet rampart]

You can always discuss with dalle. That way I learn things.

[velvet rampart]

Could you show it here? I’d like to compare.

[late blade]

I just posted it, had to edit the post

I always laugh about how people add they want an image with one quadrillion megapixels and all in ultra high definition that only gods can create and what not. just say high quality

that's just adding noise to the prompt

[velvet rampart]

Yeps. Works. Still learning the art of prompting. A bit of hit and miss.

[late blade]

There are two boats!

[velvet rampart]

Well, it does not matter. By the way, you said magic word ”noise”. I tried really simplistic prompt in my customgpt for dalle and it work like magic. Thanks!

[late blade]

Glad to hear it

[clever phoenix]

My prompt philosophy

[sterile magnet]

I don't have ChatGPT Plus 😕

[late blade]

Keep in mind, the discussions are done with GPT model, the images are done by DALL-E

That's the drawback of how I work, I got to define it that way

[echo walrus]

My prompt philosophy is to just have Chat GPT write it. I want to make a Stable Diffusion version and I had been attempting one, but then Stable Diffusion V3 was announced and now I'm not so sure.

[late blade]

That's also a good idea, sometimes you do stuff you did not expect to do

[clever phoenix]

I'm mostly the same way, the prompts I share in #daily-theme are what chatgpt comes up with, mine are much more basic and a mess lol

[late blade]

but I'm not that creative to be honest

[clever phoenix]

You're very creative, there's a plethora of evidence on the server to prove you are 😉

[late blade]

hehe, prob need the right images

I do love my cat-girls and my fox

and I haven't taken Merlot and Rosé out for a while

[echo walrus]

I gave up on Stable Diffusion because it seems like Stable Diffusion's prompts are based on arbitrary tags, while DALL-E seems to function well with actual human language descriptions.

But I'm going to try out Stable Diffusion again now that I've gone from Intel ARC to NVidia (dual RTX 4060 Ti 16GB). It'd be nice if Stable Diffusion could use both GPUs at once.

What I really want to do is get a collection of high quality Stable Diffusion prompts, then feed those into the playground version of Chat GPT with 128K context and reverse engineer some Stable Diffusion prompt creation instructions.

[late blade]

How does Intel ARC perform?

[echo walrus]

I suspect that Intel ARC has a quality issue because my results were always garbage, nothing like I've seen other people post with Stable Diffusion.

But for actual performance numbers, I don't remember. I sold my ARC A770 16GB GPUs.

[late blade]

I was just wondering, considering I've sent many nvidia ones to heaven

[echo walrus]

It was also a gigantic pain getting Stable Diffusion to actually utilize the Intel ARC GPU.

It just seems like unless you're really skilled with Stable Diffusion, DALL-E is far, far higher quality and adheres to the prompts way better. But if you're good with Stable Diffusion, the images are way better than DALL-E.

This was my experiment on making a Custom GPT specifically for Stable Diffusion. But no matter what, I couldn't get good quality, which I suspect was Intel ARC issue.

I was also experimenting with making it use danbooru tags for anime models, but it just made up tags lol

[late blade]

seems very finiky

considering there's no real support with tensorflow nor keras, at least not officially

[echo walrus]

It was based on the idea that Stable Diffusion understands prompts in chunks, where after a certain amount of tokens, it kind of starts over? Something like that. I forgot exactly.

The idea was that by splitting into chunks, I could make long prompts usable.

DALL-E is by far more user friendly, but maybe that will change with Stable Diffusion V3

[late blade]

prob, but if we gonna talk about that, we should move to #ai-discussions

[viral moth]

If you ise stabediffusion XL trubo you shoud turn off Half VAE and set steps to 2-3 yeee

[echo walrus]

Like Dys Topia said, let's move to #ai-discussions

[sterile magnet]

Hi guys! I'm trying to generate a very realistic photo, but although I added keywords and kept my prompt simple, Bing Image Creator keeps generating AI-looking photos. How can I fix that?

Here's the prompt:

a photo of a blossoming pink cherry tree spring timeframe. Branches covered delicate white flower petals contrasting azure sky backdrop. Seated upon branch orange cat, emerald green eyes trained skyward attentively. Cat's orange-white fur appears soft, fluffy texture. Overall scene emanates serene, picturesque springtime ambiance. natural, real, clear, sharp, high quality. Natural sky

Here are the original and generated photo:

[late blade]

something like this?

[sterile magnet]

coincidence?

I tried black cat just before lmao

[sterile magnet]

No, it still looks like AI

What did u modify tho

@late blade

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[late blade]

how about these?

[sterile magnet]

Wow! They're much better! Can you make the the luminosity more homogeneous and soft please?

Especially the color of the sky

I don't want a gradient

[sterile magnet]

I want the sky to stay blue (like in the original photo). And there is still the sort of gradient with the lighting, and the blur effect. 😕 But the cat pose is great!

@late blade

[sterile magnet]

Almost that. The cat is slightly weird tho. But the sky is good, just remove clouds

[copper light]

since the theme changes every 24 hours for dalle, what time does it change to a new theme?

[late blade]

in 1 hour and 10 minutes

[late blade]

The customer is demanding too much!

[copper light]

so 1 PM Everyday?

[late blade]

I dunno where you live, I'll say yes?

[copper light]

(im EDT/EST US so I put 1 PM)

[late blade]

in 56 minutes

[late blade]

My last try

[sick flax]

Hi, there! I saw your exchange with @late blade and wanted to give it a try, because I've had similar issues with realistic photographs lately. Here is my best result. "Professional photography, a blossoming cherry tree in the springtime, its branches covered with delicate pale pink flower petals against blue sky in the backdrop. An orange cat with soft fluffy fur is seated on a branch, its emerald green eyes trained skyward attentively and bathed in warm spring sunshine. Utilize bokeh in the background."

[late blade]

Love it

But the light is overpowering

Here's my feedback:

Photography: implies and can, but not necessarly, introduce light artifacts of studio montage
backdrop: suggests a set, a studio, can be a greenscreen, or a fake montage element.
bathed in sunshine: lighting spotlight effect

[sick flax]

I kinda like it that way (I live in California lol), but yeah, it can be easily toned done, like, "diffuse/soft sunshine".

[late blade]

fair enough, then it's all good

My prompt for the image was:

A casual serene scene with a blossoming pink cherry tree during spring. Branches covered with delicate white flower petals harmonizing with the surroundings. A clear sky. Resting upon a branch, a small cat. The overall scene blends as an everyday life scene.

Tho the global settings of the chat instance are set for real life images

[sterile magnet]

Thanks!

[sterile magnet]

Love the 3d effect

Thanks

[rose laurel]

How to reference Dall E images?

[sterile magnet]

Wdym

[rose laurel]

Im using a DallE image for coverpage

for a dissertation

so i should probably cite/reference it

[dim cradle]

for school? ask your librarian

[late blade]

For DALL-E within ChatGPT you can use the In-Text Citation of either the Prompt or the Revised Prompt. List the current model and the knowledge stand when you used this. The date you used the Chat session. If relevant, input prompt and output info. If the Chat can be shared (which isn't the case for DALL-E Chats currently) also provide the link and share it to the end reader of your work.

For DALL-E with the API just Input Prompt, Output Prompt, Stand of the DALL-E model as of the date of your image generation.

Also check for the required citation format you are using if there's already a set rule for Generative Text citations.

APA and MLA afaik have already introduced a format of citation if you use GPT or similar sources

[sick flax]

Look up the style guide for the citation format you are using. Major ones should have specific instructions on AI generated material.

[rose laurel]

lemme check endnote quick

nah it doesnt

not that i can see atleast

[sterile magnet]

How did u make that?!

Can u tell me please?

[fiery salmon]

I'll give you the first 5

Impressionistic Motion, grim dark, bottomland, grainy, vintage

[sterile magnet]

@fiery salmon I'll try to make an image of your style and you'll tell me if I approach it or disgress from the style

[fiery salmon]

ok

[sterile magnet]

@fiery salmon

[fiery salmon]

acccurate

[sterile magnet]

Love it

A scene of a big dark mammoth zombie wandering in a rocky valley, Impressionistic Motion, grim dark, bottomland, grainy, vintage, visible strokes, comic videogame

[fiery salmon]

Impressionistic Motion, grim dark, bottomland, grainy, vintage, 80s cartoon style, simple, low quality, rough sketch, sketch, abstract, surrealism

[grizzled loom]

this was fun #image-bot message

[late blade]

@sterile magnet are you posting those images on purpose to induce me vertigo? I'm getting scared...

[late blade]

That one is cute

[fiery salmon]

Any requests?

[late blade]

Yes, please don't get in trouble

[clever phoenix]

Drink plenty of water and take care of yourself

[fiery salmon]

no I mean ai

like what do you want to see with my style?

[vale dust]

A person riding a horse into sunset

I gave it a go and made this

[fiery salmon]

any other?

something complex

[late blade]

I would recommend to you guys, if you want to post images that can be scary, do mark them in the future with spoiler. It is ok to post them that way. It's something moderators control and it would be a shame if they erase cool images.

I know for some images are not scary. But it's how it works around here. Keep it safe, don't want you guys getting into trouble

[vale dust]

Ok, thanks 🙂

||hi||

I did it wrong, lol

[late blade]

For some weird reason Discord is using unusally large ammounts of resources on my PC...

[vale dust]

I made a non-scary image

[fiery salmon]

Mad God

[vale dust]

Very mad

0o0

[vale dust]

What was the prompt for this one?

[fiery salmon]

The one I used for the last few, but I just put the word "Rastaman"

[vale dust]

Interesting 🤔 I rly like that style

[quartz vale]

please don't forget to spoiler any images that may be scary!

[fiery salmon]

Peach with Notorious B.I.G

Princess Daisy with Eminem

[late blade]

and Bowsette?

[fiery salmon]

Dr dre with Lara Croft

[late blade]

Bowsette or deal breaker

[fiery salmon]

Fine Bowsette with Tupac

[late blade]

close enough

[fiery salmon]

What other rapper?

or musical artist?

[clever phoenix]

2B with Biz Markie please

[fiery salmon]

It doesn't know who that it

[clever phoenix]

Depressing, but thanks for trying

2B was pretty close, but not the best I've seen

[late blade]

it's ok for the art style I guess

[fiery salmon]

black 2b

[late blade]

Coipilot used to make cool images about 2B, now it's blocking it

[fiery salmon]

Nuh uh

[clever phoenix]

Really? I can her pretty consistently get it with ChatGPT, but I don't really use copilot

[late blade]

it might be triggered by something else tho. MS Content Policy is so weird

[fiery salmon]

I'd eat

[late blade]

lol

[late blade]

one image out of 10 tries...

[late blade]

2 images, but not gonna waste more boosts on trying to see if it works or not

[fiery salmon]

So...have you seen the movie Ice age? Yeah, not what happened back then,,,

"so uh... Invincible season 2...not that good actually, kinda mid."

[whole trench]

Anyone know if there is any plan to add the feature we had in DALL-E 2 where you could make a giant image by combining different prompts at different places to DALL-E 3

I forgot it’s name

It was in the editor section

[gray surge]

I think you are talking about the Outpainting?
If yes: There is nothing I know.

[late blade]

Inpainting is in place, outpainting not yet

but it might be just a matter of time before that is available

[vagrant zodiac]

I am really looking forward to it if it happens

Was one of my favourite features of DE2

[sterile magnet]

Lmao

[late blade]

Here's a suggestion that may not apply for everyone, but might be worth looking into https://discord.com/channels/974519864045756446/1226506750673555486

[verbal sequoia]

<:book_icon:1171408210398289941> `` Rule 2 `` All content must be suitable for all ages.

Refrain from using explicit language, posting NSFW content, or sharing graphic images. Any unsettling or horror content that adheres to OpenAI's policies should be hidden behind a spoiler.

[late blade]

I was checking the hard core image creators in this server, there are people with so many images posted, I'm amazed

[quartz vale]

At one point, I was one of the top posters. No longer 🥲

[late blade]

still about 20k messages.... that deserves this

[quartz vale]

it might be a bit more than 20k

[late blade]

welll I cna't ssee hidden channels o.o

Too much responsability to be honest

[quartz vale]

my total message count is 33.1k

[late blade]

any hidden personal "fiendship" channels?

[gray surge]

I only have 1053 messages in this server. c.c

[quartz vale]

to be fair, I've been here since August of 2022

[late blade]

thats a good time to get that done

[late blade]

Interesting, same prompt that generated this #daily-theme message also made this weird variation of it.

[rancid fractal]

Can I use ChatGPT Plus / Dalle to do the same stuff those high-priced interior design AI services do (give it a photo of a room, ask for jazzed-up ideas in mock-up)?

[tall mason]

Donno exactly what you have on mind, but you can always try dall-e a few times a day over here. #image-bot

[rancid fractal]

Thanks. I’m talking about uploading a pic of a room and having AI give me some interior design ideas - showing me how it could look in certain styles.

[tall mason]

Sadly, right now, Dall-e 3 only has in-painting, and only on images you've generated. Used to be you could upload a photo and edit it, but that was an earlier model.

[rancid fractal]

Ok thanks

[sterile magnet]

Does anyone know a good dubbing AI? (I don't have Eleven Labs credits anymore)

[fiery salmon]

I just typed in Squidward

[sterile magnet]

Are you on Bing or ChatGPT?

[fiery salmon]

bing

Copilot Designer

[plucky hare]

Hey! Cool pics 😎 just a heads up that this channel is for discussing DALL·E, not for casually sharing images. We do have the awesome channel #images-canvas though, which is for exactly that!

[vagrant zodiac]

oh that's cool, what was the prompt/revised prompt?

the sword reminds me of the arthur sword from Hades

[glossy scroll]

A vibrant and detailed widescreen drawing of a Dungeons and Dragons style warrior in a misty forest clearing, presented in a 16:9 aspect ratio. This majestic scene features the warrior in full traditional medieval armor, ornately decorated with intricate gold and silver designs, standing boldly in a forest clearing. The armor shines under the mystical light emitted by their large, two-handed sword, which glows with magical energy. The warrior's open-faced helmet reveals a determined look, and a deep red cloak billows in the wind behind them. The lush, green forest surrounding them is filled with an ethereal mist, adding to the magical atmosphere of the scene.

[vagrant zodiac]

many thanks

[glossy scroll]

[vagrant zodiac]

is that your dnd character?

[glossy scroll]

nah i just use dnd images as a reference to tweak my custom gpts, so I frequently use dnd inspired prompts

i use to play dnd back like 15 years ago

actually more like 20 years ago

3.5 days

[vagrant zodiac]

oh that's cool

I guess a lot has changed

I will say Dall-E has been a great help in running my campaign

[glossy scroll]

I bet, yeah. Now chatgpt can make stories and everything, practically run the game itself.

[vagrant zodiac]

my players certainly expect some visuals alongside the narrative

[vagrant zodiac]

ah... except chatgpt... which IMO isn't quite there yet to write good stories.

at the very least I find some dall-e gens can actually inspire me a bit to write more

[glossy scroll]

but don't you use it even as a supplementary, to help you write those stories?

chatgpt i mean

[vagrant zodiac]

nope

I find ChatGPT to be information-minimalist: it constantly reduces inputs to their most basic, minimal form. That's useful for data retrieval and writing code, but I find it to be significantly less creative than just a few minutes of free-form writing while looking at a dall-e gen!

[glossy scroll]

Maybe it can use both! I can perhaps make a custom gpt that incorporates both picture making and story telling for ya if you want?

[vagrant zodiac]

I have tried aha

[glossy scroll]

still no good? dang

[vagrant zodiac]

for now, I am certainly happy with my little app for generating dall-e images then just writing stuff myself

I started using a Gamemaster GPT for that, and yeah while the images were great (i've used a few ofthem early on), the writing was not

I'm not an English teacher but... I'd probably give ChatGPT a B on creative writing at the 14-16 age bracket

[glossy scroll]

maybe it doesn't quite understand how to allocate its storytelling tools. I can perhaps find and attach a pdf file for story telling techniques that it can take from and use as a good example to make some neat stories.

I remember my GM was a writer/director and he would always tell me The Rule of 3 for storytelling.

[mystic totem]

???

[hearty panther]

no kissing allowed

that's sinful

what would god think

[vapid elk]

it works for me

test 1

test 2 and 3

[hearty panther]

it appears the context in which the situation is presented alters the way the "sexual" trigger filters activate

very interesting

or their "moderation" filters I mean

[glossy scroll]

You're right. I tried my best. The stories are mediocre. We'll stick to image gens. Hehe

[vapid elk]

yep, the filters depend a lot on the entirety of the content

[dim cradle]

Why does DALL.E not listen to prompts?

[dim cradle]

I am trying to get dall.e to make this sketch of a rock not as wide, but taller. It doesn’t follow any command I give it. I’m quite experienced in GPT prompts. I’m not an idiot. Can someone help me?

I want the pencil, hand drawn sketch look to remain. I only want the rock taller and narrower.

[vital gull]

do you want the actual rock to be tall or the generated picture?

[crimson helm]

Does anyone have any suggestions on how to make Dalle-3 create consistent characters?

[formal osprey]

rotate 90°. fixed

A lineart pencil sketch of a rock, emphasizing a clean and minimalist style. The sketch should focus solely on the outline and essential shapes of the rock, without any shading or texture details. The lines should be clear and precise, capturing the basic form and contours of the rock with simple, elegant strokes on white paper.

unsure if the shadow is undesired

[hearty ether]

Wait you can’t inpaint images with custom gpt?

[dim cradle]

I want the actual rock to be taller, and not as wide.

[vital gull]

i just used a simple pencil lineart of a tall spiky rock

[dim cradle]

I want my actual rock. The rock that is there, made taller. I dont want a different rock.

[vital gull]

So adjust accordingly, ask it to give it alternative shape but with a spiky top, and command it to leave everything as it is. Alternatively you can edit that particular image. Using the select tool

[dim cradle]

I literally say that, and it doesn't do it.

Im using dall.e through gpt 4, could that be the problem?

[vital gull]

the first icon on top, do you see it when you click on your image?

[dim cradle]

I mean, seriously...?

Yes I know how to select an area to edit. I literlly want it a little taller and not as wide. I dont want a completly different rock. How stupid is this thing

[vital gull]

it's not about stupidity, its just how models work, usually it's still random and can't really follow your request a 100%

[dim cradle]

I gathered that...lol

Damn this thing is in beta mode lol

[glossy scroll]

I don't even have my in-paint feature yet and I got plenty of image gens ready to test. How sadge is that.

Meanwhile you're talking about drawing a rock.

A rock!!

Out of all things

A rock!!?!!?!!

[dim cradle]

If it can't draw a rock.... well....

[glossy scroll]

...maybe it can draw a stick...?

[open trench]

Hi all

[glossy scroll]

hi nezhno

[valid mantle]

#image-bot message

[dim cradle]

This is cool

[glossy scroll]

Ty

[dim cradle]

Whats it based on? Warcraft?

[glossy scroll]

I asked it to create a DnD art style mixed with an anime style, but 60% DnD, and 40% anime.

[dim cradle]

Cool

[glossy scroll]

Yeah, those half styles are cool. I've always liked them

[vernal vessel]

making a new gen

[dim cradle]

I've lost count how many times i've told dall.e not to include any other object. How stupid is this thing? It can't follow a simple instruction

I literally can't get over how useless this is at following instructions. It shouldn't even be released to the public yet. It's horrendous .

[vagrant zodiac]

dall-e doesn't have negative prompting. if the prompt includes 'no pencils' , it might try to include it more. Best to go back to the original prompt before mentioning pencils and deconstruct it a little/re-run it.

[glossy scroll]

@dim cradle what jord said, and avoiding negative prompts also hints at how dall-e receives words from the user in general. meaning: don't always try to speak to it using normal language patterns, instead also try and incorporate arbitrary words with how you want to picture to represent, such as "Make me a solid rock, white, grey, creases and cracks, like the picture, but bolder in textures. compare and contrast."

explain all the words that come from your mental visualization and just say them. you don't always have to arrange the words properly. just try and connect as many pieces of context as possible. dall-e will pick up on what you're trying to say.

[late blade]

I assumed you are saying to dall-e to do a pencil sketch or another phrasing that implies the use of the tool to achieve the depiction. This will be interpreted by the model as "the user said pencils, let's add pencils", so keep that in mind, it's tricky in some situations when the visual style is strongly bound to the tool needed.

my prompt was a minimalist black and white line art sketch with subtle shading of a geological cube-like outcrop

[dim cradle]

Can you make it so its based on this outcrop?

[late blade]

name the type of rock in the prompt