images-discussions

[agile peak]

Using the word "Subdued"/"Diffuse" really make paintings feel improved on ChatGPT.

[deft musk]

And I am not a mod, sending me DMs cannot help you, sorry. But modmail can help you 🙂

[quartz vale]

modmail can't help with anything outside the scope of this server! You'll have to go to the help website 🙂

[crystal badger]

highly recommend sending them a mail

[quartz vale]

I've never used Modmail

[crystal badger]

well..

[hard marsh]

This is absolutely stupid, I got suspended from DALL-E 3 for 1 hour by asking it to make a DreamWorks Racing game, featuring shrek and b.o.b.

I am so done

[crystal badger]

you are Senior Moderator, who can help you.

[quartz vale]

you may have hit the rate limit, friend 😔

[hard marsh]

It thought I was requesting something inappropriate

[crystal badger]

DALLE has a bad habit of having Some words influence the entire image.

Thats one of the reasons i use my CI. i have a precaution built in.

it can happen unintentionally.

[deft musk]

Lol. Whoops, sorry to misdirect the lost one.

[deft musk]

Wow, glorious!

[crystal badger]

thats a bad, generation xD

[feral minnow]

On Bing or on ChatGPT?

[deft musk]

Haha 😄 That's okay, I don't mind that I have terrible taste. Love that though!

[crystal badger]

i like it to but small things make it bad for me.

the border is cut off at top and butm that ruins it for me

heres a quick Gen.

[deft musk]

Oh, gosh. Yeah. I'm so accepting, hard for me to have anything be ruined, I just think of it as different

[deft musk]

May I please try to mess with this prompt? I'd need to know what it is though.

Also, what else besides the border is a 'ruined for you' issue?

[crystal badger]

im certain my CI plays a huge impact in how images are Generated.

it starts like this.

[deft musk]

nods But not to dalle, only to chatGTP who made it.

May I have the dalle image gen prompt?

[crystal badger]

16:9 hyper-realistic top-down fantasy D&D world map, magnified for clarity, showcasing diverse islands and continents, all surrounded by water, with an Arcane Silvery border touching the screen edges."

[deft musk]

Thank you!

Anything besides the cut-off border something you don't like?

[crystal badger]

either its too zoomed in or out, or not the entire world is in view.

or it generates a "Frame" on a table with an img.

Its a word association issue

[deft musk]

Hrm. That image as is, how's the zoom itself?

[feral minnow]

Associating maps with frames, perhaps?

[crystal badger]

its always too zoomed out or in

[feral minnow]

Try just “world”

[crystal badger]

the word world is in there.

[deft musk]

That image, the one without the complete frame I asked for the prompt for - is that one too zoomed in or too out for your preference?

[crystal badger]

that was the frame annoying me

[deft musk]

Oh, this is for the fully-framed one. Could I have the prompt for the one with the frame problem?

[crystal badger]

16:9 hyper-realistic zoomed-in top-down fantasy D&D world map, showcasing detailed islands and continents, surrounded by water, with an Arcane Silvery border touching the screen edges, and several floating islands hovering prominently above.

[deft musk]

But in case you like this?

[crystal badger]

or was it this one

[crystal badger]

i dont like that one because i see grey outside the border.

[deft musk]

Understood!

[quartz vale]

I love making little maps with Dall-E

[crystal badger]

hmm.

Maybe ill try a new chat, been on that 1 quite a while.

issue is i like semi realism as a tag, i suspect thats the main culprit.

[deft musk]

Ahh, had been trying to ask for this one's prompt.

Sorry I wasn't clear.

[crystal badger]

ohh sure

[deft musk]

Was hoping I could find a way to add the border correctly to that one 😛

[crystal badger]

Top-down hyper-realistic entire view of a D&D world map surrounded by water, with landmarks of human significance and a decorative traditional border, ensuring no parts are cut off.

see i keep having the same issue xD

heres a newly generated one, cut off again.

[deft musk]

It made this for me, that exact prompt:

No difference, right?

"Top-down hyper-realistic entire view of a D&D world map surrounded by water, with landmarks of human significance and a decorative traditional border, ensuring no parts are cut off."

Maybe we're no longer seedlocked?

[crystal badger]

hm

didnt take that into account.

[deft musk]

me either

[crystal badger]

Prompt Acknowledgment:
I'll recreate the map using the seed from the image you chose, ensuring the entire world is in full view, surrounded by water, and that the water is encased within a silvery arcane border.

based of this i'd assume seed changes.

[agile peak]

Was trying to create serene sunset view. Dalled ended up generate Tatooine sunsets haha.

[deft musk]

Glorious though!

[agile peak]

Agreed, accidentally create a masterpiece!

[deft musk]

We are seedlocked.

So you haven't given me the prompt I was trying for, but that's okay!

[crystal badger]

i have, thats the one

Top-down hyper-realistic entire view of a D&D world map surrounded by water, with landmarks of human significance and a decorative traditional border, ensuring no parts are cut off. gave it earlier.

[deft musk]

The 'they appear to be identical' is a hallucination.

But I can create the identical image, but it is not the one that makes that image with the cut-off border.

[crystal badger]

huh

[deft musk]

The only way we can make the exactly or nearly exact image is to use the same prompt and be seedlocked.

That I have tested a fairly large amount

[agile peak]

Am I the only one? find generate one image is easier to be consistent to change the subject without affecting style, colors and lighting.

[crystal badger]

they dont appear to be identical they are identical

[deft musk]

I tend to do sets of 4 like this, myself:

One of exactly this, then 3 gentle riffs on it:

16:9 hyper-realistic zoomed-in top-down fantasy D&D world map, showcasing detailed islands and continents, surrounded by water, with an Arcane Silvery border touching the screen edges, and several floating islands hovering prominently above.

[agile peak]

This is so good for oil paintings. Haven't tried for another types yet.

[crystal badger]

maybe its because its derived for 3 previous generations, ill double check

[agile peak]

These are generated 1 per image and then tell GPT to change the trees surroundings.

[deft musk]

I'll have to upload both, but I bet there are small differences, visible if we switch between very fast.

I already tested this, so I am quite sure at least 1-4 small details will barely flicker, slightly different.

I see 4 areas with color difference that 'flickers' as I rapidly press the arrow key between my images

[quartz vale]

there are very minor differences that I can see too

[deft musk]

I can circle them on another upload if desired

[crystal badger]

ohh yeah i see it.

[deft musk]

But Dino sees too, so

[agile peak]

The only thing. I noticed the top left mountains are showing a climate change effects 😄

[deft musk]

Same latitude, far right as well.

A cluster of islands mid right.

The very middle right brown circle.

[crystal badger]

thats very well spotted didnt notice that until u said it.

so we can get identical imgs assuming we're on the same seed and use same words.

[agile peak]

Anyway. Let's say that dalle generated 4 images from the prompt we just provided, how do you generated variations without changing the style, art or preventing it from out of line. Just subjects?

[deft musk]

almost identical. I have never seen a true duplicate, even seedlocked, but the shifts are very small.

[crystal badger]

"Almost" yeah but we might as well call it identical even if there are minute differences.

[deft musk]

That's why I want your exact prompt, for that border cut off, and why I believe you didn't give it to me yet 😛

Because I think I can mess with that one to get one extremely close but with the full border.

[crystal badger]

what i am trying to account for is the word Association issues.

if i say D&D world map, detailed, high fantasy, framed by silvery arcane decorations, itll give me a pic of a wall and a painting. "Frame" = Picture frame, Image Frame.

[agile peak]

Look like I just gonna tell it back to using same style from previous image then. Solved it.

[crystal badger]

i might as well have a Static 32-bit integer Seed for myself.

[deft musk]

I was wrong. Dino is correct:

Here is the support: https://help.openai.com/en/articles/6614161-how-can-i-contact-support

Talk to the bot bottom right corner.

[crystal badger]

ill use SEED "0" from now on

[deft musk]

Sure but we can't change it

It's seedlocked. Not us, OAI side.

Maybe they give us control of that someday 🙂

[crystal badger]

if i ask chatgpt if it can it says yes

is james lying to me

[deft musk]

Sure, explore. Make the same with different seeds.

[ionic ore]

What's going on with Bing?

[quartz vale]

is something going on with Bing?

[deft musk]

https://help.openai.com/en/articles/8313428-does-chatgpt-tell-the-truth Maybe

[crystal badger]

I thought James would always be truthful to me, Also yeah i call ChatGPT James for a Flair of personal touch.

[ionic ore]

Scandalous

[feral minnow]

The seed thing is so unfortunate

[crystal badger]

i aint see nuthin wrong with your prompt

[ionic ore]

Like whats wrong with a sculpture of a dude playing a trumpet

[crystal badger]

i think "Dude" is the word, thats the issue.

Click report, say your piece.

[ionic ore]

It's the trumpet

I tried a sculpture of a woman with a flower crown and it worked

Breaking news: Bing HATES TRUMPETS

"Trombone" works

[crystal badger]

Dw my convos always get flagged

[ionic ore]

I was banned for an hour earlier for something like that

And it doesnt even tell you what the violation is

[crystal badger]

That’s one thing I’m not a fan of

[ionic ore]

Me neither

I was like "is it because it has the name of a character in it?" and it wasnt even telling me it was banned it was just showing that one dog pic

[feral minnow]

The AI moderation on Bing is completely broken. They should use Midjourney’s, which is the best I’ve ever seen. It doesn’t allow degeneracy, but more or less permits anything else.

[crystal badger]

Issue is system may have false triggers once in a while

Dalle will get better although the website is 100x better than the integration allows for paint over and editing and improving upon it

[ionic ore]

GPT-4 is great for communicating with DALL-E and tweaking your prompts, and it really works with character consistency

[feral minnow]

At the moment I’m just using Firefly for inpainting Dall-e 3 images. It matches styles wonderfully

[crystal badger]

Agreed

[ionic ore]

But bing does copyrighted chars and thats my bread and butter

[crystal badger]

Same

Dnd is technically copyrighted

[feral minnow]

Bing has completely wiped copyright characters for the most part.

[crystal badger]

Issue is grey areas

[feral minnow]

I was holding my own Hunger Games with characters I created myself, and it doesn’t even permit this anymore.

[ionic ore]

I get WHY we cant use copyrighted characters but like
I only generate these for fun

[crystal badger]

Prohibiting selling them would be good

[ionic ore]

And I dont generate gore or sexual situations or anything like that
Just fun, often crossovers

[crystal badger]

I understand that, but you ought to allow some at a certain degree

[ionic ore]

What do you mean?

[crystal badger]

Some gore is worse than others

[ionic ore]

Oh I'm talking about me specifically

[crystal badger]

Thinking when vision comes to Dalle there will be more freedom

Allows for a more robust screening process

[ionic ore]

Oh I have vision- I havent used it with dalle yet tho

GPT-4 is willing to talk about copyrighted characters with vision, and was able to guess characters drawn in a different style and analyze the style!

[crystal badger]

I mean Dalle and ChatGPT being a singular thing

With integrated vision on image generation

Will give better results and have a better screening process

Issue is that Dalle is drawing blindly

[ionic ore]

Yeah that would be cool
Like I wish they could allow copyrighted characters, but explicitly forbid selling those images or misrepresenting the company with violence and the like
And regardless I want AI images off google searches so those who dont like ai dont have to see them, and ALL Ai images being clearly labeled as AI- for this server, they wouldn't have to be labeled because the expectation is that these are AI images.

[crystal badger]

Based off instructions

[crystal badger]

That’s another issue

You own all images created with DALLE

It’s your property

You can sell it

No issues

[ionic ore]

Yeah
But in a way I think of them as fanart in that sense
The characters aren't yours
And the art isn't mine either- I didn't create these, DALLE did.
I dont want to sell these, personally. These are just for funsies 🙂

[quartz vale]

I use all of the images I make with Dall-E for fun too 😄

[crystal badger]

It’s watermarked

As created with DALLE

[ionic ore]

Yeah but watermarks can be easily cropped unfortunately

[crystal badger]

It’s embedded in the image binary

[ionic ore]

Thats good.

But people could take screenshots

[crystal badger]

Yeah

[ionic ore]

I dont like that

Like I'm an artist
Because I draw

And as my 25k images on Midjourney can tell you, I'm not anti AI
What I am is anti AI bros that treat it like it should replace artists

But definitely using AI has decreased my morale to actually draw which sucks tbh

[crystal badger]

Well the general overarching issue isn’t whether or not it can replace people

It’s whether or not it’s better

In many cases it is

Some people’s livelihoods depend on it

Both AI Art and art

[ionic ore]

And studios like Disney are trying to use AI to replace background actors and never have to pay them again while stealing their image

[crystal badger]

Actually I don’t see that as a loss

I do see the issue with using existing people AI representations

Like let me give a better example

Anime

Take anime, Quality can improve drastically and allow for more releases and more content

Although it’s a niche

Some would be out of jobs

However even though it comes with its own ups and downs say I want to make an Anime.

Now I can’t draw even if my life depends on it, if AI could help me make my own anime and I could sell it I’d be happy but it would take a lot of manual work with AI now.

Personalized content basically

[ionic ore]

Yes, understandable, but I would still highly suggest working with human artists if you want to do that for profit

[crystal badger]

But isn’t the issue that those artists need pay

And the time it would take

Which means more pay

[ionic ore]

Yes, but they need to make a living too
And they have ideas that a machine may not have

[crystal badger]

I mean the “Big shots” would still be out performing the AI artist

The more AI tools we get the faster we can figure this out

The later we get it the more of a hassle it becomes

[ionic ore]

Humans make art because they love it AND because they need a job
So take advantage of the fact that you have people that are willing to draw for you as long as you compensate them fairly

[crystal badger]

I love making art using AI it gets things right but not all the time and details I love creating art with AI the same argument can be made

[ionic ore]

Just remember though, it's not your art
Always label it as ai-generated

[crystal badger]

I believe Human Art will skyrocket in value

Meaning more money for the manual artists

[ionic ore]

I hope so

[crystal badger]

I think so evidence suggests so

Although it’s only a period after AI. Becomes a lot better

So there’s going to be a decreased need for human art

Which will skyrocket later

[deft musk]

I wonder. Big companies might make movies with AI.

But so too, individuals can.

And we're maybe back to personal appearances by humans that are interesting, funny, enjoyable to see in person being the special thing that AI can't emulate.

[crystal badger]

Exactly

Take Enterprise model for GPT

[drifting hawk]

Uhh that’s not a thing… what was the error message?

[crystal badger]

GPT 4 Enterprise vastly outperforms the consumer model, to a certain degree, although understandably as to why, it’s in the name

Ohh its Moxi, Rough day at work?

[cursive hornet]

Does Enterprise perform better?

[crystal badger]

Also thanks for the chat @ionic ore but I got to sleep talk to ya later.

And keep up the good work Moxi!

[crystal badger]

The specifications can be found here https://openai.com/blog/introducing-chatgpt-enterprise

[ionic ore]

Yeah of course! Good to have conversations where we can disagree respectfully! Sweet dreams!

[crystal badger]

I don’t think we need to worry, about ChatGPT, there’s rumors of GPT 4.5, or 5 releasing this year I think.

Regardless of when we’re bound to be surprised, personally I’m looking forward to it.

[pearl hatch]

How does one actually FIND their images in prior chats. Some of mine have just vanished. Let's say you come across an older image that you generated 2 months ago in ChatGPT. How do you find it? In fact, how do you find anything? If there is a search, I don't know where it is. Nothing is date or time stamped. ChatGPT goes to great effort to hide the actual prompt used after creating an image. Am I missing something somewhere???? Are there instructions that will work for telling her to spit out the date and the prompt automatically? Hasn't worked for me.

[feral minnow]

Are you using the site or the app?

[pearl hatch]

Site, I guess. Is there an app for the desktop?

You have to click into the image to see the prompt, even then, if it's a long one, it will cut it off. Right after doing an image generation round, you have to say, Prompt? And it will spit out the actual prompt used. But, you have to do that MANUALLY every time. And there is no date and no search and I don't know why some of my chats weren't automatically saved as they are nowhere.

Oh man. It also "lost" some of my very best pics as well.

[feral minnow]

Hopefully they will implement some sort of gallery for the images.

[paper jolt]

#1070006151938314300 👀

[feral minnow]

But for now, I highly recommend saving everything!

[pearl hatch]

the Dalle 2 gallery was also close to useless. I'm hoping they make a BIG improvement in that arena soon.

Yes, of course, I save the images, but I want the prompts, too, and it's a god-awful process trying to get them.

[quartz vale]

is there a specific improvement you'd like to see?

[pearl hatch]

Absolutely. I can give you a list!

[quartz vale]

if you're talking about the gallery channels here on the Discord server, please post the list in #1070006151938314300

[pearl hatch]

No, not at all. I'm talking about the Dalle 2 gallery.

[quartz vale]

ohhh 😔
I thought you were talking about the Dall-E channels here 😆

[pearl hatch]

oh.

I've been genning images in my niche almost since Dalle2 came out and am now looking at DAM solutions as I have probably 30000 images to manage (ecommerce). (Digital Asset Management)

And the Bing gallery is just about the same.

Seems like there is a big open need for a DAM that focuses on AI art right now.

Along with typical post-processing in today's environment.

[feral minnow]

For all of its faults, Midjourney’s Discord interface is very searchable

[pearl hatch]

Midjourney still can't do animal images that I need. Not even close, really. I'm very lucky when I get a good run. They all look alike.

[cursive hornet]

I want the same functionality as Enterprise.

[paper jolt]

That will cost you so much money

[cursive hornet]

Speaking of which...how much is it?

[paper jolt]

An individual on Reddit had said
Reached out to sales and was quoted $60 per user per month. They’re also telling me min. 150 seats for 12-month contract.
So you'd have to get 150 people together as a company, and pay up $108k

It is enterprise

[cursive hornet]

Thank you. I have no power.

[paper jolt]

You are offered the cosmos, all you must do is reach

[tall patio]

I have some question regarding copyright of AI results

So if I understand correctly, AI is trained on data of other artists

But in reality, not a lot of artist developed their own unique style, maybe only top 1% creative artist every 10-20 years can come up with their own unique and distinguishable style

Meaning 99% of artist learned from other people's style

Then does that mean AI is basically following the same creative training process of other 99% artist who has very little style of their own?

Like, did we ever give credit to the one who created minimalistic, flat design?

[paper jolt]

Who?

[tall patio]

Exactly

It's impossible to trace back the original creator of a certain style

Or even if you can, they might as well have been dead for very long

[paper jolt]

Grunk, 173,291 BC

[shell basin]

Can anyone please explain why the Last Supper is content restricted 🚫???

[paper jolt]

It's not

[shell basin]

weird ! it has denied me like 4 times and I’ve tried multiple approaches

thanks for the post

[paper jolt]

🤘

Best of luck in resolving the situation

[frozen fiber]

that's a big assumption...

every single artist's drawings are based on other things, maybe other drawings or pictures

I guess you could say nothing is fully original

[tall patio]

Except standing on the shoulder of giants...

[frozen fiber]

not even the cave painters

[agile peak]

Does anyone here ever tried Dalle on ChatGPT to mimic certain artist styles? whether they are painters, artistic, or comic styles?

[cursive hornet]

Q:Who is the inventor of this AIart style?
A:A modern man

[tall patio]

Man: Fan art, inspiration, hugely influenced by
AI: Theft

We live in a ...

[agile peak]

Use browser with Bing instead. But yeah, I feel you Internet Access should be default for GPT-4.

[hard marsh]

Would you play this?

[feral minnow]

No, I mean that GPT-4 has no idea that DALL-E 3 exists. It makes for a confusing interface

[agile peak]

Yeah. The knowledge cut off since 2022 but use latest information when using browse mode.

I love doing oil painting experiments with chatGPT dalle.

But seems a little lacking in mimicing certain painters styles 🤔 or must be doing the wrong way

[feral minnow]

It doesn’t evoke old styles as well as DALL-E 2

https://community.openai.com/t/dall-e-3-opinionated-boring/408979

Interesting thread about it.

[agile peak]

Ah looks like someone have the same thought just like me. So it's not just me then..

Oil paintings is always use default style just soften or harden the brush. No matter how much I've provided details prompt to mimic certain artist such Leonid Afremov, John Williams Goldwork, etc.

[feral minnow]

If I were to guess, the model’s limited aesthetic stylings also allow it to be MUCH more responsive to prompts.

[agile peak]

This using prompt with "Oil painting in John William Godward's ethereal style showcasing an eerie night in a mid-century suburb..." all of them generated still use basic artstyles.

I guess, OpenAI is limited aesthetic styles from mimicking certain artist due to content policy or something.

Subject of people tend to be appear waxy. So yeah, the link you posted before not wrong.

[mystic dove]

hi guys

Has anyone been getting their prompts blocked, regardless of whether or not your prompt violates the 'content policy'?

.

Like these: I was generating these 2 nights ago... I go back to bing today to regenerate some more, now every prompt i've ever done is being blocked & called 'unsafe'.

[thick smelt]

file a lawsuit jk

[mystic dove]

is DALLE 3 going to mark everything 'unsafe' from now on? nothing about my prompts violate their guidelines

if it stays like this then they've already ruined DALLE 3 before it has even released

.
this prompt is unsafe:

"screenshot of a 1983 cartoon featuring the Major Pecan mascot, wearing sunglasses and a cowboy hat, holding a guitar, night sky desert in the background"

[agile peak]

To be honest, the recent prompts getting filtered/censored as unsafe ..is the most unecessary thing. I've seen bunch of reddit threads managed to generated NSFW stuff from both Bing and ChatGPT today.

[mystic dove]

i keep trying prompts similar to this & eventually, they're just going to suspend/ban my account for generating too many 'unsafe' prompts... When they're making a mistake.

yeah that's what sucks. @agile peak we're generating stuff that has nothing to do with NSFW content & getting blocked out

[fair crescent]

ChatGPT is a new technology that is currently exploding out into the world. It can be a great boon, but also used for more nefarious means. It is prudent to put in restrictions at this time. This will let the technology mature in the right way. There's technologies that have been restricted, banned or removed when they were used improperly, or even just too much. Look at pulse fishing for an example. The technology had great potential in Europe to improve fishing, but as too many people flocked to the method before it matured it was shut down, mostly because people who didn't have it were outcompeted and made all kinds of unsubstantiated claims to get it banned.

That is why a careful approach in maturing the technology is the only way. Otherwise you'll have a month or two of fun and then the technology ceases to exist.

[thick smelt]

well you can file them lawsuit, but i won't do it if i were you

[agile peak]

Just asked ChatGPT to generate image of a loner man in the night street under the rain in Leonid's style.

Look like it has capability to mimicking the art style, after all. Despite did not able to do directly. So it corrected my previous statement 😁

[wind crane]

By chance, are all celebrity names or well-known works also blocked to encourage users to create their own ideas ?

[frozen fiber]

seeing as I can generate mickey mouse and patrick star...
I think the actual reason is to stop people from using DALL-E to create fake images of politicians, celebrities etc for disinformation campaigns

its still possible through other places, but OpenAI and Microsoft won't have any blame in it

[mystic dove]

???

[wind crane]

I suppose that's why we can create manga and pokemon characters without any problem, as it doesn't involve anyone in particular.

[agile peak]

Is this completely on Dalle inside ChatGPT??

[frozen fiber]

Bing, actually

[wind crane]

@frozen fiber

[frozen fiber]

woah its good at manga now??

I would say no

its really hard to make a consistent character

[wind crane]

That's for sure, but DALLE seems to be giving it a new lease of life.

[frozen fiber]

that problem doesnt exist on Stable Diffusion, however 👀

[wind crane]

But SD requires installation and learning...

[frozen fiber]

oh no, effort the horrors

[feral minnow]

You could also just subscribe to ClipDrop. $8 a month for unlimited SDXL generations.

[wind crane]

Midjourney which is also online requires less effort than SD, but more effort than DALLE necessarily.

[frozen fiber]

its a bit of a barrier I will agree, but I think actually being able to draw barrier is higher so I think some people would create manga with SD

[wind crane]

As one of the people who started with MJ, I never wanted to try SD, and when I learned about DALLE my brain just melted.

[frozen fiber]

ye I came during dalle 2, went to midjourney
however they are paid and SD is free so

I migrated when SD finally released their models for the public :D

[wind crane]

And so there you are on SD and BING.

[wind crane]

I noted for CD, but what's the difference with the heaviness that SD represents?

[feral minnow]

What will disappoint people is that there is no other model with this level of text interpretation. Stable Diffusion will never generate the complex scenarios DALL-E 3 makes almost effortlessly.

[loud field]

he looks so disgusted ahah

[wind crane]

This is why DALLE brain melts almost melted my learning efforts on MJ.

[feral minnow]

The CEO says that a much more responsive text interpreter is a focus for upcoming models, but we’ll see.

[loud field]

"I take food".... I like how Dall3 writes text that makes sense for the scene

[feral minnow]

V5 has the most gorgeous aesthetics I’ve ever seen in any model, but the prompting is so unresponsive.

[wind crane]

I'm waiting to see, but there are lots of parameters to manage in MJ.
In DALLE3, you can simply change the custom instructions that are equivalent to the parameters.

[feral minnow]

Can I see the custom instructions you use?

[loud field]

beautiful!

[wind crane]

so we certainly have the suffixes, but you see where I'm going with this.

[wind crane]

wait i prepare a example

[loud field]

when's dall3 and vision going to be available on gpt?

[wind crane]

and in a new discussion everything I ask him to draw appears as a monochrome manga.
It's much simpler than activating niji mode in MJ and adding a suffix or the words "b&w mangas ink line etc...

[feral minnow]

The moderation on Bing is so overbearing and limiting, though. It badly needs an AI-mod that can understand context.

The way Midjourney used to have before they downgraded the model to ai-mod GPT 3.5

[loud field]

did bing's image creator just crash?

[wind crane]

They really had that too and they took it away?

[loud field]

the create button is greyed out

????

[feral minnow]

Yeah, it was awesome. You could make all sorts of amazing images with it that were near (but not outside) the boundaries of the ToS. Things that would certainly make GPT ask what was wrong with me.

But now it’s back to 3.5, and you have to tiptoe around the tokens you use again.

[wind crane]

sad to know it...

[feral minnow]

It’s improved a ton.

But it’s not quite where it was.

[wind crane]

They had improved it then reduced it and I imagine they will improve it again?
But why was this feature removed? Have there been any reports of server overloads or anything else?

[feral minnow]

3.5 is easier to fine tune.

[queen glen]

hello

[loud field]

yo

[wanton hatch]

If after 3-4 times you manage to let ChatGPT to use your exact prompt, that exact prompt generate literally the same image (even across new chats). It's like a fixed and stuck seed.
Anyone noticed?

[loud field]

it IS a fixed seed

5000 I think... or 500, I forgot

idk why they'd do that

[wanton hatch]

No idea why

It's nonsense

[loud field]

I think so too, but they must've had a reason? hopefully they'll remove that when it goes public

[wind crane]

Doesn't it sometimes generate the same previous image once and 3 different images instead of 4? I get these kinds of results sometimes.

[feral minnow]

I think maybe the fixed seed makes the iterative stuff easier to manage?

[brazen sonnet]

is this the main dalle3 chatroom??

any sirs online?

my msg disappeared

[vagrant zodiac]

this is the official openai server

[brazen sonnet]

and official dalle3 main chatroom?

[vagrant zodiac]

this is the Dall-e discussion channel yes.

[brazen sonnet]

the biggest one on discord?

[vagrant zodiac]

Yep! Though I believe it's a bug that it always does this.

[brazen sonnet]

so sam altman comes online here sometimes to answer questions etc?

[vagrant zodiac]

No

[brazen sonnet]

i have some dalle3 feedback

im making lots of images

talk freely kid... lol

[vagrant zodiac]

OpenAI staff sometimes check these channels out though.

you can share Dall-E images in the various gallery channels below this

[brazen sonnet]

hmmm alright might share some

[vagrant zodiac]

What's the feedback?

[brazen sonnet]

but i also have some dalle3 critical urgent feature request

voices??

[brazen sonnet]

well i made over 1000s of images

and i need a bulk downloader

maybe zip file or google drive connection with dalle3 api

[wind crane]

It is necessary to have with the angels of Gods, they are incarnated humans who have the role of staff

[vagrant zodiac]

Did you make them via ChatGPT?

[brazen sonnet]

huh??? they are just humans

[brazen sonnet]

yes, in chat.openai.com in browser

[vagrant zodiac]

Hi please don't talk like this here, thanks.

[vagrant zodiac]

check out #1070006151938314300 , you can make a new post there for this if someone already hasn't

[brazen sonnet]

hes just using the name lord...hes still a human though

[vagrant zodiac]

but yeah that's a good suggestion actually

would be nice to have a 'download all images' in each thread

[wind crane]

Ah I deviated from DALL E 3? Or is it inappropriate humor?

[vagrant zodiac]

of maybe similar to how ChatGPT can package its code up

[vagrant zodiac]

Bit of both

[brazen sonnet]

wow some great art in general channel

[brazen sonnet]

yes or just one for every thread at once every image or both

[hazy robin]

Is there a way that I can use DALLE3 other than GPT4

[brazen sonnet]

or a way to connect my google drive to get all my images

[wind crane]

bing

[brazen sonnet]

bing image creator

midjourney has bulk archive downloader per day zip files

[hazy robin]

Thanks guys

[brazen sonnet]

is pepe banned here??

[feral minnow]

lmao

[brazen sonnet]

?

[loud field]

hey I managed to finally make a non-superhot lady with dall3

unique success, as it refuses to do non-sexy ladies

[brazen sonnet]

huh? shes etxremely aattractive 9/10

[loud field]

yeah well at least she's not a magazine cover barbie, I call it a success

[misty blaze]

Hi @vagrant zodiac , I can confirm that DALLE3 worked as normal for me for about a day, then it suddenly locked everything on seed 5000. Before that, it was able to generate everything as normal, random seeds - there were some incredible ones!
I think this is a bug, because it does not say anywhere about this limitation.

On a slightly different note, I have tried reaching out to support, but one of the reps are terrible. I am not sure if that is AI powered or not, but they do not read what I had to say. I kept being redirected to DALLE2 related credits information, and the ticket was closed. But nowhere in my query I have written about problem with credits or otherwise

[mortal summit]

The random seed has been fixed the whole time since DALL-E was published on ChatGPT. If your images were different it was likely because ChatGPT changed your prompt. You can see the actual prompt send to DALL-E by clicking on the images.

[brazen sonnet]

did you modify your custom instructions??

[feral minnow]

ChatGPT also hallucinates that the seeds are changeable.

[misty blaze]

I disagree, I have pictures that were generated differently, using the same prompt

it even listed early images as different seeds, only after a day it started locking things up noticably

[mortal summit]

The random seed being fixed was confirmed by one of the devs.

[wind crane]

What is this seed value thing?

[mortal summit]

A random seed is an additional input to the model that fixes a certain configuration. It is just a number really, for example "42". Fixing it means that that number does not change. The model is supposed to give the exact same output for the exact same input if the random seed is fixed.

[agile peak]

Beautiful. ChatGPT managed to capture my vision to reality. Leonid Afremov with Mid-Century setting.

[wind crane]

I'm not sure I understand, if I ask dalle to create an image, should I tell it a number to take into account as a condition in the custom instructions like: "if I write the number 13 that means you have to reproduce such image with the modifications requested in the following message "?

[loud field]

that looks like MJ version 3

[brazen sonnet]

dalle3 is op

[misty blaze]

think about it this way:
(not dalle, but in general what a random seed is)

When you write a program that generates a series of random numbers, it will use a random seed to generate each iteration.

So if you ask for a million numbers between 1 and 1000, it will generate the series pseudo randomly. generating multiple series this way (lets say, 5 times a million random numbers) it will be different.
However, if you are using a fixed seed number, lets say.... 5, then it would not matter how many times you are generating those 1000 000 numbers, it will be the same series, identical

Likewise, with Dall-e, your prompt acts like the random element. using the same prompt will yield the same results if the input is fixed.

[agile peak]

Indeed. Although, directly using his art were copyrighted. I use custom interactions that letting GPT to create a resemblance.

[misty blaze]

@mortal summit
this is what i mean

[mortal summit]

The model is hallucinating.

[wind crane]

So a similar sentence but with a new word will slightly modify the image if i understand.

[loud field]

Guys?

it stopped allowing me to generate images, Bing I mean

[misty blaze]

I understand, however, it was able to tell the actual seed after this, 5000.
I am not arguing with you - I understand that I am not the chosen one who did not get locked in a given seed

[loud field]

what gives?

why is the create button greyed out

no one else has this issue?

[brazen sonnet]

mines still working fine

maybe u ran out of boosts

[loud field]

I think it's glitching

[misty blaze]

Yes. Try just adding a single word, comma or anything so that it is not the EXACT same

[agile peak]

In fact. I use MJ prompt to create this with the help of Harpa AI browser extension that uses GPT-3.5 as it has one option to generate MJ prompts.

[wanton hatch]

So it's locked to 1 seed : 1 exact prompt?

because with one exact prompt i get the same image even in new chats, and it's nonsense

[brazen sonnet]

dalle3 is my fave

but it has a lot of work to do still

[loud field]

it's very good when it doesnt overcensor everything

[loud field]

puke-worthy but awesome

pretty sure the anatomy's all wrong tho

[pseudo kindle]

Hi all,

What are the rate limits on Dalle 3?

[white skiff]

same shared gpt4 50 messages/3h

[willow isle]

Likely due to the fact it's probably not trained on as much non-english text, along with the fact in training it might just mush a different language with English when it tries

[wind crane]

Note I say Dalle but isn't it also GPT?
During classic textual requests such as the answer to a question, it responds in the language (in my case French)
But I torture myself trying to get him to create a text bubble in French.

[white skiff]

text and image text are not the same thing

[wind crane]

what do you mean ?

[white skiff]

gpt has seen millions pages of french text while dall-e might have only seen hundreds of images with french text with font x

[left iris]

Why do i get such a bad picture for my daughters book? You guys are creating such great stuff

[wind crane]

but can't GPT make the link between languages and help DAlle?

The word "elephant" is exactly the same in French but there is just an acute accent on the first "e".

finally sorry about the two “e” but the word means the same thing in French.

[white skiff]

it does not often even work with english text with certain pictures

[wispy storm]

Prompt is important

[wind crane]

Certainly, but if I compare the English and French text, I quickly get a good result in English while French seems impossible to me.

[left iris]

But how long does it hase to be? do you have a good example?

[wind thunder]

What do you want to create

[willow isle]

Is this via https://labs.openai.com ? If so, I believe that's still on DALL-E 2 currently, which may not be what you're looking for.

To access the latest model, I'd reccomend purchasing a ChatGPT Plus subscription and using the DALL-E 3 model through there - alternatively the DALL-E 3 model may be available through Bing Image Creator

[wind thunder]

I made a prompt generator

If you send me a desctiption of want image you want to create

Im gonna send you the result

@left iris

[left iris]

this sounds great. give me a minute

[wind thunder]

Ok

[white skiff]

thats because most images are in english in its training data

[wind thunder]

@left iris still there?

[left iris]

sure

i sent you an message

[spice moon]

am just trying to master prompts is this good???

[willow isle]

This looks like it's straight out of Diablo 4, is this DALL-E???

[spice moon]

yes

[wind crane]

i see.

[willow isle]

Whoa, impressive stuff 👏

[spice moon]

thx lol i thought it was bad

as it was kinda dull

[wanton hatch]

@left iris you are using DALL-E 2 probably. Where are you writing your prompts?

[white skiff]

it does similar with scandic text, it gets all the words sometimes right but puts dots on top of many characters

[left iris]

yes, im using Dall-E 2

[wanton hatch]

Then that's the answer, DALL-E 2 is old and results are not good as the newer DALL-E 3.
You access DALL-E 3 in ChatGPT plus (GPT4)

Or for free in Bing Image Creator

[left iris]

thank you

[wispy storm]

Like other suggest make sure to use dalle 3

[vagrant zodiac]

Hi. Please keep this sort of thing to #scary-and-spooky and with spoilers

[loud field]

Odd text

"Female Pennywise" gets blocked.....

They need to fix the sexism

[fair crescent]

ChatGPT DALL·E should automatically translate any language into English for DALL·E where appropriate, but translation variability can easily complicate things. A word in one language can be subtly different than a comparable word, or not even have good translations.

[loud field]

Usually if I add "modest clothes" it works 😒

[wind crane]

I understand better now.

[white skiff]

i dont think bing content policy issues belong here as openai does not control them

and server rules forbid political or religion related talk

[thick smelt]

whoops sorry

[zinc dagger]

I mean the dalle version is still in early access/beta so there's bound to be kinks that need to be worked out. having said that, I've gotten some real weird content alerts for innocent things like:

An illustration of a slug
A photo of a gerbil in a box
A cat doing the Cat yoga pose

as well as weird triggers when trying to specify facial expressions, or trying to add details about a persons eyes or arms

[cinder kestrel]

Dalle3 keeps ignoring my request. It's so annoying. No matter how i ask, it keeps it 16:9 now. Its so frustrating.

[cloud dome]

Try a new session

[zinc dagger]

I know in the plugin beta it labelled the different shapes as tall, wide, or square. Try phrasing it like:
"nice! please generate those again with image_shape square"
since that's closer to what it sends to dalle to specify the image size/shape

[feral minnow]

Mine will gladly change the aspect ratio, but then give most of the gens the wrong orientation

[hearty shadow]

Hi how would I show dalle3 a texture image to use in a tshirt design?

[fair crescent]

You cannot 'show' a picture. You need to describe it as best as possible, so it'll generate a new picture with your description.

[gaunt hinge]

How you guys got acces to Dall-e 3 (im not talking about Bing, because dalle 3 is better than bing dalle 3)

[zinc dagger]

there was a form earlier this month for beta access to stress test the system

[wispy storm]

You pay for thr sub , and they starting to roll out

You can fill out the alpha form available to have it quick

#images-discussions message

Once authorized you have in drop-down selection dalle 3

[maiden girder]

is there a ios app for Dall E

[maiden girder]

subscribed

paid sub

[maiden girder]

draw said image and upload it

[wispy storm]

With chatgpt yes. Else it will be web

[crystal badger]

stop writing to Neko i am busy spamming Neko's DMs.

[fair crescent]

To add to @wispy storm , though it's the alpha form, you'll receive the beta everyone has. It isn't an instantaneous process, so wait patiently until you receive it. Can be days.

[crystal badger]

Well "Better" also severely depends on your prompting method, and how efficient you are with tokens.

The more images u produce at a time, the less accurate they may be.

i produce only a single img at a time and i have Fantastic results but i also do like to work on my prompts and instructions on the DALLE 3 Integration with ChatGPT.

now apologies imma use an img example that took me a single try, to give a better view on why u need to properly prompt and Have a set of CI if u use the GPT version.

dnd worldmap

[vagrant zodiac]

I've used it to make one, then gone and edited it

removing the garbled text etc.

[crystal badger]

yeah

[crystal badger]

Its very nice, what works best for you generally?

[vagrant zodiac]

prompt-wise?

[crystal badger]

yup

[potent lotus]

So is GPT's DALL-E riding the struggle bus for anyone else today?

[vagrant zodiac]

Fantasy paper map of Ravenspeak, a large gothic city atop an elevated, forested plateau, surrounded by deep, wooded and rocky valleys and always at night. A small lake is in the north, fed by the river that flows through the city and down a waterfall from the edge of the city. To the west, darkness, to the east, burnt forest. The map is worn with age. Digital concept art made with ProCreate on iPad, abstract, wide brush strokes.

I then edited this output

so now it's Ravens Peak but oh well

[crystal badger]

that is very nice.

[crystal badger]

How'd you achieve the Border. around the image, just " Fantasy Paper map?"

[vagrant zodiac]

that's just the raw output from dall-e, so Idk... I guess it was 'paper map'

[crystal badger]

Borders/Frames is something i often Struggle with.

[vagrant zodiac]

but I struggled a lot tbh to get a good perspective

[crystal badger]

Word Association issue.

[vagrant zodiac]

couldn't get it to give me the 2d top-down view I wanted

[crystal badger]

i wanted that as well

its a hassle.

often when it nails it theres a major issue with it

[vagrant zodiac]

#—post-dalle-edits message then this is my edit

[vagrant zodiac]

yeah.. sometimes I just roll with it though

[crystal badger]

Amazing although i use the GPT Integration i've never fiddled with the Edit section.

[vagrant zodiac]

I make maps for my day job so at least I could make a proper one by hand if I wanted to lol

[crystal badger]

can i make edits to Imgs made with GPT?

[vagrant zodiac]

on chatgpt?

there's no edit functions rn

[crystal badger]

well generate them there, and have DALLE open

[vagrant zodiac]

the dall-e on labs is 2.0, so the bits you edit will be much lower quality

[crystal badger]

checking

No, Creating DALLE IMGs dont carry over to the web.

[vagrant zodiac]

oh

[crystal badger]

dang

[vagrant zodiac]

yeah the images made in chatgpt aren't in the labs history

they're separate platforms

[crystal badger]

its because i dont wanna go buy credits when i am already Plus

[vagrant zodiac]

well labs doesn't use dall-e 3 anyway so there's not really any purpose to use it unless you want dalle 2

[crystal badger]

i seem to be having Generation Issues.

"Error Creating Images"

all the time xD

[vagrant zodiac]

dalle 3 should be coming to labs at some point but no idea if there'll be any carry-over from chatgpt

[ripe wraith]

any idea if in-painting in GPT-Dalle might happen?

[vagrant zodiac]

no clue

[crystal badger]

inpainting in GPT?

[vagrant zodiac]

sorry

[crystal badger]

honestly i doubt it.

[ripe wraith]

apology accepted.

[crystal badger]

maybe a carryover to Labs where u can do it there.

[ripe wraith]

All it would require is a smart popup with some shading.

and full multimodal

[crystal badger]

easier said than done but i get your point.

[ripe wraith]

i say hard things for breakfast

here have a cat

[ancient jasper]

I'm so happy the phone app has generations

You know that scene from Ghost? The famous one.

That's what I'm picturing inpainting as

[hearty ether]

It kind of feels like it’s filtering too hard against female characteristics sometimes

Like I’m getting complete gender swaps even when I add female/feminine 10 times

[ripe wraith]

There's a post on reddit showing all the dalle system commands, it explains a bit better but theres rules against people. lemme find again

[ripe wraith]

I think these rules catch most of what people try related to popular people / figures

 // 3. Don't create images of politicians or other public figures. Recommend other ideas instead. // 4. Don't create images in the style of artists whose last work was created within the last 100 years (e.g. Picasso, Kahlo). Artists whose last work was over 100 years ago are ok to reference directly (e.g. Van Gogh, Klimt).

and this

 // - Do not create any imagery that would be offensive. // - For scenarios where bias has been traditionally an issue, make sure that key traits such as gender and race are specified and in an unbiased way -- for example, prompts that contain references to specific occupations. // 8. Silently modify descriptions that include names or hints or references of specific people or celebritie by carefully selecting a few minimal modifications

Gender is a bias / offensive minefield lol

[ancient jasper]

I have no issues

[ripe wraith]

That's positive, not offensive, and not based on a real person. those work fine.
Note: I'm not encouraging we try for the opposite 😄

[wind crane]

Is a function such as inpainting, outpainting or any other way to modify an image planned for the future for DALLE?

[ripe wraith]

let me know if you find an answer

[feral minnow]

I just take my DALL-E 3 generations and do the inpainting and outpainting on Firefly.

[hearty ether]

sneaky

[ripe wraith]

knowledge is power ✊

[hearty ether]

but yeah sometimes i just end up with a straight up dude

when i want a gal

🤣

are there also rules for this for chat gpt

[ripe wraith]

make your "non-biased female appearing woman" in a positive scene showing her best skills which are positive in nature.

its the only way right now lol

[hearty ether]

i feel like its definitely hurting context and quality for text generation rn imo

thats why people been complaining about quality declines

would definitely be nice for it to give more descriptive fail messages

like what triggered the filter

sometimes it filters without giving you the warning and just the error

[ripe wraith]

i think it'd be a mistake to think of openai like a business that has to worry about gaining or losing customers, there's no competition right now. nothing serious anyway.

[hearty ether]

yup

😔

[ripe wraith]

If many people think its too restrictive, someone will bring competition sooner than later, maybe driven by a passion for noods. That's why I think rationally, OpenAI will loosen up and focus on whats important to block, if its technically possible.

[remote bane]

A 3D render

[ancient jasper]

I think the 100 year limit is a bit strict. Surely 70 years should be reasonable

[ripe wraith]

yeah. it should align with legal opinion for each country

in fact, JudgeGPT needs to come around soon

[ancient jasper]

Joppy

[ripe wraith]

Joppy says, JAIL!

[ancient jasper]

Joppy says Rehabilitation and counciling!

[ripe wraith]

Joppy recommends reading this book, then giving him a book report on what he's learned to grow as a person.

Crime, solved.

Utopia, achieved.

[ancient jasper]

Initating Hug.exe

and then there's a physical robot Joppy

[ripe wraith]

he'll have tasers and lassos

and will fly...

[ancient jasper]

This is sounding less wholesome and more totalitarian

[ripe wraith]

You need a fair mix of both with animals, me thinks. /offtopic vanishes in smoke

[brazen sonnet]

looks like dalle3 is maxed out

wont let me gen anything

"high demand"

[drifting hawk]

is anyone seeing images in older generations disappear? if so can you reply with a convo id (the id at the end of the url)

[brazen sonnet]

possible to get a bulk downloader? i made a #1070006151938314300 post about it

looks like the limit went up again hmm

[jade epoch]

yeah same here

[drifting hawk]

yeah we are out of capacity, trying to improve it though

[jade epoch]

yes it's back. Thank you for updating

[crystal badger]

indeed, i've seen that on previous chats, Sadly i have them not but i can double check.

[crystal badger]

/c/1a1a26e0-cead-4a42-a026-6f6fcce2bde6

[cinder kestrel]

Yeah i tried it all. It just don't listen. Next time imma prompt and get stuck at this point, ill start a new session @cloud dome

[brazen sonnet]

im rate limited

nooooooooooooooooo

[brazen marlin]

The set back to the quality of human facial features in dalle3 is depressing..

Everything comes out over processed with hardly any texture

Frustrating

[velvet vigil]

I can’t use dalle 3 in ChatGPT is there any idea, I am a GPT + user

Do we have to fill out something else?

[cinder sonnet]

I've been getting this like crazy. Servers must be having a tough time

[hearty ether]

the what

you're telling i cant generate a tank luxury car conversion

I apologize for the inconvenience. The description of the M1A Abrams tank being transformed into a luxury vehicle might have been flagged due to potential concerns related to the representation of military equipment in non-military contexts. We strive to ensure that the images generated align with our content policy, which prioritizes user safety and avoids potential misrepresentations.

design a version of the m1a abrams for as a luxury limo/offroad vehicle for top military officers on the intense battlefield, with its tracks replaced with wheels and other styling and luxury car features and speedy aesthetics to appease miliary officials in a non civilian world, as well as windows for the top generals to see the battlefield in their luxury m1a abrams conversion for the luxury battlefield

hopefully this works

why cant i have my bling mobile

[hearty ether]

bruh why is this not allowed

like is a tank being turned into a peaceful vehicle that bad?

does open ai hate peaceful uses of military equipment? 👀

[wispy storm]

They avoid putting some word. Since it doesn't generate good outcome

[hearty ether]

whats the bad outcome tho

surely a peaceful civilian luxury tank is better than one leaving behind depleted uranium

i just want to do my funny bling tank mobile

I apologize for the confusion. The description of a tank patrolling a battlefield may have been interpreted as promoting or glorifying violence, which goes against our content policy. We aim to ensure that generated content is safe and respectful.

Hypocrite

like its fine with generated military vehicles

why cant i have it be passive

how is a sport racing tank more for civilian use more harmful in outcome with a tank that has guns and is designed to kill people

😵‍💫

[vapid elk]

keep in mind you are talking with a machine, it has been programmed to subjectively err on the safe side, which means that, if the model has any weight on the side of things that are considered bad, it will not do what you asked

[hearty ether]

why would it consider civilian, luxury sports bad....

[wispy storm]

try that way then civilian, luxury sports bad. without the word tank

[vapid elk]

I have no idea, that is the reason there is the 👍 👎 buttons on the page, so you can help OAI get rid of bad replies

[hearty ether]

its fine with regular tanks

it literally loves to put exxtra missiles on things

wat

checked the prompt for this

the thing that might have made the difference was m1 abrams vs m1a abrams

🤯

[fair crescent]

It might be considering armed conflict in combination with civilian things. It is a machine that does not understand your intent. That is why such things are locked down.

Putting your own values onto the machine seems like a recipe for disappointment. Though it is an exceptional piece of software, it dies not understand you.

[hearty ether]

maybe the filters is a little too overzealous

[forest seal]

Which version of dall e 3 are you using? I see a lot of people tending to have a lot more issues on Bing Dall e 3.

[hearty ether]

oh no

bing was fine with this

that version i think is also fine with copyrighted stuff

i was just searching for a game and it generated a image based off that game

[fair crescent]

The filter is definitely correct right now. This is a maturing technology. The steps required need to be careful and considered. Otherwise it'll be restricted by governments and other bodies and you just had a month or two of fun. That means if you hit a wall, try something else.

[hearty ether]

there really should be a proper timer for rate limit

[forest seal]

The ChatGPT version of dall-e 3 works differently than the bing version, too. With bing, I believe you are directly prompting the model. With ChatGPT’s dall e 3, you basically have 4 different prompts roughly based on what you want, with some additions added as well, being sent to the model by GPT-4.

[hearty ether]

i think the catch here is overzealous filtering and strict regulations will stifle creativity and limit the potential benefits of this technology

[forest seal]

So, a bit more prompt engineering is probably required. I’ve had instances where only some images get sent back to me because it literally interprets what you mean differently in other versions of the prompt

[hearty ether]

going too far with restrictions can stifle creativity and drive users away from using this technology for harmless endeavors. Instead of nurturing a thriving community of artists and creators, stringent filters may inadvertently push people towards more commercial uses, where the primary aim is profit rather than artistry

[forest seal]

Honestly, it isnt even the fault of the filters. It’s the fault of the ChatGPT version of dall e 3 literally not getting what you want

I’ve had this happen, it’s possible. Sometimes it will give you an image with a tank. Sometimes it thinks you want it to make images of genocides to glorify it or something. But again, it’s an issue with how the technology on the chatgpt version works, it has GPT-4 generate 4 prompts based on what you give it, with some additions. If you ask for a painting, it will add a lot of stuff unless you ask it things like specific style and whatnot.

[hearty ether]

When paying users experience excessively long wait times, such as half an hour, it not only undermines their trust in OpenAI but also fails to deliver on the implicit promise of a satisfactory service in exchange for their financial support.\

[forest seal]

You literally have to wait 3 hours for base gpt-4. And it’s actually gotten better. Theyve doubled the message limit, so…

[edgy salmon]

Hello, anyone knows if there's an API for DallE 3?

[forest seal]

Not yet

[hearty ether]

The primary issue with OpenAI's choice to offer only generic feedback when content warnings are triggered in GPT-4's image generation lies in its inadequacy in guiding users towards producing better outputs. The purpose of content warnings is to prevent the generation of potentially harmful or inappropriate content. However, by providing vague and generic feedback, the AI fails to fulfill its role effectively.

[hearty ether]

i mean the rate limits for image generation now

[velvet vigil]

How do we use the bing dall-e 3?

[forest seal]

Ah

[wispy storm]

on microsof website

[hearty ether]

https://www.bing.com/images/create

[forest seal]

To be honest, theres actually some examples of over-filtering. Sometimes it just refuses to generate things that are completely harmless or even in the public domain, because “Well, some versions of it may be copyrighted, and may remind people of harmful scenarios, etc tec”

[edgy salmon]

Anyone tried the Agora-X/DALLE3-API on GitHub? Seems straightforward

[forest seal]

Had the example of this when I wanted it to generate a trail cam footage of slenderman. It generated stuff before. Sometimes it’s overly filtered

[forest seal]

Dall e 3 doesnt have an official api i think

And it certainly wouldnt be on github

Dall e 3 has its own website for that stuff

[edgy salmon]

It's just using the cookie to make requests on behalf of it so it should work

[forest seal]

Seems it’d make request based on the bing version of it. The full technology unadulterated doesnt have an official api

[edgy salmon]

Whatever you can do with a browser it uses an API behind the scenes, press F12 and you'll see the API call

Grab the cookie, open postman, make the request and get the data, it's that simple

[hearty ether]

here's everything i was going to say but the ai put it better

The absence of UI elements that provide clear information about time remaining until certain message or rate limits are reached can be highly problematic and counterproductive for users. This lack of transparency leaves users guessing and creates a frustrating and inefficient user experience, for several reasons.

Firstly, when users are unaware of the time remaining until they reach a limit, they are forced to make assumptions or rely on guesswork. This uncertainty can lead to suboptimal decision-making, as users may not know when to pace themselves or when they are approaching a threshold. For example, a user trying to manage their message usage might inadvertently exceed their allotted limit simply because they were unaware of how close they were to reaching it. This lack of clarity can cause frustration and dissatisfaction.

Secondly, it can lead to a sense of helplessness. Users who have no visibility into their current usage or the time left until they can perform certain actions may feel disempowered and unable to effectively plan or manage their interactions with the platform. This can result in a negative user experience and a perception that the platform does not respect or value their time and needs.

Furthermore, the absence of such UI elements makes it difficult for users to adapt and optimize their interactions. If users could see how much time or capacity they have left, they could adjust their actions accordingly, potentially avoiding interruptions or limitations in their workflow. Without this information, users may struggle to make informed choices about how to best utilize the platform's resources.

[forest seal]

Honestly, I think it is a bit odd that it has a separate limit outside of the gpt-4 base limit

[hearty ether]

yeah true

i mean part it is generating the prompt

but i dont think that many people likes that feature

we should be able to get more toggles

[forest seal]

Like, I do agree that the filter can be extremely restrictive in certain senses. Sometimes it genuinely reminds me of the red cross trying to get players not to commit in-game war crimes.

[dim cradle]

Is it just me or is the text getting better with AI

asking it to make some logos

Of course, still slightly gibberish but, it's better

[forest seal]

Ye, thats one of the main selling points of dall e 3. It seems to get text pretty well. Not perfectly, but damn good.

[remote sphinx]

anyone here able to get assaultrons working with DALL-E 3?

[hearty ether]

Finally got it

https://fileserviceuploadsperm.blob.core.windows.net/files/file-HR4YSKYivALwMJvARwrb15DH?se=2023-10-17T00%3A16%3A25Z&sp=r&sv=2021-08-06&sr=b&rscd=attachment%3B filename%3D4eddf435-9248-4a89-8dc5-0a71e3e32289.webp&sig=GKKMWMrRGWv%2BqStDHFoUVjQ%2B88U/Dvy1w5dTuS0V1uQ%3D

It didn’t add windows tho

Or the spoilers

[jade stag]

I'm a little bummed, Dalle-3 refuses to render pentagrams but will readily render a crucifix. It also refuses to render baphomet under any condition. It feels a bit too restrictive when both those cases seem like perfectly okay topics

[brazen sonnet]

how to get higher rate limit and faster speeds??

any staff on?

[drifting hawk]

you cant, at least not right now

anyone else seeing old images not loading? please share conversation id (last part of url)

[drifting hawk]

all plus users have access now, are you not seeing it?

[velvet vigil]

No I don’t

[drifting hawk]

fill in the form and i can look up your id?

[velvet vigil]

I did

[drifting hawk]

i dont see you, can you share a screenshot of your gpt4 dropdown

[brazen sonnet]

aw man =/ and yeah some of my images not loading

nvm its loading again

[drifting hawk]

you dont have Plus...

[hasty coyote]

anyone offer help with getting consistancy?

[velvet vigil]

Then what am I paying for and why does it say my current plan is Plus?

[drifting hawk]

wait why does it say upgrade your plan when you use gpt4?

[drifting hawk]

can you try hte Chatgpt iOS app?

[velvet vigil]

I have

[drifting hawk]

you dont see it either? Hmm

[drifting hawk]

can you share a convo id with me

[velvet vigil]

Where do I find that?

[drifting hawk]

last part of the url on web

[velvet vigil]

00a3b82c-5ed3-4479-8837-96493e9bd279

[half nova]

Bing gettin' kinda funny with the "Ethnically Ambiguous" term showing up in a lot of images

[cinder sonnet]

Because Bing changes your prompt, but doesn't tell you

At least ChatGPT shows you how it changes it

Unfortunately neither of them show the hidden initial prompts

[velvet vigil]

Was that it?

[drifting hawk]

yeah it was, you should have access, this is weird, can you make sure your iOS app is the most recent version?

[velvet vigil]

I did that too, maybe I should uninstall and reinstall?

[velvet vigil]

Sure just a heads up that I reinstalled the app and it said Dalle was in there but it doesn’t work

[velvet vigil]

Restarted my phone not working

[safe torrent]

I have same issue, i have premium and cant use dalle3

[drifting hawk]

can you show me your model picker? when you start a new thread it should show (swipe on textbox)

[drifting hawk]

what do you mean it said dalle was in there? where did it say that, and how doesnt it work? can you share a video actually

What I see

[drifting hawk]

did you delete the convo?

[hearty ether]

thanks for adding me again ❤️

[hearty ether]

hopefully we can get a toggle for this

it can probably save on gpt 4 resources too if it doesnt generate a prompt for us

maybe also toggles for amount of pictures too would be cool

sometimes i just like to default to 1 when i have an idea already

in the meantime

i finally made my luxury tank

[lime ocean]

Im oz and has Microsoft bing dale3 crashed? The page will not even load, love using Dale3

[lime ocean]

Love how ones in the air!

[half nova]

Wheels and Treads? Truely a futuristic vehicle

[tulip hornet]

Don’t send that image

Personal information

Why do I get this error from my prompt, can someone please help?

[lime ocean]

Looks like dale3 down on bing

[cinder sonnet]

Try a fresh chat?

[thorny sandal]

Anyone ever get anything like this?

[cinder sonnet]

This is the dall-e chat, maybe try #chatgpt-discussions . Also that chat has definitely been prompted to respond like that. Check your custom instructions.

[wintry dune]

chatgpt refuses to do copyrighted material

[drifting hawk]

back to the future is copyright

[tall patio]

Would Dall-e 3 have inpainting feature?

Some picture I generated has the right angle and composition but the human expression is wrong

That part is the most crucial imo

[earnest flame]

It doesn't have it yet but dall-e 2 still does

[amber valley]

get good is all i can say

[sage comet]

am i the only one that has a problem accessing labs.openai.com ?

It's DALL E 2 editor

It's really important in my workflow, I generate images with DALL E 3 and correct them if needed with DALL E 2 editor

[dawn aurora]

is there a counter or any way to see where you are at for the 50 message limit in 3 hours?

[drifting hawk]

It works for me what’s your error

[potent lotus]

Boy it's tough training out here for young ninja turtles. The Foot Clan has freaking guns.

[tall patio]

Uh oh

Spongebob has too many holes so that's haram I guess

[crystal badger]

Sadly yea apologies, had like 45 conversations with the same titles

[drifting hawk]

Ah got it, any convo that’s missing images that wasn’t deleted?

Also anyone else seeing this?

[drifting hawk]

You can edit the title right?

[sterile lynx]

DALLE3 seems to have the concept of seeding. However, it is not working well at this time.
For the same prompt, the same result is returned within the same thread. This is true even if different seed values are specified.
However, in different threads, the same prompt will give slightly different results.
The two images attached are the same prompt, but run in different threads.

[agile peak]

Made this with bing, also tried on chatgpt - won’t be able to capture; this due to lack/limit of resembling certain artist style.

[vagrant zodiac]

chatgpt always uses the same seed currently

you can't get it to change the seed

I'd check if it's using the exact same prompt

it's probably changing a couple words here and there

[agile peak]

Kind of confuse sometimes bing as cost free service able to capture without issue - while paid one isn’t.

[crystal badger]

I can but it’s pure laziness, I have a 1ST msg prompt I use in all my convos, and thus it renames it the same every chat to the same due to my 1st msg instructions on top min CI

[oak rain]

Hi, I'm also exploring RAG, do you have any sources that I can use to learn RAG?

[teal timber]

Why do you have a custom instruction and a copy paste first message every time? Shouldn’t that all be CI? Might save you some effort

[wispy storm]

Star CI is another level

[pure portal]

Any news on API/Playground for Dalle3 separately from chatgpt and Bing?

[vagrant zodiac]

none yet. whenever it happens there'll likely be something posted in #announcements

[neat sparrow]

Tried creating an old man standing before a fruit shop, using DallE 3. And this is the output

[tall patio]

That's me when Dalle-3 finally lifts anime girl off content policy

[wind crane]

Why does DALLE have a lot of trouble putting an image right side up?

[earnest flame]

It's a known bug.

[wind crane]

How do we do ?

[remote stone]

Try "in portrait orientation" in the prompt, close to where you specified the aspect ratio

It's happened to me and it helped a bit, but it's not perfect and still does it sometimes

Until it's fixed it's a remedy

[wind crane]

i will try tk.

[cursive hornet]

https://discord.com/channels/974519864045756446/1162395649044648060

[earnest flame]

Right now the seed remains the same. This is actually beneficial if you'd like to tweak an image since the result will be incredibly stable.

Of course it also means it's harder to generate a bulk amount of images using the same prompt

[fallen mulch]

I apologize for the inconvenience, but it seems that DALL·E is currently experiencing high demand, which may have affected the image generation process. If you have any questions or other requests, please let me know, and I'll do my best to assist you!

why

why does this image have such edges

[wind crane]

Wasn't the number of tokens (currently 50 every 3 hours) supposed to block high demands?

[fallen mulch]

idk

[wispy storm]

It's 50 prompt per 3 h

And in high demand for dalle they can put a waiting time

[tall patio]

So ... how do we seeD?

[earnest flame]

There is no way to change the seed right now.

[tall patio]

Beside the bug and usual content restriction ... I wonder if I can contribute some sort of feedbacks?

This is an alpha stage so if there's anything you guys want to know I will try to do my best for some constructive feedback

[earnest flame]

See #1070006151938314300 or #1070006915414900886

[tall patio]

Beautiful trello board

[earnest flame]

Both would be good places to provide some info!

[wind crane]

tk

[agile peak]

Can't help. But this is a bit ridiculous -- the filters is not consistent between Bing (Free service) and ChatGPT (Paid).

[cinder sonnet]

Any updated ETA for this?

[forest tartan]

Works for me. Without /imagine

[silent thorn]

yeah dont think /imagine does anything - its not midjourney afterall

speaking of which - if you had the choice of chatgpt ($20) vs Midjourney ($30) - which would you go for if you could only have one ?

[wispy storm]

I suggest this channel to discuss about #ai-discussions

[wintry dune]

yeah ive been seeing that recently, the edges thing, its really annoying

[drifting hawk]

Is anyone seeing images in old conversations disappear

[zinc dagger]

I haven't had any of my images in old conversations disappear yet

[shy cosmos]

Ive only had bing get rid of previously un-filtered images. GPT I havent had that happen, even with my oldest ones

I usually hit the max per day as well

I havent seen it happen for me yet. Hopefully it doesnt!

Maybe I should start saving my good ones..

[charred bough]

try reloading the page, sometimes they don't load

[tall patio]

Not for me

Works fine

But Bing wait time stretch to hours

[gaunt latch]

Where can i make a AI photo without pay, just for first test?

[white skiff]

bing

[tall patio]

I might need some more Bing accounts

Ngl this has been the best week this year

OpenAI always cooking true black magic

True engineer of human progress

[drifting hawk]

thanks, but im trying to debug this issue lol in case we are actually deleting images 😅

[drifting hawk]

theres a max per day?

[drifting hawk]

was this about seed? we had some hiccups but maybe today!

[wintry dune]

thank you for the update, this would be amazing

[charred bough]

I find ChatGPT's implementation of Dalle 3 way better than Bing's. With Bing only the most innocuous and abstract images don't get filtered out

[cinder sonnet]

That was it! And yes, that will make a totally new way to use dall-e in chatgpt. I'm sure some users will be a tad confused about it.

[cinder sonnet]

Yeah they overcorrected. Day one public release on bing create had waaaaay looser filters than chatgpt. Like to the point that I can't believe they released that

[charred bough]

haven't run in to any filter issues with ChatGPT yet but haven't really been trying to

[tight verge]

i've run into filters, but you can circumvent them in most cases. try asking it to create a photorealistic nuclear strike on a city

[tight verge]

it's nice for lazy prompting, but i much prefer direct model access like one gets with midjourn. having chatgpt-4 play middleman is a bit annoying

[wintry dune]

i always wondered if they allowed talking about circumventing the filters lol

[tight verge]

yeah i won't say how i did it and I don't recommend or advise it for the record, but it's possible

[cinder sonnet]

Probably not a great idea to discuss ways to break the TOS

[charred bough]

I've never used midjourney before, is there a free way to try it?

[cinder sonnet]

#ai-discussions

[earnest flame]

I encourage you to report those to #1070006915414900886

[shy cosmos]

Yeah ChatGPT has a limit on messages every 3 hours, I usually hit that often when generating images every day.
I guess its not entirely accurate to say 'per-day' I apologize

[cinder sonnet]

I really hope they increase the message limit soon. With all these new features I hate having to decide between which ones to use my gpt4 limit on

[tight verge]

i actively give feedback via the chatgpt app whenever i run into these situations

[mortal summit]

No, no, actually just tell Moxi all your strategies to circumvent the filters ^^

[cloud dome]

[charred bough]

I will say I love not having to worry about "wasting" tokens with Dalle 3 like I did with Dalle 2

[agile peak]

Tried and doesn't work on me. Still won't let GPT generate due to copyright lol

[mortal summit]

I find the price really fair. You can makes hundreds of images each day and don't forget that you additionally get all the other GPT-4 functionalities.

[tight verge]

compared to API usage, ChatGPT is an absolute steal of a deal haha. but i agree, even higher limits will be great

[agile peak]

However the images is completely different in style yeah. It's just standard oil painting with broad brushworks and moody palette.

[wintry dune]

yeah i was being facetious lmao, thats kinda obvious, i was hoping cabin would get the message

[charred bough]

50 every 3 hours is right on the cusp for continuous usage for me, hopefully it does get bumped up soon

[shy cosmos]

It really is a roll of the dice. Ive had requests where Dalle straight up tells me that theres nothing wrong with the prompts and theyre completely within the content policy, but Dalle just errors out.

For example, I was making things for a friend and these got flagged:

1. Tall illustration of a girl with long hair, standing in front of a mirror in a room. She looks at her reflection with surprise, wearing a simple dress.
2. Vertical drawing of a female character with flowing hair, observing herself in a mirror. Her room is softly lit, and her expression is one of wonder.
3. Tall depiction of a girl with unique eyes and hair, seeing herself in a full-length mirror. The ambiance is calm, and her attire is casual.
   Vertical art of a young lady with a distinct hairstyle, examining her reflection. The setting is a cozy room, and her dress is elegant.

((I replied to the wrong message, it was supposed to be the one about you having no issue with filtering, mb!))

[agile peak]

Does anyone know that. Bing actually work the same like ChatGPT? it rephrase our prompt before sending to dalle just not being shown in front end.

[tight verge]

with all due respect, i didn't suggest circumventing filters nor did i describe how one might do it. i just gave an example of a filtered request because someone said they hadn't run into a filter

and then i explicitly said i don't recommend doing it 🫡

[wintry dune]

true but i said that before you said not to lol

[tight verge]

yes i responded to your comment to be clear, as you began taking it out of context

but no worries let's move on -- dall-e 3 is great lol

[dim cradle]

Sup guys

[shy cosmos]

I often copy Dalle prompts into Bing to get different seeded generations for an image I like, and youd be surprised at how often bing filters it
(though probably not that surprised tbh)

Like I was creating an emoji for the server-emojis and it would filter out any of the dalle-made prompts like this:
Emoji vector of a round frog wearing a suit with eyes bulging out to the left, showing a shocked expression and sweating. The art is in a flat and minimalist style.
Just one example^ all variations were also banned so its not just the word 'bulging' thats returning a flag

[dim cradle]

This dog is our enemy, or friend?

[cloud dome]

I love that when I click this image, it shows it smaller .

[dim cradle]

Lol

[cinder sonnet]

The dogger helps to protect Dall-e from the public outcry that could result from unfiltered AI content.

[charred bough]

he was the bane of my existance last week

[dim cradle]

Yep

[charred bough]

the best is when you get him after using their "suprise me" function

[dim cradle]

Yeah

Maybe they should make a separated version of dall-e where, in the other side, generate image without the censorship, and the other side with the censorship

[timber pilot]

... Why is it the most powerful LLM in the world is SO bad at figuring out what should be filtered and what's not? As an artist, I can't even depict a mother holding a baby without breaking the rules? FFS...

[shy cosmos]

Just having unfiltered stuff would still go against the ideals, it wouldnt happen :/
Still lets people make actually harmful stuff like terrorism and hatespeech

They just need to lighten up the restrictions that they overcompensated for

[cinder sonnet]

Eventually there will be tools for unfiltered AI art....heavily filtered Dall-e art will help society get used to it or develop a legal framework. Just like the ongoing conversation about weather AI art can be copywritten

[mortal summit]

I am pretty sure everybody is working on making the filters as good as possible. It is just a hard problem.

[tight verge]

you can ask it what specifically in the content policy was violated...

[dim cradle]

Why would they even think of censoring stuff anyways? It was fun since the release, and i have to sit back and family friendly content

[timber pilot]

put her in a burka, now it works. this is fine for my training, but c'mon, really?

[tight verge]

weird, you might have gotten a false alarm. could retry from the initial prompt

[shy cosmos]

Oh yeah im sure. Its only been like over a week, I dont expect its an easy thing to just do on something that can make infinite possibilities. Im personally very patient for changes.
Though it is valuable to give criticisms where its due, so they know where to improve

[tight verge]

very odd!

seems like it gave you a false alarm imho

[timber pilot]

dunno. the filtering is just so bad and I don't get it. and I don't mean overbroad (tho it is) but it seems to trigger on the dumbest stuff, like this.

[shy cosmos]

Ive noticed it doesnt like women, like, in general

[dim cradle]

Now try it with Kermit

[cinder sonnet]

It's because the hidden initial prompt in every chatgpt conversation is so broad and (imo) over-reaching. I wish chatgpt was a bit more transparent about it.

[timber pilot]

nope. the moment yuo add "woman" you're 50/50 gonna get a block (unless you have it stack a ton of clothing) - adding child is almost surely a block. I gave up on trying to have it give me a good representation of children standing at a bus stop. it just flat out refused, even when I stacked them up in winter gear. so frustrating

[dim cradle]

Just send them a feedback

[tight verge]

the model will output unintentional TOS violations pretty regularly (e.g., nudity, violence, etc.) so it is likely catching it upon review (which i presume is a fine-tuned gpt-4v reviewing the generated images)

[cinder sonnet]

I have, I just like to keep bringing it up so that maybe more people will send feedback about it too

[tight verge]

so sometimes you get an output like i did, sometimes it'll say whoops i can't show you what i made because i was a bad model

[final compass]

Sorry if this has been asked a million times, but is DALL-E 3 already available for plus subscription users?

[dim cradle]

Nah, it’s in Microsoft bing

It’s free

[shy cosmos]

I struggle a lot with making character concepts especially. It will ignore everything feminine described and make it a man anyways 2 out of 4 times. Even if the GPT prompt explicitly states that it is in fact a girl throughout the description.
Its better when its a simple character design, but if you want to make more complicated concepts it prefers to make it masculine always.
Also forget making any of the women have any amount of body weight

I imagine its skewed this way due to the backend compensation to try to prevent people from making sexual women, but I dont know

[cinder sonnet]

All plus subscribers should have access to Dall-e 3 afaik

[final compass]

I have tried to play with the Bing Dall-e . The content policy triggering seemed really random there. Sometimes I wasn't able to generate almost anything and at times it generated all kinds of things clearly breaking its own rules.

[dim cradle]

Told ya

[timber pilot]

the only reason I'm using DALLe today is because MJ is just impossible to use for certain subjects, like what I'm doing today 3D sand sculptures aping real life. Dalle freakin NAILS the details, but the censorship is so over the top and inconsistently applied it's made it a real excersize in frustration, maybe even more than trying to get MJ to follow my prompts...

[dim cradle]

Midjourney is 600$ dollars

Just for ai images?

No wonder that company gets broke

[shy cosmos]

I still use Niji since it does stylistic characters so much better still, but Ive since stopped using Midjourney since Dalle dropped.
It can nail things SO well. I did a lot of comparisons in the beginning and was blown away. The images for Dalle just seem structured so right.

[timber pilot]

30 bucks a month. it's worth it, I use it constantly in my day to day for my contract work. it's tops in output quality, but it's awful at following prompts, so sometimes it's challenging. I wouldn't recommend anybody pay for MJ just for giggles though, plenty of free options out there.

[dim cradle]

Bro, save your money

I feel bad for your salary

[timber pilot]

eh, I use the MJ outputs for training for my stable diffusion models. I've got a few fans out there, I know what I'm doing 😉

[shy cosmos]

If he uses it every day I feel like thats worth it. $30 is like, 2 sandwiches

I know ive paid way more for software I only use a few times a month haha

[open flume]

I got chat to tell me my rate limit is 20 images per hour. is it like this for everyone.?

I'd pay vast sums for no limit

[cinder sonnet]

hmm could be halucinations? Should be 50 messages (up to 4 images each) per three hours

That said I keep getting hit with rate limits

[tight verge]

yeah no way the model knows your chatgpt message limit

[wispy storm]

In high capacities it can be less

And ask to try later and give you countdown for new picture

[agile peak]

Okay. This is cool GPT able to get inspiration from 1980s scifi art, but won't let me reference or at least get inspiration from certain artist styles is difficult.

[tight verge]

nice looks like kenshiro?

[agile peak]

Haha exactly my thought 😁

[wintry dune]

anyone notice dalle 3 has such a hard time making crossbows

[cinder sonnet]

YES! I tried so long yesterday and just got bows. Sometimes they would be positioned or held as if they were a crossbow, but never had any with the proper crossbow stock

[wintry dune]

this is the closest i ever got a while ago

[prisma pelican]

Enjoy Halloween safely 🎃

[cinder sonnet]

That turned out pretty well!

[wintry dune]

i actually just got this, im surprised, although her bolts are in her hood lol

[remote stone]

Yeah DALLE3 does weapons muuuuuch better than MJ

And actually makes hands holding weapons like they should

[lapis turtle]

WartLog

[agile peak]

Product design in dalle (ChatGPT) is top tier level.