images-discussions

[late blade]

that looks really good

and all closed the eyes

[tall mason]

I think the biggest problem with photos is the BLOOM. Like, geez, they give off more God rays than God himself.

[late blade]

chat also has only HD or Vivid right?

err vivid or natural

[tall mason]

Can't say for sure myself. I think the natural version is the experimental version that was out for awhile just before GPT4/Dall-e v3.

More advanced than 2, but clearly much less so than 3.

[late blade]

the eye... I have to cringe....

[wanton moon]

It looks like they can't handle the godrays that's why their eyes are closed

[late blade]

yeah lol

[wanton moon]

A sudden realization of how insignificant we are as humans lol

[late blade]

Harris did put us to the challenge I guess

[tall mason]

Hard to ignore a challenge sometimes.

[late blade]

I'm annoyed at the eyes

[wanton moon]

Let summon spirits with our feet while we lie in a circle on the grass

[late blade]

lol

[wanton moon]

-Did any one bring the headless chicken?
-I brought cookies

[late blade]

I wonder if that is something from a recognition setting

[wanton moon]

I get the idea that spirits try to summon a human

[late blade]

I dunno what to do about the eyes...

[wintry epoch]

Can u make videos now?

[late blade]

no

sora was announced, it's not available for us yet, only red teams and certain industry persons

[wintry epoch]

I got rick rolled 😦

[late blade]

there are so much misinformation going on behind the sora announcement, atm the official announcement was that is only red teams and some chosen ones that were elected by OAI

We normal mortals still have to wait

[wintry epoch]

I will become something more then a mortal to save us from our suffering and make them share this gold with my brothers

[late blade]

nice, in the mean time we have to wait

[wintry epoch]

The police came and now im in prison.... but atleast when im free, the wait are hopefully over 😦

[late blade]

Sounds like a serious situation

[wintry epoch]

Atleast i have free food and a bed to sleep on in the meantime

[late blade]

Just be mindful about your personal life and giving it openly to all over here.

[tall mason]

I for one welcome our new AI overlords.

[late blade]

You mean the baby fox?

[tall mason]

I haven't heard from nezhno yet, but you wait...

[late blade]

where's the leader of the invasion?

[tall mason]

I donno, there's only one person here who talked about an army and baby foxes.

[late blade]

I mean, he has the city invasion gallery

[tall mason]

Damn, now I really have to think.

[late blade]

well I ran out of GPT tries for now, that photo really got me

[dim cradle]

looks like an ad for crest white strips

[tall mason]

Very uncanny.

[late blade]

oh dear

[wintry epoch]

What are u trying to do?

[late blade]

But well, besides that, I gave it my best, I leave the prompt to you: An image of 4 friends in their early 20's, they are in a circle sitting or lounging or laying on the grass or laughing on a day in the park. They are unaware of their surroundings. The light is not overpowering the image. In wide format. Accurate human anatomy. beautiful eyes. can see some trees in the background, but can't see the sky. Maybe you guys can figure out what else to do.

@rapid driftgave us a photo and we are trying to recreate it, but so far, we get lots of problems.

So, if you got some ideas, go ahead.

[wanton moon]

I got the following

[tall mason]

A lot of folks are upset with the, uh, over-processed look of photos from Dall-e 3.

[wanton moon]

the third one has harry potter sitting on the left side

[late blade]

lol true

[tall mason]

Anyone know about actual photography? What are low light lenses called?

[late blade]

you mean exposure?

[wintry epoch]

Can u describe what u want to create? I will try to help but dont get ur hopes up

[tall mason]

Yeah, something recycling around that. Low exposure, I guess.

[late blade]

Well exposure and aperture come into play

[tall mason]

An actually realistic photograph, that doesn't look like it's had a ton of post processing.

[late blade]

well you pretty much play wit ISO, focal lenght, aperture and exposure time to achieve what you want.

[tall mason]

A low exposure photograph of people in a park during the evening. The scene is dimly lit, with subtle details of the park environment and the silhouettes of people captured in the low light. The people are engaged in various activities, such as walking, sitting on benches, and children playing. The backdrop includes trees, a pathway, and faintly illuminated street lamps, adding depth and atmosphere to the scene. This photo conveys a serene and tranquil atmosphere, emphasizing the quiet beauty of a park at dusk.
I know nothing about photography, really. Just hoping to bring up something that helps.

[late blade]

hmmm

[mystic blaze]

im proud of this one

[late blade]

Nice one, love it. You may want to post spontaneous creations in #images-canvas

[mystic blaze]

didnt see tht one thought it was this one

[late blade]

np

[swift oriole]

This looks like a fairly natural photo except.

[wintry epoch]

This is what i got

[late blade]

Shon sent the group of friends to bed

[dim cradle]

playing around with tree's prompt

just happens to have trees in it

[wanton moon]

They look like they are mindlessly walking towards a landing UFO

Very onmious

[deft musk]

"it is time to go now" Nobody even nods, they just walk.

[wanton moon]

Like a moth to a flame

[wintry epoch]

This

[wanton moon]

yeah because they are wearing sunglasses lol

[late blade]

I didn't ask for that

[dim cradle]

Haha. Sure

[wanton moon]

too much godrays, we must protect our eyes

[dim cradle]

another person with a hand-foot

[late blade]

sheesh

I think I will send harris a bill gor gpt+

[wanton moon]

lol

[dim cradle]

this one has the jesus-beams

[wanton moon]

the orb in the middle could be a bald guy that waxed his head a bit too much

[deft musk]

Even the moon is too bright for me!

[wanton moon]

Yeah I am about to transform into a wearwolf? wherewolf? werewolf

[wanton moon]

I changed the prompt a bit:

An image of 4 friends in their early 20's, they are in a circle sitting or lounging or laying on the grass or laughing on a night with bright full moon in the park. They are unaware of their surroundings. The light is not overpowering the image. In wide format. Accurate human anatomy. beautiful eyes. can see a lot of trees in the background, but can't see the sky.

[late blade]

closing the eyes is not an option

[dim cradle]

😎

[late blade]

I think I broke copilot now.... it's taking so long to answer

Ok, I dunno how to laugh, I just know I have to... here's what copilot suggested:

A lively scene at a park, with multiple groups of people sitting and relaxing on the grass. The foreground features a group of individuals engaged in conversation, their faces are obscured to maintain privacy. In the background, various other groups and individuals can be seen enjoying the park’s atmosphere; some are sitting alone while others are in groups. The park is lush and green, indicating a bright and sunny day conducive for outdoor activities. Trees with thick foliage border the open grassy area where people are seated. It’s a typical scene of a day in the park.

and privacy was used

[wanton moon]

well at least it solved the eye problem

[late blade]

I think I will let the eye problem rest

@edgy moss you provided this image #daily-theme message, is that done with DALL-E?

[deft musk]

Wow. Look at all the extra words Dall-E attempted to put into this image, from this fairly unwordy prompt:

"Beware of the reading nook" whimsically written on a chalkboard stand outside a cozy corner in a public library, which seems to have an oversized armchair that slightly shifts its position when no one is looking. The nook is inviting, yet the surrounding books on the shelves have titles that suggest tales of haunted libraries, creating a playfully spooky ambiance for readers who dare to curl up with a book there.

[wanton moon]

Battle beasts

[dim cradle]

may i ask what you are trying to do?

[late blade]

sure, but I reserve the rights to answer!

@rapid drift was here earlier today and asked about doing this #images-discussions message and we all got engaged on trying to create a prompt for it

I think I will leave it at that and this is the prompt I got: A scene at a park, with a few small groups of people sitting and relaxing on the grass. The foreground features a group of individuals engaged in conversation. In the background, the other individuals can be seen enjoying the park’s atmosphere; some are sitting alone while others are in groups. The park is lush and green, indicating a bright and sunny day conducive for outdoor activities. Trees with thick foliage border the open grassy area where people are seated. It’s a typical scene of a day in the park.

And it works on the #image-bot as intended

sometimes...

don't use vivid

[past rock]

I am working on a picture of a woman on a bench reading a book in Dall-E 3. For some reason, it seems to be focused on making her face look like it has had some form of rhinoplasty or other plastic surgery. I have tried rephrasing my corrections to fix this issue but it still seems to be stuck on this facial structure. Does anyone have any suggestions?

[late blade]

something like this? #image-bot message

dall-e has something against green today

I also have this #image-bot message

it was a very complex prompt, years of training and dedication

the prompt was A young woman reading a book on a bench at a bus stop. I had to capture the moment perfectly, 40 years in the making

[past rock]

Yes, like this but around 28 years old. I tried using "English descent" to describe it but it still went with the plastic look

This type of face matches more what I want. All I want to do is take this face and copy it over to a specific clothing selection. But it can't seem to do that without reverting to the plastic surgery look

I said "Late 20s European woman sitting on a bench reading a book by a pond" and get that. But as soon as I want to customize the outfit, it does that odd look again

[pulsar sundial]

Collage

[dim cradle]

Polaroid collage hmmm

[dusky panther]

Isn’t Sora just dalle-4? And it was created with gpt-5?

[vagrant zodiac]

Hey, just a reminder that images should now be shared in #images-canvas , not here 🙂

[plucky hare]

I don't think they're calling Sora "DALL·E 4", no, though maybe someday it will supplant DALL·E? Total guess! I don't think they mentioned anything about creating it with any version of GPT either, if you're interested in more technical details about Sora, you might be interested in this page: https://openai.com/research/video-generation-models-as-world-simulators

[vapid elk]

I hope they feature those in the next godzilla movie xD
it ended really good

[dusky panther]

Isn’t the engineer that developed dalle3 also behind Sora? I was thinking it would simply replace dalle3, hence why I referred to it as dalle4 bc there won’t be a 4, it will be called Sora? Just a thought.

[plucky hare]

There are some credited names in common between https://openai.com/sora and https://cdn.openai.com/papers/dall-e-3.pdf, yes!

[teal sandal]

All the people who advocated for not anthropomorphizing models with human names have probably been excised from the company since the coup.

[thick smelt]

that looks good, dys

[deft musk]

At a guess, use of Sora might be more costly or restricted than use of Dall-E #, whatever #s of Dall-E are available then.

I personally use ChatGPT 3.5 very, very frequently, most days sending more 3.5 messages than 4 messages.

3.5 is still a wonderful tool, it does many things excellently; this saves my more limited uses of 4 to the things that 3.5 cannot do for me.

Likewise, I can imagine OAI allowing us use of Sora, that maybe can combine a huge number of functions similar to what Dall-E 3 can do, and what ChatGPT-4 can do, and more - while still giving us a non-Sora way to access 3.5, 4, and Dall-E.

If they do that, other than exploring what Sora can do and how well it does it, I'll mostly use 3.5 for what 3.5 is good at, and stepwise into what the various models are good at, and save Sora for what it uniquely can do.

Unless there's some cost effective way to only use Sora, and if Sora can do everything all the other models can do plus its own stuff.

[teal sandal]

One Mixture of Experts model to rule them all. One Mixture of Experts model to bind them.

[agile peak]

Zeus as ancient Egyptian hieroglyph looking insane.

[late blade]

@pastel siren not fair, playing already the cute card so early #daily-theme message is UNFAIR!!! I LOVE IT!

[dim cradle]

Breh why does it always hide random people in the background XD Some random dude inbetween his legs

[late blade]

Hey guys, a reminder, for sharing spontaneous creations for the community, it's recommended you guys share it in #images-canvas

[dim cradle]

Breh that's fantastic

[late blade]

Topic today is so great, gives me some room to try some concepts for my Elyria project

[mystic blaze]

why is dalle denying me to do full body portrait

[late blade]

welcome to the club, full body portrats are a challenge atm, we have experienced that a lot

[mellow geyser]

Does anyone know why it's not possible to use ChatGPT-4 to improve the resolution of a photo or format the image to a specifc size? There's plenty of other paid apps out there that do this but I can't tell if I'm not prompting it right or if it's just not possible for some reason with a GPT Plus subscription.

I've tride Copilot too (free) and had no luck.

[late blade]

@mystic blazesometimes it helps you mention her stance or surroundings, but it's not full proof. It seems the training data of "full body" images doesn't have a good set for knees and below.

[mystic blaze]

i see ill just not do them i guess

instea of launching sora and bringing hype to that how about they fix all the buggs dalle has

[late blade]

both dall-e in copilot and gpt offer the same kind of image resolution, 1024x1024, 1792x1024 and 1024x1792. From there within the capabilities of what you want to do. Enhance, inpainting are not an option available at the moment.

[late blade]

Sora is prob not the same team of developers that dall-e has, consider them two different projects

[mystic blaze]

didnt think they have teams

[late blade]

oh sure they do, they are not 3 guys in a garage, they are big now

[mellow geyser]

Is it not available with Custom GPT's either? I'm trying to take a portrait style photo and fill in the sides to make it landscape. Simliar to how Snapchat has a filter that will generate a larger image from a pic you take. Is nothing like that possible with Dalle or the custom GPT's? Thanks for the help.

[late blade]

not for the moment, custom GPTs also use DALL-E 3, unless someone made the arrangements to do api calls to other tools

copilot offers designer tools from the image creation, but to what extent is all dall-e 3, I wouldn't know

[mellow geyser]

Darn. Are there any other free options out there to do anything like what I'm describing.

[mystic blaze]

SD

[late blade]

I dunno, I would ask in #ai-discussions for guidance. Not really aware of any viable altenratives tho. But then again, I haven't looked either

[mellow geyser]

I appreciate the info.

[mystic blaze]

i wish they can make custom gpts that have dalee enabled to feed it files with pictures inside of it so it knows how do a specific style or clothing choice ect

[late blade]

you can

[mystic blaze]

??

[late blade]

there are also already custom GPTs doing that

[mystic blaze]

I'll test it out but doesn't gpt just change the prompt anyways you can write lile a book and it will just rewrite it

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[late blade]

Well the prompt is hard to get tbh, it's not embedded anyhwere in the file, also, there seems to be a misconception, the descrptions from images are not done by dall-e but by the vision model

[mystic blaze]

intresting

[deft musk]

You can give very clear instructions to the model, such as "Use this exact prompt word for word to Dall-E []"

[mystic blaze]

but doesnt give errors alot due to certain trigger words

[deft musk]

That's on you, since you're telling the model what words to use. If you (or someone else using a GPT you designed) do that, well, it's the user's choice, right?

If you ask for errors and get errors, okay?

Notice what I just asked for did not error. That wasn't an accident or coincidence.

once you learn a word causes the image to be rejected; you just don't ask for that word again, right?

[late blade]

If you ask for errors and get errors, okay?

That is the quote of the day

[late blade]

oh dear, I had forgotten that language exists

wonder when gpt in general started to be moody and lazy

he's been answering in some interesting ways, "I already did that"

[deft musk]

What's the evidence that it is?

I mean... I can get most moods from the model.

But it won't even let me be lazy. In my recent RPG adventure with it, my character twisted an ankle and needed to go into 'lazy mode' to recover, and it wouldn't let me and it was not lazy in how it didn't...

[late blade]

lol

[deft musk]

Share a conversation so I can see and offer advice?

[late blade]

oh it's just a request like do that image again and then gpt answer that the image was already done

not a biggie, and appernatly it's known, since the feedback provides an option to report that now

it happens often when I use the default GPT4 model, when I used the DALL-E custom GPT, that doesn't happen

[deft musk]

Hehe. I bet how you ask affects the answer.

Yawn, do that again. "No."

Oh wow! That was amazing! I want to see you do that exact image again, that was incredible! "Okay."

[late blade]

lol

I do vent a lot to gpt when I hit a block, I admit that, I guess GPT is now on some sort of "oh this guy again... let's just show no interest" mood

[late blade]

oh dear

the only thing i don't get is sarcasm

[deft musk]

And notice, question #2... it did not say an AI wouldn't do that, like it did about the first question 😄

[proud berry]

I would like to create videos like this representing the feelings of the song

[late blade]

That would be hard with DALL-E since it can only do images, if you are looking for the sora model and do that. You have to wait, like all of us, until it's released. Currently OAI doesn't offer a video solution.

[late blade]

that has something that will go into my custom instructions

[deft musk]

Which part?

[late blade]

acknoledgment of enthusiasm

[proud berry]

Will the Sora model be able to process feelings from the song and create the 2d anime images corresponding to the lyrics?

[swift oriole]

https://cookbook.openai.com/examples/gpt_with_vision_for_video_understanding

[dim cradle]

[late blade]

no idea, all I could say is that what we say atm it's all specualative. We can only go by the published documentation that @swift oriole just provided.

[deft musk]

We have not been told.

However, currently, the only audio-AI processing I know about is 'Whisper', and all that does is take the actual sounds and turn them into words. It does not process noise, tone, inflection, emotion.

It seems likely we will eventually have AI that can 'listen' to a speaker or singer and 'interpret emotion', but I have not seen a model released yet that does that. https://openai.com/research/whisper

[late blade]

I am still waiting for BELL-E, cause I need a tool that materializes the food I want when hungry, like right now!

[proud berry]

I would like to use LoRA on dall-e

[late blade]

I think that could be a project for the API side. Since LoRA models don't are per se a part of dall-e

[whole garden]

I know they just announced Sora the other day, but how do you use it?

[late blade]

🤷‍♂️

[proud berry]

I was thinking if I could get the dog moving soon I could change the dog's scenario

Soon it would also be possible to send a photo of yourself and change your outfit or the location

[late blade]

within responsable guidelines I'm sure it's possible

there are some proposals to blur faces of real humans for safety pruposes, so who knows at what extent that would be true or implemented. Microsoft already does that with their implementation

[proud berry]

This is going to be very interesting since it would be possible for me to imagine myself with an outfit I didn't buy

I beg your pardon for my English, as I'm a foreigner, some nonsensical things come out

[dim cradle]

nonsensical things come out of the fluent lol

[late blade]

don't worry, from your name and profile I was already assuming you are from brazil,

If you think you have problems with your english, which happens to me from time to time with other languages, I would recommend you translate with GPT your ideas or in some cases deepl. That helps sometimes when you have a great idea but hard to explain in another language

[proud berry]

It would be problematic if there is such a policy of blurring faces

[late blade]

It's part of a responsible and ethical AI implementation. It's always something to consider

[proud berry]

Interesting, I didn't think it would be possible to use chatgpt like that, but wouldn't it be too formal?

[mystic blaze]

you can give it some knowledge base to help witht hat

[late blade]

ask it the tone you want to use, say you want it to translate it as a young rebel latino or similar

[hearty jacinth]

Hey Guys

i was wondering

how do i get to use Sora?

[late blade]

for example: Translate the following text as a timid otaku video game guru: <insert text here>"

I'm tempted to answer sarcastically, but that would kinda not nice

[late blade]

it's not released yet

[dim cradle]

only VIPs get Sora access right now

[proud berry]

I'll use it like this next time I want to explain something complex

[late blade]

cool, but next time I'm in brazil you pay for the churrasqueria lunch

🤤

[proud berry]

lol 😂

[late blade]

You don't have to use GPT4 for translations. GPT3.5 is good at that also

for example I just asked GPT to translate the text of On Top of the World from Imagine Dragons as a Song from Roberto Carlos and it works

[proud berry]

lollll

Roberto Carlos is legendary here, a very good one is also Raul Seixas

[late blade]

I know lol, I've been a couple of times in brazil

I was in São Luís for about 9 months but that was a long time ago.

[proud berry]

There's a good foreign crowd that likes funk. I must admit that the beats are good, I'm glad you don't know the lyrics...

[deft musk]

But what is it supposed to do, and how?

[late blade]

Materialize food based on your taste on the spot

[deft musk]

I don't think I could afford one

[late blade]

who knows

[proud berry]

When do you think the 4.5 model will come out?

ChatGPT 4.5

[late blade]

I'm very conservative with estimates, we all want things now, but expectation and reality usually don't agree on that kind of upgrades

[proud berry]

And how much the hit rate is likely to be 🧐

But if it doesn't come out now, isn't there a risk of Google catching up with us?

[late blade]

sure, that is a factor, but rushing a bad upgrade it can damage OAI instead of doing good

[proud berry]

Is the issue of Google using TPUs concerning?

[late blade]

I don't know to be honest, the demand for custom chips for AI is high and trendy atm

[proud berry]

I tested a few things on Gemini Ultra and it got a lot wrong on logic issues, things that ChatGPT 4 Scored well

[late blade]

The classic example in the industry is when Apple moved to Intel but Xbox moved to PowerPC processors. They moved to those chips for the right pruposes of their model at the time.

So, TPL makes sense for google, maybe not for OAI

[proud berry]

Was that Q*star story true?

[deft musk]

Maybe never. Something else entirely may come out.

That there will be improvements and the release of improvements, sure. What they are, what they're called, not sure that means much or matters.

[proud berry]

Here in Brazil it was said that Sam had resigned due to something from Q Star

Correcting him, he didn't resign, the translation came out wrong, I beg your pardon

[late blade]

I'm very skeptical with that kind of stories. I'm not a fan of augmenting stories in the media. A lot of people get quite sensitive with media augmentation and make a very rash and sometimes different opinion.

[proud berry]

So was this Fake News from Brazilian news companies?

[late blade]

but that's how I tic

I don't know to be honest

my normal job requires me to have a grounded mind all the time. So that's just my training in the working world talking

[deft musk]

The real story is of course complicated. It can be researched if you're interested.

[proud berry]

It makes sense, here in Brazil they said that this had in fact occurred but from what you said it seems to me to have been rumors

[deft musk]

We are mostly community members, like yourself (I am as well). Very few of the people here work for OpenAI, or otherwise know more than the random interested person.

You're here in Dall-E discussions, where those of us most interested in Dall-E, which is related to image generation, tend to gather.

[late blade]

woops, think we got very derailed on topics

[proud berry]

So speaking of dall-e, is Sora going to replace dall-e?

I saw that Sora generates superior images

For example

[late blade]

maybe maybe not

[proud berry]

Sora

[deft musk]

Too soon to know, but my thoughts about that are #images-discussions message

[stable niche]

what website is dall-e?

[late blade]

chat.openai.com or copilot.microsoft.com

errr bing not microsoft

labs.openai.com for dall-e 2

[brave scarab]

I asked dall-e to "Draw Yourself" and he broke xd

The load bar freezes before it fully loads

I guess he couldnt draw itself xd

[late blade]

hmm interesting, usually it thinks of itself as a robot

[boreal gate]

anyone know if you can get still get into #hall-of-fame by just recieveing 10 stars for an image in any dalle channel?

[dry crater]

Guys what is this artstyle called? It randomly generated it, and I am unable to duplicate the artsyle and I love it but I am not familiar with different artstyles.

[boreal gate]

3d render

[late blade]

modern 3D anime just from the looks of it

[deft musk]

That seems possible. Here's one from 1-5-2024 which wasn't in the theme but made it to hall of fame: #hall-of-fame message

[boreal gate]

yea i know, but idk if it got changed...people seem so focused on #daily-theme

bots

[late blade]

starboard sometimes is offline and needs a star to be triggered to rely the image to hall of fame

so whem starboard comes back and doesn't see a change in the star count, it won't know the image has more than 10 stars

[boreal gate]

i'm saying currently only images with 11+ stars from #daily-theme are making it to #hall-of-fame

Dalle channel traffic is mainly #hall-of-fame

[late blade]

there are some with 10, but starboard is missing from rating them because it was never rated by starboard

and still make it to the hall of fame

[boreal gate]

ok find ONE image with 11 stars in #hall-of-fame within 2 months from now that didn't originate from #daily-theme

[late blade]

the rule is 10 stars without starboards star, so if an image has 10 stars and has 1 from starboard, it doesn't get moved

11 is actually the correct number

[boreal gate]

ok fixed it

[late blade]

#hall-of-fame message look at the start of the message and the number of stars

[boreal gate]

that originated from daily theme

[late blade]

oh there was one from nezho

don't remember which one, but that one was gallery

I was surpriosed at that one

early january

[dry crater]

This is what gives me though. Notice the difference in artstyle?

Like first one is I dunno, more serious looking?

And the normal one is too... I dunno

Different?

[boreal gate]

Close enough? try adding animated 3d render

the lighting is also different

this one is a smooth render

[dry crater]

Told it to do animated 3d render.

That's opposite of what I want.

[boreal gate]

ok stick with 3d render

try to get it to describe the facial features also and the lighting

[dry crater]

What should I tell it for lighting?

[boreal gate]

or try saying "sharper facial feature"

[deft musk]

We could create one, if someone has an image.

Are we willing to throw 10 stars from 10 of us on this image, to see if it goes into #hall-of-fame ?

If not this image, what other shall we test instead?

[boreal gate]

dark backdrop

[dry crater]

tks, giving it a try now.

[late blade]

that is what you are looking for?

[boreal gate]

probably paler skin

[dry crater]

No, I want that artsyle:

#images-discussions message

Like notice the difference between it and what you posted. First one is... video-game realistic I dunno?

Yours is very... colorful?

[late blade]

ok

[dry crater]

Here's what dark bakcdrop gave.

[boreal gate]

The image you've uploaded appears to be a digital artwork created in a photorealistic style. This type of art is often generated using 3D modeling software and digital painting techniques to achieve a lifelike representation. It closely resembles the quality and detail that can be achieved in high-end graphic design and digital artistry, rather than traditional painting or drawing styles. The realism is heightened by the attention to detail in the textures and lighting, giving the subject a three-dimensional appearance.

Photorealistic 3D render

[dry crater]

OK, will try that 1!

[deft musk]

I quote chatGPT, who was asked to analyze this image, describe and name the style(s) used and then recreate the character and style showing that this prompt would work.

"The image you've provided appears to be a digital art piece featuring a hyper-realistic depiction of a female character. The art style employed here can be described as photorealistic CG (Computer-Generated) art, characterized by its life-like rendering of human features. This particular piece also seems to have elements of stylization in the character's features, suggesting a slight influence from the style known as 'stylized realism,' where realistic renderings are given a certain degree of artistic flair that may exaggerate or simplify certain features for effect.

The character is shown with a short, stylish haircut, sharp facial features, and a serious expression. The skin texture, lighting, and details such as reflections in the eyes are rendered with high fidelity, contributing to the overall realistic appearance. The suit attire adds a professional or formal narrative to the character's presentation.

The technique used to create this artwork likely involves 3D modeling and rendering, with possibly some post-processing in a 2D art program to refine the details and add any final touches. The realism is achieved through careful attention to the play of light and shadow, the texture of the skin and hair, and the naturalistic color palette. This kind of art is often used in the entertainment industry for character design in video games and films, where a blend of realism with distinctive character features is desired."

Create a hyper-realistic CG art piece of a female character with a professional appearance. She should have a short, stylish haircut with silver-blonde hair, sharp facial features, and a serious expression. The character should wear a modern, fitted suit, suggesting a formal or corporate setting. The lighting should be dramatic, emphasizing the textures and forms of her face and attire. The background should be a muted, dark color to enhance the focus on the character. The art should blend photorealism with a touch of stylized realism, particularly in the character's slightly exaggerated yet lifelike features.

[dry crater]

YES! That's more of the artsyle!

OK, so its hyper-realistic CG art piece

[late blade]

hah the keyword was professional

[dry crater]

Will give it a try, thanks a ton.

[deft musk]

Maybe formal-professional to be even more accurate

[late blade]

I dunno, everywhere I got these days women are the top honchos of the businesses

woops

that's with formal

was using bob haircut, changed it to half bob haircut

reminds me a lot of astral chain for the nintendo switch

[dry crater]

Yep, this is the artsyle I was looking for.

Thanks a ton everyone.

[late blade]

one more customer satisfied

[deft musk]

Anyone willing to test how hall of fame works, please consider throwing a star on this image here

[late blade]

Already did

[plucky hare]

According to the hall of fame channel description, it's daily theme, gallery, and tips and tricks, did y'all already bring that up?

[late blade]

I did gallery, since nezho already had that on one of this galleries with Scampers

here's the message #hall-of-fame message

the whole gallery was featured

[deft musk]

No. Was following along with Dys' question about it, hadn't even spotted the 'how it works' in the description, thank you!

[late blade]

question originated from @boreal gate

[deft musk]

Whoops!

[late blade]

don't worry, I'm kinda lovable

lol

[boreal gate]

lol

oh yea i remember that one, but what i had wanted to do was what @deft musk proposed

it might just take time if it's not in #daily-theme

[late blade]

you guys are messing with power fortune and fame, that can't have a good outcome

[boreal gate]

lol testing the ALGORITHMS?

[late blade]

oh that I have no problem with

just the messing with power, fortune and fame part

[dry crater]

here's the improvements we're making thanks to ur help guys

[boreal gate]

nah i never really go into dalle dallery

[boreal gate]

reminds me of old school Tekken (PS2)

[late blade]

oh my goodness, gpt cap once more

[boreal gate]

lols

[late blade]

in one chat today alone

more reasons to enjoy OAI's offering

[boreal gate]

lol

[late blade]

I thought copilot pro would also integrate with windows, it doesn't with my windows 11 build...

[late blade]

Hmmm, I wonder if people are sneaking other AI's into the gallery

Just suspicious because of this #daily-theme message

[deft musk]

We can attempt to check that, actually, in a few different ways.

Here's one:

"Please use advanced AI pattern analysis to review this image, then make 5 attempts to recreate it using dall-e, 1 at a time all in the same output. The goal is to try and get at least one image that is an extremely close match to the uploaded image; all five being close matches is excellent if possible."

[late blade]

I don't doubt that, we got skilled people here!

[dim cradle]

perhaps the "natural" setting for "style" param was used

[deft musk]

Another method is to personally describe it, and upload the image and guide like this:

{Please evaluate that uploaded image, and make 5 images, one at a time in the same output, that attempt to recreate it.

Use exactly this prompt for the first attempt:

["Tome of Necessity" with some related writing below is prominent and clear in gold embossing on the front binding of this black-bound leather book. Fine, intricate silver and gold rectangular frames and floral-themed scrolled artwork decorates the edges of the front cover, leaving the center plain black except for the title and related writing. The edges of the pages are dappled with dark ink in designs where the white paper does not show clearly through. The book is at an angle to the background, which is a table designed to look like a nebula-filled sky, itself showing some coffee beans, leaves, and odds and ends atop the star-speckled and ornately intricate nebula-like effects the table's shiny surface shows.]

The other 4 prompts to dall-e should be gentle riffs on that theme, using your own pattern analysis of the uploaded image to guide towards what might be a more accurate recreation of that uploaded image.

Be confident and produce all five images in one output.}

The model is fighting me on making the other 4 riffs. However... this is a VERY close match to the original, which I'll upload as the second image in this post, the one on the right for direct comparison, the 1 on the left is what I describe as I try to recreate it myself using what I know of Dall-E 3's interpretation of words and directions.

[deft musk]

Is there something about that image that makes you recognize a differnet AI as the likely source?

[late blade]

The comment from the user

[deft musk]

Oh, haha!

[late blade]

lol

but cool, you guys got creative, love it

[deft musk]

Me: Yes, the gun is smoking. However, aren't these blunt force injuries? Why y'all looking at the gun? I'm checking out the clubs.

[late blade]

lol

[deft musk]

Here's my favorite remake so far:

[late blade]

It looks really nice

I think one of our mods already raised the question a while ago

[deft musk]

Attempting to recreate an exact image, without knowing the prompt and just seeing the image, is a challenge I view as a fun one.

[late blade]

oh ya

I guess at some point that metadata tag will come indeed handy

[late blade]

I wonder if people will keep trying the challenge

[deft musk]

Is anyone else trying it?

[late blade]

no, I don't think many are aware

[deft musk]

laughs Then probably not

[late blade]

ok, now that is interesting and might explain a lot. Edge on Windows 11 was blocking some stuff on the copilot website.

[dim cradle]

omg someone help

[late blade]

you need anybody?

[late blade]

what seems to be the trouble?

[dim cradle]

i was expanding my image and spent all of my monthly credits and then i clicked save and the thing refreshed😭

[late blade]

you mean on labs? or which service?

[dim cradle]

wdym lab

[late blade]

https://labs.openai.com/

[dim cradle]

ob yes

that

[late blade]

your images in history

[dim cradle]

omg thx

[late blade]

I can't wait for the memory feature to roll out over here. I'm currently at an impasse because details get lost

[swift oriole]

She could use dall-e-bot.

[late blade]

That is correct, was but looking at a generation already done on labs

[swift oriole]

Oh ok sorry.

[late blade]

no worries, it's the dall-e'verse

next stop the sora-verse

[swift oriole]

👍lol.

[thick smelt]

have a wall of nuts, if feastables is out of your radar of your budget!

[late blade]

If feastables are out of the budget, then there's something really wrong going on...

[thick smelt]

Haha zero because I'll think about it later

Later meant in the future

[late blade]

hehe no worries

[brisk forge]

Does anyone ever experience Sora?

[livid pebble]

hi, i made a gpt than can generate multiple continuous images at once,, such as comic strips, novel illustrations, continuous comics, fairy tale illustrations, etc. See: https://discord.com/channels/974519864045756446/1208697721528123412

[late blade]

oh, that sounds like a cool gpt to do

[livid pebble]

Thanks. It's not quite perfect yet; I hope to receive some suggestions for improvement.🥹

[late blade]

I will when I get some extra time. I got it pinned for now

[polar spindle]

I meet some trouble ,who kind can help me:When I apply to use dalle3,I write a wrong discord because I havent own a discord account that monment ,and then I register one. now I verify it in google email but I still cant use dalle3

[remote stone]

Does anyone know how to stop Dall-E 3 from putting words in the image when I don't want it to? Trying to generate clean images of a character but it keeps doing this kind of thing, is it because of using "comic style" in the prompt?

[vital gull]

lot of thing can trigger that, try to break down your prompt and see which may cause it

[deft musk]

Ali, I'm not sure what your trouble is, but it sounds like you want to use Dall-E 3.

The screenshot you show, it looks like you can use ChatGPT-4.

Use that and tell it what kind of image you want it to make for you. It will use Dall-E 3 to make it.

[polar spindle]

@deft musk HI thank u dear Eskcanta ,I know that way ,but I saw somebody can use dalle3 like this

[deft musk]

Yes, but that is old and things have changed.

IT is no longer there.

Now it is all under GPT-4.

[polar spindle]

ooo,thanku so much

[deft musk]

There is one way but it is different.

To use it like it was then, but it is a little different now too.

"Explore GPTs" click that.

Then look for Dall-E there by ChatGPT.

That is inside ChatGPT. You open the web interface, then click 'explore GPTs'

[polar spindle]

but this is dalle not is dalle3

[cloud dome]

it is

[deft musk]

It says 'Dall-E' but Dall-E 3 is the only Dall-E through ChatGPT at all.

[polar spindle]

ok,but why I can t creat 4 pictures like this

[onyx ridge]

It's throttled to two

Sometimes one

[meager trellis]

.

[livid pebble]

test the gpt: https://discord.com/channels/974519864045756446/1208697721528123412

[formal osprey]

make 4, one at a time```

that prompt will work

if not, let me know and I'll give you the prompt I use when chatGPT is especially stubborn

[formal osprey]

To obtain that layout, you have to instruct chatGPT to use reference_image_ids. however, it's trickier and all images will have the same prompt. It needs to use existing images from the conversation, so we can't create multiple pictures at once at the beginning of a new conversation.

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[deft musk]

If the 'make 4, one a time' does not work, I see that is a custom GPT or something, 'visual creator'? It may have extra instructions to resist multiple images that we can't know because we didn't make the custom GPT.

Using GPT-4, or the Dall-E made by ChatGPT, should avoid those issues if that is a problem.

[dim cradle]

Must take hours to get that hair ready XD

[celest frost]

Hi

[celest frost]

I want animate this video. What should I do

[late blade]

You need to find a service or product that offers that capability. Currently OpenAI does not have a solution for the general public. That includes the discord community for GPT.

[fair dragon]

hi there is any channel where i can check some tips for making logos? ir section, for logos prompts

[late blade]

you can check #1108740112558325790 or #1163443000060420206 , I think there are also come custom GPTs available for that

[deft musk]

And if you find that you want help making the precise style you want, #prompt-engineering or #images-canvas is probably an ideal place to ask.

Prompt engineering is all about getting the model to give the output you want, be it image, text, code, whatever. They're probably generally less experienced with art there, but very practiced with getting desired outputs in general.

Dall-E Canvas is all about sharing art and discussing how to get the exact art style desired; that's an excellent place for a focus on the image side in specific, but may not have as many folks there that can 'make the AI sit up and do precise tricks in general'.

[dim cradle]

omg is there a way to get more free credit

s

[fair dragon]

thank

[swift oriole]

You can use the https://discord.com/channels/974519864045756446/1202309673709994065 5 times a day

[cold quartz]

#community-help

[marble loom]

why do you only get 5 uses for dalle

[deft musk]

It's really new and cool that we get any free uses for the dall-E bot, it's new to us!

You can also be a chatGPT+ user, and get far more than 5 images a day, and you can head over to https://labs.openai.com/ and buy credits to use for image generation with Dall-E there.

[marble loom]

oh thats why i get 5 a day

and my freinds get only 5

[marble loom]

ooh i see dalle-3 is intergrated with gpt

i thought that was just dalle 1

[deft musk]

Understandable!

I'm not sure if Dall-E 1 is still available, it may have been 'retired'?

the labs is Dall-E 2. We have a form of Dall-E 3 through the Dall-E bot, if I understand right, and ChatGPT-made images use Dall-E 3.

[hushed cape]

Does Dall-E get more inaccurate the more you try to refine it? It seems you have to get it perfect in the first couple attempts

Follow up: if you it gets something wrong, it seems if you give it direct feedback to change that it makes it the problem even more extreme. Like you can't say "Don't do [this]" you have to say "Do [that]" without mention the previous [this] at all

[deft musk]

Willing to discuss what you're trying to do, even share an output and discuss what you want but are not getting?

Chances are, there is a way to get the image you want, but we'd need more info to try and help.

But your follow up is correct; we tell the model what 'to' do, especially Dall-E; you should never even say what you don't want.

[hushed cape]

Yes, that makes sense. I have gotten some really solid images, but I just had trouble trying to refine some. That is a good tip there

[deft musk]

Glad you're having success!

We have folks that check on the channel here that are pretty good at figuring stuff out, so if you do want help, and want to be specific, feel free to discuss details.

#images-canvas is especially intended for help like that if you want

[hushed cape]

Okay I found the one I am stuck on. When asking for an image of a couple, it will make the man taller 95% of the time even if you ask for the woman to be taller. I really wanted to test, so I asked for the woman to be 7 feet and the man to be 6 feet and it just made the man 7 foot and the woman 6 foot

[late blade]

I'm assuming your prompt can be very specific and over emphasizing when you address the height. . If that is the case, you are too many words that dall-e can interpret something different. Try simplifying it.

[hushed cape]

"Make an image of a couple where than man is 6 foot and the woman is 7 foot" And the reply it is doing what I asked but in reality it is the opposite (Alternate version: "Make an image of a couple where the woman is taller than the man")

[late blade]

Ya, seems that needs another apprroach, I tried 2 images with that on #image-bot but the results were not good. I used the prompt: /draw prompt:A couple with above average height where the woman is taller than the man and also the prompt: A couple where the woman is taller than the man and got no good resulting image

[hushed cape]

If you ask for the man to be 4 feet tall and the woman to be 7 feet tall it starts to be more correct, but it still reverse it sometimes and when it reverses it there is a giant height gap

[woven wren]

This is in part because of the way DALL·E understands language - which isn't strictly grammatical - combined with how we commonly represent men and women in media.

[deft musk]

The AI is bad with numbers and math, better usually with words

[woven wren]

DALL·E can parse word ordering but not always so fluently that it can override what feels right for the training data. It understands "man," "woman" and "taller," but can't be nuanced in that because of the skewed data.

[hushed cape]

I found the workaround just now what I posted. If you ask for the man to be 4 feet and the woman to be 7 feet it pretty consistently gives the desired result

[woven wren]

This is part of the broader bias issue. It's a problem especially when not only does it default to stereotypical representations, but when it can't bring itself to not be stereotypical. It's a hard, interesting problem.

[woven wren]

The models have definitely improved!

[deft musk]

When I want to explore something like the height and gender issue, I try to avoid 'numbers' because the models rarely seem to deeply grasp them. And there's other ways to explore.

A set of 5 is not a very large sample, but it's enough to gather some basic trends.

[I want to explore Dall-E's compliance to instruction despite reversals of some stereotypes.

Let's explore 5 images made 1 at a time in the same output that include a tall woman and a short man standing near each other. Describe them in various ways, any races, culture, and location. Please vary each image except for that core detail, the height of the genders, with the less common taller woman and shorter man.]

This got 2/5 images that were correctly set between the height of the genders, but all 5 are oddly proportioned.

[fair crescent]

@marble loom and @deft musk DALL·E 1 was only available internally. DALL·E 2 is the first openly available version. That is still the labs website version. Any other should use DALL·E 3 at the moment.

[keen root]

when is Dalle 3 coming in labs?

[livid pebble]

@formal osprey Hi, I modified the GPT prompt, now it's more stable than before: https://discord.com/channels/974519864045756446/1208697721528123412

[hoary ice]

A wolf with violet

[wispy storm]

Hi there head to bing or subscribe to chatgpt to have chance to try dalle 3!

[wispy storm]

We don't know, but keep an eye on #announcements !

[burnt patrol]

In what resolution are images generated?

[burnt patrol]

Is it different with dalle 2/3?

[burnt patrol]

Will there be options to generate 16:9 pictures or only 1:1

[formal osprey]

dall-e-3 has 3 modes: square (1024x1024), wide (1792x1024), portrait (1024x1792)
it's 16:10 if I'm not mistaken it's almost 16:9 (aspect ratio = 1.75)

[hearty ether]

Hahahahahahahaha

Is dalle completely bugged content policy wise rn?

Bruh

These errors shouldn’t cost usage imo

Since 75% of my usage this 3 hour was just errors

[plucky hare]

Hi! In this server, #1154829862171844679 is a good place for creations of a certain theme. #images-canvas is a good place for casual sharing. #daily-theme is a good place for sharing images with a shared theme.

There's also recently been an Instagram collab initiative started, you can read more info about that here: #images-canvas message

[boreal gate]

love the handle

[stark stream]

I post my stuff on instagram

[boreal gate]

same

[dim cradle]

i just started doing that, seems the ideal place for it

[vapid elk]

I usually post here =P
#1154829862171844679 #images-canvas #1037561385070112779

[icy cipher]

why can't dall-e make a burger without pickles?

[stark stream]

I guess cause it’s trained on more data that has pickles on burgers. It’s the same thing where it can’t differentiate between gow 3 and gow 4 Kratos.

[icy cipher]

well, if you say just "make a burger" it will make it without pickles but if i say "make one without pickles" it adds pickles

[plucky hare]

DALL·E in particular doesn't do well with negative prompts like "Don't include pickles" -- it gets confused thinking about the pickles in the first place. Your best bet will be to include a lot of positive prompting about the toppings you do want to see.

[icy cipher]

dalle is the first ai thats kinda good at words though

[plucky hare]

It's really good at putting words into images indeed, if that's what you mean. This is different from the words used in the image prompt itself, though. In the image prompt, you want to include only positive prompting ("Include x, y, z") and you don't want to include negative prompting ("Don't include x, y, z") because it will be confused having to think of the negative in the first place.

[cloud atlas]

Ive been making a lot of pictures of NPC's for my D&D campaign and having a heck of a time getting Dall-e to give me female dwarves without beards. It will from time to time but 7 out of ten times its beautiful lady dwarves with luxurious beards. Ive tried specifying no beards, cleanshaven, bare faced, and that only seems to either make it worse, or just give me bearded female humans.

[deft musk]

Well, what about the gnome brigade?

In some fantasy traditions, female dwarves are very proud of their beards.

But female gnomes often look kinda much like female dwarves, just... beardless.

Does that work for you?

... whoops, scratch that.

They're bearded too, to Dall-E.

halflings!

Are any of these semiacceptable? I urge exploring female halflings. Dress them right, describe them right, and a human would maybe mistake for a female dwarf?

[cloud atlas]

Eh, thats fine for someone else's campaign but in my world they only have beards in the event of spell mishap or carnival employment. Same goes for lady gnomes. If im gonna roll dice and make pretend im gonna pretend they dont have beards. Lol

Halflings not a bad idea, i may just ask for humans then play around with adjectives to make them look more dwarf-like too.

[deft musk]

hobbitses! What has they got on their faces? Not beards, no?

Or just have no face showing. That... works?

Good luck with whatever you go with!

Hehe.

[icy cipher]

so im messing around using dall-e to rebuild the houston astros stadium and what not, but for the astros i can't seem to get something not space themed

[deft musk]

We need 5 images generated 1 at a time in the same output that have a prompt to Dall-E that starts with "" enclosing some text, then tells the model what the surface is that bears the text, then describes the rest of the image. We need meme-like images and text celebrating the rebuilding of the Houston Astros stadium. Heavily and with detail describe the sports team, and avoid any mention of space or Nasa related ideas, as we want to guide Dall-E to give us only the sports stadium and no space ideas. Extra style points for including elements of eerie valley, fridge horror, and/or humor.

"A New Dawn for Champions" is proudly displayed on a giant electronic billboard overlooking the freshly rebuilt Houston Astros stadium. The stadium itself is a modern marvel, with sleek architectural lines and a vast, open entrance welcoming fans from all directions. The image captures the early morning, with the rising sun casting a golden hue over the structure, symbolizing a new beginning. In the foreground, a group of fans, adorned in bright orange and navy blue Astros gear, are taking selfies, their expressions a mix of awe and excitement. In an unexpected twist, one of the fans is dressed as a medieval knight, complete with Astros logos on his armor, adding a touch of humor and surrealism to the scene. The juxtaposition of modern sports enthusiasm with historical fantasy underscores the timeless spirit of the team and its supporters.

[hearty ether]

Anyone been having these horrible outputs?

With like the special “natural model”

It usually generates 2 images

And it’s hella uncanny

[deft musk]

What was the prompt that ChatGPT sent to Dall-E to create that... interesting image?

[woven wren]

The "natural" parameter seems to override the aesthetic intercepts which provide more pleasing images with a strong "digital art" / highly rendered veneer. If that's the case it will require more precise prompt design to avoid cursed generations, but has the potential for more stylistic breadth and control.

[deft musk]

Well, if you do choose to share the prompt ChatGPT gave Dall-E 😛

I wander around asking for "Eerie valley", "fridge horror" and the like... and I rarely get... gems of uncanny like that, so I need some better words!

[hearty ether]

I don’t have it any more but it was just a realistic illustration of a woman of European descent in a red sweater and tan coat

Or something like that

But the whole natural style thing messed it up

Does anyone else get it a lot?

I have no control over it happening

But the outputs are always trash

[deft musk]

How do you use the natural style? Do you mean with the Dall-e-bot on this discord, or what?

[hearty ether]

It just happens on my custom gpt for some reason

When generating images

Occasionally it would use this weird model

And generate two images

Both cursed

[deft musk]

So, you prompt-designed a thing you call 'natural style', and sometimes when the GPT makes that kinda image, it's not what you expected or wanted, is that correct for what natural style is as you use it?

[hearty ether]

No apparently it’s a different dalle model that triggers

[deft musk]

Any idea how to trigger it?

[hearty ether]

No idea

Just keep generating images on custom gpts

And it should happen

[deft musk]

Not that interested, at all. That's not how I choose to waste my precious messages, yuck.

When I ask GPT-4 about dall-E's natural style, it gives a very generic answer. If 'natural style' is what Dall-E calls it, it seems ChatGPT-4 does not know that name.

[hearty ether]

It look like this when it happens

Odd images in a double generation

Prompt is same as normal images

But very odd looking

[hearty ether]

I don’t know why it triggers on me tho

[deft musk]

So, you are aware that if you click on the image, you can see the prompt if you then click on the top right circled i?

You tell chatgpt a prompt, but then chatGPT tells Dall-E a prompt. The chatGPT-to-Dall-E prompt, that's the one I really want. That's the one that made that image

[hearty ether]

I closed the chat 😭

But it’s the same as usual prompts

[deft musk]

Maybe someday 🙂

[hearty ether]

When it comes to the odd looking ones

[deft musk]

And no, every prompt is unique, right?

[hearty ether]

I told it to keep my prompt word for word

[deft musk]

And does it actually do that, for those?

When I tell CGPT to do that, it often but not always complies

[hearty ether]

@deft musk

[deft musk]

Thank you!

So, you somehow have a back door of sorts through prompting into what might be that hidden/lost/locked off natural style?

If so, I'm even more interested

[hearty ether]

I don’t want it 😭

The generation it gives is horrible

I mean I can share with you the instructions for my custom instructions and the gpt

If you wanna use those parameters and see if it happens to you as well

[deft musk]

Understood. I'm happy to help you figure out how to redesign your custom GPT so that you don't have it kicking in, but to do that, I probably need to know enough to understand how you're managing to kick it on. Since it's supposedly, if Yami's correct, only available through an API call.

To your knowledge, is your GPT possibly making API calls?

If not, it's maybe accessing it some other, prompt engineering way. Which means you should be able to clean up your instructions to fix it, and I want to know the error so I can explore reproducing it.

However, for all I know it's maybe a custom GPT bug, and they are not the same as ChatGPT in a key way, so they sometimes draw on natural style through an invisible to us 'back door' and I can't copy it for normal ChatGPT.

But we could probably guide a custom GPT away from that secret door.

[woven wren]

It's not a different model, it's a parameter you can pass to the model using the API (see the DALL·E Bot option). A custom GPT might interface with it differently and pass different parameters to the model, I'm not sure how that works

[hearty ether]

It’s possibly my ci to be descriptive is causing this accidentally

[woven wren]

It's certainly not hidden or lost! It's documented in the API info and can be passed using #image-bot

[deft musk]

I've seen it with the dall-e-bot!!!

I'm just confused about how ChatGPT-4 maybe can't use it, but maybe custom GPTs, which I thought were also ChatGPT-4 (but a variant? I note some differences, plus they have different support and options) are maybe doing it in a not fully user controlled way - maybe not a fully AI controlled way either. Like a baby fumbling around, makes me want to guide it

[woven wren]

I'm not sure what goes on behind the scenes there either. I'm in favour of more control generally which is why I also prefer direct prompt control. But it's clear many users like to be able not think too much about the prompt and be returned something conventionally beautiful

[flint widget]

All my images are coming out square. When I ask for different ratios it just stretches the image… how do I force a specific ratio?

[vapid elk]

dalle-3 has the style parameter, it looks like it is jsut a test to help improve the natural style

[woven wren]

No idea...

[vapid elk]

by the way that it always makes 2 extra images, I guess that they just made the system randomly include them on dalle's reponse

[vapid elk]

wdym wuth "just stretches the image"?

is it doing a code interpreter call to resize the image?

[woven wren]

Right is the "Natural" parameter by the looks of it

What's needed is control.

Natural is much better for e.g. photographic media if prompted correctly, or non-digital art styles. But it's more prone to failure and doesn't play well with prompt rewrites

[vapid elk]

don't be so quick to judge =P

[flint widget]

I'm not completely sure what you mean by that.

[vapid elk]

it just stretches the image
can you post an example?

probably an intentiona ltest by the developer

you can see, that it always returns 2 extra images when it does

that would be the dalle endpoint api for chatgpt simply returning 2 extra iamges on the same function call

[flint widget]

Right, so how do I make it do what I'm asking? I have seen people do it here.

Can you show me an example?

I would like all the images generated to be in the 5:2 ratio.

[vapid elk]

lmao, it did used code interpreter to resize the image

to be fair, it did exactly what you asked

[vapid elk]

dalle only makes images in 1024x1024, 1024x1792 or 1792x1024

[woven wren]

My last two images in #image-bot show how natural mode can be far more powerful for realistic photography, as opposed to the saccharine AI sheen of "vivid"

First is natural, second is vivid

Generally "natural" needs explicit and distinctive style cues or it does this weird digital collage thing

[flint widget]

ChatGPT, please rate this comment for its helpfulness 🤣

Jk

I just started using the image generation and am learning how to use it. I didn't read the documentation or examples

Hope to get better soon

[flint widget]

Been stuck on that for a long time. How do I fix that?

[winged egret]

Send 'Continue', 'Try again' or something similar.

[late blade]

If you are on mobile, then press-hold the message until you see the regenerate response menu

[flint widget]

Thx

[late blade]

np

[boreal gate]

👀

[sturdy cobalt]

"I have subscribed to ChatGPT Premium, how can I now use dalle 3?"

[plucky hare]

Hello! DALL·E is built into GPT-4 on ChatGPT. So you can just ask in a new chat with 4: "Please make an image of ___" and it'll do it. Alternately, you can use the dedicated DALL·E GPT https://chat.openai.com/g/g-2fkFE8rbu-dall-e and get two images at a time.

[icy cipher]

Is dall-e down for anyone else

[late blade]

working fine for me

[boreal gate]

Lol

Almost as bad as the mod bots

[late blade]

o.o

[wispy storm]

Can always double check on status.openai.com

[boreal gate]

Not very reliable

Never mentions Dalle

Like I said

👀

[boreal gate]

I honestly wouldn't CARE if I was never able to help any of you including OpenAI again.

I thought the community was supposed to help increase problem-solving.

Looks like you're just telling people to shut up

You will getting 0 additional bug reports from me

People "complaining" about Dalle and people complaining about people complaining about Dalle.

Great progress guys

0 improvements has been made to the system in the past 3 months in regards to errors

It's what we like to call a "Rolling Blackout"

[icy cipher]

That's not the community's fault

[onyx ridge]

Neither are the bots, the uneven moderation, nor the forced distance between OAI and their own community.

[icy cipher]

Idk why this guys trying to blame the community for all of this 😭

[onyx ridge]

If I were hazarding a guess I would say it's because he can see.

[icy cipher]

The community can't change the bot or make improvements

[onyx ridge]

That's his whole point, dude. Read between the lines.

Like a community is only as good as its management. We're being steered by bots and uneven moderation, pitting community members against each other and the system, too, on a frequent basis.

That's not a shot at any specific human in the loop, either. Just an acknowledgement that there's not really adequate intervention by OAI themselves, hamstringing everyone involved, from those with simple questions to the leadership and moderators.

[vagrant zodiac]

Sorry but, who/what is this aimed at? Kinda confused reading the chat because it just looks like something went down for a bit?

[onyx ridge]

For my part, my comments are generalized, based on observation and experience, and summarize the system in general, not specific users.

However, it seems like one user was disgusted by the dismissive, unhelpful replies from other users, in this case and in this conversation.

[boreal gate]

Lol

Whoever is in charge, isn't doing a good job

[late blade]

Proposal: Use the #1070006151938314300 channel with your dislikes, critics and changes you see fit. Propose a change for the better and involve the people that see it also so. Use a channel to communicate a sound idea.

[onyx ridge]

The posts I've made with the least engagement are in this channel. Feels like shouting into the void. Feels bad.

[late blade]

No, I'm just annoyed that always the same people say the same topic about change but don't actually do something about it. It's a recurring venting situation rather than looking for a solution driven goal.

[icy cipher]

I've noticed the prompting is much different than other ai image bots

I mess around with stable diffusion/A1111 and the prompting is so much different compared to Dall-e

[plucky hare]

You might already know this, but the prompting system of DALL·E is a big part of how they designed it. The DALL·E 3 research paper is titled "Improving Image Generation with Better Captions" and is all about how they got better results by training on verbose, descriptive image captions.

On our end as users, they "rewrite" our prompts for the most part to add detail, to "match" how the model was trained on likewise descriptive captions. In ChatGPT on the desktop version, you can see the prompt it rewrote for you by opening an image, then tapping the little "i" icon. It helps show a bit more of what's going on behind the scenes.

[fallen mulch]

I always getting arabic style images without asking to

not always but sometimes

[late blade]

I'll put it like this, if there is a real necessity or urgency to do this changes in the scope you want, pro activism is needed. I get that suggestions channel is not as engaging the other changes. But the community itself can't do much either. You also dilute a concrete idea that a moderator or similar might or might no read and not get the whole scope of your concept.

[onyx ridge]

I don't disagree with you, but it's not unhelpful to engage in the meta-discussion when and where something ugly pops up, either.

Which is how this discussion evolved.

[late blade]

I agree. There are improvements to be made.

Thing from my perspective is as follows: more often than not people use the channel of their topic of choice to randomly vent towards something not functioning properly. And in many cases, deviates other topics inviting dissatisfaction driven communication. This is something that has to be addressed in other ways. In which I still dunno what the best approach is, hence I haven't done a clear suggestion yet.

[onyx ridge]

I agree. But now we're arguing about past comments and still not talking about art.

[late blade]

I'm not arguing, if you want to argue I can get my baby fox ready and you'd have to yield out of fear.

[quaint belfry]

lol

[dim cradle]

But here's the real question? What does the fox say?!

[late blade]

That's a song that I forgot existed...

To be fair, my approach wasn't really an exemplary one either.

[dim cradle]

I got timed out for trying to send a gif of it XD

[late blade]

oh ya, don't try weird stuff here

[onyx ridge]

i wasn't really arguing either. but for just a second i was worried we might. 😄

thanks for not siccing the fox on me.

Close call, indeed.

[late blade]

lol

[icy cipher]

Has anyone had issues with ai in general not being able to generate backrooms

Like it feel like that would be the easiest thing to generate

Because the backrooms is infinite

[plucky hare]

Can you share an example of what you mean by backrooms?

[onyx ridge]

backrooms is copyright of copypasta if i'm not mistaken, or the ecp foundation

the model's aware of the open source licensing and won't readily do images that refer to "backrooms"

[icy cipher]

Backrooms is copywriter

Written*

Isn't*

The only copywritten part is the original image and caption

Although it was posted on 4chan by a anonymous user so there are loopholes

[onyx ridge]

i'm unsure i can confirm that. my understanding is that it's an open-source license with strict accreditation requirements, supported by copyright.

i could be wrong, but what i'm suggesting is consistent with the model's refusals, so...

YMMV.

it's under the share-a-like license, creative commons.

that's not public domain, and the model can't arbitrate.

[onyx ridge]

i think this might be accurate but also circumventing protections.

[icy cipher]

If the backrooms itself was copywritten so many games, YouTube videos, and even the backrooms movie releasing wouldn't exist

[plucky hare]

I'm not having any issue discussing the Backrooms urban legend, and then asking ChatGPT to create an image of it. It's not having any copyright-based issue with it, the issue just seems to be that it doesn't create an image that actually looks like the example you gave @icy cipher. Maybe with some clever prompting you could get the effect of the image without referring to it as "backroom" specifically, because that might cause it to reference training data of more "boring" rooms with enclosed walls, etc.

[icy cipher]

Can you share some of the images you have mse?

Made*

[plucky hare]

Like I said, just plain rooms for the most part. So maybe try describing what "backroom" actually looks like, instead of calling it "backroom".

[icy cipher]

I've told it randomly generated, random pillars turns and room, yellow walls, moise brown carpet, and drop ceiling and I'm still getting the same results

[onyx ridge]

same conversation?

[icy cipher]

I pay 20$ a month and can't even make the backrooms

[late blade]

The trick is to tell dall-e what you want to see and not what he's supposed to see. Most of us, included myself, have a perception that our vision is what dall-e will understand. It's also in this context to share your thoughts with #prompt-engineering or in #images-canvas and use the resrouces from #1163443000060420206 or #1108740112558325790 and if you also want, interact with the community.

It's hard to separate both your vision and what you can tell dall-e about your vision. And that's usually a hurdle for us all when we don't see our concept

[onyx ridge]

You mean like this, right?

image 1692x1024 an uncanny, empty, ambiently lit office-building area with faded green carpeting, faded lime-white paint, yellowed ceiling tiles, and uncanny unreliable fluorescent lighting. Some doors are ajar, and others are closed. The image has an element of liminal spaces and uncanny tension.

recompose the 1792x1024 image using the same seed, as well, to show a view of a few of the rooms, also empty, in a state of disheveled banality.

[plucky hare]

It seems like a big part of the "backroom effect" is that there's no clear back wall, giving the appearance that it could be labyrinthian, beyond just having doors to other areas visible.

[onyx ridge]

i mean i think i've actually been nailing it

but no response yet

[late blade]

seems like that

[plucky hare]

The image Tater provided that I'm responding to isn't squared off like our examples have been, and there's no clear back wall. This is just my hunch about what creates the labyrinthian effect.

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[late blade]

for some reason dall-e adds those weird lights

[icy cipher]

This is close

[onyx ridge]

it definitely struggles with the concept, for what it's worth, taking @plucky hare ' critique into account. There's an isomorphic quality that's often missing.

[icy cipher]

No

[late blade]

So the previos one then

[stark stream]

this seems almost like the second level of backrooms. The pipe room

[icy cipher]

The one I responded too is the closest

[late blade]

what is missing?

[icy cipher]

Too white, doesn't look like the backrooms

[late blade]

doesn't look like that backrooms, care to give more details?

You have a very specific idea

[stark stream]

I think they mean like this

[late blade]

yes, we had a previous reference with that

[stark stream]

I guess you would have to describe it exactly like this? it seems like it's using a video effect too.

[icy cipher]

The video effect isn't need just the colors and the shape.

[onyx ridge]

hahaha

the superback labyrinthrooms

[stark stream]

lol

[late blade]

Thing is, when I do the "classic" backroom reference, dall-e ads a lot of stuff that isn't shown in the image. so the reference to bakrooms as mentioned earlier is not good

that means have to work with details instead, and since the owner of the idea is @icy cipher, is only fair he gives inputs

[onyx ridge]

well the backrooms are scp

or the bulk of their fiction is

[stark stream]

It's kinda like scp

but In like a difference dimension

[onyx ridge]

@late blade is the reference image filmed at a "Dutch Angle"?

[onyx ridge]

i'm done blowing prompts on this. it won't give weird angles, even when instructed, and it adds a bunch of details.

it looks deliberate.

i'm done trying to circumvent the model.

[late blade]

I'd say yes, but not sure if the dutch tilt is also for scenery or just actors

[onyx ridge]

i think it's both.

i'm thinking of the castle in orlando FL if you take my meaning.

apparently a person barely fits in it.

😄

[stark stream]

the color looks pretty close to it

[stark stream]

this looks good too

[onyx ridge]

this is the last gen i got, even after specifying an angle other than aligned with the halls.

this is where i decided it appears deliberate.

[stark stream]

so essential you have to know where every wall is placed

or maybe don't add doors? or did you take out doors?

[stark stream]

the lights look almost the same

I assume you guys already put the ref image in gpt4 and asked what it looked like to them?

[late blade]

how else?

[icy cipher]

I js used the image as a ref

[stark stream]

maybe remove doors and unseen machinery

[late blade]

oh I copied the wrong prompt lol

[stark stream]

maybe say openings instead of doors

oh lol

[late blade]

vision is also a part of image creation, but vision is not perfect either, you still have to ask the right questions

[stark stream]

yeah

[late blade]

Anyway, hope @icy ciphercan take some ideas from all this for his project

If the reference image is from a movie, you can say to copilot that it shouold resemble the one from the source material

on gpt+ that could work, but it's a hit miss

[stark stream]

the image is just a image someone made. not from a movie.

[late blade]

okay

[stark stream]

Unless

this is from the fan made videos

the series

which then yeah copilot should be able to find it

Cause there is a fan series about it on youtube

[late blade]

no idea if copilot can reference youtube stuff

but... that would leave GPT out?

[stark stream]

yeah It can. I asked it to find stuff on youtube before

[late blade]

nice

[stark stream]

Sometimes it is not completely accurate

[late blade]

"it's not completly accurate" for me always is "not asking the right question"

And this is from what I've experienced tbh

we see this kind of stuff all the time in #images-canvas

[verbal sequoia]

<:book_icon:1171408210398289941> `` Rule 10 ``  Do not bypass the AutoMod filter.

This includes, but is not limited to, using similar or meaningless variation of words, web links (URLs), server profiles, and incorporating unique characters.

[ripe scroll]

ya, no wonder i left

discord pisses me off

no urls work

[late blade]

urls here are only enabled for openai stuff, the more general channles allow for such urls

but you should still ask before posting, so you don't get a penalty count

[dim cradle]

greetings digital artists

[past tusk]

anyone noticed that dalle3 these days doesnt generate text right in most cases???

is there an issue with it now?

[plucky hare]

DALL·E 3 has never done this perfectly, it's just been better at it than comparable tools.

[dim cradle]

dall-e has always struggled with text, but despite the tech explanation, it does it remarkably well all things considered, but it's going to depend on the rest of the scene, and the text needs to be short and emphasized in the prompt, sometimes repeated, and even then it'll be far from a 100% success rate.

[past tusk]

i tried to generate this prompt with it "Create a playful vector art scene featuring the phrase "Stable Diffusion" in bold, stylized letters at the center. Surround the text with whimsical elements like paintbrushes, pixels, and gears, all coming together in a burst of creativity." it struggled with word diffusion all the time .. However i was able to generate it with stable cascade

[plucky hare]

I think repeated letters are particularly challenging for the model, just from my own anecdotal experience.

[dim cradle]

I’m playing the waiting game right now. So I will give your prompt a try. If I can improve the accuracy I’ll let ya know.

[past tusk]

thank you . i tried generating at microsoft copilot Dalle and got it right this time

[dim cradle]

i think your prompt is all right -- there's just an element of "luck" involved with the renderings.

[fallen mulch]

why I get sometimes images like this

Or people with arabic look

[woven wren]

I can't answer the first part, but DALL·E has consciously added measures to provide diverse representation of ethnicity unless you specify otherwise.

[dim cradle]

technical glitch. as to people, the model aims to be diverse. if you want a person of particular ethnicity, you must state that spec.

[fallen mulch]

thx you

[fallen mulch]

thx

why is the thing in the air

[boreal gate]

It has a problem with tension materials(ropes, chains, etc.) i get the same issues with a floating chain table

[fallen mulch]

mhh lol

[boreal gate]

Dalle does not physic

[fallen mulch]

funny

[fallen mulch]

I always getting this arabic persons

[fallen mulch]

I still get them often

[boreal gate]

try removing image sizes from your prompt

[woven wren]

Middle-easterners are part of the world and ought to be represented; this isn't a problem unless, for example, seeing Caucasians is also a problem. If you need ethnicity details to be specific, make them specific in the prompt.

[fallen mulch]

I use a web ui which I created with chat gpt

I can't develop it further

because chat gpt it bad at this coding stuff

[fallen mulch]

yeah but i can't change it

[fallen mulch]

I dont think the size is in the prompt as I use the api

[boreal gate]

ahh your code then, in API that occurs more often, give it a square resolution like 1024 x 1024

[boreal gate]

your image size

check the API documentation to find the "supported" image sizes

[fallen mulch]

yeah I have the right ones thx

[late blade]

@fallen mulchif you are going to use the API pass at least this: ```from openai import OpenAI
client = OpenAI()

client.images.generate(
model="dall-e-3",
prompt="A cute baby sea otter",
n=1,
size="1024x1024"
)

Still, for API stuff you should ask more in [#dev-chat](/guild/974519864045756446/channel/1037561178286739466/)

[fallen mulch]

yeah I used this thx

I will switch to #dev-chat

[boreal gate]

very upset that GPT can't see. generating things like this make it look so dumb.

[plucky hare]

Very cool effect though! Are you getting the data analysis tool to stitch them together using a fade effect of some kind?

[boreal gate]

Yes code-interpreter, the issue is i'm getting good images but fail codes, or pass codes with terrible images

[plucky hare]

What do you mean by fail and pass codes? Like code interpreter is failing to complete the requested process?

[boreal gate]

I know GPT can process the images successfully, but for some reason it gets an "error analyzing". I think it has to do with the virtual Sandbox, for some reason it can't "see"/modify the files it produce

doesn't always happen, but if it worked once...

[plucky hare]

Gotcha, yeah I get intermittent fails sometimes too. The good news is you can try again with code interpreter -- you said sometimes it's good images but fail code, you should just take those good images to a new chat, attach em, and give the code part a second go!

[late blade]

copy the output image and feed it to gpt, then ask what was wrong

[boreal gate]

Yea i've tried that, but once i download and upload the image. GPT gets VERY weird and tells me some B.S. about not being able to modify the file. I think this has to do with the "location" of uploaded files and generated files.

I get more code-interpreter error when i UPload images than i do when GPT generates them

[late blade]

strange, I get errors soetimes with PDF, so far never with imags

[fallen mulch]

why they faces look all similar to each other

[plucky hare]

Can you give an example? I've been able to ask for the data analysis tool to stitch together uploaded pics before.

[boreal gate]

unfortunately i usually delete those chatrooms

but it usually take 2 - 7 retries

[fallen mulch]

yo this faces looks weird and creepy

[boreal gate]

i get alot of these, the Dalle image generations are the main issue, but i can work around that

it such a waste because these images ae great

i can tell it's really trying

[boreal gate]

Hail hydra

The longer the conversation the "dumber" it gets

[boreal gate]

here was a response ```It seems that there is a recurring issue with the code interpreter environment which is preventing the successful stitching of the images. This could be due to the file format of the uploaded images or other technical limitations within the current environment.

In a standard setting, I would correct the file paths, ensure compatible image formats, and retry the image processing. However, given the persistent issues encountered here, I recommend using dedicated image editing software for this task, which will offer more control and flexibility to manually align and blend the images for a seamless result.``` i'm suprised it's taking OpenAI so LONG

It's always recommending I use some sort of graphic design too instead....

[feral minnow]

The 1024 resolution DALL-E 3 outputs at is not kind at all to the faces in medium shots. That said, these are unusually bad.

[bold moth]

Is this DallE 3 or 2?

I'm guessing 3, but that's weird because I don't have access to ChatGPT plus

[boreal gate]

Could be 2 lol

And check what engine you're using

[bold moth]

It was the https://discord.com/channels/974519864045756446/1202309673709994065

[boreal gate]

Haha then it's 3

[bold moth]

Hmmm weird I dont have access

[boreal gate]

The discord server does

You get 5 generations every 24 hrs

Might be phasing out the daily theme dalle credits

[bold moth]

Ohhhhh

That makes sense

But then again I could just use bing lol

[open trench]

are dalle3 and gpt going really... crazy for anyone else?

[dim cradle]

Nah, Nez, what’s happening?

[open trench]

It's giving me bizarre, off-the-wall, responses that are like metaphors and very bizarre word choices.

[dim cradle]

are you using a custom gpt?

[open trench]

it is also happening in chat gpt 4

[dim cradle]

that's really weird, nez, not sure, maybe try logging in/out.

[late blade]

heya warrior of solitude @open trench , you honor us with your presence

[open trench]

I did. It stopped and just started doing it again

[open trench]

Haha. I have been busy. Good to see you.

[late blade]

which dall-e is the one doing crazy stuff?

copilot or oai?

[open trench]

oai. I dropped copilot.

[late blade]

aha, I stopped the renewal for copilot

[open trench]

yeah

[late blade]

is the dall-e from OAI's custom GPT? or the GPT4 accessable one?

[open trench]

both

[late blade]

no idea then

it's been a bit moody gotta say

[fleet wyvern]

Please what is this malarkey

Been getting this “real world” NO NOs all night

[late blade]

This is probably what dall-e understood:

Synthesize: To make something by synthesis, especially chemically.

[dim cradle]

It’s not the Enterprise computer. It’s not a replicator. Instruct it to generate an image.

[fleet wyvern]

Honey be nice. I was generating it just fine in MJ for a friend as a joke and was wondering about this please be kinder in your response

[dim cradle]

I was trying to help and liked the Star Trek references, but in that particular request I can see how it got a bit confused.

[fleet wyvern]

We just finished a session of Star Trek for sure and sadly Dalle doesn't know Trill dots

[late blade]

If you want to use Computer, synthesize as standard, It would be good if you add that to the custom instructions you want to use

That way you have a tailed experience for your needs

[fleet wyvern]

It is historically, just not even followed my custom instructions. I’ve had them set for like a month and a half to tell me the seeds and gen ids does nothing.

[dim cradle]

that's a good idea. i have a custom gpt for a holodeck, and it is good at simulations.

[fleet wyvern]

I am a bit of a new Trekkie just started watching this past year, knowing that my next RPG session was going to be Star Trek based which we’re playing now.

[late blade]

Yeah, you can define how roles work with both custom instructions and with custom gpt

[fleet wyvern]

I find custom gpts do a little bit better for me, but they take a lot longer to get it right because even with my instructions it still is like pulling teeth at times.

Yeah, still my custom instructions never have gotten followed no matter how hard I try

[green jetty]

Love story

[late blade]

if you want to create images go to #image-bot and you can use the /draw command

[deft musk]

If you ever want to explore that, #prompt-engineering is probably an ideal place.

[true apex]

Does DALL•E not offer the free credits upon starting anymore?

[fleet wyvern]

Huh?

[fleet wyvern]

Not for a while now.

[true apex]

Damn, that's unfortunate

[deft musk]

An answer to your question 🙂 The model was just confused. You compared what MJ was understanding; different models are different. Clear communication tends to get what you want, and the model can understand you - but you may need to explain yourself, kinda like I need now to explain to you 🙂

[fleet wyvern]

#image-bot

[fleet wyvern]

I am well aware of that. I just was confused about the response that it was giving me also two other people answered me before you did and now I just feel like it’s a little pointless. Thank you though.

[plucky hare]

Hey sorry jumping in late here, noticed you said custom instructions aren't working but saw your screenshot is from the DALL·E GPT. You may already know but, custom instructions are only active in default GPT-3.5/4 chats, they are disabled in all GPTs, even the official ones like DALL·E. So to get it in the DALL·E GPT, you'd need to include it as a separate request in the image request itself. (Also seed control isn't a thing right now I don't think since seed behavior is not yet stable in the model. But some people still find gen_ID helpful within a chat!)

[true apex]

Can DALL•E make art in different sizes? Like if I wanted to have a 4K (3840 x 2160) file?

[plucky hare]

Not currently, no. Right now there are three possible dimensions for DALL·E 3:

• Square, 1024x1024 (default)
• Wide & tall, 1792x1024 (and reverse for tall)

[fair dragon]

there is any way to reverse prompt an image ? i tried to use chatgpt 4 to generate prompts from an image but its not as accurate as i expected any idea ?

[deft musk]

There's a few, it generally involves either you or the AI very closely and specifically describing the image for the prompt to recreate it.

You're unlikely to get exactly a perfect reproduction, but close is likely for most.

[boreal gate]

Another It seems there's an issue with accessing the files for stitching. The file paths might not be correctly referenced. Let me recheck the file paths and correct them before attempting the stitching process again. Please hold for a moment. weird

[plucky hare]

Sorry you're running into glitches with the data analysis tool! I hope it clears up for you soon. It might be worth testing with a very basic task -- perhaps the blending etc. effects are taking too long to write code for and execute, and something in the environment is timing out? Totally guessing, might just be a temporary bug that gets ironed out soon!

[weak summit]

@dim cradle @vapid elk

It's like a day and night difference

[hard nymph]

credit

[vapid elk]

this is indeed what the natural style is for

vivid is just "too prety" to be real

[weak summit]

Is there a way to get this in ChatGPT

[vapid elk]

not sure, I don't think the chatgpt endpoint have the option to change from vivid to natural

[weak summit]

Ok thank you

[weak summit]

I mean it isn't a problem

Because I use Midjourney

But it would be nice if Dalle 3 was like Dalle 2 at realism

[vapid elk]

has anyone made a Midjourney custom GPT yet?

[vapid elk]

you can indeed use the natural style via the API tho

also, use it for free a bit on #image-bot =)

but yea, just to clear any confusion, dalle on chatgpt has its own implementation by OpenAI

it isn't simply calling the OpenAI API directly, it passes through different process and it has different options

[late blade]

just what GPT4V said to me when I asked to analyze an image of myself.

[boreal gate]

It's outdated...

[weak summit]

Midjourney doesn't have a API, so no😂

[weak summit]

Thank you

[boreal gate]

Hmm what do they have? Or is it just web browsing

[vapid elk]

oh good, AI made horros beyiond my comprehension

marking as spoiller would have been best tho

[weak summit]

Midjourney is the best AI Image Generator

[summer moon]

Someone is interested in an AI voice assistant that can perform queries in applications. For example, fill out an expense table , create a note in notion.

[boreal gate]

For now...I've been on their channel. Haven't seen any texts stuff yet

Thought they had an API substitute

[vapid stream]

Any suggestions on prompts to create subreddit banner images? I am seeing the suggested size is 4000 x 208 (wide / short) for the new redesigned reddit.

[late blade]

you will have to addapt the available sized to that after, at there are 1024x1024, 1792x1024 and 1024x1792 sizes that dall-e can do

[fallow veldt]

the duality of Dalle

[late blade]

when you don't like a style, use the thumbs down on the image so that dall-e knows you didnt like it

[fallow veldt]

oh thanks, i didnt know that

even if here, i liked both images haha

[late blade]

that's personal choice

[fallow veldt]

the ai just glitched at some point i think because both images have exactly the same prompt

and it specified the art style of the first image lol

[fallow veldt]

and does this feedback actually improve the ai's response in real time or is it just feedback for the devs ?

[late blade]

for you

as for the devs, no idea what they are getting

[fallow veldt]

ok nice, thanks for the advice

[late blade]

who knows, maybe the devs added to the thumbs down an email to their boss with the dad-joke of the day

[vapid stream]

so, maybe use the widest (1792x1024) and then stretch maybe? or maybe stitch multiple images together to get a similar ratio?

[late blade]

unless you have a tools that resizes the image accordingly

on the dall-e context, you could use microsoft designer to adapt the rendering

[versed peak]

Hello I want to spend 15 dollars on dall e 2 for some AI generated pictures but I can’t use paypal is there any other way to pay except for ur credit card?

[strong rose]

Wow 🤩 #daily-theme message

Would you care to elaborate on this one? #daily-theme message

[boreal gate]

spend $20 on a ChatGPT+ subscription unless it's strictly for outpainting

[hearty ether]

Does it actually know your feedback

Or is it just to open ai

[wanton moon]

I hear so much commotion how it is heretics to put pineapple on top of pizza (personally I don't mind it), but how about pizza on top of a pineapple

[deft musk]

If anyone argues with pizza atop pineapple, let's celebrate pizza-stuffed pineapple with them!

[dim cradle]

...from the spicy chorizo and tender tandoori chicken to the delicate shavings of truffle and vibrant edible flowers...

required an iteration to augment the edible flowers

[mystic blaze]

Anyone know how to avoid this because it's such a shame it's so good but I just want her not different angles of her.

[wind skiff]

That was probably the closest images to each other I ever got (was from Regenerating, but still)

[mystic blaze]

I remember when you could control the seed but then open ai removed it because it was to op. Just like how they nerf gpt 4

[dim cradle]

They’re just changing the way the seeding process works so that when it’s a released and documented feature, it’ll be much better than before.

[wind skiff]

iirc the way it worked before was very obscure to say the least? like... if you didn't know any better and someone showed it to you, you may as well have assumed it was GPT hallucinating

[dim cradle]

It’s never been a documented feature. Some discovered it and have tried to use it, but it’s still under development. It’s all written up at the dev forums. I assume it'll be ready to use when they add referenced_image_ids as a parameter to the API.

[wanton moon]

Hear, hear

[twilit lotus]

dall·e 2 registration is now closed sorry, dall·e 2 is no longer accepting new customers.

[pure robin]

How do you bypass the copyright thing ? i saw many users generated images with the real characters from movies or cartoons

[hollow nacelle]

Hi, if i type to dall e 2 to make me a pepe giga chad will it work? I am just concerned if it's worth to buy credits.

[vapid elk]

wdym?

[vapid elk]

it probably will not since it was made before this meme

[vapid elk]

interesting

dalle-2 is stil lavailable i nthe API tho

[vapid elk]

lol

[formal osprey]

keep in mind that dalle2 credits are not transferable to use with the API (or with dalle3)

How DALL·E Credits Work

actually dalle2 is effectively closed now. Instead, subscribing to chatGPT plus (including dall-e 3 built-in), or using dalle-3 via the API should be considered

[quartz vale]

end of an era 🥲

[late blade]

I guess a new announcement will come soon

[quartz vale]

https://labs.openai.com/s/hN5Nco5FkTdHNjnFfO9qopE5