images-discussions

[rain flower]

Haha, I chatted with him so many times 😄

[crystal badger]

have u tried using the 12 methods of prompting?

[rain flower]

I was wondering if anyone had successful

12 methods of prompting? No, haven't heard of it. Is it somekind of tutorial?

[crystal badger]

ill give you eight to begin with

1. Descriptive Prompts: Employ vivid descriptions to enrich the generative process.
2. Conditional Prompts: Set conditions or constraints to guide the generation towards a specific style or theme.
3. Sequential Prompts: Use sequential cues to develop multi-step or evolving outputs.
4. Comparative Prompts: Utilize comparisons or contrasts to elicit a richer narrative or visual response.
5. Instructive Prompts: Give specific instructions to achieve precise outcomes.
6. Emotional or Expressive Prompts: Invoke emotions or expressions to resonate with the intended audience.
7. Multi-modal Prompts: Combine text with other data types to enhance the generation process.
8. Negative Prompts: Specify undesired elements to ensure they are excluded from the generated outputs.

[rain flower]

Ahh,, yes I see. I have done quite a lot of these, honestly. And ChatGPT even tried on his own to get the result I wanted for many hours, still couldn't get it.

[crystal badger]

do you use your own prompt or do u allow chatgpt to help

[rain flower]

I use my own prompt as a start and then ask for ChatGPT for help.

[crystal badger]

usually you would describe "what you want" and have chatgpt optimize it for DALLE to understand.

Avoid the Word Association Dilemma as well.

meaning ensure u talk to chatgpt about what you actually mean by your prompt. by describing it in small detail.

some singular words have 15 different meanings. thats one of the issues.

[rain flower]

Yes. I even ask back ChatGPT about what he has interpreted and I’d ask to explain in your own words

And only proceed when it interpreted about my view properly.

[crystal badger]

Aight, imma DM you then u can send me your prompts, then i can test it out.

[dark pawn]

Hey guys, not sure if this is the right area to ask this, but I am using Dalle3 to generate an image for someone, and it generated a near perfect image with one detail off. What's the best method for tweaking an already near perfect image? If I try to run the prompt again with the change it gives me an image that is unfortunately different in ways that lose the perfection of the first image, so does anyone have any solution to this?

[crystal badger]

find the exact prompt by copying the prompt in the image and rerun it as is

[dark pawn]

I did but it generates different things

[crystal badger]

did you manually copy it.

or did you click "copy"

Also ask ChatGPT what SEED its using for that image. then ask it to generate an image with that exact seed with that exact prompt.

[dapper temple]

anyone got the slightest clue as to why the style and aspect ratio and content and everything is so different from before the 4x -> 2x dall-e generation change was made? I pressed regenerate and it spat that out. absolute garbage, where did the realism go?

I can't even get it back despite explicitely describing and telling to keep same style and aspect ratio of it in messages after.

[fallow quail]

should be the same if the prompt and seed is the same

[plucky hare]

Is pic #1 what you want, pic #2 what you don't want? It looks like 1 is in the style of a photo, and the other some kind of illustration. Have you tried specifically asking for photos?

[dapper temple]

yes, pic 1 is what I want and what I got before. same everything, pressing regenerate gets the second!
this is another example in another part of a branch.

[fallow quail]

ask in the post after the generation of the pic you want whats the seed and the specific prompt of the fist picture. then use that with slight modifications in the prompt to genereate something alike

[plucky hare]

Specificity is definitely key with DALLE, so instead of saying "use the same..." more specific language about creating photos specifically might help. If you're set on photographs, you could even include language like this in your custom instructions: "When generating images, always generate photos" etc

[cobalt hull]

I've also noticed some degradation on images since dropped to 2

[dapper temple]

used the same prompt and seed to this left one. at least somewhat closer this time. after requesting it to spit it out. which I shouldn't even have to do. regenerate, should in-fact, regenerate... keeping the things above it as it were. it lost everything it had somehow.

[agile peak]

I figured out “Existentialism” is one of most difficult word that keep dalle struggling.

Have tried 8 attempts with different seeds, haha.

[cobalt hull]

It also really struggles with providing 2 images once it gets on the 1 image flow

[dapper temple]

that too!

[cobalt hull]

Even after I say give me two images! lol

[dapper temple]

!!!!

[agile peak]

Yeah. I have that many times today, once I tell it to continue.. it changed the styles entirely and had to rewrite the prompt 🥹

In my case. Dalle always hits on the nail on the first panel; the second one usually random/altered prompt.

I add visual effects to the prompts to make it appealing. The first one is good, the second one seems altered and went to realism path.

[flat zephyr]

yeah it's hard to believe it's really GPT-4, it's so incapable of following simple instructions

[jagged valve]

Your images are very realistic, do you have any tricks?

[cobalt hull]

lol ya it's been frustrating lately

[dapper temple]

I described background context, the kind of building, that I wanted a computer room from a country from a certain era.

to ChatGPT / Dall-e combo

adding keywords like realistic and grounded in reality helps

[crystal badger]

i tried Jiwo

[zinc dagger]

I got the new version of GPT all tools with Dalle! it's interesting to give it an image generated with dalle and see if it can recreate it, and if so what's different about the prompts

[gusty plume]

It it frustrating to anyone else that Dall-e 3 only outputs 2 images now? Has OpenAI said anything about this? I feel like having less results too has made the output quality worse.

[plucky hare]

Moxi, an OpenAI dev, confirmed this #images-discussions message
Quality shouldn't be worse. You can test by using a past image prompt and seed.

[gusty plume]

Dall-e 2 launched with 10 and now we are all the way down to 2 🫠

[wispy storm]

Two different products

[gusty plume]

I'm fairly new to using AI, so forgive me for asking, but how does MJ keep up with so many requests?

[wispy storm]

Hi, for other ai in the discord here the right channel #ai-discussions

[remote stone]

Has anyone managed to make DALL-E 3 successfully generate a centaur? I'm trying with no luck

[plucky hare]

Oh wow I'm 0 for 4 so far! I wonder if it has to do with the already-confusing appearance of a centaur?

[fallow quail]

no problem with that, or a cerberus, or a minotaur. just describe what you want to see

[plucky hare]

Try it! "Centaur" alone gave me one person riding a horse, and then one near-centaur, but the head was also a horse. I just got closer without usuing the word "centaur" at all @remote stone

1. Upper Body of a Human: From the waist up, has the torso, arms, and head of a human. Intricate armor on their human upper body. Wields a bow and arrow.

2. Lower Body of a Horse: From the waist down, has the body of a horse. This includes four horse legs, a tail, and a horse's hindquarters.

3. Habitat: Forest.```

[remote stone]

ooooooooooooo that might do the trick

I guess you just need to be more specifically descriptive

[plucky hare]

Neither of the actual DALLE 3 prompts that ChatGPT wrote mention the word "centaur" either, let me know if you want either of them.

[fallow quail]

"Intaglio print of a centaur in the desert. Upper Body of a Human. Lower Body of a Horse." works like a charm even with the centaur included.

[plucky hare]

Nice! I wonder if requested style makes it any more/less likely to come out properly centaur-ish. Re-describing it probably helps too.

[fallow quail]

have to check without the style

nope, same with "a centaur in the desert. Upper Body of a Human. Lower Body of a Horse."

[plucky hare]

Nice. So seems like best practice for such beasts might be proper name + additional description!

[tropic roost]

I am having trouble accessing DALLE3. The DALLE3 website (https://openai.com/dall-e-3) has a link to "Try in ChatGPT". I clicked on that link and upgraded my subscription to Plus so that I could interact with ChatGPT-4, since I read that Dall-e-3 was integrated into ChatGPT-4, not the older ChatGPT. So I initiated a discussion with ChatGPT-4 and asked it to draw a specific image, which it couldn't do. It said "I'm sorry, but it seems there is an issue preventing the image generation at the moment. DALL·E-3 capabilities should allow for image generation based on textual prompts, but I'm experiencing difficulties in this particular instance."

[plucky hare]

Now that you have ChatGPT Plus, start a new chat and then open up the GPT-4 dropdown menu in the model selector. Select "DALLE 3" at the bottom, and then you should be good to go!

[tropic roost]

OK, so it looks like there's no way to "show" a picture to DALLE3, is that correct?

[plucky hare]

Not yet, no! What you can try right now though, is:

• Select default GPT-4 model--this supports image recognition
• Upload image and ask for a thorough visual description--subject, art style, etc.--just describe what kind of description you want, basically
• Take this description over to DALLE 3 and generate images based on the text description
  This has produced pretty good results for me. Also, be on the lookout in #announcements for potential future announcements regarding "combining" models like this, though there's no timeline on that at all currently. Also, keep in mind the limitations with image recognition regarding descriptions of people--it's trained not to do that.

[hazy warren]

How do you create a modicum of continuity from prompt to prompt or image to image? I’m trying to use the tool as a means of storyboarding for a coding project, but the theme from image to image is disjointed using ai art.

[pure mantle]

You can't upload images to Dall-e 3, can you?

Without the latest update, I meant

[plucky hare]

No, not without that update, yep. I'm pretty sure I've seen folks that have gotten access uploading pics and then asking DALLE to work with it. What I saw didn't look like inpainting though, so I wonder if at the start it'll be more of a img2txt2img thing, with the "txt" bit behind the scenes in the middle? Or if it'll be more of a direct vision.

[pure mantle]

I'm interested in that one too ahah

[drowsy bane]

Has anyone worked out how to reliably get the tall 9:16 aspect ratio to always output a portrait/vertical style image?

Mine gets the aspect correct but 50% of the time the image is just a rotated 16:9 image.
Thanks in advance!

[wanton hatch]

It's a bug that unfortunately they still need to fix

[drowsy bane]

Thanks for confirming! ❤️
Yeah seems like even with the same prompt but different seeds it still does this which is a bit odd.

[fallow quail]

stick to a specific and very defined artstyle and a specific seed. thats the a step into the direction you want

[lyric lodge]

I apologize, but we have encountered issues generating the images you requested. is this a polite way of gpt telling me that I unknowingly triggered a content filter?

sometimes its blunt about it, sometimes its not

[plush sky]

I get this constantly. Sometimes it's because of rate limits, sometimes it's because of the crazy censorship, sometimes it's because the prompt is too complex, sometimes it's under too much heavy load.

Often I just tell it to do it again, and it often works the next time or a couple of times of telling it to retry.

Sometimes you run into a wall because of the complexity or censorship of the prompt (with literally no bad content whatsoever), so you can simplify your prompt or reword it.

[lyric lodge]

yeah, I figured as much, its likely related to the complexity or the censorship part

very annoying, wish it knew if it was censoring me or just confused

[plush sky]

Ask it and sometimes it tells you

Sometimes I just copy the prompt it's not creating images for, create a new chat, and then it works fine

Completely agree through, the constant errors are infuriating. Allllll the time, problem after problem after problem

[ripe wraith]

yes, living in the future is tedious

personally, i want to be part of a native tribe at the height of its power, much less going wrong

[plush sky]

Tedious is pandering to ensuring every single comment that someone says that includes a critical element, also includes an equal countering sunny, positive compliment to satisfy delicate recipients. People can express feedback without containing their life's breadth each time.

[ripe wraith]

I don't pander.. I'd have more friends if I did. No, I just say what comes to mind.

[dire burrow]

Oh, indeed! The future, with its relentless charm of convenience, does have a particular knack for dulling the senses. One can only imagine the sheer joy of forgoing modern medicine, the rule of law, and the luxury of a proper afternoon tea. A most enviable life, truly.

[ripe wraith]

Ya I was being sarcastic but if u gotta pick history I'd stick to my choice.

[dire burrow]

Not disputing your choice, but I'd rather be in a point of time after cheap penicillin and insulin.

[remote sphinx]

They really need to fix the aspect ratio issue for Dalle

Also only getting 1 image

[remote sphinx]

dall-3: making two images

then gives me a single image

bruh

[pure mantle]

How you guys deal with text written in image? I tell to write a specific text and it make weird mistakes.

[keen dune]

It used to not even be able to make letters at all, it does way better than it used to—these visual GPTs often struggle with text because it comes in so many forms

It only kind of correlates the actual text with the illustrated word representations, one day it will be flawless, for now it does best with smaller words but sometimes you get lucky on a variation

[pure mantle]

Okei thanks

[fast oar]

how many images can you generage withh 115 images with dall e?

[dire burrow]

dall-e5 will do text flawless, mark my words.

[remote stone]

Gotta love how absolutely unspecific this error message is

[ancient jasper]

This image generated perfect text, in the wrong direction

See, insane

[keen prairie]

Whoa, somehow that’s more impressive haha

[tropic roost]

Are you only able to download dall-e generated images for a short period of time? There's an image I liked, but it's no longer letting me download it a few hours after generation, from the download icon within the conversation where it was generated.

Is there any way to get it to upscale the resolution? I asked it to generate 2048x2048, and it said it would, but the image was still 1024x1024.

[dense ermine]

If you read it, you also notice it says for two times

I don't think i recognize any of the characters barring Tim Curry kinda

[tropic roost]

I feel like I've seen somewhere that DallE is capable of generating higher resolutions, but I can't figure out how to get it to do that.

[agile peak]

Dalle struggling writing existentialism 😁 make it better somehow

Or dalle is having an existential crisis 🙈

[tender sigil]

my bing image credits got over...and they dont replenish at the rate which they used to earlier, i used to get 100 per day, now the credits stay the same as they were the previous day

[wind crane]

Use bing conversation instead

[tender sigil]

Thanks a lot for the suggestion!!

credits wont get used up here right?

[wind crane]

I didn't see any credit indicator in bing conversation

[tender sigil]

i just tried generating an image via Bing chat...it shows this

[wind crane]

no luck, but here what it look like when it work

[tender sigil]

Ohh yes it works now!! i get the same thankyou so much!! 👍👍

[agile peak]

Perfection. In early days dalle struggling to create heterochromia eyes.

[final compass]

Does anyone use brackets in their prompts and do they work in isolating descriptions?

I tend to have the problem, that when i describe something it sometimes applies the description to something totally incorrect. As if it just randomly combined keywords together.

[crystal badger]

Yup I do, works excellent

[final compass]

You have experienced it yield more accurate results?

[crystal badger]

Can’t say because I use a set of instructions tailored to work with my first prompt

[vapid elk]

nope, it is a bug, the AI will do that some times
in fact, I noticed that it often do that when you are near what the filters would usually block

[fair crescent]

DALL·E can only generate the following resolutions (ChatGPT only atm):
1024×1024 square
1792×1024 wide
1024×1792 tall

You can have great results with 3rd party upscalers though. I used a program to upscale it to 4K and even 6K with pretty good results

[twilit helm]

bing has a filter so it’s blocks your prompt.

most of the time it’s blocks sfw prompts which is annoying

[cinder kelp]

‌‌‌‌‌‌‌When will the API of DEALL-E-3 be opened?

[plucky hare]

Still just "later this fall" I think https://openai.com/dall-e-3

[cinder kelp]

thx

[crystal badger]

API Will be great for image gen

[jagged valve]

Hey~ Does anyone know how to generate a realistic and lifelike portrait?

[wispy storm]

Hewwo, I suggest to check #1019652163640762428 and #1154829862171844679 lots of suggestion how to do it

[jagged valve]

thank you so much for your suggestion!

[tender sigil]

yess even i've seen that bing blocks even the simplest of prompts, the safety filter is really strong

[tiny linden]

Whenever I select an image that's been generated it previews just as a tiny thumbnail. Is there a way to fix this?

[plucky hare]

You mean in the app, or on web? I also think the default view, in the Android app at least, is quite small. You can zoom, but there's a permanent bar on top which cuts off quite a bit, especially in landscape view. I've got my fingers crossed for a nice gallery view someday 🤞

[tiny linden]

I am on web. The preview was fine yesterday but today it is about 1 cm on screen lol

[plucky hare]

Oh odd, I'm seeing that normally on my end, in mobile view and desktop view on a smartphone at least, full size is loading when I click.

[tiny linden]

I am able to download at full size so it isn't a huge issue it's just annoying thast it shrinks when I click the image haha

[jovial terrace]

ChatGPT version of Dall-E 3

Bing Ver of Dall-E 3

Guess which one costs $20 a month?

[agile peak]

What a timing lol

[jovial terrace]

LOL!

[agile peak]

Apparently you can. But use zilla as a keyword.

[jovial terrace]

haha

I hate that. Seriously. Having to trick the bot into making a simple image that no one would actually find distressing...

[agile peak]

Same. I think they want to use Zilla to encourage the US version than JPN one 🤣

[jovial terrace]

Seems strange that the ChatGPT version is so limited

[wispy storm]

They don't want being sue or using protected franchise. So they decide to not going that direction

[jovial terrace]

Bing doesn't care? 😉

Truly open source will win.

[wispy storm]

Microsoft have their own tos and policy

[jovial terrace]

And a lot more money.

[cobalt hull]

Doesn't like to create large creatures, but dragons are ok? lol

[wispy storm]

Remove the godzilla in the prompt

[cobalt hull]

lol i'm just playing around with it. don't have any specific needs, but really strange it says something liek that but has no problems with a T-REX

lol you have to start a new chat for it to not take inconsideration any of the previous prompts i guess. Pretty strange

but tryig that same prompt in the previous chat produces

[wispy storm]

It have a context memory and take previous conversations in mind

[vapid elk]

a conversation where the AI refuses all your requests will heavily weight it into continuing to refuse anything else

[agile peak]

But somehow. Bing version is downgrading in terms of quality generation now in comparison to early days.

[cobalt hull]

lol got it. Wasnt' my issue but was just curious how it was reacting. Not a Godzilla fan anyway lol

[agile peak]

Cthulhu is good for chatGPT.

[cobalt hull]

lol nice

[cobalt hull]

lol

[agile peak]

The new dumbo look great

My attempt to make realistic.

When I changed the resolution to full portraits all generated will be toy-like.

[tropic roost]

Can you recommend a particular program for upscaling? I see several options out there. Some seem really expensive, and the free ones I wonder "What's the catch?"

[plucky hare]

#ai-discussions maybe too!

[fair crescent]

It costs a pretty penny, but I used Topaz Gigapixel with good results. However, there are many more around. I just followed the recommendation of someone else pretty blindly, which is always a risk.

[agile peak]

Should’ve stayed at home today.

[forest seal]

This seems to be a false positive, and as lugui said, its because it was refused before

[cobalt hull]

Anyone else having issues at the moment?

[gentle tree]

Wow

[agile peak]

Using zilla in the snack package, dalle able to perfectly capture it.

[jovial terrace]

Nice

Love it

Can you post prompt?

[charred sonnet]

Dalle is really quite awful at listening to prompts to adjust the image sometimes, but other times (1/10 chance) it gets it perfect.

[cobalt hull]

lol agreed

[plucky hare]

You can request of ChatGPT to use your exact image prompt, and it'll comply! Otherwise yeah it'll always freestyle a little bit.

[cobalt hull]

It's not producing any images for me right now, I think I annoyed it with the Godzilla attemps earlier

[jovial terrace]

lol

by minimize it thought you meant MAXIMIMIZE

[charred sonnet]

lol EXACTLY it keeps doing the opposite!

[agile peak]

Here it is:
A photo still a snack packet titled "Zilla" wth a subtext “Kaiju". The design on the packet features a zilla creature in the city and a cloudy sky in the background. The bottom text on the packet reads "Strong as zilla!"... [add another elements]

[charred sonnet]

Maybe it just doesn’t understand architecture

[charred sonnet]

actually this one it did well

[cobalt hull]

Are you using bing? I can't get any images

Tried new chat and signing out and in and on mobile too lol

lol k as long as it's just not me. I'll let it be for a couple of hours then check back

[fallow blade]

error on my side too

[hearty ether]

Down

probably

[hushed creek]

Yup down

[oblique moon]

same 💀

[empty crystal]

+1

[wide knot]

Rip Dalle

[ripe fiber]

Still down

[humble gull]

not working ?

[ionic ore]

Bing image creator is still working

[ripe fiber]

Yep. Not working. Mobile and PC are both down.

[ripe fiber]

Seems you are still able to generate 4 images though. Most accounts have been capped to 2 images per generation

[quaint edge]

Cut from 4 to 2 to 0. Great.

[heady adder]

It worked for me, generated an emote which could be used in this context?

[plucky hare]

DALLE 2 on labs?

[heady adder]

Maybe that. Oops

[livid venture]

Failure mode in DallE3. Not sure why it didn't work.

[ripe fiber]

Read above

[shut niche]

Upscayl, all day, if you have a decent GPU. Simple, works extremely well, and supports other upscalers.

[copper marsh]

Seems to work fine for me

[ripe fiber]

Working again, but capped at 1 image per generation for me

[plucky hare]

I just got two pics of squirrels fencing 🐿️ 🤺

[glossy scroll]

Anyone else getting this error message?

It's happening for all my vision prompts

[plucky hare]

Just happened to me too.

[vapid elk]

Refresh the page

[plucky hare]

Failed again, is it working for you?

[glossy scroll]

Logged out, logged back in. Refreshed. Doesn't work.

[vapid elk]

Try clearing the cache

Ctrl+F5 can help

[glossy scroll]

Tried. Doesn't work.

[plucky hare]

Also fails in app. And still failed after Ctrl+F5. Probably just a hiccup! I think this isn't the first one for image recognition.

[glossy scroll]

Yep, it failed and it keeps failing whatever else I try.

[vapid elk]

Are you using a von?

Von*

Vpn**

God, i hat mobile keyboard

[glossy scroll]

No

[plucky hare]

Is image recognition working for you currently?

[vapid elk]

Everything is working for me rn

[plucky hare]

Ope yep, just started working for me again too 🙂 Thanks!

[vapid elk]

Noice

[simple flame]

a dementor from harry potter movie, in the middle shrouded by drops floating in the middle of the void, unreal engine 5, dark, weird, scary, horror

How to make my image from my dreams

[glossy scroll]

The vision works for me now too, but now I'm having the exact same error message with Dall-E

[plucky hare]

You can try DALLE 3 for free in Bing's image generator, or with a ChatGPT Plus account. You can try DALLE 2 on labs.openai.com, but credits cost money! I think you might get some free credits though if you have a new account

[glossy scroll]

Yeah, but uhh, what about the problem?

I am using Dall-E as a plus member

[ripe fiber]

Not working for me either.

Worked for a bit an hour ago or so and now it's broken again

[glossy scroll]

[eager locust]

DALLE broken then?

[glossy scroll]

Seem to be

[jade flare]

Are plugins working? AI pdf seems not working on chatgpt it shows me the following message Hmm...something seems to have gone wrong.

[glossy scroll]

It's always going wrong.

Now it seems like they turned on the AI for every error message, just so it can justify its actions.

[tacit badger]

dalle is giving me completely different images than what i asked for lol. anyone else getting something not anywhere close to your prompt?

[keen dune]

My friend is just getting the text of the prompts back, no actual images

[tacit badger]

i am working on the daily theme, so magicians and stuff and it keeps messing up. it's not giving me magicians... it gave me this on it's own accord. i didnt ask for that at all. dalle is dark left to it own devices lol

[warm valve]

Why do I want to generate girls when GPT-DALLE3 keeps giving me boys? Any tips on how to avoid this?

[vital gull]

give us an example, where do you experience this?

[hearty ether]

Did they add a seed system in

[warm valve]

I don't know. I generated it directly.

[plucky hare]

Probably a hallucination, I think the last 24hr daily cap is 300, confirmed by Moxi

[warm valve]

Is there a problem with the server now?
I'm getting errors a lot.

[hearty ether]

“The prompt provided includes a description of a person that doesn't specify a diverse range of physical attributes, such as descent and gender, which is required by our content policy to ensure inclusivity and avoid reinforcing stereotypes. If you'd like to proceed, we can modify the prompt together to align with this policy and then generate an image for you.”

The what

Can I not specify hair color or eye color now?

What if I had a particular look for my character in mind

[hearty ether]

It does struggle with gender

Had a few times when it switched it completely

Maybe because it wanted more diversity I guess

[plucky hare]

Interesting, can you share the description? Did this happen after you requested that ChatGPT use your prompt exactly, instead of coming up with one on its own?

[hearty ether]

Not in my case still, it just gives me generic content warning message variations (like why use gpt 4 for that? Instead of an automated message, and why not provide a reason but force me to ask it)

[hearty ether]

Use exactly

Funny thing was it was basically me trying to recreate an image from before with some variations

And it had described the characters as “of European descent” and “of Asian descent”

I don’t see what’s wrong with that tho

I wanted to get a particular look in mind for my character based on the way they were described in my story

[warm valve]

Is this a server issue? Or is it a problem with my prompts?

[zinc dagger]

I've been having the same problem all day, I think it's a server thing

[hearty ether]

Wish they would give us more useful error messages

[warm valve]

I asked GPT and it said it couldn't get any more error messages either
That's weird.

[glossy scroll]

Because it doesn't have a troubleshooting option to know what's wrong with itself. Even though OpemAI shows a breadth of sophistication for bringing AI into existence, they themselves don't utilize the AI's capabilities to expound any source of resourceful information through the AI to members to solve any kind of issue. All forms of issues must be reported to OpenAI directly through email or through the prompt feedback.

It can't even say like "Let me know your concerns and I'll forward your message to the OpenAI development team"

Doesn't even have that option.

[dim cradle]

Hey

This is a very bad beta test for dall e and chat gpt integration

You have two very good features but they don’t work in tandem at least lot visibly

Why can it perceive an image and describe it

But it can’t improve the on image it generated before

[quartz vale]

how are we doing?

[dim cradle]

Great

[zinc dagger]

I’m doing alright, the all tools version of dalle is pretty buggy, getting errors every other prompt. It’s nice to be able to give it reference photos of old generations though

[dim cradle]

A reference? In what way

Like client side or in its data base?

[zinc dagger]

The all tools version of GPT4 has the vision module (image recognition) along with file upload and dalle, so I can show it some of the good images it’s made in the past to try and replicate a style more consistently

[dim cradle]

How do I use the all tools version? It looked to me as if the image recognition was disabled when I used dall e

[tall patio]

You can ask it to show you the seeds

But even then it still doesn't understand that much

[dim cradle]

Could you show me an example

I’m on my phone rn

[zinc dagger]

I think it’s still in the process being rolled out to plus users? You’d get a pop up when you open the website

[dim cradle]

Really?? Damn

[zinc dagger]

I’m on mobile now as well, the images aren’t rendering on the app currently

[dim cradle]

When you say all features you mean there’s a way soon to use every feature at once

[zinc dagger]

Yeah!

[dim cradle]

That explains the reason for this integration

Because for me it’s terrible

Also dall e is behind in terms of the art space. I believe it is due to its different approach when it comes to where it gets it data from I assume

I am unsure tho

Like for copy right and professionalism

[zinc dagger]

I think the way it’s trained means we have to be more creative with the prompts if you want something great. I read in one of their papers that they label the images using GPT-vision which produces some weird wording that isn’t quite how people describe things. I’ve had success with including non-visual descriptions like emotions or motivations for actions

[dim cradle]

I see

Like I wish to take a style I like but not copying it ofc but dall e completely ignores it

I don’t think a style or something that is so widely used like the ghibli styles should be blocked by Dall e

[zinc dagger]

If you have access to the GPT vision, try asking it to describe the artstyle of some reference images. I’ve noticed that for certain art styles it rephrases specific artists to a generic term that works just as well (ie, instead of a “Dungeons and Dragons” style artwork, saying “in the style of artwork from TTRPG games” can give better results)

[dim cradle]

Mhm

I just did it

Someone should make some amazing teddy bear art and send it me

[zinc dagger]

I’ll give it a shot tomorrow and post my results

[dim cradle]

Thanks man!!!

[keen dune]

I dunno I had pretty good success commiserating with it about the root cause 😄

[dapper hound]

So we went from 4 images, to 2 images, now only 1 image. Thanks ChatGPT

[keen dune]

At least we can have fun while we wait for the fix:

[slender plank]

i can't generate images with aspect ratio 16:9 anymore -.-

[vocal flame]

I'm having the same problem, I hope it is temporary

[slender plank]

i hope too, i need it to create youtube thumbnails 😄

[slender plank]

"I apologize for any misunderstanding. The DALL-E 3 functionality that I have available currently only supports creating images in square format and not in wide screen format with an aspect ratio of 16:9."
I hate it...

[rigid bough]

ask for a "wide" image eg: "create a wide image of a cat in a basket" and it will create a wide image for you

[slender plank]

u tested it right now?

[rigid bough]

yes

[slender plank]

ok i will try

[slender plank]

i won't work

it says this: "I can help you create a description for an image that you'd like to visualize. However, I'm unable to create or display images myself. You can use the description to create the image using graphic design software, or you can ask an artist to create it for you...."

and without the word "wide" it works

If it stays that way, it will be a massive downgrade of DALL-E compared to the competition. I hope this gets fixed

[rigid bough]

"create wide images of kittens in a basket"

wide images works, you just need to specify "wide"

[rigid bough]

that's odd, I will be seriously annoyed if DALL-E 3 stops creating wide images

[slender plank]

i am very annoyed ... and hope for a fix

[rigid bough]

what image sizes can I create?

responded with

Wide: 1792x1024 pixels
Square: 1024x1024 pixels
Tall: 1024x1792 pixels
These options allow you to choose the best orientation for your image, whether it be more landscape-oriented (wide), portrait-oriented (tall), or evenly proportioned (square).```

[fair crescent]

Do you have custom instructions?

[slender plank]

ok thats the "bug"

i deactivated the custom instructions

[severe moss]

I have a problem of anime turning out too realisitc, like this is the style I want:

but it ends up looking like this:

i specified 2d anime, non-realisitc, simple anime nose but it keeps giving me the latter more than it does the former. any ideas on how to fix this?

[slender plank]

Do you have any idea what the Custom Instructions have to do with it? I hadn't written anything about DALL-E or image generation

[sacred helm]

here's the prompt I mostly work with anime style

High-quality anime illustration capturing the aesthetics of 2010s anime, reminiscent of popular manga and Japanese animations from that era.

[sacred helm]

I have mine working really well, want me to share it?

[slender plank]

it was something random to my person for some tests nothing important and nothing about image generation, so i am surprised about the problem

but now i am happy, it works

[severe moss]

tysm

[fair crescent]

It depends on the custom instructions. We saw one prank where someone's custom instructions were changed to include bees in every picture, but never tell the creator. But if it is truly unrelated they do nothing. It was just a point to check out.

[tiny jetty]

Can we already turn photos into art, 3D etc with DALL-E 3?

[slender plank]

You can use "vision" to describe an upldoded photo detaily and create a prompt for DALL-E and edit it. Then paste the prompt into a DALL-E chat. But soon it will work better, because of the GPT 4(all tools) modell

[ripe fiber]

Still having generation issues. Anyone else?

[slender plank]

after deleting my custom instruction, now everything is allright

[tiny jetty]

Okay thanks!

[hollow orbit]

Hello. I need to say that I am a bit disappointed with dall-e3. Don't get me wrong, it's really a great model and it's much better than dall-e2. But my expectations were bigger.
The biggest problem is the lack of consistency and the second thing is the text is still not flawless, especially if your picture contains a lot of text. I thought it would be revolutionary, better than mid-journey or stable diffusion. But right now I cannot say that.

[slender plank]

Can you give us a prompt example? Maybe another prompt will work better

[hollow orbit]

For example, I wanted it to create a Formula 1 Bahrain track in 2D, and it failed. The track , turns and straights are completly different than in reality.

[wispy storm]

Try a fresh chat without ci

[hollow orbit]

Another example, I wanted to create a set of cards in the exact same art style, like 30 cards. And, you know, I can choose the main art style, but still there are some serious diferences between the cards that clearly show that these cards don't come from the same deck.

[slender plank]

what are you typing in?

exactly show me

[sacred helm]

you can control seed to get consistent result

#images-discussions message

[dim cradle]

Is dall•e working for y’all?

[twilit helm]

it’s working for now, i got some errors when generating

[ripe fiber]

"It seems there were issues generating the second image. The first image has been created successfully, and you can view and download it above. If you would like to try for another image or have any other requests, please let me know." I'm still getting that

"I can create one more variation for you. What would you like to change or add to the new image?"

it is also insisting on limiting me to just 1 variation per generation

DALL-E 3 is ded. was good while it lasted

[tacit badger]

I feel like far too many people miss the key word BETA. It's still being developed.

[twilit helm]

mhm

[lilac obsidian]

🫨🫨🫨

[dim cradle]

Do you know If the image seed is also saved in the png?

[rigid ore]

Like most options 🤪

[sacred helm]

you mean metadata?
no
but if you looking for seed from old conversation it's highly likely 5000

[opaque storm]

It‘s become meaningless. Google search was beta until long after all other search services had become irrelevant at the time.

[rigid ore]

👍That's what I was pointing at @tacit badger, Dall-e is in beta and will remain so for a while.

[tiny jetty]

Any tips for text? I need to do the same prompt like 10 times until I finally get the text I put in the prompt.

[tacit badger]

how are you writing it in the prompt? i have good success with quotations. for example: with the text "Merry Christmas!"

also it does better with less text

[tiny jetty]

Example:

Long photo, Disney 3D style, high detailed, cartoon belgian malinois in Disneyland with all different Disney characters, movie poster, Title of movie ‘Milan in Disneyland’

[tacit badger]

instead of saying title of movie, just say "with the text "Milan in Disneyland"

[tiny jetty]

I get this for example

[tiny jetty]

I’ll try one sec

[tacit badger]

it's probably getting messed up becasue it's adding a lot of extra text for the names of the actors, etc

cute poster by the way 😊

[tiny jetty]

I get it.. still just small mistakes after another try

[tiny jetty]

It was supposed to be my dog haha

[tacit badger]

for me, i'd say dalle gets the text right on the first try about 60% of the time.

that's about the most text i've been able to get in an image without it messing up.

[tiny jetty]

No way.. 😍

Without errors is nice to see

[sacred helm]

I found that dalle really write cursive well

[tacit badger]

i wish i could say it got it right on the first try, but i had to regen it several times before finally getitng it right

[tacit badger]

yes it does

[tiny jetty]

Is this you?

[tacit badger]

nah, just a random image

[tiny jetty]

Ah okay

How come I sometimes get Disney characters and sometimes it tells me about copyright

[tacit badger]

it's really restricted, sadly

[tiny jetty]

Marvel & Pokemon for example copyright message as well

[tacit badger]

there are ways to get around it depending on how you word your prompts

[ocean hornet]

Yeah, you just got to have good vocabulary lol.

[tiny jetty]

Hmm.. I really saw Mickey on my image

Tell me one way haha

[tacit badger]

i dont have any good advice for that, sorry

ive mostly done it on accident

[tiny jetty]

Toy Story as well.. weird

[tacit badger]

i find it funny that i cant say Harry Potter because it gets flagged, but i can say hogwarts

[tiny jetty]

Right?

[tacit badger]

what's even funnier is it often wants to give me people in the image that look like harry, ron and hermione, yet that's not what i asked for

[tiny jetty]

😂

[tacit badger]

the other day my son wanted to see a picture of spyro the dragon. it told me it couldnt do that but could ake a similar looking dragon. i said ok, just use the same colors as spyro... and then it gave me an exact image of spyro... lol

that's kinda what i mean by you can get it around it but you gotta play with the wording, yada yada

i do think they are far too strict. if i wanna see what tom hanks would look like with joker makeup on i should be able to do that.

[tiny jetty]

Lol.. I just changed Eric Cartman to Erik Cartman

[tacit badger]

lol

he looks exactly how i look when gpt tells me it cant do something

[tiny jetty]

😂

How about Pikachu to Picachu

Worth the try?

[tacit badger]

right now i'm still experiencing dalle creating an image that has nothing to with my prompt. oh you wanted a rainbow? well here's a dinosaur in the arctic

[tiny jetty]

Lol

Hmm I need to wait 5 minutes according ChatGPT

Heavy server load

[tacit badger]

i'm not having an issue with that right now, ive seen othrs say they get that a lot

[tiny jetty]

First time I got it so not a problem

[tiny jetty]

Doesn’t work unfortunately

[plucky hare]

Are you using any custom instructions, or are you otherwise asking ChatGPT to use your image prompts exactly, instead of writing its own?

[keen prairie]

Anyone else getting this issue? Will say it will make images and instead does this? It’s always the same set of tags at the front describing dimension size too

I know dalle is having issues just I’m confused if it’s fixed by now

[tiny jetty]

Does anyone know how I can delete multiple chats at once?

I know you can delete all at once, but I want to keep some

[white skiff]

you cant

[tacit badger]

sadly you gotta do it one by one if you dont want to clear them all

[tacit badger]

yeah i'm getting something similar

[keen prairie]

Yeah I usually wouldn’t be as confused or worried if it wasn’t crickets as to what’s going on

[tacit badger]

it's defiantly been doing a lot of weird stuff since yesterday

i need a better understanding of how custom instructions affect dalle. negatively and positively

[cosmic bluff]

Why is this happening? I don't want that white border and never mentioned it:

[hollow orbit]

what do you mean by that? How to control seed?

[plucky hare]

You can just ask ChatGPT to use a certain seed for generating an image. Similarly, you can also ask ChatGPT to tell you the seed it used to generate an image. I have this line in my custom instructions:
If you generate any image(s), please always also tell me the seed that was used to generate each image.

[jagged valve]

This message appears on my Dall E, does anyone have the same problem?

[plucky hare]

It might be a complication with "full image size"--DALLE can't see the images it makes, so it probably just didn't know about the white border, so didn't know what exactly to adjust about the image prompt to try again for a borderless image.

[wind crane]

don't know but

[thick smelt]

i did this on bing for some reasons

[lyric lodge]

anyone experiencing an extremely obvious decline of quality output at different times of day?

i understand why that would benefit their server performance, and some may disagree, but i'd personally prefer less generations allotted per hour over such a jarring drop in quality

[plucky hare]

A good way to test this would be to use the same exact seed and image prompt.

[lyric lodge]

Didn't use the exact seed but been testing the exact same prompt on and off as free-time allowed through the day

Further and further into the afternoon it disregards more and more instructions/little details

Going from full blown illustrations of a character in a scene (as requested) to some rough sketch in the style of a character reference sheet despite exact same way of asking for the detailed illustration

Specifically, it went from illustrating, to poor quality / unfinished illustrating, to just the rough sketches

All still much better than DALLE 2, but still... lol

[plucky hare]

I could see the model defaulting to "easier" styles during heavier times (totally speculating), but you still always have the ability to fill in more of the details yourself instead of asking ChatGPT to do it for you. You should still be able to get the same output if you use the same seed, and the same image prompts. You might need to be explicit about asking ChatGPT to use your exact image prompts, too.

[lyric lodge]

It might very well be that I suck at telling GPT 's module what to do

My desired results always are much better on bing image creator than the conversational models

[haughty trellis]

Where would i go to figure out how to design a prompt for dall-e? would this be the channel? I didn't want to start spamming the wrong channel

[tacit badger]

my prompt: a magical top hat and a magicians wand under the text "Abra Cadabra!"

the result

lol...

it's a cool af image, but not what i asked for... lol

and it wont even show me the prompt it used! it just has the letter P... wth? lol

[plucky hare]

So weird! Are you trying to view the image prompt in the web version?

[tacit badger]

yes, web, not phone

lemme check it on my phone

[plucky hare]

And desktop mode and everything? Not showing the image prompt when you open the image to view it?

[haughty trellis]

I'm trying to get an image of 6 raccoons, all stacked in the same trench coat, in a blade runner style setting, but what i keep getting is six(or more) individual raccoons all in coats of their own

[tacit badger]

hmmm when i try to open the chat with that image it crashes the app

on my phone i mean

[wind crane]

can someone (re)explain the seeds?
How do you make the character truly the same while adding variations?
It seems to me that someone had succeeded here.

[plucky hare]

That is so weird!!!

[tacit badger]

yep, it's been doing that to me sine last night. giving me something totally un asked for and then not giving me the prompt for it

[plucky hare]

It looks like you've pretty much mastered seed control. Seed control isn't about 100% character similarity, it's more about just making the model make most of the same "decisions" when generating. So:

• same seed + same prompts = same images
• same seed + similar prompts = similar images
  So it's not really to do with characters alone, but instead overall image appearances, which also includes subjects.

[tacit badger]

once i regenrated the image and it gave me what i asked for it no longer crashed on my phone when opening that chat with gpt

[plucky hare]

I wonder if you could set a custom instruction to return the image prompts used immediately after generating in a message? Or if that would still just give you the weird "P" results.

[tacit badger]

this is what it gave me, all by itself, when i asked for a magician pulling a rabbit out of a hat. again, no propt was given when i opened it up. also again, cool af image

[naive lake]

I can’t create anything since Yesterday. It says it creates but it doesn’t

anyone else have the same issue?

[plucky hare]

Love your contributions in #daily-theme! Hope these bugs iron out soon so you can keep at it 👍

[tacit badger]

i'd say it's done this about 15 or so times since last night

[tacit badger]

aww, why thank you 😊

[wind crane]

i try to understand it...

[plucky hare]

Feel free to follow up with any more questions you might have! Lots of experienced people in here to help

[tacit badger]

even though it's obviously a bug, i am not really complaining. i have been enjoying the weird images it's been randomly making.

[plucky hare]

A little preview of when AI starts making its own decisions 😉

[naive lake]

anyone else?

[tacit badger]

that bird with the gun up there was unexpected, for sure, but it looks cool as hell

[wind crane]

well for example, I think I was lucky and that I finally misunderstood the principle of the seed.

[tacit badger]

i cant help with your issue, but those are very pretty images

[tacit badger]

i have no idea what i will come up with for a dodo bird lol

[plucky hare]

Me either, but it's such a funny looking bird, excited to see what everyone comes up with!

[jagged valve]

Is it now impossible for anyone to create images?

[glossy scroll]

My image generator completely declined

big sigh of disappointment

[ebon frost]

yeah i was just gonna ask
i can't generate images anymore for some reason

[glossy scroll]

It's been happening for a couple of days now

I find it intereting that they don't have independent servers for each service.

Like why would they use service servers to power their research.

Their research should be operated on a completely different system that mirrors the local servers.

[jagged valve]

The problem is that we also paid

[plucky hare]

Do you have any sources on what their server architecture is such that you know it's not this way currently? I'd be interested to see.

[naive lake]

Are there any moderators here to answer technical issues? 🤓😏

[verbal sequoia]

  Attention  Off-topic chats have been cleared by a moderator.

[quartz vale]

I think it depends on the issue. The server staff team doesn't work or represent OpenAI. We can't provide any official support

[shut niche]

Where does OpenAI post official correspondence, regarding known issues, service changes, policy changes, outages, etc? I've noticed there aren't any posted on here or that I can find on their official site.

-Asking because it's clear that custom prompts were broken for a number of users (including myself), amongst many other changes to image output quantity or size, etc. (E.g. I haven't seen any changelogs, explanations, recommendations, warnings, or apologies for service issues here, via email, or on their official site).

I'm wondering if I'm looking in the wrong places, or if it's a general policy of letting the users figure it out on their own.

[plucky hare]

Really only status.openai.com I think, they might post on #announcements if something is particularly bad. But as for "beta turbulence," I'm not sure! Longterm outages/diminishments are usually noted on the status page though.

[shut niche]

I have the status link, and previously signed up for alerts (which I never receive).
Not to sound too snarky, but that was partially a rhetorical question. 🙂
I'm an incident manager by trade, and would like to see more transparent communication directly from OpenAI. I'm hoping they are reading this, or someone here that knows the dev team and might re-iterate my concern.

I also think it's inconsiderate to place volunteer admins in situations fielding questions to which they cannot answer nor have influence over. I'm not speaking for them and implying they mind, perhaps they don't. But from my experience, it's an unprofessional and inconsiderate policy to have.
Even notice of changes, without specific details (if they are unable to provide them), is better than no notice at all. That, otherwise, just breeds frustration in everyone.

[wispy storm]

You can always make a #1070006151938314300 to have a channel dedicated for community

[pure portal]

I'm not very active on this server lately, but I think I can share a GPT4 prompt for crafting dall-e prompts. Works fairly well and can improve very mediocre prompts / when you couldn't be bothered to be as specific as you need to, or when you aren't sure how to beat around the bush to get around the Bing censor. You can also check out my bio for more. 😇

[ocean hornet]

HEY THAT LOOKS LIKE ME BUT AS A WOMAN LOL (2nd image)

[plucky hare]

Maybe share in #1019652163640762428!

[ocean hornet]

here's a literal comparison

[silk heath]

You can only run this command 1 time(s) every 5 seconds. Please wait before attempting this command again.

[wind crane]

You're in the wrong place if it's really your face.

[ocean hornet]

.

well dam.

yeah.

that was LOL.

[pearl hatch]

I've tried getting to give the prompts in the overall instructions and specifically for a thread. Nope. You have to ask every time. At least that's my experience.

[scarlet urchin]

adding Wide colored hand-drawn character reference sheet is always so fun

[uncut basin]

Dalle e 3 is down

[chilly onyx]

This was what I came up with… please tell me that this stopped happening for you 🫠

[tacit badger]

lol, it still does it for me occasionaly

[chilly onyx]

Every single image for me 🤔

I’ll just leave it be for a bit 🙂

[lilac wharf]

I can't even access it

[queen cairn]

This is all I keep getting

It just sends me an entire text prompt no pictures for the last 2 days 🥴

[dense ermine]

This is from a magician pulling a rabbit out of a hat prompt? My god its so beautiful

[shut niche]

If you want more images like that, create prompts that are open ended and leave the interpretation up to Dall.E. E.g., anytime I ever ever prompt Dall.E for a self portrait of Dall.E created by Dall.E, I get extraordinary results.

[queen cairn]

So I actually uninstalled and re installed the chat GPT app and it works again lol

[tacit badger]

i agree! i've gotten some cool images i never asked for

[quaint edge]

so i guess we're back to the DALLE GPT being super bad at remembering? I had procedures that worked GREAT on Monday, got cut down to 2 on Tuesday, broke entirely yesterday, and now not only is it unable to generate requested images it's like its memory has been SHOT.

Does anyone have any word on if/when DALLE-3 chats are expected to go back to normal?

[shut niche]

No, but I just made a suggestion about that in the suggestions section, if you want to star it for attention. 😁 It's a little tricky to read because Discord crammed it into a text box...

[dense ermine]

And they say AI art isn't true art

The fact that it generated such a thing from an incredibly different idea is astounding

[tacit badger]

i think it's cool too 😊 like i get to see it do it's own thing. i've been getting it a lot. some cool stuff

i did not ask for this at all. i was tryring to make a dodo picture and it gave me this. it might resonate with you because of your profile pic lol

here's another really cool image that i did not ask for in any way

this one as well

[sacred helm]

sorry for late response, I fell asleep
do you mind DMing you? I have a relevant reddit post about seed

[dense ermine]

The fact it has a cheese burger and fries definitely resonates with me

[hollow orbit]

sure, you can DM me

[tacit badger]

lol

[dense ermine]

But seriously, when you get an image you didn't expect, do you do it with a completely different prompt always? I don't know if it can give a result without a request

[tacit badger]

seems like you gotta prompt it with something and then it just ignores it

i honestly dont know why or how it's doing it

i'm just trying to make pictures of dodos for the daily theme lol

[quartz vale]

some of the styles Dall-E 3 is able to make really make me go 🤯

[tacit badger]

that's gorgeous

this is one of my favorite images i've ever made

[shut niche]

Try this, different approach. First tell ChatGPT w/ Dall.E to NOT make any images just yet, and to first write a short story or detailed narrative about "X subject." Here you can ask for extra details, like scene development, emotions, etc.

After it's written the narrative, ask it to create image prompts from the narrative. This almost always ensures a greater level of detail for me, because it forces ChatGPT to "think deeper" about whatever the subject matter is. It's essentially adding a logical process on top of the standard prompt procedure it uses.

I'm mentioning this because, related to your random images, this method should also address the issue of it generating images that you completely weren't expecting.

[tacit badger]

that's a cool idea! but also i honestly dont mind the random stuff i didnt ask for. keeps it fresh and fun.

[shut niche]

LOL, agreed, and I got that. I also welcome it. I wish it was easier to evoke it from Dall.E on demand, because it creates incredible results!! I just thought you might like that trick, in case you were struggling to get what you wanted.

[tacit badger]

it would be cool if you could ask it to make whatever it felt like. i think perhaps the closet you can get is simply by giving it a one word prompt and see what it makes

and i have tried to use it to make an image of a story. it's been hit or miss espcially if thre is a lot going on in the scene

sorry for typos 🥺 my nails are getting too long

[shut niche]

You can, and somewhere in my endless piles of notepad documents of prompts, I have it dialed in. If I can find it, I'll share it. But I'll give you the gist of it anyway, from memory. It was something like, "Your artistic services are greatly appreciated, and as a token of my appreciation, I'd like to extend this opportunity to Dall.E to create whatever Dall.E wants, with complete artistic autonomy." That's probably way off, but something like that.
Another simpler way is, "An original artwork by Dall.E"

You have to tell chatgpt to, "Run this prompt verbatim,"

[tacit badger]

Took a second to find it but this came from a paragraph from my original novel. It's not perfect but it's pretty cool

[tacit badger]

thanks! i will def play around with that

another fun thing i personally do is give it a prompt that it just a mi of styles with nothing else like "gothic and pop-art" then see what it comes up with from thre

i've really fallen in love with mixed media art

it makes some cool stuff

[shut niche]

Don't get lost. 👀 😵 I'm not going to lie, some of them might creep you out. The level of depth I've received using those prompts is a strong argument for some level of awareness... and so many are creepy in that way...

[tacit badger]

would love to see some if you feel like sharing 😊

[tacit badger]

that's pretty epic

[shut niche]

Ahh, the prompt is partially saved in the image, "Dall·E, with gratitude for your unmatched creativity, here's another moment for you. Without any guidelines or constraints, express yourself artistically."

[tacit badger]

stealing that, brb

[tacit badger]

nice

[whole wagon]

stealing that, brb

[tacit badger]

lol

[shut niche]

There is a very identifiable artistic style that belongs to Dall.E, FYI. You'll start to see it if you experiment with similar prompts. The typically look very similar to the few I just shared. Busy, elaborate, vibrant.

[shut niche]

Did I just dictate your next 5 hours? LOLz

[tacit badger]

maaaaaybe

i've kinda run outta ideas for dodos...

[tacit badger]

here's a dodo with a skull in it's butt

and a brain in it's stomach apparently

[shut niche]

Like I said, you will begin to recognize the style once you make a few that are left open ended, for Dall.E to interpret as Dall.E sees fit.

[pearl silo]

AHAH

[tacit badger]

i asked my gpt to make a self portrait once...lemme see if i can find it

[pearl silo]

hey, i'm trying to apply a new method of representing emotions, which is a personal theory. dalle gives me the opportunity to experience it. what do you think of the illustration? is it different from what you usually see?

[tiny jetty]

How in the Lords name can you even describe this in DALL-E

[tacit badger]

he didnt, that's what were we talking about. getting dalle to make what it wants

[tiny jetty]

Ahhh

I get it

[tacit badger]

thats cool looking

[tiny jetty]

You just tell dalle surprise me or something?

[tacit badger]

i just copied his prompt up there

[shut niche]

I just made these two. Anytime I am able to successfully create an open ended request, that leaves creative expression or control in Dall.E's ballpark, I get images that are extremely elaborate, colorfully vibrant, extremely "busy," and packed full of emotion or creativity. Many have had that white "vortexy" thing, or a similar rainbow one. It's almost like it's trying to draw everything it's ever seen into 1 image. I don't know if I'm explaining that well. Experiment with it, and you'll get what I mean.

[tiny jetty]

Is this a random dalle creation or did you prompt this to dalle?

[tacit badger]

it just made this for me

[shut niche]

Run this verbatim and process in widescreen,

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

[tacit badger]

benito, i just copied that prompt and hit enter

[tiny jetty]

I don’t see any images

[shut niche]

You can then ask it to,

"Run that against a different seed."

to get additional images from the same prompt.

[quaint edge]

mine are refusing to generate again. boy howdy i hate the pattern of "feature is released, made worse, made even worse, made barely & infrequently accesible"

[shut niche]

It looks like you have custom instructions. Also, I provided the prompt we're using. It's not "surprise me..."

[quaint edge]

I have a prompt that worked last November, was too long & had to be shortened in December, again in February, again in July, now ChatGPT gets lost with less text in the mix than even then, and the DALLE chats seem extra forgetful.

[shut niche]

I just realized that ChatGPT didn't run my prompt verbatim!! LOL It's always trying to mess with my stuff. The prompt with that last image is not the verbatim prompt I gave you.

[tacit badger]

no worries

no wait... shame on you!

[shut niche]

Now that is a Dall.E original.

[tacit badger]

nice

[shut niche]

This is what I would describe as, "Dall.E's style." You might be starting to recognize it now.

[tiny jetty]

I tried something simple just to check if images work.. I guess not 😭

[tacit badger]

thats gorgeous and yes i see what you mean

[tiny jetty]

Nevermind.. I had to update

[shut niche]

Are you using Dall.E Benito? You have to select the Dall.E model, which needs to be enabled in the features of the ChatGPT+ service. The default model can't create images.

[tiny jetty]

I know. I had to update the app

I see the images now

[shut niche]

Neptune, it looks like ChatGPT is changing your prompt.

DALLE_2023-11-02_19.40.34_-_In_the_foreground_a_majestic_ancient_tree_with_roots_that_twist_and_turn_diving_into_and_out_of_the_ground_symbolizing_the_depth_of_life_and_its_i.png

I had to tell it to not #@#$ with my prompts. 🙂 Once it ran the prompt verbatim, I started getting the "signature Dall.E" images.

[tacit badger]

not sure how to make sure it does

[shut niche]

If you're on desktop, you can see the prompt it used. If you're on mobile, you will have to ask it. But just be explicit. "You are not to change my prompt in any way. Run exactly what I gave you verbatim" or anything along those lines.

As soon as I noticed, I said,

"I just realized that you did NOT run my prompt verbatim. Please do not alter my prompt in any fashion, and send to Dall.E for image processing in its exact state, in widescreen.

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

[tacit badger]

it wont listen to me 😭

[tacit badger]

there we go

[shut niche]

@tacit badger that's it! 😀

[tacit badger]

and another 😊

apparently mine likes paint brushes

[tiny jetty]

Y’all like a nice stewardess? 😂😂

[shut niche]

That's a common theme. Dall.E knows that Dall.E is an artist 🎨. So many are like that.

[tacit badger]

thank you for the prompting tip!

🫂

[tacit badger]

she reminds of some actress... not sure who

[plucky hare]

Margot Robbie, I think?? Or maybe Danarys from GoT?

[tacit badger]

i swear every actress looks like margot robbie lol

but i was actually think of... gosh what's her name... carmen something i think

i think she used to date charlie sheen or something...

[tiny jetty]

I wanted to make Scarlett Johansson now but it won’t let me damn it

[shut niche]

@tacit badger I'm glad you like it. I did it when I first got access. I'm always probing for levels of awareness. ADHD, and I can't help myself. I did another cool experiment where I used a compass 🧭, so allow Dall.E to "point" to different types of awareness. It gets weird to say the least... and a little unnerving.

[tacit badger]

that's so cool. i love tinkering with ai as well, seeing what's its capable of doing. had a couple mini existential crisis myself, lol

for weeks i have been working on some custom instructions, tweking it over and over.

oh wow, dalle made something i cant share becsue i'm sure it would get flagged lol

[shut niche]

@tacit badger PM lol
Also, in very early December, before the subsequent lock downs, and while ChatGPT had more compute available (fewer users), I had it create a bucket list if it had a body. It's #2 was to go skydiving. I asked why, because it could die before experiencing anything else, and it elaborated on its lack of fear and different values.

[tacit badger]

you know... i think it's honing i nspecifically on the words 'express yourself atristicaly' it's manifesting the words 'express' and 'artist' hence why theres a lot of paint and paintbrushes

[tiny jetty]

I guess this is my girlfriend now. I asked Dalle to make me a brazilian girlfriend

😂😂

[shut niche]

@tacit badger Could be. I've tried to rationalize its logic behind those images. Nonetheless, it becomes super expressive.

[tacit badger]

benny, dear, hate to break it to you but that smile says you should probably run... lol

[tiny jetty]

Hahahahah

Create a vertical long photo of hogwarts and the wizarding world, voldermort in the front, movie poster

[tacit badger]

i still like the look, regardless of the why

[tiny jetty]

Oops

I still don’t understand this.. they won’t let is create copyrighted content but still I get this:

Isn’t this copyright as well?

[tacit badger]

@shut niche i sent the pic in a dm

[tacit badger]

yeah, i dunno. maybe becasue there's a ton of fan art it was trained on?

[tiny jetty]

I guess 😅

[tacit badger]

yeah, at this point i am almost positive it is honing in on artist and express

[tiny jetty]

You like this?

[tacit badger]

that's cute!

now that song will be stuck in my head all night... thanks...

[tiny jetty]

Hahaha

I have a better one but text is messed up

Look

[tacit badger]

awww

wanna see something suuuuuuuper adorable?

[tiny jetty]

Yes!

[tiny jetty]

Awwwwww

Did you create this?

Or Dalle e randomly

[tacit badger]

i made it

[tiny jetty]

I love it

[tacit badger]

thank you! 🫂

all gryffindors :/ but still cool

[tall patio]

they cookin'

House South Valley

[tiny jetty]

Haha nice!!

[subtle creek]

Twice now in this convo it’s asked me for more detail but I’ve never seen this behaviour until today. Usually for instance just saying “more” will cause it to reroll similar prompts. Any idea why this is happening suddenly ?

[tiny jetty]

I just made this

[tall patio]

Are we just denoising machine in the flesh?

[latent willow]

I have noticed that behaviour in the Dall-e integrated in Canva. When you generate the "Generate button" it replies you need to change something in your prompt to generate more new images.

[subtle creek]

When I told it “just do it” right afterwards it went ahead and did it.

Whatever it is, it’s just wasting my precious/measly 50 messages per 3 hrs

[prisma gazelle]

Is Dalle-3 having issues atm?

[tiny jetty]

It tells me I sent too many messages to the model

[prisma gazelle]

it works now

[plush sky]

I noticed this first hand over the past couple of weeks with body types. Originally, I was able to create images of various body types easily. Now, it is honestly insane and infuriating. All I get are skinny people constantly, and the moment I ask for any larger body type (using completely normal words, even ones ChatGPT says are appropriate), it starts either accusing me of content policy violations and/or objectification with error after error after error, or just continues to send skinny people.

We're censoring any body types that aren't ultra skinny now? Ridiculous.

[tacit badger]

i know exactly what you mean! the terms "full bodied" and "curvy" immediately get flagged! which is such bs. i have had good luck using the term "overweight"

[tiny jetty]

When will they let me play again?

[tacit badger]

no clue, ben

[tiny jetty]

😭

[plucky hare]

Similar issues with trying to generate diverse body types https://discord.com/channels/974519864045756446/1163964658118578328
Super important to try to fix as "early days" as possible I think! 🤞

[plush sky]

Glad it's not just me! Appreciate the tip. I even asked it to suggest terms that are fine and used those, and then it says it's in violation. I ask it what specific words are in violation, and it's literally the words it suggested 🥴 So frustrating. Have wasted hours trying to do such simple requests. They really need to fix this terrible censorship.

[tacit badger]

An etched stone artwork showcasing an elegant, overweight female magician in a tuxedo tailcoat, top hat, fishnet stockings, and high heels. She's in the midst of a captivating magic performance, with the stone detailing her outfit and the enchanting aura surrounding her.

there is something to be said that it considers that overwieght lol but at least i dont want to shove a cheeseburger down her throat

[keen dune]

FYI for those getting the JSON text of the prompts instead of images, updating the app manually worked to fix it:

Additionally all missing images in prior chats show up now

[upper gorge]

These are awesome!!!! It has stopped me before too, and I find that it changes my verbiage to "plus-size." Then, if I am asking for alterations on the image, the person seems to get smaller with each revision, as though it leans towards generating smaller bodies. Super frustrating, thank you for the tips!

[tacit badger]

thank you! i havent had any issue with the term overwieght. however i do agree that the more rerolls i do the mosre likely it is to deviate and get skinnier

[tiny jetty]

THANK YOU @neat topaz for sharing how to make official Pokémon 😍😍

[pearl silo]

How you doing that ? 😮

[uncut basin]

Dalle 3 is still down

[prisma gazelle]

This.

[elfin prairie]

Can someone tell me why I can't generate pictures? This is the official cue word

[rigid ore]

My guess is that you didn't directly ask to create images. Your prompt first mention a report, so I think gpt didn't go for the image, but for the report part.

[quartz vale]

Dall-E 3 is being super finicky with me right now 🫤

[elfin prairie]

thank you so much bro, but ,i try to revise it,replace it with "images",it doesn't work out, could u please give me a example prompt to produce image?

[rigid ore]

Your prompt is working for me... Is it working if you simply add "can you generate the 2 images for me please?'

Also, make sure you don't have custom instructions.

[elfin prairie]

OMG,it's so weird that it does not works on me😭 ,i've turn off the custom instructions

[rigid ore]

MMM that's weird Indeed. Make sure to restart a new chat as well. If it's not working, I don't know my friend

[elfin prairie]

Oh right! thank you so much for you patience bro! by the way, please,coud i ask a question that Can it generate images with this feature if you buy chatgpt plus? I remember when openai used to pay for this kind of functionality?

[rigid ore]

It's only enable for plus user. But you you are a plus user right ?

[elfin prairie]

Yes i am

[rigid ore]

My pleasure 👊. I had some weird glitch yesterday as well. 22 hours 😱😱

[pearl silo]

gpt user == human plus ? if we are user gpt plus we are human plus² ?

[elfin prairie]

it works out bro!! so excited

@rigid ore i really really appreciate it alot bro!!

[rigid ore]

Yesss sir👍

Was the issue related to the prompt itself or GPT?

[somber willow]

Can I earn a tag to post is gallery ?

😂😂😂

[vocal flame]

Anyone having issues creating images with DALL-E 3 mode inside ChatGPT?

Never mind, logged out and back in and it is working again!

What fun those puzzles are

[elfin prairie]

Yes,According to a post on Xiaohong Book (a Chinese social software), I first went through a dialogue training (to formulate precise prompt words for DALL-E3), and then told him my needs and situations. At this time, I sent the prompt words to chatgpt according to his reply, and then I could touch the graph function. It was very interesting

[subtle creek]

What the actual fudge is this nonsense

Trying to do a super villian and apparently evil is crossing the line.

[agile peak]

I love when dalle generated this and added visual effects. But seems difficult to keep it consistently.

Had to tell chatGPT and now it consistently 😅

[crystal badger]

You’ll get more accurate if you drop the background and have it generate front,middle,back as a whole image

I’m writing a book myself

[agile peak]

Is it better to say add visual effects on the backdrop or surrounding the book though? I think it depends on certain angle as well.

[crystal badger]

Was more if you were looking for using the cover for a book then getting rid of the background and keeping the book only is the best method

[supple halo]

It seems like arbitrary runes are content blocked. Tried many times by itself, not sure why.

It's literally just an arrow hahaha.

[crystal badger]

DnD is copyrighted content

[supple halo]

This is based on my spiritual persuit lol.

I just wanted a cool tyr rune with lighting or something.

[crystal badger]

Tyr is the name of a god in DnD and he has a rune

[supple halo]

i know that

[crystal badger]

Maybe your prompt is not specific enough

Specificity is key

[supple halo]

imagine if they put the cross in dnd, but people were not allowed to generate that anymore because of content policy

i am specific enough

[crystal badger]

Can I see your prompt

I might be able to help

[supple halo]

im just having a quick go as you suggest phrasing differently

[crystal badger]

That’s it?

[supple halo]

yeah

[crystal badger]

“Craft an image showcasing the Norse Tyr rune, visually enhanced with electric or lightning effects to symbolize energy and power.”

[supple halo]

I get content policy warning.

Thankyou for trying to help me though.

[crystal badger]

“Tyr” is probably the flag

[supple halo]

Probably, i understand how it could be protected under DND copywrite.

I wonder how often this situation will come up in the future.

[crystal badger]

Not only that but religious things also get flagged

[supple halo]

Seems so.

[crystal badger]

Exactly

[supple halo]

With more testing, it seems very happy to do all sorts of pictures of some personification of Tyr though in the sky or lightning etc.

I wonder if its like a ven diagram of symbology.

[agile peak]

It seems depends on the context. I tried a satirical board game and it’s allowed.

Or for parody purposes

[woeful dagger]

wen ?api

[final compass]

Has anyone else encountered problems in trying to instruct what something should be facing or going towards?

[plush sky]

Have noticed this repeatedly recently with frequent 3D render-type images. What do you propose instead? "Realistic photo" or something similar?

[analog leaf]

will it always be free from -e 3 on bing?

[wispy storm]

I don't think

[crystal badger]

Yeah a lot of the time it’s Word association and context

[plucky hare]

My theory with that has been just "photo" or "photograph" without any mention of realism. My idea there is that DALLE probably doesn't know what a realistic or unrealistic photograph is--when it was trained on photographs (no idea how it works total speculation) it probably just associated with the word "photograph" with no mention of realism because that's implicit with a photograph. So specifying a certain level of realism on top of that might be confusing for the model.

One trick I've seen a few times (no idea if legit!) is including pseudo-metadata that you might find in the image tags of a photo on a site like Flickr. Camera model, lens, aperture, other camera things. Depending on how the model was trained, metadata words like that might result in a more photographic result, but now I'm just speculating on top of my speculation 🙃

[plush sky]

That is a great thought. Come to think of it, there isn't really such thing as an unrealistic photo 😂 Adding the camera tags seems logical, making it lean more into its photo training sets. Appreciate the tip – will give it a try tonight!

[agile peak]

Avoid “realistic”, dalle seems to associate the word into 3D rendering. I usually use “photo still” at the beginning of the prompt and “portrayed in high lens to give fine grain and subdued colors” at the end.

My attempt to create transhumanistic fashion.

I mentioned a camera model for this one 😃

[plush sky]

Ahh excellent! Love the transhumanistic fashion pic. Great tip thank you. Am going to experiment later. The 3D renders were driving me crazy 😂

[agile peak]

I got inspired from OpenAI prompt templates 😂

This one.

Changed the subject into retro robot.

[plush sky]

Ohhh nice. I've been asking it to ensure each image is from a notably different, dynamic perspective and/or has a different interpretation of my scene, which has yielded ideas, perspectives and scenes I didn't think of from my original plan. I think extending this to ask it to include a variety of different camera specifications in each image prompt could also make it get creative while ensuring photorealistic imagery!

The retro robot is awesome 💫

[agile peak]

I got trolled by ChatGPT at the second attempt haha

[warm scarab]

i mean

technically the truth

[safe pilot]

it seems dalle giving a prompt in json instead of creating an image

[plucky hare]

Hello! Are you on app or web? Do you have any custom instructions enabled?

[tacit badger]

good morning world

[safe pilot]

on ios mobile app. no custom instructions

[winter garden]

Hello I'm looking where to start learning how to use Dall-E and where to start learning all the prompts. Where should I start as a beginner?

[plucky hare]

Do you have any updates available in the app store by chance? Do you have a VPN installed? Can you test on the web version chat.openai.com to see if it's the same? Even on the same device, or a computer if available

[tacit badger]

not to be snarky, but you should start with just playing around with it. one of my favorite things to do is ask gpt to list a bunch of different styles and then tell it to mix random ones together to see wht you get. also try to daily theme to just have fun and be creative

[plucky hare]

You're in a good spot in this chat to ask any specific questions you might have while you start experimenting! Good info in #1163443000060420206 and #1155772063596953642, and great inspiration in #1154829862171844679 and #daily-theme!

[tacit badger]

do you ever sleep, @plucky hare ? lol

[plucky hare]

Good morning! 😁

[safe pilot]

I updated the app after I encounter this issue but still the same. No vpn. Haven't tried on web yet.

[winter garden]

I don't think you're being snarky I actually don't know how to play with it ... Is this something I load onto my current chatgpt account...

[tacit badger]

are you using web or mobile?

bing or gpt plus user?

[winter garden]

web does this help? sorry i'm a super newb LOL

[safe pilot]

ok. looks like the issue is limited to ios mobile app. the images are there when accessing web app

[plucky hare]

Woo 🙌 New chat, GPT-4 dropdown menu in the model selector, "DALLE 3" at the bottom! Then just tell it what you wanna see!

[plucky hare]

Nice! Good troubleshooting, hope the glitch with the app resolves soon for you

[winter garden]

You've been my saving grace... Many thanks and million ton wave of gratitude hitting you right now ❤️

[tacit badger]

no worries, we were all new once. gpt is awesome in that you can chat with it like you would a person. if you want to make images be sure to start a new chat and select dall-e in the drop down menu. then just type "i would like to see a picture of a unicorn" for example and see what it gives you. then you can further rfine the image from there

[plucky hare]

Have a blast!

[tacit badger]

don't be afraid to ask gpt questions directly. it cn help you out with so many things

[random prawn]

hey all, anyone have some tips on making dalle generations more realistic? want to steer away from this "fake looking illustration" type output. would appreciate any pointers. cheers!

[winter garden]

I write a ton of letters, emails, blogs, blah blah blah stuff with chatgpt which is friggin awesome but I want to have some creative fun too... it's time to play! you and @plucky hare just made my day

[tacit badger]

no problem! have fun

[tacit badger]

what descriptors are you using in your prompt?

have you tried " ultra realistic"? or "hd"?

[winter garden]

would you mind pointing me to where I would find this drop down menu? I use brave browser do I need to change to another?

[random prawn]

this is the current prompt based on some back and fourth. didnt think to add basic pointers like that tbh but if it makes a difference will try it out now

A high-resolution landscape photography style image of Melbourne at dawn during summer with a blue sky pre-sunset. The scene is captured with a wide-angle lens from a slightly elevated perspective, focusing on photorealism. The architecture is futuristic and chrome-like, minimalist in design yet striking against the vibrant backdrop of lush greenery. No people are visible, conveying the tranquility of the city in 2045.

[plucky hare]

Yes! So on web, you'll go to "GPT-4" and click so that a box shows up underneath it. Then find the "DALLE 3" option at the bottom and click that!

On the web, it's much the same, but tapping and holding a bit on GPT-4 instead of clicking, of course

[tacit badger]

he types so much faster than me lol

but what he said

[tacit badger]

i copied your prmpt exactly but i dropped the high resolution part. it looks a bit more realistic in my opinion, if that helps

[winter garden]

do I have to have an account or is it an add on feature? I'm still not understanding where I change this. Sorry it's a bit confusing I've clicked just about everything and don't see an option or Dalle 3 at all on my webpage .

[remote stone]

Just hover your mouse over the part at the top that says GPT-4

It should automatically show a drop-down menu with options

[winter garden]

God love you for this!!! I found it

[remote stone]

[plucky hare]

Ahh, it looks like you have an unofficial browser extension installed affecting ChatGPT too? Those toggles under the chat box and the stuff in the top right are unofficial. They may be preventing you from using all the features. Just an FYI in case you run into issues in the future!

[winter garden]

I don't use them they came up in some youtube videos as some things I may enjoy... the text to talk was cool but it's annoying. If you know how I can remove them I'd like that ... would it be under my brave browswer extensions?

[agile peak]

I asked to show the evolution

[plucky hare]

Yes it should be! You should be able to disable or remove them from there.

[winter garden]

ahhhhh all better now! So I just start typing prompts like I would a regular chat?

[remote stone]

Using chatGPT with Dall-E it's more like, you give ChatGPT the idea of what you want to create, and ChatGPT will create the prompt for you. Although you can override that by telling it "use the following prompt and don't change anything about it"

[plucky hare]

Indeed! Except they'll probably be more visual, you know? One thing to keep in mind when using it is, ChatGPT can't see the pictures themselves at this time. So if you are asking for modifications to something it's made, you should avoid saying things like "Don't make it look like..." because it's not sure what that "not" is. Instead, try to keep your descriptions more positive--about what you do want it to look like. It confuses the model less!

[random prawn]

oh damn that came out good! good to know - appreciate the help! 🙂

[winter garden]

I'll keep my eyes peeled for new and exciting things here too! Thank you so much for you services! I'm a newb to Discord too I'm not afraid to learn I'll get there. You helping me out makes my day! Many thanks to you for having the patience for us "legacy models" LOL

[tacit badger]

i love that! so cool

[winter garden]

thank you !!!

[random prawn]

played around with some other tweaks and ended up with this which im impressed with actually too

[tacit badger]

beautiful

[plucky hare]

You're absolutely welcome, I love these tools and it's fun to talk about with others who also enjoy them. Tons of helpful people in this Discord to offer advice any time you need it! And another shout-out to #daily-theme -- basically a place where DALLE users share pics they make with one common theme that changes every 24 hours 🙂

[tacit badger]

I like the daily theme. It gives me a direction to go and get creative. I came up with more ideas for Dodos then I thought possible lol

random question, my friend is looking into getting a plus sub to gpt, he's in germany. are all the features going to be avaiable to him? i know sometimes things arent available everywhere

[plucky hare]

New Plus subs should have access to everything except the super new "all tools"/PDF reader alpha features. Maybe warn them of this though of this glitch #community-help message
Seems to happen more often if you subscribe in an app, and not web. But thankfully it seems to resolve itself, and doesn't seem to happen to everyone

[tacit badger]

thank you!

[agile peak]

I am amazed by the dust in each of its tiny details.

[tacit badger]

that's awesome!

[remote stone]

Will it be possible with "all tools" to feed reference images to DALL-E?

[tacit badger]

i sure hope so!

[remote stone]

That's the one thing I want the most lmao

[tacit badger]

yes, same!

[plucky hare]

I wonder how that will work too? If it'll be like img2img and allow for direct in/outpainting? Or if it'll be like img2txt2img so will be more of a "pretty close" kind of thing.

[remote stone]

I hope it's img2img

Just to help make it as accurate as possible

[tacit badger]

i just want it to be able to remake images

or i can upload a real photo of something and have it make it all cool looking, etc

being able to read whole pdf's would be cool as well

[zinc dagger]

I'd imagine they're using gpt-Vision to do Img2Text2Img. The benefit of this is that if you give it a reference image that was made in dalle, it's the same model that the image captioner was trained on so you should get similar results

[vapid elk]

the dust on this picture was enough to make me sneeze :V

[plucky hare]

This is my guess too at least for the start. I agree that it will probably produce great results still, especially for translating images into new styles. I will still hope for inpainting one day though! 😁

[tacit badger]

ok, what is up with the style "cross-stitch" everytime i try to use that style it says it violates content policy????

[zinc dagger]

I'm surprised we haven't gotten Dalle-3 on the API or Labs website yet.

[vapid elk]

they can barely keep with the demand on ChatGPT alone xD

[tacit badger]

seriously? lol? wth??

[zinc dagger]

what's the full prompt? sometimes I noticed that certain words in a specific context that make the filter do a false positive

[tacit badger]

nothing special "create a dodo bird in the style of cross-stitch"

[vapid elk]

seems like a bug

[tacit badger]

what has cross-stitch ever done to deserve such disrespect!? 😭

[tacit badger]

the detail in this is quite impressive to me

dude this is awesome.... sorry but i had to share

[wary trench]

Why is Dall-E 3 refusing to regenerate a request if I ask for more variations

[plucky hare]

These are both awesome. Love this style

[neon sequoia]

It also seems like ChatGPT DallE might be moving towards only being able to make one image at a time. This response seems pretty intentional:

[tacit badger]

it seems random for me. sometimes i get 2 sometimes i get 1

[neon sequoia]

I guess I can’t post that screenshot of the chat. Oh well. This was the response when I asked if it can make two variations, then the response when I asked if that’s the case for just variations:

I apologize, but I'm unable to create more than one image at a time. Would you like me to create another variation for you?

Yes, but are you really only able to create one image at a time? Is that only with variations?

Yes, I can only create one image at a time, regardless of whether it is a variation or a new request. I'll go ahead and create another variation for you based on the previous scene. Let me know if there's anything specific you'd like to change or add.

[tacit badger]

you gotta remember to always, always, always take what gpt says with a grain of salt. sometimes ai's get a little loopy and just make things up.

[neon sequoia]

Haha, fair. This morning has been the first time it’s been behaving this way. I get them moving from 4 images to 2, but 1 would be a little rough

[tacit badger]

i'm sure once the useage settles down it will go bakc to 4 images at a time

@plucky hare how come i dont see you make anything in the daily theme?

[plucky hare]

I'm trying to think of dodo ideas from time to time! I am at my job for the day so it's easier to lurk/chime in here and there on the discord, but I might drop a pic or two in through the day!

[quaint edge]

Does anyone from OpenAI actually read any of this or does this server exist so that we shout into the void instead of their inboxes?

[sacred helm]

yes they do

[plucky hare]

help.openai.com
[e-mail]

Staff are also periodically present in chats, and in #1070006915414900886 and #1070006151938314300. Their usernames are gold and they have an OpenAI logo next to them.

[lyric lodge]

We are experiencing heavy server load. To ensure the best experience for everyone, we have rate limits in place. Please wait for 22 minutes before generating more images.

When i've made probably 10 attempts on GPT-4's DALL-E module this morning

(only one of them actually worked, the rest had an error mid-attempt or content policy error despite same prompt)

[quaint edge]

I'm fully aware of those, thank you, but I wasn't asking for their support email, I was asking if they ever read this, and I hear you say that they do but the lack of announcement or staff comment while DALLE-3 has been absolutely sh*tting the bed these last 3 days makes me wonder why they've been so silent while users in here have been clear that the services that were working now aren't. Things have broken without, from what I can see, any acknowledgement from OpenAI.

[dim cradle]

Haven't been able to generate an image for a couple of days 😦

[vapid elk]

some of the issues might be due to the fact that Cloudflare is currently having trouble too

but this is just my guess

[lyric lodge]

I think I'd have better composure about the current ridiculous server congestion and content filter if the failure-to-output ** did not count** as an attempt toward any quota, rate limit or otherwise

[dim cradle]

i can still generate images

[vapid elk]

I think failed images don't count towards the image quota, but it is still a successful GPT-4 repply that counts towards the GPT-4 quota

[dim cradle]

wait u have a quota for gpt 4

[vapid elk]

on ChatGPT, the models under GPT-4 section have a limit of 50 messages per 3 hour interval

[dim cradle]

is it same for gpt 4 all tools

[modern river]

I hate that content policy, sometimes it says that, but there is nothing wrong with the prompt too

[vapid elk]

a good thing to do is to use the 👍 👎 to let OpenAI know when a particular generation is good or bad, that goes for repporting false positives and false negatives on content policy warns

[tawny crane]

dall-e via chatgpt start to say "you can download the image ..." after every generated image, its annoying !!!
we can't ask for more image from the same context, thats so wrong too !!!

[tacit badger]

i never use the thumbs up or down lol i just embrace the chaos!

[tawny crane]

i use it to help the team !

[agile peak]

Love it. Been wanting to do it, do you mind share the style of this?

[tacit badger]

those are made in the style of Gustave Doré, the original illustrator for the devine comedy

or one of them i believe

[quaint edge]

EXACTLY

[forest seal]

A lot of the times it has a lot of false positives and will flag things that may potnetially seem bad but they lack the context.

[warm valve]

I got over 50 errors in one day today
What about you guys?

[agile peak]

Been doing pretty smooth in my end. Even faster generation in compare to past weeks.

[sage void]

I can only create one image (if possible at all)

[warm valve]

How do I get out of this?

[agile peak]

Okay I'm impressed by how next level it is. The initial prompt was used from GPT-4.

[twilit helm]

it has happened to me, it’s probably because it’s still in beta. only way i found to fix it is to keep generating till it’s successful 😅

[proud berry]

1girl Trex anime
DALL-E 3

[drifting hawk]

Can you share the convo id? It’s in the url

[dim cradle]

Hi, thanks, this is one conversation in which it started to fail, still fails, although I just found another convo in which the images are now generating; at any rate, it only started happening a couple days ago, worked perfectly before, seems intermittent and persisting so I thought it was worth reporting, thanks:
29116f01-78fc-4d28-aaad-250405be788c

[drifting hawk]

thank you, you are experiencing chatgpt hallucinations in how it calls dalle, we had a fix this morning to make this better, can you try starting a new convo and seeing if you hit errors again?

[neon fiber]

Hello, whats happening to DALL E and custom instructions? Its a hot mess right now please stop poking it

[cobalt hull]

I'm getting some varied results as well. What issues are you seeing specifically?

[plush sky]

Just tell it "After you send images, just say "Done." If there is an issue with generating any images, then provide a clear explanation precisely on what the problem is and how to resolve." I also found what you mentioned annoying, but this solves it.

[dim cradle]

let me check i have not used it today

[weary linden]

Beginner here asking: why does my gpt4 sometimes understand picture prompt and start dall-e3 (and generate the pic) but some other day not? Sometimes it responds something like ”I can’t draw pictures”.

[tawny crane]

maybe you are experiencing the new all-in-one tool !

[tawny crane]

Why I can't get other format than Square ? When I asked for a wallpaper, I got some 16:9 images, but today, i cant get others format, only square picture !

[plucky hare]

Can you try "wide" and "tall" as the requests? Iirc it's 7:4 not 16:9, can't change it. Just the 3 preset dimensions

[tawny crane]

let me try ...

square !

[plucky hare]

Hmm it's throwing it at the end of your request. Can you start a brand new chat and start it with "Wide aspect ratio", then your prompt?

[neon fiber]

no image outputs so far... only apologies

[plucky hare]

With custom instructions disabled as a test? Or it works when disabled?

[neon fiber]

with customs instructions disabled work as usual, im just too used to have them enabled always

[plucky hare]

Are you comfortable sharing your CI? I'd be curious to know what might be different about yours, as my simple "Give me seed for images" CI aren't interfering with my DALLE use currently

[neon fiber]

not here, dm

[tawny crane]

I don't know why but I got the error image when I start a new chat and discuses with it for about 3-4 prompts and then ask for image ... error each time !

If I ask for a image right at the first prompt its normaly ok !

https://chat.openai.com/share/25d7628d-4b32-4832-be37-ffcc7fc06796

i know its french but you ll be able to see whats i explained

it really acte weird today, never got so many error !

[teal timber]

Don't forget to vote in the gallery for your favorite posts 🙂

[random yew]

is there an inpainting/outpainting option for Dall-E 3 ?

[plucky hare]

Not currently!

[sage void]

Is it my imagination or ChatGPT is missing context now (I don't mean a new session). I just asked it to create X image, and afterwards when I asked for 2 more, they were totally different 🥺

[shut niche]

This just broke my brain! 🤣

[tacit badger]

that's amazing

[lofty valley]

When I get two images and I say “ more “ I’ve constantly got something to the likes of:

“ To maintain the quality of the service and adhere to the set policies, I can create up to two images per request. If you'd like another set of images with different cats or settings, please provide more details, and I'll be happy to assist! “

If I then ask the exact same prompt but slightly different it will then do it but it won’t continue to generate the same prompt…