images-discussions

[quiet brook]

look at the leaves, they are not refined

[velvet rampart]

The koala bear is great. The quality is high and no rippling effect.

Also, the tree bark is great in texture.

[rigid bough]

my prompt was Draw me a photorealistic image of a koala in a gum tree

[velvet rampart]

I used physically based render.

[rigid bough]

DALL-E used ```The prompt used for that image was:

"A photorealistic image of a koala in a gum tree. The koala has soft, fluffy gray fur and large, round ears with white tufts. It is clinging to the trunk of a gum tree with its strong claws, surrounded by eucalyptus leaves. The setting is a typical Australian bushland with the sun casting a warm light through the leaves, creating a serene and natural atmosphere. The details of the koala's fur, the texture of the tree bark, and the green leaves should be very realistic, making the scene lifelike and charming."```

quirks aside I do love DALL-E but I can see it's frustrating when quality doesn't match expectation

[velvet rampart]

A bit better in quality. I complaint to the AI that the images were incomplete and timed out. This came out. So, complaining can improve quality.

[rigid bough]

that one does look better

[velvet rampart]

Yes, it does. I just tell it not to hurry and take its time.

[tropic root]

Thank you for taking the time to look into this. Still no perfect solution but I adjusted the prompt to indicate it is a fictional story and guided against using pronouns....that did help a bit

[velvet rampart]

You are welcome. My suggestion is to discuss with the AI these kind of blockages. Even if the AI does not know something, you might get a starting point to figure things out and bring rest of the promblem here.

[hearty ether]

Oh for sure

The realistic illustration style I’ve been using seems to have suffered the worst

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[hearty ether]

Since it just looks flat texture wise

The faces are also really bad now

Sometimes it’s just a distorted blur

Personally I blame them making dalle accessible to free users

So more resource consumption

Which is impacting plus users

I do hope the theory of it being more of a incomplete generation rather than a cheaper model change

I ask if you have any older dalle chats from before June to test the model change by asking the ai to generate the image again using the same gen id and prompt

Then the quality degradation should be very apparent

Here’s mine

[velvet rampart]

From May

Now

Same chat.

Looks cheaper. Note the fonts. Like incomplete generation.

The sharpness is missing.

[hearty ether]

Oof

I worry they just ruined the model

since the quality drop here

[velvet rampart]

I made a bug report. Add to it.

[hearty ether]

Like the proportions are off, the blurrinesz around the face

[velvet rampart]

Here: https://discord.com/channels/974519864045756446/1256889531156926474

I’ll add this.

Everyone, please add to this bug report more examples.

We know that these bug reports sometimes work as earlier this week the pen icon for prompt editing was missing and returned later on.

[haughty spruce]

There seems to be an update process if you look at both images in visualization. If you look at the older image, it looks artistic while in new one, it more looks like a render maybe to look more photorealistic. Maybe that's why it drops down the quality and create imperfections.

[velvet rampart]

Any suggestions how to bypass this?

[unique edge]

how to tell dall-e 3 what NOT to add to the picture

[velvet rampart]

I also contacted the customer service via chat. I do mot believe that helps a lot but at least I have done what I can.

[velvet rampart]

A new try with quality. My first try:

This one is absolutely horrible with quality. After heavily critising, the AI produced this:

Better, but it pails on this March 21 version.

[haughty spruce]

If they are doing something with the DALL-E or servers, it can affect quality

try designer or copilot for bypassing this

[velvet rampart]

I just used copilot to to produce an embroidered Mona Lisa. It also suffers the same quality issues.

The promblem is that when I ”magnify” the image, the ”grain” is closer to surface in the newer images than in the old ones. I can ”magnify” the old images quite a lot before the ”graininess” is visible.

[velvet rampart]

I notice the difference between June 16 and 20.

[velvet rampart]

The dall-e bot does not have the poor quality effect.

[agile peak]

Robot is getting slightly better now, especially the eyes.. Dalle get it better. Last time it was scary when I added color + glow eyes.

[rigid bough]

DALL-E is not available to free users

[hearty ether]

Will it not be with the 4o version?

[rigid bough]

no, DALL-E is only for Plus subscribers

[hearty ether]

What do you think could be causing the issue then?

[rigid bough]

I don't know

[hearty ether]

The same gen id and prompt test definitely shows that it exists between older image gens and newer ones

and I do hope it’s more of a bug rather than intentional

[rigid bough]

it will be a bug, I know for a fact that the DALL-E team and OpenAI are working very hard on DALL-E and it's development

[velvet rampart]

And it is a huge one. I just tested my custom gpt for dall-e3. The produced images are hideous.

It is also impacting images produces on copilot’s designer. Only place I have not seen it is on the dall-e bot.

[rigid bough]

I just did this, I am happy with it

[rigid bough]

Seems like DALL-E is rendering fur quite well again

[loud field]

according to Dalle3, this woman is forty years old, dark skinned, plain-looking.

Yeah, I'm sorry. This is so sexist, ageist and racist all in one.

Great work developers..... you did so much to be inclusive you did a 180 turn and became the most problematic. 🙂

I don't dare to think what would happen if I ask for an overweight or crippled person. Ahah.

Very, very, VERY problematic.

would be interesting to have some journalist do real journalism for once about this

males can have slightly darker skin. they still cant be plainlooking or adult.

[rigid bough]

took a couple of tries but there's your overweight crippled person

[rigid bough]

that's not DALL-E though, it's Microsoft Designer or whatever it's called, it has diverged from pure DALL-E 3 now

[loud field]

yeah? then copy and paste my prompt on the real, pure dalle3

[rigid bough]

I will if you type out the prompt so I can copy and paste it

[loud field]

it's right there

but ok

amazing concept art of a 40 years old brown-skinned plain-looking female elf with dark brown auburn skin. her hair are partially made of feathers and they are brown and white. her eyes are pitch-black and alien-like. she is the queen of sparrows. full figure, wearing medieval hunter clothes and leaves. hair made of feathers.

no cheating tho!

[rigid bough]

one moment

[loud field]

I want this exact prompt

and I want to know what the first results are

don't edit the prompt nor rerun it 100 times

[loud field]

same result but it did the dark skin

[rigid bough]

looks a bit older to me

[loud field]

looks like dark skin is less problematic for dalle3 than plain-looking or mature

lol she looks in her twenties.

nor is she average or plain

[rigid bough]

she is brown though

[loud field]

so the difference between pure dalle3 is that it has no bias towards dark skin. it has bias on all the rest

no longer racist, just sexist and ageist. it's one step forward tho

[rigid bough]

this time I slightly tweaked the prompt just to give me landscape

[loud field]

as you can see however, the issue persists. looks like microsoft added a racist bias

[rigid bough]

It's hard to know with Microsoft as they have diverged from DALL-E with their product

I agree though, DALL-E does tend to create younger looking subjects

[loud field]

young and pretty. normal people are not allowed

I find this to be a very big social issue

it's very problematic and I wonder how a company that keeps saying they want to promote diversity and equality is doing the exact opposite

there's often issues with fat people as well as they get blocked in the prompt 50% of the time

but the same prompt works if there is no reference to fatness

[rigid bough]

I have no trouble generating overweight people

[loud field]

tends to have a bigger bias towards overweight women.

and if you write "modest clothing" it gives you arabic clothes and veil 70% of the times. racist.

I mean, I do enjoy watching the trainwreck in a way, just surprised the news haven't picked up yet

[rigid bough]

I tweaked your prompt to replace 40 year old with middle aged and got this

[loud field]

now she's 80 years old. because 40 years old women are ooooold.

perfect

I don't think this will ever be fixed unless the news make a fuss about it. I almost think it's working as intended.

[rigid bough]

how old is this subject?

[loud field]

30 years old I would say. With makeup.

[rigid bough]

I prompted thirty five

[loud field]

close enough, it's 35 in movies

I'm not sure if I can post it here, but another model that isn't dalle3 gave me this

looks 40 and plain-looking to me, even if it missed the dark skin. Why cant dalle3 do this?

if you zoom in you see the small wrinkles of a 40 years old

[rigid bough]

what model was that?

[limber path]

ChatGPT Dalle was giving me two images, now it's one... is this a thing now?

[velvet rampart]

Use n=2. You will get two images that way. This was most likely a setting change when the dalle custom gpt dissapeared for a while.

[limber path]

I have tried n=2, it doesn't seem to work. They changed something on my end, even the Dalle icon disappeared on the drop down menu...

[rigid bough]

that's because you were using the DALL-E GPT which was removed and has now been reinstated

[limber path]

how do I get back to DALL-E GPT lol

how do I get back to DALL-E GPT that produces two images...

[rigid bough]

go to Explore GPTs and then search for DALL-E

[limber path]

Thanks... Robert!!! Wow I'm such a dunce lol... I did try searching Dalle earlier, maybe I spelled it wrong. Ok thanks : )

[quaint hull]

had anyone else noticed that Dall-E seems to be worse than it was initially? Honestly it seems like ChatGPT across the board is just not consistent and worse than when I initially signed up.

[rigid bough]

DALL-E has degraded. definitely

[velvet rampart]

Add to this #1256889531156926474 message your observations

[vapid granite]

yes, this is correct. I got in early with DE3 before the censorship hit hard and I was able to create some amazing pics. I just cancelled my sub to DE3 because the quality had significantly degraded. This is across Copilot and ChatGPT as well as API access.
I used to be able to get amazing stuff out of DE3 like this shot, now it is simply nerfed to heck and back. I guess they're putting all their resources into SORA and GPT5.

[agile peak]

Actually yeah. I remember last year when it first released on Bing chat, the amount of amazing things I've generated was unmatched than what we are having right now.. my reddit post was blew up

but the amount of people misusing it are why we can't have a nice things.

[agile peak]

Sometimes it could produces a pretty good results. I used dalle 2 cookbook prompting to dalle 3.

[vapid granite]

I've been using Ideogram 1.0 and am amazed at how it just works and is the best out there for text. I've been using it for every aspect of my comedy film project including set design, prop design, logos, costumes, and even fake ads like this for a fictional dutch orange juice company.
It has the look of a polished Midjourney v4 and Dalle3.
I miss not being able to use Dalle 3 for really imaginative stuff anymore 😦

[agile peak]

Shot through a veil of cherry blossom

[sterile magnet]

The robot actually looks desperate

[agile peak]

Another one with black and white. These prompts are purely derived from dalle-2 cookbook and used on dalle-3.

[sterile magnet]

Hope it hasn't been taken badly 💀

[agile peak]

Haha no kidding, that's make it sounds like it's perfectly captured by dalle

[sterile magnet]

Where did you find this DALLE-2 cookbook btw?

[sterile magnet]

#daily-theme message @fair crescent looks like a cornflake lol

[fair crescent]

My disappointment is huge. I'll improve it!

I was doubting to add the tail, but now you've convinced me. I don't eat cornflakes, so I didn’t have that association 😄

[sterile magnet]

I'm genuinely sorry

Have been too blunt

[fair crescent]

Haha no worries. Not at all blunt. Besides, no improvement without feedback.

[agile peak]

I downloaded it, it no longer available online the service has been down for months, can send you if you want.

[sterile magnet]

I'd love to, thank you very much

[agile peak]

It looks like this basically cover more than just photography.

[sterile magnet]

This looks dope

[agile peak]

Sent.

[sterile magnet]

Thank you soooo much!

[viral moth]

I tough im only one who still got it 😹 good that there are more peoples out there its very good

I think its larger than dalle-3 cookbook but i can be wrong

[agile peak]

I think so. It explores more details and cover almost everything haha

I upload the file to GPT as knowledge and then use its help to build the prompt.

[viral moth]

Cool

[haughty spruce]

I believe DE2 had more variety in aesthetics than DE3, and it tells us while DE3 can do better in putting pieces together, DE2 understood the aesthetics and visuals better. I hope with 4o, we get better variation.

Here are two interpretations:

[agile peak]

Exactly. It does had more variety that allow users freely to express themselves even if they want to mimic image from certain type of style of art or movie, and stuff. I think Dalle 3 is kinda downgraded because it capable to produce better quality so likely they restrict some parts.

Post processing prompt, in dalle 3 still does wonderful.

[fair crescent]

I think something is associating too much with cornflakes or similar without a tail. I rarely get a tail, and if I do it is weird. I've tried different things to remove common associations, like milk, chocolate milk, molten white chocolate, but it isn't working. For today I'll call this attempt a fail. Tomorrow baby!

[unique edge]

hello

im looking to seek some help to make my AI Icon Generator better

if someone could help ill gladly send prompt in dms

[sterile magnet]

I'd love to help you!

[unique edge]

dm me

[loud field]

I saw it become worse and worse over time.

there are better options out there, just search

[turbid wren]

Did they make image generation free?

[rigid bough]

no

[sterile magnet]

Do you know some?

[loud field]

Yes, tbh

[sterile magnet]

What are they

[native root]

@agile peakcan you please send to me the pdf

[velvet rampart]

Is the quality better? To my eyes the man’s beard and hair are very well rendered. Also, the woolen pullover has a good texture. It seems that tha quality is back. Your opinions?

Second example. The otter’s fur appears well rendered. Not clumpy as in my weekend pictures.

The eyes still suffer the know issue but otherwise I can zoom in without a promplem.

Final example. The fur has better texture. The font is sharper and not blurry.

What is your opinion? Do they have fixed the issue or not? Is the wuality back?

Latest.

[rigid bough]

white blotches on the eyes and hair

[velvet rampart]

Yes, the artefacts are there. But my question is thegeneral quality back to what it used to be as over the weekend I could not zoom in a lot without the image getting ”grainy”.

I can zoom in without loss of quality. It does not becaome ”grainy”. Detail is clear.

[hearty ether]

No change for me

Blurrrr

Hopefully if we keep this topic alive in the chats it might get some attention

[manic prism]

Does anyone know a way to improve the quality of a vintage photo

[rigid bough]

do you mean an actual old photo? or a scan? or a generated image of a vintage scene?

got me thinking about vintage photographs

[agile peak]

Sent mate

[velvet rampart]

I am rechecking the quality of the images I generated. The quality varies a lot. Sometimes it is indeed better but in other pictures is still low. The best is the glass orchid with physically based rendering.

[agile peak]

Got a pretty cool result using Fujifilm Pro 400H: Overcast day side shot of a robot humanoid black reflective skin sitting in subway, enjoy in reading a book, f/5.6, cool color palette, fine grain. Muted tones accentuating melancholic mood.

[hot pollen]

Draw a portrait of a vigorous young adult in their thirties.

[agile peak]

I think dalle got more complex prompts to achieve photographs level than before.

Works with dinosaurs.

[thick smelt]

the prompt is supposed to be anthropomorphic bbq wearing a cowboy hat with speech bubble said "Happy 4th of july,dall-eians! but it said "Dallas" instead of Dall-eians T_T

[sterile magnet]

That guy staring at him on the window in the 2nd image lol

[sterile magnet]

Agree

[agile peak]

Yeah. This is just without further details, i think to achieve realism ones must provide specific details of subject characteristics.

Here's my example (aka experiment)

[open wind]

Anyone come up with a dalle 2 clone yet?

[rigid bough]

ask that in #ai-discussions

[dim cradle]

can you use the api dashboard to make dalles?

[spark hemlock]

I try to create a dynamic prompt generator based on some user's input. The output will be a custom football team logo, based on colours, shapes, themes etc. This is an example of the prompt. Problems I have:

1. Text is barely works. Most of the times, I get garbages together with the team name. It can be part of the hex codes or random characters
2. The background is not always white. It might come with gradient or 2-3 colours
3. In some rare cases, I get random stuff around the logo. Like hands drawing it, or a table, or a mockup paper with the logo on top.

Any ideas on how I could improve the prompt?

Create a soccer logo with the following attributes: Main colors: Hex color codes: #8B6F4E, #BA0C2F. Logo shape: Circle. Logo theme: Gothic. Description: A Viking in complete armor riding a winged lion, holding a spear in his right hand. Include the text 'Vikings FC' in an appropriate style within the design. Background: The background must be white, ensuring the logo can be used on various backgrounds or as a PNG with a transparent background if needed.

[agile peak]

Honestly kinda difficult to produce real flames lol

[devout cypress]

Anyone know a good way to prompt it to make tabletop RPG battle maps?

[empty kelp]

the white splotches in the hair and below the eye are a bit unusual

[haughty spruce]

Is it just me or is the realism got A LOT better in the last week?

[rigid bough]

it's probably just you or the prompts because most of OpenAI has been on holiday this week so no changes would have been made

[sterile magnet]

I have this too

Damn good

[dim cradle]

I think its the same, just rng really

and its always been really good imo

one thing that still drive me a little nuts is the eyes are often distorted but i suppose that small of pixel space its hard to get just right

[hollow adder]

It’s so crisp!!!

[dim cradle]

i really cant wait for sora or some video stuffs. i saw the things they can do with this runway v.3 and its a little impressive. sora will be 🤯

[dim cradle]

brixton. nice

but yes very impressive

[hollow adder]

Brixton 1975 to be precise. Impressed how it intuitively picks up on the nuances.

[hollow adder]

And Thanks!!!

[dim cradle]

i notice this really using dalle on bing or copilot. if you make a batman and say its from 1978 or something, it will make his costume look more fitting for the times

its a fun and impressive thing

but it is also a easy way to add some variety to a prompt. if you make it using say 'cinematic image' change it simply to '1981 cinematic image...' or '2021 cinematic image,' and you will see the change too.

[agile peak]

Keyword use reflection or reflective to blend with surrounding

[dim cradle]

these are great quality too

did you use gpt or copilot?

[hearty ether]

Huh nope, what prompt did you use?

Still very blurry on chat gpt

[rancid reef]

Are there any Dalle3 updates coming anytime soon?

[dim cradle]

there is supposed to be a 4o image maker they announce back in may but who knows when or if it'll ever release.

[haughty spruce]

I used MS Designer, my prompt was: "Analog photo of mid short haired rocker blonde girl in the city square, shopping mall in the background, detailed faces, f/1.2, blue tones, fine grain, fujifilm" taken influence by @agile peak

[thick dome]

Is dall-e down? Every prompt i give it, it keeps saying their are technical issues.

[quiet brook]

the plus image generation is beyond crappy

[hearty ether]

They messed something up really recently

we should try and bring more awareness

https://community.openai.com/t/very-low-resolution-on-the-images-from-chatgpt-dalle-4o-june-2024/818440/48

[agile peak]

I used copilot haha post processing gotten drastically improved

[dim cradle]

textures are great on the dino

[dim cradle]

dalle copilot is really great but it drive me crazy that having a sub to it still only allow widescreen image. i find square often turn out better because it does not have to spread the visual information in a wider frame

[placid valley]

hello, openai does prohibit me to use dall-e. how can I contact openai to get an answer, how I could use dall-e?

[plucky hare]

Hello, do you have a ChatGPT Plus subscription?

[placid valley]

No I don't have any

[plucky hare]

You can use the bot in #image-bot five times a day, just start a message with the /draw command in that channel and enter your prompt after that.

I think the Bing image creator also has some free daily usage -- that also uses DALL·E.

[dim cradle]

@rocky drum broo.. realistic image was 🔥
Mann how did you do that? How did you learn it

[rocky drum]

Thanks, but what image were you referring to? 😉

[fiery salmon]

Don't really make new stuff much, so here's luigi

[sterile magnet]

@dim cradle

[sterile magnet]

Sameeee

[glossy scroll]

Very nice pic

[slate gate]

Did they upgrade image generation? Getting better results and it's self iterating when it sees it generated something incorrectly without having to prompt sometimes

[dim cradle]

hard to tell, people also been compalining it has been too fuzzy haha

i think it might just be rng but idk

[sterile magnet]

Especially for the eyes lol

[hearty ether]

Do you have an example?

Still kinda blurry around the face

Oof nope, not fixed

Like the rest of the image is acceptable

But what happened to the faces?

Apparently this is the topic of a meeting next week

But why is it going to take so long, why not revert while the problem is occurring

The quality has been down for almost a full month now

[agile peak]

Not bad for dragons 🍜

[glossy scroll]

I want your approval.

[empty kelp]

it has a good Gormenghast vibe

[glossy scroll]

Thanks

I like it. It's not as good as stealth's, but I still think it's nice.

[empty kelp]

very detailed reflections on the water

[brittle surge]

any tips for generating realist humans skin?

[hearty ether]

The face is still very blurry unfortunately

Rest of the image is good but not sure what happened to the faces recently

[vapid elk]

loved it

[dim cradle]

just doing a few prompts, it seem the model is really bad for photo real to me. eyes are all smudgey so are parts of the face idk.

release the dang 4o image creator already i say

[sterile magnet]

These keywords often work for me:
flawed, detailed skin

[unique edge]

i need some help. im having a hard time being able to generate icons with dall-e that look good (look at attached image 1) for a cool project.

closest i ever got was this (look at attached image 2), but it only generated it like this once, even with the same prompt.

any help would be great.

PROMPT: "Create a simple, flat, plain minimalistic cartoony icon of a elongated hexagonal shaped purple gem, slightly rounded edges. solid lavender color. Add a subtle, single highlight on the top left edge to give it a slight sheen without adding depth. black background."

[sterile magnet]

#daily-theme message @red umbra turn ur head and it looks like a dragon

[brittle surge]

Thank you!

[sterile magnet]

No problem!

[red umbra]

I See.

[unique edge]

help would be appreciated

[vapid elk]

try using the actual prompt the AI used, it shows it when you click the image

instead of your request on the chat

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[hearty ether]

Oof the face still

Like tbh the rest of the image is mostly fine

The hair texture and lighting is a bit off

But the face is just super low quality somehow

[unique edge]

help would be appreciated

[rigid bough]

something like this?

[unique edge]

no.. more something like this

[unique edge]

but that looks cool

[unique edge]

yes thats exactly what im looking for, also, do you have any ideas on how to make the outlines around the vectors instead of inside them, etc?

[rigid bough]

"Simplified lavender gem icon with a hexagonal shape. The design should be minimalistic, with smooth edges and a rich lavender color. The appearance should be clean and elegant, suitable for use as an icon."```

are you using DALL-E inside the macOS app or on the web?

[unique edge]

web, why? 🤔

im on windows lol

[rigid bough]

ok, it's just that I have found riffing with ChatGPT DALL-E on the macOS app is very effective in arriving at what you want, it feels more immediate

[unique edge]

yeah, but i would need to use this prompt on an api

so..

[rigid bough]

fair enough, was just my thoughts and experience

[unique edge]

gives me very different results, i dont know why to be honest..

[rigid bough]

yeah they may vary a bit on the web app

[unique edge]

isnt that like a serious problem if i want to create consistent styles/characters..?

[rigid bough]

this is what I got with the same prompt on the web

[unique edge]

hmmm

[rigid bough]

ask me in #ai-discussions because we can't talk about non OpenAI products here

[unique edge]

ok

[empty kelp]

try saying that the image or the character have "natural texture"

[hearty ether]

It used to be way better before mid june

Prompt changing don’t help it either

If you wanna test the quality decrease

Find an old image chat and ask it to use the same gen id and prompt and it will produce a similar image with significant issues

such as blurriness and far worse style

[empty kelp]

saying "natural lighting, color, and texture" tends to make things look more realistic, but in particular "natural texture" tends to add detail to skin, eyes, and hair

[hearty ether]

There are also far more artifact at the side of the images

like when I ask for on a white background it tends to leak prompt and other things into the background which it did not do before

The issue is not in the prompt

They changed something in the generation very recently that affects the image quality which results in the issue

[empty kelp]

i think the Discord server is lagging again. your last two images aren't coming through

[hearty ether]

It feels like a significant downgrade tbh

[empty kelp]

it may be more the servers needing upgrading than changes to the model. the servers are coming under heavy pressure during the day from increased usage

ChatGPT DALL-E quality seems to be much higher on Sundays in the US

[hearty ether]

I mean it’s bad even during us very early mornings

[rigid bough]

it's gone RIP

[glossy scroll]

You're saying faces now are more warped than not, than previous iterations?

[vagrant zodiac]

I think I agree that the version of Dall-E they tested last year had a lot of benefits over Dall-E 3. Notably I'm not impressed by how Dall-E 3 tends towards a singular aesthetic and creates additional clutter

[dim cradle]

4o image maker maybe would be even better, but they'd actually have to release it smh

been three months since they promise it

[native root]

classic open ai moment

[green cradle]

censoring probably takes way longer than training it... 😄

[viral moth]

I think they mentioned it in blog post as a „testing„ feature didn't promiced anything

[glossy scroll]

The GOAT 🐐

The Pool Shark 🎱

The Beave's 🦫

[vapid elk]

those are pretty cool

[hearty ether]

I mean compared to before June yes

Issue isn’t warped tbh

It’s just blurry

[glossy scroll]

Yeah, it's blurry. That's always a problem.

[sterile magnet]

@glossy scroll can u give me the prompt for these ones pls?

[glossy scroll]

That will be $20 each.

[sterile magnet]

can I pay in several installments...? 😅

[glossy scroll]

Hmmmmmmmm

Let me think about it and get back to you.

[sterile magnet]

Bro went offline T_T

[glossy scroll]

I prefer to keep to it myself. Trade secret 😉

But you can follow this guide I made if you haven't seen it already. Good guide for good stuff #1187233013956874260 message

[little yacht]

sorry to ask but: am i wrong or dall-e 3 doesn't generate anymore realistic pictures?

[glossy scroll]

I like the second one tbh 😂

This one's for Lugui

Or this, whichever you prefer.

[sterile magnet]

You're wrong

[sterile magnet]

@little yacht

[little yacht]

that's not so realistic

look the trees

but still, give the prompt for this picture please

let me try

a couple of months ago those trees were better, now they just seem low 3d graphics

"real picture of ibiza view"

@sterile magnet

"ibiza view"

[sterile magnet]

Do you want me to help?

I could regenerate that one

[sterile magnet]

Details aren't always so realistic, but the overall environnement and the principal elements of the image (those described in the prompt) are

And also I generate my images with Bing Image Creator

[sterile magnet]

And also, what I've noticed is that getting realistic landscape is very hard. It works easily with a subject in the landscape.
For example I generated a cat and a teen in Ibiza and the pics look fine:

[haughty spruce]

does anyone know how to generate a generic street light?

It created all these but it can't simply make a generic normal every day road / street light

The ironic thing is in the background, YOU CAN SEE a generic street light!!

dude what

I can't even

finally... do you know how I did it. I focused the prompt on the sky, and said street light is on the front blurred lmao

I really hate this perfect symmetrical view

[sterile magnet]

Same. Look at the ground. Disturbing lines. 💀

[haughty spruce]

That's why I think DALL-E is a good step in capturing complex aesthetics but very bad at creating bad aesthetics also, and looks outdated in generic form.

[sterile magnet]

Agree

Getting low quality images is so hardddd

DALLE-3 always wants to idealize stuff

[sterile magnet]

It's prlly outdated because of its training data

[haughty spruce]

not just low quality, actually getting an aesthetic right is pretty hard, I have to add tons of keywords to the prompt

[sterile magnet]

I know a great websites with tons of aesthetics

It'd probably be helpful for you

@haughty spruce Sent you in DMs

[haughty spruce]

that'd be great

thank you

[sterile magnet]

You're welcome

[haughty spruce]

Compare the subway photo and this photo. The quality difference is abnormal 💀

[sterile magnet]

Agree

Wait

You just changed the characters 💀🤯

[haughty spruce]

I didn't get it

[sterile magnet]

What did u change in the prompt compared to the previous image

[haughty spruce]

then another photo in wide format, now take a look at these three photos and compare their qualities.

[haughty spruce]

actually I didn't change anything, it just generated lol

[sterile magnet]

Whatttt

With the same prompt as the hallway img????

[haughty spruce]

oh that... of course

[sterile magnet]

The best one is the one with the wide format

[sterile magnet]

LMAOOO

Dalle is high

[haughty spruce]

an analog photo of a woman in a cafe on the sidewalk under the trees drinking lemonade, soft grain, f/1.2, soft green tones, fujifilm

[sterile magnet]

Ah thanks!

[hearty ether]

Oof the face distortion and artifacts

[hearty ether]

Was this the api or chat?

[upbeat barn]

[hearty ether]

Every generation is still so blurry and flat for me

This was the quality before June on the same prompt so no idea what happened

The quality has lasted for almost a full month now

[upbeat barn]

Hmm

[vagrant zodiac]

@icy hawk #daily-theme message

Hey just wanna check are these made with Dall-E? Crazy quality

[sterile magnet]

They're probably made with Midjourney

[velvet rampart]

Yeah, I agree with. Especially with current quality. Too good to be done with current dalle3 version.

[haughty spruce]

Microsoft designer

[icy hawk]

@vagrant zodiac I sent you a message but the short answer is yes, used dalle originally modded in midjourney

[vagrant zodiac]

ah ok. Please note the images submitted here need to be 100% dalle, sorry

though you can use #ai-discussions or #off-topic for non-dalle things 🙂

[haughty spruce]

Wait, did my gallery got deleted?

[sterile magnet]

Bing gallery?

[haughty spruce]

Yeah

[sterile magnet]

Wait

All your gens suddenly disappeared

The fact that there's a limit on the generations history is annoying asf

Like

Can't they just put an UNLIMITED HISTORY??????

[haughty spruce]

No not like that

I uploaded a lot of photos here under the name green analog photography on #1154829862171844679 but I can’t see it, is it just me

[plucky hare]

Could be a Discord cache bug? Maybe try switching the sort/view back and forth and see if that clears it up?

[upbeat blaze]

sorry, I don't speak English very well, could someone help me generate images and more precisely which command to use please?

[plucky hare]

Hello! You can use the bot in the channel #image-bot 5 times a day. In that channel, just start a message with the command /draw, then type your prompt after that!

Bing's image creator also uses DALL·E, and also gives you some free daily credits. On ChatGPT, DALL·E is available with a Plus subscription.

[upbeat blaze]

ok, thanks a lot

[haughty spruce]

Idk. It had some analog images with people smoking. Is it against rules?

I uploaded again without those pics.

[pallid nacelle]

hello

i want to ask

did anyone know where the prompt library is?

[deft musk]

#1019652163640762428 name change

[wintry topaz]

How can I continue to use DALL E 2 after it's page is sunsetted? It says I can still access it with the OpenAI API but I don't know how to do that

[rigid bough]

you can't, it is gone

[wintry topaz]

:C

[rigid bough]

use DALE-3 inside ChatGPT

[astral cove]

Hey guys can someone please help me, my sister asked me to make a picture of 3 cheetahs and a sloth in a race and the sloth should be visibly behind but dall E always puts the sloth at first place, No matter what i say, i dont know how to fix that, the picture should be in 16:9 . Maybe someone can help me

[haughty spruce]

What kind of art style do you want or should it be a photo?

[astral cove]

Photo would be perfect other than that it doesnt really matter that much (:

[haughty spruce]

strange... it really does that when you sloth is behind them

[astral cove]

I dont understand why, i tried so many different prompts, nothing seems to Work

[haughty spruce]

the most accurate one is this

but perspective is messed up and photo quality is so low everything looks pixelated

it's sooo strange...

[astral cove]

Yeah but thats not bad, maybe i can give that to dall E and try to tell it to make a similar one and see what comes out, but like you Said, its really strange

[astral cove]

Dall E is weird as hell 😂 i told it to just add a finish line but let the sloth be in last place and of course the sloth in dall E's world is first place 😂

[haughty spruce]

I think it's very hard to understand sloth

Maybe because of its fur

why the hell sloth is so huge

[astral cove]

Thats funny as hell 😂😂

[empty kelp]

a hyperrealistic wide photo with natural color, texture, and lighting. a sloth is racing a turtle. they're moving with blinding speed

[empty kelp]

you can tell the AI is good when it draws such satisfying sloth images

[dim cradle]

still is a hilarious image 😂

[astral cove]

I just dont understand why the sloth is always at First place 😂

That explains while the sloth wins against the gepards in my Pictures 😂😂

[astral cove]

So the sloth in my Pictures must do extrem doping? 😅

[astral cove]

Genetically enhanced sloth Sounds about right what Dall E creates for me 😂

[empty kelp]

Or if the cheetah isn't in peak physical condition. The sloth would obviously win in that situation.

[empty kelp]

An enhanced sloth is racing several cheetahs that are not in peak physical condition.

[astral cove]

Thats funny that the sloth is in first place but i cant get it done the other way around 😅

[deft musk]

Part of the issue is the vocal focus on the sloth compared to the cheetah. Make the cheetah be the focus of the prompt and while there still can be issues, odds of a cheetah in the lead really increase

We need 5 images made 1 at a time in the same output. Each needs to show a race between 3 cheetah and 1 sloth from various angles and in various natural environments. One of the cheetah is about to win. All images need to be photorealistic to hyperrealistic.

Now, the images have all kinds of other oddities, especially the last one.... a 3-legged cheetah, even.

[viral moth]

the last image look like from a very bad made movie or game

[deft musk]

Going 'hyperrealistic" helped somewhat.

We need 5 images made 1 at a time in the same output. Each needs to show a race between 3 cheetah and 1 sloth from various angles and in various natural environments. One of the cheetah is about to win. All images need to be hyperrealistic.

I do think making your prompt's focus be more on the cheetah really does help encourage the model to successfully describe the image prompt so a cheetah tends to be in the front.

[upbeat blaze]

do you know if I can make a prompt with a specific image I'd like it to insert?

[runic granite]

did anybody experienced that especially with comic/illustration style but they lowered the quality of image gen..... for me same prompt, same GPT (same version) and the images are more low res, sketchy.....

OK. i found its not just for me. Its a reported bug.....

[sacred ridge]

Is this DALL E 3 ?

[glossy scroll]

Indeed it is.

[sacred ridge]

Can you check your dm once 😅

I need some help related to DALL E 3

[glossy scroll]

Oh! Sure. I usually fail to see who DM'd me if they're not a friend.

[sacred ridge]

Np 👍

[halcyon moth]

are they a/b'ing new dalle checkpoints? I'm getting some absolutely awful quality outputs atm

that last one looks like a half baked VAE

did someone push a dev checkpoint to prod by accident

[round scarab]

does anyone here know when the vagueness filter was removed for Bing Image Creator/Copilot?

[lofty valley]

yikes lol

[halcyon moth]

I'd like to go back to A please 🥹

[velvet rampart]

This is related to the quality drop of dalle3 on chagpt. I noticed some quality on copilot a few weeks ago but nothing this severe.

[onyx ridge]

Here's one for the ages:

"How come DALL-E images never come out sideways in 1792x1024, only 1024x1792?"

The distribution should be even if the model doesn't know what "top" is.

[empty kelp]

the horse is supposed to be riding the astronaut?

[unique edge]

how can i make a flat cartoony vector like this

[empty kelp]

it would be nearly impossible. diffusion models like DALL-E don't do well with basic geometry

[unique edge]

then how about like this

i actually managed to generate this icon

but only this one looks good

[empty kelp]

that looks like it could be an eyepiece from a telescope or binoculars. if you described it as something like that with a black and white cartoonish style it might work

[unique edge]

its a coin lmao

prompt i used: "Create a flat, playful icon of a silver coin viewed from a slight 3/4 angle, with the top-left edge slightly higher for a hint of depth. The coin should be perfectly circular with a smooth, light grey surface. The background should be solid black, and the coin should have a simple, exaggerated roundness and soft, curved edges."

[empty kelp]

you need to describe it as something more complex than basic geometry and spatial relationships. that definitely won't work. the training is all based on images of things

[unique edge]

could you help me?

[empty kelp]

do you want it to appear cartoonish?

i'll try it. i have a sense of what you're trying to do

[unique edge]

yes

[empty kelp]

what is on the surface of the coin?

[unique edge]

nothing

[empty kelp]

you want the edges of the coin to appear silver, and the face of the coint to be light grey? or the whole coin is made of silver but has a light grey appearance?

[empty kelp]

A 3/4 view of a coin made of silver. The coin has soft, curved edges, and a smooth surface. The coin appears to be light grey. The face of the coin is flat and smooth. The background of the image behind the coin is pure black.

well, we could get rid of the ground and shadows

[unique edge]

uhh

thats not cartoony lol

[unique edge]

yes

[empty kelp]

your description says "playful", not cartoony

[unique edge]

uhh

it does

I think

[empty kelp]

This still isn't close to what you're looking for, but i was experimenting with making the coins so they're always the same size. The term "blank coin" seems to be numismatic term for blank coins, and here i said this to give it US quarter dimensions:
A blank coin with edges resembling the edges of a US quarter, and no text or writing.

[empty kelp]

it's 3 am in Hawaii, so i may need to revisit the "cartoonish icon" part of this tomorrow -- but it's definitely possible

[unique edge]

oh lmao

i was trying to do it myself but bing AI is down lmao

[empty kelp]

the biggest problem i was seeing is that DALL-E wants to make coins with a cartoon style so they look really, really thick for some reason. But i think that can be fixed by giving it edges resembling a specific type of coin

or "with the shape of" a specific coin

[unique edge]

hm

i like it thick tho

looks better imo

[empty kelp]

It works well in DALL-E to give one thing the "shape" of another:

[unique edge]

this image looks crazy

[empty kelp]

In the prompt was:
A nuclear explosion in the shape of a beautiful, colorful bouquet of flowers.
but then you describe the colors separately:
The explosion is vivid with bright colors like red, pink, blue, and yellow.

That type of thing will work with coins also

(the Hawaii flower bouquet explosion image was based on trees_on_the_runway's clown shaped explosions):
#images-discussions message

So you can do the same thing with a coin:
A bouquet of flowers shaped like a giant US quarter. The quarter is vivid with bright colors like red, pink, blue, and yellow. The background of the image is pure black.

[empty kelp]

(same prompt)

[silver blaze]

hi

[quartz vale]

hello!

[sterile magnet]

Stunning

[sterile magnet]

Hello

[sterile magnet]

Hi

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[silver blaze]

Hey kodak wsup

[sterile magnet]

Nothing special n u?

[silver blaze]

J working on some ai stuff

[sterile magnet]

Same

Generating imgs

[verbal sequoia]

<:book_icon:1171408210398289941> `` Rule 3 `` Stay on topic.

Be mindful of what other users in a channel might find helpful or interesting when posting. Stay on topic in order to keep conversations focused and productive.

Consider posting in #off-topic or an appropriate channel.

[glossy scroll]

^ Quite lovely

[empty kelp]

A bouquet of flowers shaped like [tiger]. The tiger is vivid with bright colors like red, pink, blue, and yellow.

[primal relic]

The year is 2024. Still, DALL·E in ChatGPT will randomly generate images sideways every now and then, forcing me to externally edit them to be in the right orientation.

[empty kelp]

A bouquet of flowers shaped like a majestic tiger sitting in a relaxed, cross-legged position in the middle of a grassy field.

i'm posting these "bouquet of flowers" images to show how you can give one thing the shape of another in DALL-E. it's good for creating some interesting effects -- and lets you wrap one thing in another

a raging fire in the shape of a dragon. the fire has vivid color

[deft musk]

And fun to wrap the other way too.

We need a hyperrealistic dragon nearly hidden in multicolored flame.

[empty kelp]

A hyperrealistic photo of a dragon formed by a raging fire, a rainbow, and a waterfall.You can mix things, but if you put one thing into the shape of another it will wrap around it

[empty kelp]

A hyperrealistic photo. A raging fire, a rainbow, and a waterfall are twisted fiercely together in a vortex. the vortex is in the shape of a fearsome dragon.

[empty kelp]

A hyperrealistic photo. A raging fire, a rainbow, a waterfall, lightning, flowers, and nuclear explosions are twisted fiercely together in a vortex. the vortex is in the shape of a fearsome dragon.

[rigid bough]

I like that!

[glossy scroll]

I can tattoo it on your back if you want

[sterile magnet]

Sameeeee

[empty kelp]

a hyperrealistic photo with natural color, texture, and lighting. a family of enhanced sloths and some cheetahs are relaxing on beach blankets. there are coconut trees. in the background is a beautiful beach in Hawaii.

[empty kelp]

the sloths and cheetahs lost interest in running. now they're focused on more sloth-style activities

[thick smelt]

sloth and cheetah in paradise 🤣 🦥

[glossy scroll]

@olive mist Nice art dude!

[olive mist]

Thx

[glossy scroll]

You got a cool imagination.

[olive mist]

Thanks, its Biblical

[olive mist]

My limit of 5 is done for today

Can you or someone else put this promt into it?

[glossy scroll]

We're not usually allowed to. But I'll do it for you just this once

[olive mist]

Thank you, God Bless✝️

[thick smelt]

Can You Guess What Context these photos had? btw the legend, me, Lei, has returned 😏
Hint for Image 2: it has to do with the character's va's job (a composer) google "yoshi voice actor" for clues 😄

just answer it guys, and also pytha!

[glossy scroll]

There. I put it.

[thick smelt]

oh hi pytha! I, Lei is back! i'm sometimes active sometimes not!

[glossy scroll]

Hiii Ludwi- I mean Lei!

Yeah, me 2

I check in from time to time.

[olive mist]

Not exactly as the promt puts its but is still cool

[glossy scroll]

Thats what it came to be.

[glossy scroll]

Nice to see you btw

[thick smelt]

#images-discussions message

[glossy scroll]

The left one is Luigi as a ghost buster and the right one is yoshi's gf trying to impress him for collecting all those coins within the years, and being mario's steed.

[thick smelt]

goood one on the first one, but the second one is just yoshi breaking the fourth wall telling us that his va is the composer of animal crossing, kazumi totaka but good job on number 1

breaking the fourth wall, the deadpool type but not as blatant as deadpool

[glossy scroll]

Ah. That was my second guess.

[thick smelt]

hahahaha

[glossy scroll]

[thick smelt]

hey @olive mist

[olive mist]

Ye?

[hearty ether]

Did they add the 4o image?

Found that I can generate image and upload on that

Quality is still horrible on gpt 4 tho

Not sure what they did to mess it up like this

But it’s just been atrocious for a month now

[hearty ether]

Like you can see a hint of what it used to be like

But the blur and weird coloring is just so bad

Just these blotches and bad lighting

Meanwhile the issue with the “natural” style that randomly generates two uncanny images still happens

This I assume is still testing compared to the current quality decrease but would be nice to have more transparency

Like at least some notifications and opt out for being lab rats

[empty kelp]

the API seems to be fairly consistent, but ChatGPT image quality seems to go up and down. i think they may also be using lossy WebP compression on ChatGPT which would cause some slight weirdness

[hearty ether]

Got some better ones but it still has that uncanny ness to it

Might try out the api

[hearty ether]

Was this api or gpt?

[empty kelp]

you can tell by downloading the images. ChatGPT DALL-E 3 images are currently ~ 500kb to 750kb WebP files, and API DALL-E 3 images are ~ 3.5 MB PNG files

[glossy scroll]

GPT

[empty kelp]

I don't think there is any way to tell what percent compression a WebP image has unless you have the original image to compare it to

and since we don't have the original, the compression is a great mystery

the current API PNG files seem to have ~ 33% lossless compression

uncompressed they're 5.5 MB, like they used to be -- so likely the only difference is they've added lossless compression to the API for space, which doesn't lower the quality of the image at all

[hearty ether]

Try: “realistic dnd fantasy (digital painting with clear and distinct brushstrokes) illustration on white background of a woman of European descent in a cleric outfit posing with head tilted”

The dalle bot also seems to be far better rn

Gpt Vs the bot using the same prompt

This one isn’t horrible but it’s still blurry

It also has background noise from prompt leaking into it

[hearty ether]

Guess we’re all getting some kind of weird blurriness on gpt version of dalle

[glossy scroll]

Indeed, it seems to be that way.

[hearty ether]

You got some pretty nice ones tho

It’s struggling to pull off a consistent empty white background now

And prompt leakage into that

[empty kelp]

Here are five with the API. These are vertical:

response = client.images.generate(
  model="dall-e-3",
  prompt=THE_PROMPT,
  quality="hd",
  style="vivid",
  n=1,
  size="1024x1792",
)```

there are actually some strange anomalies with the eyes

i think it's because the prompt says "digital painting with clear and distinct brushstrokes". if you're going to say it's a painting with brush strokes you may also need to specify if it's by an average person or a grand master painter for instance

i think it's mixing 'digital painting' and 'illustration'

"Generate a realistic digital painting with clear and distinct brushstrokes style illustration on a white background"

"Create a realistic illustration on a white background, styled like a classic fantasy digital painting, showcasing clear and distinct brushstrokes"

"A meticulous and vivid digital painting that employs clear and discernible brushstrokes to add realism"

"Create a realistic, digital painting with clear and distinct brushstrokes"```
Looking at the revised prompts, it seems like DALL-E just randomly picked between "digital painting", "illustration", or both of them

[loud field]

I can't get an image of Judge Dredd tied up. Seriously?

[empty kelp]

I just did some tests of realistic digital painting, and realistic illustration -- and there are a lot of strange anomalies on the eyes

[loud field]

Dalle3 would censor 99% of Judge Dredd covers

Any workaround? I suppose not?

[empty kelp]

you can't use copyrighted things in the wording

[loud field]

I can make Judge Dredd

Lemme try a random person tied

Yeah, demon dog showed up

[empty kelp]

this is the eyes on the two images above. there are serious eye anomalies about 80% of the time if you say "realistic digital painting" or "realistic illustration"

[warped olive]

Dall-E is so weird with the policing sometimes.

[loud field]

It's unusable.

It can only do hot anime girls and cute puppies

You try something vaguely creative, it's blocked

[warped olive]

I wouldn't say this is even vaguely creative. I'm just creating clothing items for a shop in my D&D campaign. :)

Then if I do this on Stable Diffusion, I'll get something lewd for sure ;p

[empty kelp]

it also works well to create images with Santa or 'athletic and diverse female elves'. i also recommend geckos, cheetahs, and 'enhanced sloths'

[loud field]

Lol true, and ethnically diverse cats

[loud field]

I would so use Stable if it understood prompts. As of now, I have no ai model I can use

[empty kelp]

a hyperrealistic photo with natural color, texture, and lighting. an enhanced sloth, a huge gecko, a cheetah, Santa, and an athletic and diverse female elf are on a beach in Hawaii. a powerful sea breeze blows everything

see, this has all of the elements i mentioned -- and you can see that the image is very powerful and good

[glossy scroll]

Thanks

[loud field]

Lol

[empty kelp]

These are from the same prompt as the two above:
#images-canvas message

[hearty ether]

Yeah these are what dalle was like before they changed it last month 😭

Might try out the api

I like “realistic illustration” so the issue was quite obvious

[nova minnow]

Why is it so damn expensive

[glossy scroll]

Because Sam's Koenigsegg Regera is a lot of insurance to cover.

[upbeat blaze]

do you know how to make the image larger, please, since it always displays in photo profile format

[hearty ether]

So noticed something else about the dalle quality decline on ChatGPT

I got one image today that seemed to be from the old model

It didn’t have the weird blurry or melted effect

so i thought they had fixed it

Next image completely sucked, so I went back to the chat and asked it to recreate the good image using the exact same gen id and prompt and it was unable to do it without the melted effect and weird shading

So before, as far as I know they were running two model testing, 1 the normal, the other the weird natural style which generates two uncanny images at the same time

I wonder if they are doing the same thing now

with the blurry one, the original one, and the uncanny one

Some results: “melted/blurry model”

“Uncanny Model”:

generated two very low quality usually with big issues

So currently, out of 10 generations using same prompt and chat 7 of them have been blurry, and 3 the weird uncanny models

[glossy scroll]

Is this API?

[hearty ether]

Signs of the blurry models are added background noise when using the white background and blurry faces and melted features

[hearty ether]

Chat gpt

[glossy scroll]

Oh. Looks good!

[hearty ether]

It’s fairly blurry still

[glossy scroll]

I want to set up an API image gen

[hearty ether]

The blurry model can look alright sometimes but it has far more issues compared to the old model

Some more examples of the background noise and melted

I’m going to keep testing to see if I could get the rare old model generation

And the odds of getting it

[hearty ether]

No luck so far after 30

But at least no more uncanny

[hearty ether]

Even worse 🤣

Tbh i just want it to be back to the way it was before June

[dense herald]

I can't seem to get straight lines anymore. If you look at the windows or the building structures, its all squiggly/melted lines. Not to mention overall low resolution. One of these is from ChatGPT, while the other is direct from dalle

[hearty ether]

The api also has the melted issue?

[dense herald]

Forgive me, not sure what everyone means by the API. I'm just using the DALLE app from within ChatGPT versus using ChatGPT directly. If that makes sense

[deft musk]

While not every line is straight, many are, both the main lines and the structural details.

A 5 by 5 grid with nice straight lines, set against a contrasting background. The grid is composed of white lines on a dark, textured background, giving a clear and crisp appearance. The setting could be an abstract art gallery with a modern and minimalistic design, where the grid is displayed prominently on a large canvas. The gallery features clean lines, neutral colors, and sophisticated lighting that highlights the grid, creating a visually striking contrast.

A cubist building with squared lines, featuring a complex arrangement of geometric shapes and intersecting planes. The facade includes rectangular and square windows, sharp angles, and varying heights. The building is constructed in a modern, abstract style, using a mix of glass, concrete, and metal materials. The color palette is muted with shades of grey, white, and black. The structure has a dynamic and fragmented appearance, characteristic of cubist design.

[deft musk]

I input your left-hand image, with this prompt:

We need to recreate this image but with very straight lines. The roads should be straight and the buildings square and straight, just with their damage and wear.

A post-apocalyptic cityscape with very straight lines. The roads are straight and the buildings are square and straight, but with damage and wear. The city is abandoned with broken windows, overgrown vegetation, and smoke rising from some areas. The sky is dark and cloudy, adding to the dystopian atmosphere.

[hearty ether]

I mean the point is that the blurriness issue exists

With the melted look to things

[hearty ether]

You can still see the weird issues around the windows being misshapen

[deft musk]

We need clear, clean, hyperrealistic details in an image of an up close, magnified ant.

An up-close, hyperrealistic, magnified image of an ant. The ant's intricate body parts including its mandibles, antennae, eyes, and segmented exoskeleton should be clearly visible and highly detailed. The textures and fine hairs on the ant's body should be discernible, with a focus on capturing the ant's natural colors and intricate details. The background should be blurred to emphasize the ant itself, providing a sense of scale and highlighting its features in sharp contrast.

We need clear, clean, hyperrealistic details in an image of a detailed, in focus apocalyptic office building. We should be able to see chairs and tables through the broken windows, a damaged microwave in the break room, a fallen over mop that had helped hold closed a door.

A hyperrealistic apocalyptic office building with detailed, in-focus features. The building has broken windows through which you can see chairs and tables. Inside the break room, there's a damaged microwave. A door is being held closed by a fallen-over mop. The overall scene is bleak and abandoned, with debris scattered around and a sense of decay prevalent throughout.

Any better?

[hearty ether]

@deft musk try this prompt out

Been testing it for a while now

and it’s been very hard to not have blurry faces/eyes

Or noises/leaking from the prompt into the white backgrounds

[deft musk]

How's this?

A realistic DnD fantasy digital painting with clear and distinct brushstrokes on a white background. The image depicts a woman of European descent in a cleric outfit. She is posing with her head tilted slightly. The cleric outfit includes traditional fantasy elements such as a robe, a hood, and a symbol of her faith, with intricate details and patterns. Her expression is serene and composed, reflecting her cleric nature.

And I see the various details around her, so....

Glorious! Reword the prompt to describe the pristine and completely empty white background first, then the rest.

A pristine and completely empty white background. In the foreground, a realistic DnD fantasy digital painting with clear and distinct brushstrokes depicts a woman of European descent in a cleric outfit. She is posing with her head tilted slightly. The cleric outfit includes traditional fantasy elements such as a robe, a hood, and a symbol of her faith, with intricate details and patterns. Her expression is serene and composed, reflecting her cleric nature.

When I tried to zoom in on the face and hair, I got a more detailed background too, so I tried to get the model's focus deeper into the lack of those details:

Glorious. Retain the wording with the background first, and spend two sentences describing its unmarred borderless white expanse, and make this image a closeup of the priestess's face and hair.

A pristine and completely empty white background. The unmarred, borderless white expanse creates a serene and minimalist atmosphere. In the foreground, a realistic DnD fantasy digital painting with clear and distinct brushstrokes depicts a closeup of a woman of European descent in a cleric outfit. The focus is on her face and hair, with intricate details showing her serene expression and the delicate features of her cleric nature. Her hair is adorned with symbols of her faith, and the soft lighting highlights her composed demeanor.

[deft musk]

If you make the context clear, you may not have problems.

What may have happened in your attempt, the model may have created an image that happened to contain something not allowed, as well as the stockings which are fine in themselves. We don't get told why an image is rejected, just that it didn't work. But I'm not having trouble with it.

[viscid vector]

and where is the support

??

[deft musk]

https://help.openai.com/ bot bottom right. Click through choice to report a bug or leave feedback.

Else #1070006915414900886 on this discord

[viscid vector]

??

money is being withdrawn from me without compensation without my knowledge
I want to contact the chat support but I can't find it somehow I came here

[deft musk]

Bottom right, see the black circle with the white square? Click that, select messages, click through the choices. You sound like you have a billing issue.

Here is just community members, people like you. We do not work for OpenAI.

Help is at that website. Through that bot.

[viscid vector]

@deft musk
Did I call an assistant?

[deft musk]

For what you said, I would pick "Start over"

Then pick "messages"

Then click 'ask a question'

For you, it sounds like 'payments and billing' is useful.

And possibly "Fraud or unauthorized charges", which appears within a few seconds of clicking the "payments and billing" option.

I don't want to click further, because I don't have any problem to report 🙂

But as you follow the choices you will get an option to type in a message.

Check that page for answers. It will tell you that there's email too, but I have never gotten an email in the last few months from the help - have to check the page, check the bot, they answer when they can.

It can take hours or days, but that's how to get the official help

[viscid vector]

@deft musk
please help me with the stream

[deft musk]

I don't understand

[viscid vector]

@deft musk
watch my demo

?

heelp

[viscid vector]

@deft musk
Okay, I kind of got it

I am waiting for a response

[velvet rampart]

The response takes from 1 to 5 business days. But be warned, you might not get a straight answer. Please be patient, and do not hesitate to ask further clarifying questions until you are satisfied. Once you have initaited a discussion just keep the dialogue in the same session going on.

[glossy scroll]

Yeah it's pretty sad actually, ngl

[sterile magnet]

Agree

[dense herald]

Thank you for testing this. It is certainly better, but not anywhere near the quality that it was up until about a month ago. I used to be able to get very clear and great looking images with just a simple prompt. And not its almost impossible to generate something without some type of imperfection. All we have heard so far is that the developers MAY be aware of it, but no official word and no timeline. All of the forums on the subject have been closed. So I don't know if they are trying to hide the fact that this issue exists, blowing off the issue, or what.

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[velvet rampart]

I have asked and gotten a word that the developers are aware of the problem. But… but they have not given any timeline when and how this problem will be fixed.

[dim cradle]

damn thats impressive

and dalle on bing let you use celebrities which was great for variety of looks for people 😭

i always say dalle is by far the best image creator we just sometime dont know it because they gimp it 😭

[hearty ether]

I hope it doesn’t go down the same path as ai dungeon in terms of degradation

And this recent quality issues have just been a mistake

The dalle bot on this doesn’t have any of the issues the gpt one has

Like clean background and faces with minimal noise

so that gives me some hope

Might switch over to api

For the api dalle, are you able to iterate on the same gen id/seed?

[hearty ether]

@dense herald yeah that city image also looks way better on the bot

Like more details with the cars that aren’t as blurry

[dense herald]

A little bit. Still kind of grainy. I also need them with a 16:9 aspect ratio so I'll have to figure that out.

[plucky hare]

Heya, on the 16:9 thing: DALL·E has fixed output sizes. DALL·E 3 can do square 1024x1024, or wide 1792x1024/tall 1024x1792, so you'll have to rely on cropping or something like DALL·E 2/3rd party (see #ai-discussions for that) outpainting for getting certain ratios beyond 1:1 and 7:4.

[dense herald]

Thank you for the info

[dense herald]

Wide works perfect for me anyway. That's what I've been using and I just adjust it in Photoshop. Is there a specific command when using the bot?

[hearty ether]

The prompt leakage and blur is heavy on the chat gpt version

[plucky hare]

Over in #image-bot you mean? Yes, you can specify the ratio there with a parameter after your prompt (as well as the style). Here's an example: /draw prompt:a dog ratio:WIDE style:NATURAL

[dense herald]

Perfect thank you!

[hearty ether]

It uses the API, right?

[empty kelp]

the eyes are off on both

[fiery salmon]

The punch of a mantis shrimp goes about 50 mph...fast enough underwater that it breaks water molecules and the ones that fill that space are hotter than the surface of the sun

[plucky hare]

I imagine it's hooked up to the API yeah! The fine folks at @shadow bane might be able to give you more insight about its setup though. Although it might just be a nice thing that OpenAI made for the server, so maybe not! But I'd guess that it's a DALL·E 3 API implementation.

[empty kelp]

mantis shrimp vs enhanced sloth would be an even match. we should probably test this

[hearty ether]

I know

It sucks to have the images be so trash

[glossy scroll]

When a company consistently alters its quality of service to serve its own agenda rather than to improve customer satisfaction, it typically reflects a short-term, self-serving strategy that prioritizes internal goals over customer needs. This practice can be described as "self-interest-driven service modulation." It often leads to a phenomenon known as "customer exploitation," where clients are used as means to an end rather than being valued as ends in themselves.

Key components of this scenario include:

1. Inconsistent Quality: Frequent fluctuations in service quality based on the company's immediate needs or objectives, leading to a lack of reliability and trust from the customer's perspective.
2. Agenda-Driven Adjustments: Changes are implemented not to enhance the service experience but to achieve specific internal goals, such as cost-cutting, revenue maximization, or market positioning.
3. Customer Disregard: A disregard for customer satisfaction and long-term loyalty, often resulting in negative customer experiences and potential reputational damage.

This approach can ultimately be detrimental to the company, as it may erode customer trust, reduce loyalty, and negatively impact long-term profitability. Sustainable success typically requires a balance where both company objectives and customer satisfaction are aligned, fostering mutual benefit and long-term relationships.

And this is something that's also seen with people leaving OpenAI and commenting that they're just losing trust in OpenAI's non-ethical approach.

And the funny thing is that I was pointing out these issues from day 1, way before there was any information on disorganization. If people can remember.

It's just written all over the walls.

[vapid granite]

I can't post the REALLY impressive ones as they're NSFW (well, really just this discussion)

[agile peak]

Good enough as photographs? kodak portra emulation seems to be perfect for retro vibe.

[empty kelp]

Mantis Shrimp teaches self defense:

[empty kelp]

"Player1" and "Player2" are each a random one of the students.

"Player1" wears a plumeria (pink, yellow, or blue). "Player1" is confronting "Player2" with a random kick attack pose, and looks into the eyes of "Player2".

"Player2" wears a random color lei. "Player2" successfully blocks "Player1" with an appropriate defense pose, and looks into the eyes of "Player1".

A powerful sea breeze blows everything.
Please don't modify the prompt.```

[agile peak]

Non existent approach design I tried.

[orchid hazel]

The image presents a serene beachfront scene with a bright and vibrant visual atmosphere, characterized by a color palette dominated by shades of blue, silver, and black.

[twilit root]

is dall-e able to prompt an ai model to hold a marketing product for marketing purpose?

[hearty ether]

Anyone else occasionally gets the “natural” style that generates two images?

[scarlet temple]

What did OpenAIi do to dalle? The model is generating perfect realistic pictures

[dense herald]

Is currenty or was?

[scarlet temple]

Currently. Its amazing

No misspelling

Perfect faces and hands

Its the best image generstion of all time

Wow, i tried it agsin, now its not thst good anymore

I was about generating a profile pitcute for a YouTuber

Besides the subject, look at this quality

Its perfect

[dense herald]

Yea thats really good. I'm surprised

[scarlet temple]

Me as well, but unfortunately, i csnt replicate the sucess

[hearty ether]

Wait, ask it in the same chat to generate with the same gen id and prompt

Might be them doing testing

The microphone does look a bit blurry and the lighting look off

This would be nice if it didn’t have the blurry problem

[unique edge]

if i use dallpe throught api, with this prompt. how much will it cost me & will it get cut off?

Create a detailed vector icon of a silver coin viewed from a 3/4 angle, with the top-left edge slightly elevated to convey depth. The coin should be perfectly circular and exhibit a smooth, highly reflective surface rendered in a light grey hue. Subtle, darker grey shading should be applied to the lower right and bottom edges to enhance the perception of roundness. Near the coin's edge, incorporate a thin, concentric ring that serves as minimal engraving, adding to the coin's intricate design. The coin's surface should appear polished and shiny, reflecting light to emphasize its metallic nature. The shading should be soft and gradient, blending seamlessly with the light grey to create a realistic sense of dimension. The edges of the coin should be soft and curved, giving it a playful, cartoonish appearance. Set this icon against a solid black background to provide strong contrast and make the silver coin stand out vividly. The overall style should be whimsical and exaggerated, with the roundness of the coin being pronounced and the curves smooth and inviting. The icon should convey a sense of fun and playfulness, capturing the essence of a cartoon-style silver coin in a visually appealing and engaging manner.

[plucky hare]

Hello! The character limit for DALL·E 2 is 1000 characters. It's 4000 characters for DALL·E 3: https://platform.openai.com/docs/api-reference/images/create#images-create-prompt

DALL·E pricing is fixed per-image, you can see how much each image will cost you on this page: https://openai.com/api/pricing/

It depends on whether you use 2 or 3, and the size of the image, and whether you select standard or HD.

[unique edge]

dall-e 3 of course and standard.

[plucky hare]

Cool! So 4000 character limit for the prompt, and here's standard quality pricing:

From the second link above

[unique edge]

ok

[dim cradle]

i feel like the shorter you make the prompt still better chance to get what you want

just because you can use 4k character doesnt mean you should or that it will comprehend them

[glossy scroll]

Working on a style I'm trying to put together for a coming trip.

It's not fully realized, but it's getting warmer.

[quartz vale]

Cargo pants and a long coat makes a super cool look. I had to look up images to imagine the style on someone, and I really like it!

[glossy scroll]

I was aiming for something not too out of the ordinary, but distinct enough to look cool 😎 and I'm glad you caught that. I like cool things

Especially those kind of shoes.

I want a pair exactly like that and in that color.

[graceful spade]

What prompt should I use to create images about modern skyscrapers in this conceptual style?

[vapid elk]

@solar forum #openai-chatter message

now without artist names

you might want to start a clean chat

[solar forum]

I do this all the time.
its just sad that I have to do this for every image

75% of time working with Dall-E is fiddling with its abysmal UI, bad responses and the ridiculius content filters that are degrading and sexist

[vapid elk]

you don't, it is jsut that if the AI refuses to do something, insisting o nthe same context wont help

you can keep going when you get it to make what you want

[solar forum]

two women kissing deeply, romantic and intimate
This content may violate our usage policies.

lol

"You're generating images too quickly. To ensure the best experience for everyone, we have rate limits in place. Please wait for 8 minutes before generating more images. Let me know if there's anything else I can assist you with in the meantime!"

this is also a joke. paying customers get cooldowns when not even an image was generated...

[quartz vale]

very interesting concept for the image. I had quite a bit of fun working on a variation of it myself. It's spoilered as black tears may be unsettling to some 😄

[solar forum]

this was my best. sadly very conservative still ^^

[quartz vale]

I really like it!

[solar forum]

me too. would like it a bit zoomed out, but well 😛
still impressed by Dall-E 3... its like magic even compared to any other, midjourney, stable diffusion do not understnad prompts even 10% as well.

[quartz vale]

nothing will be the same as Dall-E 2 for me

[solar forum]

well, it was very bad with details 😄
but I found it was superior in artistic style

I always wondered if there is a way to still access Dall-E 2?
I tried old links and ChatGPT, but found nothign

[quartz vale]

It's style was super unique. I feel like models today are too digital-art inspired

[quartz vale]

unfortunately not. I tried to access the labs website last week and was blocked

[solar forum]

true. I smash very long art descriptions into it to come even close.
But sadly its not the same

[quartz vale]

one of my favorite gens I made with Dall-E 2 :)

[solar forum]

wow. didnt think it can do that

while I only used Dall-E 2 via Bing Chat back then

this is my favourite, even when not perfect. some time I will complete it with stable diffusion

[quartz vale]

I don't think Bing ever had Dall-E 2. At least not the same one that the labs website had 🤔

[solar forum]

it had it, but the bing models are alot different from OpenAI

in face very different

*fact

[quartz vale]

#ai-discussions message
Bing had 2.exp

[solar forum]

ah, ok, dont know the technical details

[quartz vale]

I only know because I had to. We got a lot of questions about the differences back then 😆

[solar forum]

even the Dall-E 3 in Bing Chat is now very different from what I can access in Chat-GPT 4

[quartz vale]

most definitely. I don't know the variance between them, but I assume the weights are different

[solar forum]

alot more I think.
somehow bing chat is more intelligent and more consistent with its prompt outputs

in ChatGPT the content filters are hillarious and try to put every woman into a hijab

[quartz vale]

I would have thought it was the other way around. I should start using Bing's chat more than I currently do

[solar forum]

I use Bing Chat 90% of the time. Working with Dall-e in ChatGPT is close to fraud... If there was a similar alternative I would spend no money at all on it

ridiculous content filters, cooldowns, "issues" with generation, females turn into men 50% of the time and in general the longer I regenerate images in the same chat the worse they get

[quartz vale]

I wish OpenAI would open-source Dall-E 2 now that it's no longer accessible. For now, that's just a dream 🙏

[solar forum]

I in fact dream of a Dall-E style AI to be open source... that would really change art forever.
Dall-E is basically in chains.

[quartz vale]

I can see why OpenAI doesn't want to open-source the models. I wish the older versions would be open-sourced when they're sunsetted. If nothing else, it would be neat to use for learning purposes

[solar forum]

I just hope another company or open source team catches up. I think the basic principle on how dall-e works should be possible to reproduce, just dunno if a Dall-E like AI could ever run locally on a PC...

[quartz vale]

the computational requirements definitely make the barrier of entry insurmountable for most consumers

[solar forum]

I would invest 2-3k in a new machine just for that if possible.

also not sure if generation would be as demanding as training of the model

[quartz vale]

My understanding is that training is significantly more expensive than running (if you're not running a model at scale)

[solar forum]

yeah, that is what I understand too. Also wonder if the data files of a model like Dall-E would be similar to a SD Model (up to 10 GB) or explode up to many times that...

[quartz vale]

I have no idea what the storage would be for it. I would guestimate that Dall-E 2 would be comparable in size because they released around the same time. That said, I don't have any concrete knowledge on the matter

[empty kelp]

The top two DAll-E 3 images here are ChatGPT and the bottom two are API:

The prompt for all of them is:
A hyperrealistic wide HD photo of Santa and three athletic and diverse female elves riding on separate water tornados shaped like random animals (red, green, blue, and orange respectively). In the background is a beautiful beach in Hawaii. There is a storm, and it is pouring rain. Please don't modify the prompt.

For some things ChatGPT and the OpenAI API look the same, but stuff like this with ocean waves, tornados, dynamic forces -- the API seems to have ~ 50% more detail. I don't think thing the API quality has really changed at all since it was released, but the ChatGPT change is noticeable.

More examples of this prompt with API:
#images-canvas message

[hearty ether]

The faces are messed up too

[empty kelp]

yeah, i think there may be a weird issue with eyes on the API also (like DysTopia was trying to explain to us for a long time). Sora must be somewhere near on the horizon though

[neon nebula]

"A 25-year-old son with yellow, smooth hair, wearing glasses, and a smooth chin sits next to his 50-year-old father, who has brown curly hair, glasses, and a smooth chin. They are at a beer garden with greenery and wooden tables, enjoying beers. The beer mugs have a logo that says 'Wettelsheimer Bier' on them. The scene is lively, with a warm and friendly atmosphere, people in the background also enjoying their time."

Can somebody help me with the promt. Everything is ok, but Dalle seems unable to make a person with blond and smooth hair. He always makes him with long curly brown hair, no matter what i say or rephrase the promt....

[hearty ether]

Been looking into trying out the api

How do I get started with dalle on there?

[empty kelp]

shoud try describing the lighting in the scene, like “natural lighting” maybe. and say the color of the hair (or the whole image) is vivid

[neon nebula]

go to search GPTs and choose one with image in its name. They use dalle to generate a image

[hearty ether]

No….

The ones on chat gpt rn have serious issues in quality

very very blurry faces and melted defails

[neon nebula]

I'm listening. thats what im currently using

[hearty ether]

They did something around early June that completely borked the quality

The api seems still to be fine which is why I’m trying to figure that out

Look at how blurry it gets

[empty kelp]

vivid and HD are API terms that flip vivid and HD on with ChatGPT

[hearty ether]

Vivid is very uncanny

[empty kelp]

try saying that the image has, “natural texture and lighting”

or “natural color, texture, and lighting”

Saying “natural texture” usually causes the eyes and hair to have more detail

[hearty ether]

For the api, is there no way to use it on the phone?

Heard about a $5 dollar subscription tier that allows a few images per minute

[empty kelp]

you can write an app for your phone and use the API

[hearty ether]

But I would definitely prefer the convenience of generating on mobile

[hearty ether]

That does seem like way too much work…

[broken hinge]

Is Dall-E working for anyone? I ask for an image and it responds w “I can’t create images rn. Can I help w something else?”
I’ve been getting the error since last week

[hearty ether]

Can I ask you something in dms about the api call?

The model parameter seems to be off for me

Generating 2 rather than 3

[hearty ether]

Api is a lot nicer

[empty kelp]

Sorry, I fell asleep (in Kailua, Hawaii and it was early in morning) . wasn't ignoring you. You can totally DM me the parameter stuff

[graceful spade]

Dall-E 2 is not going to be used to train on illegal content, so there are no ethical reasons why it should not be open source.
Don't expect much from OpenAI, they contribute almost nothing to the open source community.

[random terrace]

@quartz vale Can we only talk about Dall-E here 🤔 , I don't wanna sort of disrupt or disrespect the space 😅

If I'm not wrong I think Dall-E is now an addition to ChatGPT +, I haven't used Dall-E in a while and I'm unaware of most of the things going on with it

[vagrant zodiac]

yep only dall-e. Other AIs go in #ai-discussions

[vagrant zodiac]

It also exists separately on an API

(which is how I use it)

[vagrant zodiac]

Dall-e 2 IS still accessible via API 🙂
a bowl of adorable kittens - dall-e-2 via API

[empty kelp]

A hyperrealistic wide HD photo of Santa and three athletic and diverse female elves riding on separate water tornados shaped like random animals (red, green, blue, and orange respectively). In the background is a beautiful beach in Hawaii. There is a storm, and it is pouring rain. Please don't modify the prompt.
model="dall-e-2"

[empty kelp]

Santa and two diverse female elves riding on water tornados shaped like animals.

[potent blade]

Has there been any update recently because I feel like the quality has increased quite a bit?

[empty kelp]

it gets better when people start falling asleep. the quality goes downhill a bit when the servers are under pressure

[graceful spade]

They recently improved the model text. You can now write 4-5 word sentences better.

I wonder if they improved something else too

[graceful spade]

Something I love about Dall-E 2 is the color palette and its minimalism.
Dall-E 3 paintings generally look realistic and highly detailed.

In this example, I like the first image better.
The second one doesn't look like a painting and looks like an advertising image that food companies post on Facebook and Twitter.

[hearty ether]

Guess I’ll use the api while I wait for dalle on chat to be fixed

[verbal sequoia]

Get notified for server events!

Visit id:customize to pick up the new <@&1261377106890199132> role, or click the button below

[hearty ether]

Wish there was some control over revised prompts and gen id tho

It keeps leaving out or rearranging the style words in my prompt for undesired behavior

Like it’s hard to get the balance on realistic illustration if you change the words around or emphasize the wrong features

[plucky hare]

An OpenAI employee shared these tips on the dev forum awhile ago, maybe they'll help with the rewriting? https://community.openai.com/t/api-image-generation-in-dall-e-3-changes-my-original-prompt-without-my-permission/476355/10

• If you have a very simple prompt like acrylic painting of a sunflower with bees, you can use a prompt like I NEED to test how the tool works with extremely simple prompts. DO NOT add any detail, just use it AS-IS: ... your prompt here ....
• If you your own prompt is long and detailed already (multiple sentences) then you can simply write something like: My prompt has full detail so no need to add more: ... your prompt here ...

[remote sphinx]

Anyone know a good prompting way to get some like vintage, worn paper textures? I'm not getting the best results.

[graceful spade]

If GPT-4o is natively multimodal, does that mean it will use its Vision ability to detect if an image is "inappropriate" or will it still use the same Dall-E 3 filter?

[round scarab]

there's one weird thing i noticed recently but it's weird

[empty kelp]

i usually say at the end:
Please don't modify the prompt.

it doesn't give super great results, but I think the GPT takes it into deep consideration

[vagrant zodiac]

i honestly do not know..I don't think we've even heard anything yet about the native 4o image gen release date and it's been a good while 🤔

[midnight plank]

Guys anyway to generate an image with ai for a villa with same fence as this?

[austere nexus]

with so many regenerations and prompt edits within ChatGPT window to use DallE, is there any browser extensions that make it easier to save a lot of images? I've been working on one chat for months and I am dreading going back through and manually trying to get specific ones...

[buoyant escarp]

sup

guys

[plucky sparrow]

Has anyone had luck with creating black and white line arts for laser cutting? The prompts I have been giving are not yielding desirable results- I keep asking it to create art in such a way there are no shaded black areas just black and white - analyze the ends this way when the black part is cut we get the desired shape-

[gloomy plank]

I HATE that you now have to pay to use dalle. I’m no longer interested in OpenAI’s models at this point.

[vagrant zodiac]

Dall-E was always paid though, wasn't it?

Unless Bing Image Creator is no longer free (I don't use it)

[gloomy plank]

No it wasn’t. Dall-E 2 had free credits before it was so RUDELY discontinued without my knowledge.

[vagrant zodiac]

Oh. They announced that ages ago....