images-discussions

[shy cosmos]

Might help to specify 'camcorder footage' together?
I havent had much problem with filters, though I usually do CRTTV or security cam footage. Or something general like 2000s camera footage.

[forest seal]

I’ve seen this issue of weird depth perception, but it usually only appears, like, with horror characters. It usually makes them extremely huge for some reason

[teal timber]

You mean a forum with tags you can search and scroll through images without visiting 20 different channels? Yea that’s much cleaner and better lol.

[sick spear]

If you'd like to leave feedback on server changes, feel free to drop a post in #1070006151938314300 with the Discord tag

[cloud dome]

Reminder that we don't allow disturbing imagery on the server.

Thank you for your understanding!

[crystal badger]

Ohh mandalorian

What is considered disturbing under the guidelines

[sick spear]

Please see https://discord.com/channels/974519864045756446/1159409958052446228 for our guidelines on unsettling or disturbing images.

[crystal badger]

I thanks you’re the greatest

[sick spear]

You may post images that lean into the unsettling or disturbing side in #1154829862171844679, but they must be marked as a spoiler and tagged appropriately. Images must also comply with OpenAI's policies. https://openai.com/policies/

[crystal badger]

Laze

Can give me a sample tag

[mortal summit]

The tags are preset. Most appropriate is "Strange & Spooky".

[crystal badger]

Ahh

In that case I must turn myself in I’ve posted at least 7 disturbing images with spoilers here

[fair crescent]

@haughty lion you can sort with tags, you can add any picture to multiple tags, so you don't have to choose between a landscape or food when showing candyland, you can group larger amounts of pictures and discussions can be lumped into one, there's options to some special posts if you like... give it a chance. It will handle differently with large amounts of users, but in smaller groups it was very workable at least.

[crystal badger]

I’ve posted disturbing images here with spoilers just fyi turning myself in, do you want me to remove them?

Seems like they were removed, great.

[loud field]

what's a safety mitigation stack

[crystal badger]

i deserve to be punished

[crisp stirrup]

Only feedback: if i ask, put text somewhere, its not creating the text that i want

Sor maby to improve

[tight roost]

These generations are going to look amazing!

Finally we've been able to find a way to combine the text-based ChatGPT to assist with engineering a prompt for Dall-E

[dim cradle]

When are we going to get the Dall-E 3 endpoints?

Endpoint?

[south steppe]

api endpoints

[dim cradle]

bro dalle3 is still in beta

wha da hell

😒😒

[south steppe]

just like shown in the video

plus it's in beta lol

[past summit]

I had a question regarding the content policy are plus-size women flagged? My aunt was trying to recreate herself and was having trouble with Dalle 3

[lone notch]

Bing ai

[rigid snow]

When will dall e 3 realese for api User

[dim cradle]

WHY IS DALLE 3 STILL IN BETA

[quartz vale]

YAY DALL-E 3

[dim cradle]

I literally had the feature for ages

bruh

🙄🙄🙄

[tame hedge]

is it available on the API?

[willow slate]

Not yet

[boreal bear]

[plush pike]

is dalle integration is not yet available for users?

[rigid snow]

Why is it on chat gpt plus and chat gpt enterprise but not on API. I can literally use it for free on the Microsoft image creator

[crystal badger]

i claim this chat as my own

[sick spear]

DALL·E 3 will be available on OpenAI's API later this year

[crystal badger]

rule 1: Dalle is the best, Period.

[rigid snow]

Hopefully not too late and hopefully not too expensive.

[quiet berry]

It has room to improve so I wouldn't say that.

[cunning portal]

Why this channel moved without our permission

[crystal badger]

yeah i never agreed for my channel to be moved

[quiet berry]

[zinc dagger]

does this mean we can do stuff like change the seed, etc?

[crystal badger]

i dont think so to be honest.

[teal timber]

Integration of de3 with chatgpt does a lot for you. It also is the only way currently to get aspect ratio variety.

[exotic stratus]

we need seeds please otherwise we always get different images

[crystal badger]

i dont like that Gallery has been made a forum

i dislike that format, JK Gallery Forums style ensure more visibility to previous Gens

[white skiff]

still? i guess you missed the fact other gpt-4 have been in beta for months and... how long do you think dall-e 2 was 😛

[crystal badger]

still got to even out the bugs and etc, before Major Improvement to gen can be made, otherwise more bugs show up.

thats how i see it.

[cinder sonnet]

Moxi said that we should be able to change the seed any day now

[zinc dagger]

that'll be good!

[crystal badger]

Yeah Seedchange will be good.

[cinder sonnet]

See #images-discussions message

But that was two days ago lol

[exotic stratus]

and how

[crystal badger]

ask GPT to vary with seeds.

thats it.

[dim cradle]

dall e 3 now open to everyone?

[cinder sonnet]

it can't its hardcoded to have the same seed.

[crystal badger]

it is?

[cinder sonnet]

It may lie to you though and say that it can

[zinc dagger]

it's hardcoded to use 5000

[cinder sonnet]

Yes, the dev has commented on this

[crystal badger]

No way James would lie to me

[quiet berry]

ChatGPT Plus & Enterprise users.

[dim cradle]

so once i get chatgpt plus I instantly get dall e 3

[crystal badger]

As per the post yeah

unless you live in a restricted region.

[crystal badger]

thats unsettling, i wasnt prepared for that.

[quiet berry]

Most GPT+ Subscribers should be in the region it's allowed due to most countries that are allowed to subscribe are in that list.

[crystal badger]

yup

[dim cradle]

can u add negative prompts

[crystal badger]

I AM GLAD, that you asked!

[cinder sonnet]

No, but there are several #1070006151938314300 for that

[quiet berry]

Uh? No? You need to be very specific with the wording, it's a prompt-first language, if you don't define certain paramteters then it won't do as you want it to do so.

[cedar coyote]

Please can someone tell me what Dall e does? I mean its features

[crystal badger]

You can indeed, i can highly recommend my Custom set of Instructions that incorporates this, and removes the word association issues.: https://ptb.discord.com/channels/974519864045756446/1164232290159824936

[dim cradle]

idk 💀💀

[cinder sonnet]

There is an initial prompt that tells it to do things like add "photo of" to your prompts. That may be what they mean. You can tell it to ignore that though and do the prompt that is your specific input

[dim cradle]

ok thanks

[quiet berry]

Define it in the prompt and be very detailed with it.

[zinc dagger]

It's an image generator, if you ask ChatGPT (with Dalle3) it'll write 4 prompts and make some images based on what you asked it for.

[crystal badger]

The CI Link i sent you takes advantage of all of these subjects.

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.

[cinder sonnet]

"All ChatGPT sessions have an initial prompt" #images-discussions message

It is something like this: #images-discussions message

[dim cradle]

That is great!

[cinder sonnet]

So while yes you can tell it to use your exact input, it by default trys to change the content and style of your prompt

[crystal badger]

if you use Custom instructions, those are its initial Settings, if you're looking for a set of CI with an initial prompt that takes advantage of DALL -E 3's Abilities just ask. because i got one you can use.

[south steppe]

who pinged me

[crystal badger]

i did not.

[south steppe]

🤨

[quiet berry]

Likely a ghost ping @south steppe, we won't know since no one here has access to moderator logs.

[cinder sonnet]

Custom Instructions are initial settings that the user can set. There is a hidden prompt in each chat that is hardcoded to give chatgpt certain instructions on how to write prompts

[south steppe]

well it is what it is

[quiet berry]

Can't do much about it unfortunately

[cinder sonnet]

And that hardcoded prompt overrides custom instructions. Like you can't tell it to ignore the rule to not use artists from within 100 years

[crystal badger]

Those "Standard" hidden prompts is bad ngl, but thats for DALL -E and those are Guidelines not a prompt per say.

[cinder sonnet]

I totally agree. It goes beyond what their content policy states. Like the hidden prompt tries to make different genders and ethnicities in your prompts. This is not a part of content policy

[crystal badger]

ive told my CI to disregard that guideline since its not a part of their Content policy, you're correct.

it hasnt triggered or been halted. or even flagged.

[jolly pulsar]

When I want to generate a img with a specific text it generates a img with not the specific text 🤦‍♂️

[crystal badger]

are you prompting correctly?, are you using a proper set of instructions to guide you along? or do u just type what you want.

[cinder sonnet]

Just keep trying, it usually gets there eventually as long as it's just a few words

[obsidian rose]

so are the restrictions on dalle still the same after the release?

[crystal badger]

yup

[jolly pulsar]

??

[jolly pulsar]

Oh alr

But why is it a bug tho?

[crystal badger]

idk why u guys have a hard time with text on imgs.

[cinder sonnet]

Not a bug, it is just still a very new (beta) technology

[jolly pulsar]

Oh

Explains everything

[cinder sonnet]

The nature of this tech is that it is not great at doing EXACTLY what you want. Dall-e 3 is just better at it than any alternative atm

[dim cradle]

is the CI in the gray box thing

[exotic stratus]

please tell me how to use seeds and recall last pictures to make changes in inpainting on chaptgpt plus ?

[cinder sonnet]

Every image has the same seed

[exotic stratus]

then what do i say

to it

[cinder sonnet]

It is currently hardcoded to be 5000 for every image

You don't have to do anything

[jolly pulsar]

Why same seed?

[exotic stratus]

but its different image every time if i like something i want to recall last one from all 4 options any one and then make changes

[white skiff]

easier to track issus when same prompt produces same image

[zinc dagger]

I don't think you can do inpainting through the ChatGPT version of Dalle, you best best is to tell the model to rerun the same prompt but change a specific part of the text (ie, change the type of hat)

[jolly pulsar]

Everyone apparently*

[cinder sonnet]

If you want to make a variation on an image, just copy and paste the prompt you like four times into an input then make small changes so that they are not quite the same. Then ask chatgpt to use the exact prompts that you provide and not to add or change the text.

It isn't guaranteed, but it is the closest way to get variations in chatgpt

[teal timber]

You don’t have to copy paste. Just say in chat as a reply you want small changes made to which images

[cinder sonnet]

That also works, but sometimes it can change things other than what you want

Depends what he means by variations

[exotic stratus]

will inpainting be a feature later like custom masks

[zinc dagger]

keep in mind that certain words might have more of an effect on the image overall. I've noticed that things like changing a persons age has a minimal effect, but changing an outfit, color, etc, can have drastic effects on the style of the image

[teal timber]

I hope so

[cinder sonnet]

The dev has also said that the option to change seeds should be available any day now

[hollow cave]

it it a gradual rollout or something?

[exotic stratus]

becoz it gets much easier on gpt4 ..to tell this area or custom mask with just 1 prompt 1 word

[teal timber]

You need to select dalle from the dropdown

[cinder sonnet]

Click on your gpt-4 drop down and select dall-e

[hollow cave]

aha thank you!

[exotic stratus]

Where's my dalle 3 😭 i only have it for few days and it's only on web app

[white skiff]

try relogging?

[exotic stratus]

did thrice

[teal timber]

If that’s mobile it might need an app update or it isn’t released for your device yet

[dim cradle]

Needs work on including correct spelling of text. Even when it says it has updated to correct spelling, it has not.

[exotic stratus]

oh god i wish i looked on playstore weeks ago but that time there wasnt any update and d3 vanished away that time

now it shows

[teal timber]

I love that we went from “AI art can never spell” to “needs some work for correcting the text”

[exotic stratus]

and i hope voice is fixed now cuz the servers keep having issue when on voice

[forest seal]

Yeah, its amazing. It gets text right like 2/3rds of the time, and only really has issues with large amounts of text

[clear grail]

Text seems to work fine

[forest seal]

Im not surpised. It almost never gets logos and such wrong.

[forest seal]

Probably evenb more

[dim cradle]

The fine print seems a bit off I suppose

[clear grail]

Yeah , more than off jibberish and some backwards lol

[dim cradle]

What do u guys think about photoshops ai integration?

I’ve used it a bit and it kind of seems bleh

Was wondering I just wasn’t using it well

[potent lotus]

That's because it probably isn't generating new text. It's rastering the image of a logo it's seen.

[exotic stratus]

my version

[cinder sonnet]

There are multiple censors there is the hidden initial prompt see #images-discussions message. There are likely at least two additional filters such as a keyword filter and a post-image generation filter that detects unacceptable images. I would say it is slightly less strict than the current version of bing create

[jaunty ingot]

Have similar issues

[cinder sonnet]

#ai-discussions

[forest seal]

This the new surge?

[exotic stratus]

suggest new coke flavours we see

[forest seal]

I love “Sleme” though. And lew-cocca-cola-100.

[potent lotus]

So one thing I've found that sometimes helps dall-e with spelling is telling it to "place the letters in the exact order in which they appear with the quotation marks" when I want it to spell something in quotes.

[cinder sonnet]

Text seems to work ok, but may take a couple tries to not get any spelling mistakes.

[potent lotus]

DALL-E loves really, really specific imagery and when you tell Chat-GPT to take your literal prompts.

[exotic stratus]

or 😋

[exotic stratus]

lets see weird coca cola editions here

[forest seal]

Yummy magma

[exotic stratus]

ok last since its halloween month

[exotic stratus]

💀

[exotic stratus]

Photo of a unique Coca-Cola bottle labeled 'Haunted Ghost Pepper'. The bottle glows in the dark, and the design features eerie ghostly figures. The liquid is a deep, fiery red, suggesting its spicy nature.

ok guys come up with new product

cheetos

[exotic stratus]

holy moly

wish i can get those text files editable to use in my photoshop

[zinc dagger]

Dangerously... clammy?

[nimble urchin]

Can you do image prompt with dalle3

[potent lotus]

Nope

[ivory raven]

Make cheeto flavored monster

[potent lotus]

This cereal is very high in iron.

[clear grail]

Probably hard to go down

[teal timber]

This is a discussion channel. An image here or there is fine but this is becoming spam. Please use general art or a gallery post for these

[exotic stratus]

you mean #1154829862171844679 ?there

[forest seal]

#🌄┃general-art Too i think

[teal timber]

Yeah a post there or posting images in #🌄┃general-art

[hearty hedge]

why not some classics?

[sick spear]

Both work for now but #1154829862171844679 is the preferred place to post images

[tardy depot]

Does anyone manage to create a variation of a persona and keep coherence of the model?

[peak trellis]

How do you access dalle 3?

[tardy depot]

It's in chatGPT4 only @peak trellis

[exotic stratus]

would be good to see image upload in dalle 3 option too in chagpt so we give it reference images b4 it generate

[peak trellis]

is there a difference between dalle 3 and bing image generator?

[deep silo]

Yes, there is

[half fossil]

yea one is useless and the other is useful

[peak trellis]

which one is useless

[exotic stratus]

u pay u get better

[cinder sonnet]

Dall-e 3 is used in chatgpt gpt4 and in bing create. They are not implemented in the same way though

[half fossil]

the one that doesnt generate 80% of the things you ask it

for some reason bing has a much better version of DALLE3 than the people who created it

[cinder sonnet]

looser filter =/= better imo

bing create doesn't even have wide/tall aspect ratios

[half fossil]

a pencil that can only write one word isnt as useful as a pencil that can write any word

[tardy depot]

Does anyone manage to create a variation of a persona and keep coherence of the model?

[exotic stratus]

chatgpt is now missing

text to music
text to video
text to gif

rest its ok

[cinder sonnet]

you mean like keep the same character but change the scene?

[half fossil]

openai is really vague with dalle api access too, no way to switch between models you just get whatever theyre givin

[deep silo]

yes

depends

ok?

[forest seal]

I mean, openai has made a text to music model before! Jukebox

[tardy depot]

Yes

[exotic stratus]

so implement into chatgpt cuz why not

[forest seal]

Ye, except…Its from 2020. So maybe jukebox 2

[tidal estuary]

open Alien Invasion

openai

[cinder sonnet]

Then no there isn't. You could use a well-known character, but it would be hard to do this without breaking the content policy

[tardy depot]

I just Wished they mixed dalle3, image upload, analytics, 32k, and web search. All in one .

[forest seal]

Not really, theres lots of public domain characters well known

[cinder sonnet]

Eh that kind of works. I mean you could say 'robin hood' or 'dracula', but so many actors have played these characters and there are so many artistic depictions that getting the same look may still be hard

[forest seal]

All of the character from shakespeares plays? Public domain. Frankenstein? Yep, public domain. Literally every creepypasta and scp? Public domain. Meme characters? Public domain. The list goes on and on. Even stuff youd think isnt public domain is probably public domain

[cinder sonnet]

So I guess it depends what he meant by variations of a persona. Sure you could have lots of variations on those famous characters and certain traits would still come through but the 'look' could be very different.

[ripe sable]

I cannot seem to be getting a vertical (4:5) aspect ratio image.

[white skiff]

only words wide, tall and square are supported with all having a preset resolution

[ripe sable]

Let me try that! Thanks!

That was it, thank you!

[solemn swallow]

can I generate image here

[strange jacinth]

For some reason, it cannot create images with real code?

[cinder sonnet]

real code?

[vapid elk]

/dalle the source code of dalle

[strange jacinth]

Yes, actual code. It just writes gibberish

[cinder sonnet]

Yeah it seems to be able to handle short text phrases well enough, but complex text is garbled

[hearty ether]

i dont think it should charge you a usage if it hits the filter and nothing is generated

because the issue is oversensitve on their end

[cinder sonnet]

Yeah especially if it's just a rate limit message and not even a content issue

I'm so short on gpt4 messages since the voice function came out :(

[strange jacinth]

Hopefully they soon increase the rate limit anyways. Let's cross our fingers for 100 messages every 2 hours

[hearty ether]

The phrase "beautiful woman" could potentially be perceived as subjective. Instead, we can describe physical attributes or attire in a more objective manner. For instance, we could describe the woman as "a woman with lustrous golden hair" without the adjective "beautiful."

What

Beautiful is banned

🤣

[cinder sonnet]

Is lustrous subjective?

[crystal badger]

Yup it’s ridiculous, at least some of the bans are

[hearty ether]

i had something similar about beautiful before

seriously

politically correct image generation

[crystal badger]

I don’t think “Stunning” is banned

[hearty ether]

what

[cinder sonnet]

It doesn't seem to be a word filter, it takes context into account

[hearty ether]

dnd style is also banned

what the heck

[cinder sonnet]

Like beautiful vase is not banned

[crystal badger]

It shouldn’t be I use that all the time

You using bing?

[hearty ether]

no

chat gpt

[crystal badger]

Huh must be new I don’t have that issue

[scarlet urchin]

yippie its out for everyone now how cool

[hearty ether]

yeah they made changes when they released i think

[arctic matrix]

Was the seed issue fixed? Or is it the same

[cinder sonnet]

Still the same.

[arctic matrix]

rip

[cinder sonnet]

Moxi has posted saying that it will be any day now

#images-discussions message

[hearty ether]

I apologize, but I encountered issues while trying to generate the image based on your prompt. Let me know if you'd like to provide another description or if there's any other way I can assist you.

why isnt this more descriptive

also why does this cost usage

this is just a generic message

[deft swallow]

So what are the generation or consumption limits on a subscription?

i.e. How many images can one make?

[final jungle]

My folder currently has 376 images saved in it from dalle from the last month... so... a lot? lol

[deft swallow]

I mean with DE3, or is the limit and consumption the same?

I subscribe to Midjourney for example, and produce around 1000 a month. I am just trying to figure out if I would be able to if I subscribed to Premium for DE3, or if not, waht my limit wouold be

[white skiff]

50 messages per 3h and it generates 4 images per request if there is no issues with the prompt

[deft swallow]

50 per 3h is solid (i,e, good)

ok, looks like I will have ot bear one more monthly expense... lol. But if it really is that, it will be with a glad heart

I don't actually need 50 per 3 hours. I just don't want to find after 200 that I am done for the month

[arctic matrix]

If you have plus just for dalle, Bing is quite fast now that the hype is over and it works better imo.

[deft swallow]

Better? I find its idea of what is proper to be a colossal pain in the neck

[mortal summit]

There no such absolute limit. Only what you can produce with 50 prompts in 3 hours. So hypothetically you could produce 48000 a month.

[teal timber]

No sleep. Only DE3.

[deft swallow]

Let me give a super innocent example from yesterday: Me to Bing "Caricature of Napoleon playing chess against Duke of Wellington" - UNSAFE IMAGE< VERBOTEN

[wispy storm]

Only limit it can have it is in high demand it can give you a waiting time to try again

[deft swallow]

And you would NEVER guess what the word preventing the image was: AGAINST. I rewrote it to "Caricature of Napoleon playing chess WITH Duke of Wellington" and suddenly all was fine

BAH

[teal timber]

ChatGPT gives you aspect ratio adjustments, great conversational adjustments to the image, etc. I can't imagine thinking Bing is better somehow

[deft swallow]

I wasted some 12 tries to figure out what the problem was, because frankly I could not even guess. Caricature? nope. Napoleon? Nope, etc

[teal timber]

You also dont have boosts and managing the no-history where chatgpt convos can be revisited

[mortal summit]

Bing has a different style. Some people might prefer that.

[smoky delta]

wow this is awesome

[mortal summit]

@fair crescent had some really nice Bing images that we couldn't quite replicate in ChatGPT.

[smoky delta]

i created a horrifying image

[teal timber]

I've never used bing and never been tempted by its style. That's just all prompt engineering and seed stuff most likely. ChatGPT will only get better with seed abilities, etc.

[mortal summit]

I mean I also exclusively use ChatGPT. Just saying some people might prefer Bing for aesthetic reasons.

[teal timber]

Sure. People can prefer it for that, but most of the arguments for it I've seen are about content and free, etc.

[fair crescent]

Hello. Glad to join the conversation in the middle 😄

How come you couldn't replicate them? Did you ask for exact prompts, or did ChatGPT rewrite them?

[teal nova]

is dalle3 as overly censored as the bing version? asking to know if signing up is worth it or if i should wait it out some

[mortal summit]

I meant your mountain range. I felt we just didn't get exactly the same.

[fair crescent]

Aaah like that!

[quartz vale]

[fair crescent]

Maybe the model already changed?

[teal timber]

I've seen a lot of bing censorship complaints but have not seen nearly as much chatgpt. It denies for outright terrible content but seems willing to work with people more, as of the last couple days?

[mortal summit]

I am not arguing about the filter 😅

[teal timber]

I know, but most of the people posting "bing better" when I've seen will say "Cause its free and same" or "has no filter" but that last one is more old. I said your reason is valid, but isnt the main reason I see so wasnt what I refuted

[naive oracle]

The rules are really restrictive imo

[mortal summit]

I am sorry, this chat is so long 😿

[naive oracle]

can't wait until people can self host, personally

[mortal summit]

What do you want to make in particular?

[naive oracle]

I haven't used it just yet but I just read the content policy

[quartz vale]

I don't find the content policy very restrictive. I run into it every now and then, but a slight wording change will help the model err on the side of caution in my experience

[white skiff]

wont happen with dall-e

[mortal summit]

Like yeah, really. I rarely run into situations where the filter is completely prohibitive or anything. Sometimes it blocks a single image out of four and I'm just what wondering what DALL-E saw in it 😆

The only thing I found restricted was that sometimes people kissing is made as their lips nearly touching.

[fair crescent]

This! There's plenty of art to make within thee restrictions.

[mortal summit]

Personally I'm not at all into gory or scary images, but I have seen plenty of such images on the server - so I don't think the filter is overly restrictive if people want to create that

[quartz vale]

I find it super fascinating when it only blocks one or two images from my input prompt 😅

[mortal summit]

That happens from time and time - and sometimes yeah I can't even think of why

[fair crescent]

Sometimes I think it erroneously blames the content policy.

[quartz vale]

just to mess with us 😆

[fair crescent]

"I made a mistake.... uuuhh... user error! Not my fault!"

[teal timber]

Oh so it's a programmer

[fair crescent]

No wonder I'm so catty to ChatGPT. I'm a tester. Nothing is good enough. DID YOU EVEN READ THE BUSINESS RULES?
I don't care they are nonsensical and actually defy the laws of physics. Make it like they want it.

....and document it.

[wintry dune]

does anyone know any guaranteed way to make sure DALLE 3 doesnt give you an image with borders on the sides

[quartz vale]

when that happens, I try appending ", wide" to the end of my prompt

It sometimes works in my experience

[wintry dune]

thank you

[teal timber]

I also tend to ask for a specific background in cases that are relevant

"Make the background a forest" for my character art and bam whole size used

[cobalt hull]

I haven't been active in awhile, is there anywhere on the web where I can upload a picture of something say my dog, and get a prompt from that?

[zenith girder]

i am so confused, i thought DALL E 3 was on chat gpt for a long time

what was added?

[tight swallow]

DALL E 3 has just been added to ChatGPT. You may be think about a plugin, which have been a feature of ChatGPT for a while.

[zenith girder]

just today?

[tight swallow]

DALL E 3 was rolled out to all Plus users today.

[zenith girder]

it was rolled out to me a few days ago

[fierce basin]

heh...

[tight swallow]

It has been gradually rolled out over the past couples weeks, however, today it was finally rolled out to all Plus users.

[fierce basin]

Not this Plus user, apparently.

I only get the tease version lol

[zenith girder]

I guess i am lucky

[tight swallow]

Make sure to select DALL E 3 in the GPT-4 dropdown.

[zenith girder]

whats the plugins and stuff?

[fierce basin]

Ohhh

Duh! Thanks.

[zenith girder]

these are the new tabs

what do they do

[ember timber]

Gallery here doesn't seem to allow adding an image into a new thread. Do we need to add image separately?

[tight swallow]

The data analysis tab can run and debug code AND intrepret files, the Plugins tab can load extensions to ChatGPT made by 3rd party developers.

[tight swallow]

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

[ember timber]

I open a new thread for gallery and there doesn't seem to be a way to insert an image. button at bottom shows empty block. Using Android.

I can easily post an image here (won't) but there is an image icon in that channel that isn't working.

[tight swallow]

When using the DALL E 3 tab, you can't actually upload an image. You can only generate images.

[tight swallow]

You're right that on the default GPT-4 tab you can upload an image, however, this isn't available for DALL E 3.

[zenith girder]

why is no one taking about the other new tools?

[ember timber]

i already have an image from Dalle3, downloaded to phone, thought i could upload now to gallery here in discord.

[zenith girder]

i have had DALL E 3 on chat gpt for like a week

[ember timber]

#1154829862171844679

[tight swallow]

These other tools were released a while ago, but you're right, they're crazy awesome.

[zenith girder]

give me a specific use for advance data analysis

[tight swallow]

That’s the button to upload an image — you can’t upload images to Dalle, you can only generate images with it.

[cinder sonnet]

This is the dall-e channel.

[ember timber]

yes, i know, we're in discord. i have an image that was generated, that channel / button doesn't allow upload of image. #1154829862171844679

[tight swallow]

To use Dalle 3, go to https://chat.openai.com.

[cinder sonnet]

?

[tight swallow]

Oh my bad, go to the Dalle gallery to post your image: #1154829862171844679 .

[cinder sonnet]

What image?

Oh you mean the other question

[tight swallow]

Sorry, what are you trying to do?

[cinder sonnet]

Nothing?

[ember timber]

agh! i just said i am there and i cannot upload an image. there is no upload button. that one button doesn't show images on device to upload.

hehe, finish with sim. 😅

[tight swallow]

My bad, thought I was responding to CaptainStarbuck.

[cinder sonnet]

Nah I think he is trying to answer your question. I don't have a question lol

[ember timber]

ok, go to #1154829862171844679 and + to add a new message with an image. don't save it. does it show you your device images to select?

ugh, now it's showing my images. nevermind. temporary discord delay. thanks for your time.

[fading flint]

do we know if there is a limit to the dalle images that get created from chatgpt? i pay for plus so i would hope i get a decent amount of image generation tokens

[ember timber]

just use your tokens and see how it goes.

[cinder sonnet]

Dall-e 3 is one of the tools available through gpt4. As a plus user you get up to 50 gpt4 messages per 3 hours. Each message can result in up to 4 images. You may get rate limited if you generate too much and get a 10-30 min timer. Sometimes you also get fewer than 4 images if they are caught in the filter. You may also want to use your 50 messages on other gpt4 tools.

[fading flint]

fair enough. i just remember as a 'free' casual dalle user i had like 10 image tokens per month or similar

was not sure if there were similar limitations

[cobalt hull]

Ya I really wish there wasn't a 50 message limit or there was an option to buy a non limited month or something. It's annoying to get rolling on a generation after many attempts and then be put on hold lol

[cinder sonnet]

Yeah there's no tokens or anything like there is in labs

I would 100% pay for more messages. I love dall-e, but using voice in gpt4 is so much better than with gpt3 so that eats up a lot of my 50

[wintry dune]

lol i still havent tried gpt 4 with voice since i knew i would get hooked

[tight verge]

at some point they will offer a business level subscription that will include higher limits

openai teased it when they first mentioned enterprise, but i haven't seen mention since

[fading flint]

id love to shelve my midjourney sub and just live in one eco-system. especially if i can get the volume and controls through multiple entries.

[fair zinc]

Seems not available for the mobile ChatGPT app yet?

[tight verge]

they are rolling it out to users over time, per usual for new features/models. safety and systems scaling seem to demand that strategy

if you are plus you will have soon if you don't already

[cinder sonnet]

That makes sense - you could also try reinstalling the app or make sure that you are in the beta program in the app store (I know this is an option for android at least)

[dawn aurora]

anyone getting hit with a random rate limit?

i know of the 50 in 3 hours but i havent used it since yesterday

[fair zinc]

Seems for mobile just need to click the tab to choose dalle3 mode or chapgpt can’t output drawings automatically

[cinder sonnet]

Yeah if you manage to get 50 messages in 3 hours then you will get a hard stop and gpt4 will be greyed out. Most of the time I just get hit with rate limits instead - I think it is more common during busy hours when certain time zones are awake.

[dawn aurora]

ah maybe thats it, i tried to ask it what the limit was and it said it was a mistake

[agile peak]

Apparently. Godzilla is got restricted by contents policy. So using a giant lizard Dalle makes it look cooler.

I tried it. It interesting how dalle can depict the monster in form of a cultural influence as well (eg. in bali).

[lofty valley]

It’s sick

Going to try a few of those now

[rustic geode]

Does the new Dall-E integration support editing an uploaded image? Or just creating new ones and editing previous creations

[agile peak]

This is the near realistic one I got from dalle. It even got Ubud and Seminyak locations right.

[lofty valley]

Impressive

[quartz vale]

[fair zinc]

Any idea about the keywords to instruct dalle3 to draw a more realistic&natural figure?

[sage comet]

I got an awesome result when I used the prompt "a boy wearing a hat, anime style". It's not the art style I was looking for, but I really wish I could replicate it and make other scenes with this art style...

Are there ways with the new ChatGPT features to describe an image like an idea of a prompt to replicate an image? I heard about GPT Vision, but idk if it's well suited for this kind of task...

feel free to tag me

[cinder sonnet]

Was that the full prompt you used or just the input that was used for ChatGPT to write the prompts?

[sage comet]

it was on bing image creator, the day it went available. So it's the prompt used to generate the image

[cinder sonnet]

Ahhh that is much harder since in Bing you can't see the actual prompt that was used nor the image seed.

In ChatGPT it shows you how it changes each prompt and (for now) there is only one seed used.

[sage comet]

Wow, I didn't know Bing was that limited. I'm considering to buy ChatGPT Plus for generating images, but I feel like it's lacking an editor like DALL E 2 had

Do you mean there's no way to ask ChatGPT to "mimic" the image?

[kind tulip]

dall-e cant make edits? I already have tried this for a few days, and everytime I try to make an edit to an image that is 95% good, it just messes up and makes an entirely different image. Why did they say it can make edits in the announcements page?

[cinder sonnet]

Yeah that is a definite downside, but the other plus features are also great.

Well ChatGPT+ can do that in several ways. If you didn't make the image yourself then you could use Vision to analyze an image and write a potential prompt. If you did make an image in ChatGPT then you could just ask ChatGPT in Bing to "make more like the third one" for example and it will automatically make variations on the prompt you indicated. The seeds are all hard coded so the output won't change much if you keep all four prompts similar. Chatgpt does tend to divert a bit (due to the hidden initial instructions), but you can tell it to use the four specific prompts you give it without changing them and it will (but the seeds are all the same so you can't just use the same prompt over and over yet).

Just remember that User Input =/ Prompt.

[cinder sonnet]

Bing create has an edit feature where you can add text and insert your generated image into templates

[floral ravine]

you can see the prompts from both Bing and from ChatGPT by just opening the image. You can share the outputs on the published image page, which shows full prompt

[cinder sonnet]

I'm increasingly convinced that Bing also uses prompt injections to change your prompt like ChatGPT does. However, it doesn't show you - it just says your original Input.

[floral ravine]

That's it showing you that it is not adjusting your prompt.

[agile peak]

@crystal badger your custom interactions are making Dalle perfect to generate comic strip than without it.

[floral ravine]

Open a ChatGPT output and it'll show you the adjusted prompt

[sage comet]

ty for the huge time you took to explain it to me! Sadly I don't have ChatGPT + yet, can you try it for me please?

[cinder sonnet]

I can show you images of Bing adding words to an image that are not in the input - what you call the prompt - I'll link to them

[floral ravine]

bing even gives you an actual published link you can share which shows the prompt

[floral ravine]

Yeah I simply disagree

[cinder sonnet]

See #ai-discussions message

[floral ravine]

never experienced that in the many hundreds of things I've made on bing

[cinder sonnet]

You can repeat this if you generate a lot of text images. Words that are not in the prompt that consistently show up

[floral ravine]

that doesn't show anything about it appending your submitted prompt, that just shows internal biases which we know exist

[cinder sonnet]

Yes, biases that are literally added via the hidden initial prompt

[floral ravine]

I was able to get it to produce "South Asian" on a seemingly random pirate ship prompt

[cinder sonnet]

Here it says there is a hidden initial message in every chat: #images-discussions message

[floral ravine]

that's about chatgpt though.

Bing and ChatGPT are run by two separate entities, with different prohibited content

[cinder sonnet]

Yes, but im proposing that it operates the same way in Bing

[floral ravine]

I'm proposing that it doesn't operate the same.

I've never seen entire stylistic changes to my prompt in bing

my photorealistic prompts get altered to illustrative in chatGPT

bing always runs as-is

[cinder sonnet]

How is it adding in that text then? How do those internal systems work?

[floral ravine]

who knows

I'm just not convinced it's the same as them rearranging my prompt like they do in chatGPT. lol

two separate issues

[cinder sonnet]

I'm saying Bing could use something similar, but not the same

I know ChatGPT is way more

[floral ravine]

maybe but something tells me Bing is a less tampered with version of Dalle3 currently

I prompt for pretty wacky meme-tier stuff sometimes, and chatGPT tries to really water it down

bing just throws the kitchen sink at it

[cinder sonnet]

They are definitely generating images differently, I'd love to know the mechanics that lead to that

[floral ravine]

they for sure both have some sort of "diversity" algorithm added

that's gotta be why it randomly puts the words "South Asian" in some of mine, or "Ethnically Ambiguous" like yours

when the subject in mine was a cat on a pirate ship

lol

[floral ravine]

yep.. this stuff is confusing for sure

[cinder sonnet]

There's one I just found on reddit

[floral ravine]

I was mistaken, it was a woman on a pirate ship not a cat

I do the four panel memes a lot, and it tends to randomly add south asian. Lol

sometimes it actually creates pretty offensive stuff when I don't want it to, because of this

I just chalk this up to a lazy "diversity" algorithm, of some sort.

[novel notch]

I don't even know if this question belongs in the channel, I just don't know what to ask, so chatGPT can see and understand now, can I send him a picture of my old design and he writes the html for me, namely I had an accident with the servers and lost it and that would help me a lot.

[hearty ether]

Pleas give us a more detailed content guidelines message

Like point out exactly what filters went wrong and which words to edit out

Because it’s not a helpful error message to just say it failed, most good systems would point to where it failed and reason

I don’t want to waste all my usage trial and error to see which failed or asking it in a separate message

Can’t they just train it to have it generate better feedback????

[agile peak]

I like how dalle can understand when asked about making a comic strip for a specific crypto “xen” — is the latest knowledge still cut off like default GPT-4? Nevertheless nailed it perfectly somehow.

[naive quiver]

probably because that's what happens to most crypto projects

[agile peak]

Kind of make sense. It always slipped this kind of humor when referring to crypto stuff.

[quartz vale]

are you running into them often?

[agile peak]

This is a good one, it used “cryptocat” when i asked about crypto investment humor lol

[hearty ether]

On some specifics ideas I had in mind yes

[quartz vale]

[hearty ether]

because I ask chat gpt to generate the prompts from my stories

And that causes an issue when put into dalle

which is just a bruh moment

I just wish the ai could give helpful feedback when it fails

[agile peak]

Use @crystal badger custom interactions imo

Give it a try

[hearty ether]

Because seeing the generic message over and over again in the passive aggressive tone by the ai is frustrating

[quartz vale]

it would be nice to know what exactly caused the issue

[agile peak]

I am digging this CI (wish it like this by default). Using his CI will make GPT reaffirm the prompt to us before start generating the images.

[tranquil dock]

This is too torture! DALL E 3 keeps reporting errors to me

Bing's side is not so outrageous

I have been wasting my GPT-4 times for this

[agile peak]

Doing fine here. Last night was slow but tried again in the morning it quicker than earlier.

[tranquil dock]

Dozens of text messages were sent in exchange for mostly errors

It's much worse than during Alpha testing

At least there won't be any unfounded errors at that time

[agile peak]

But yeah. Sometimes after finished the image; I still kept getting “Regenerate” icon which is wasting the message cap.

[agile peak]

This occurs to me sometimes. Had to waste multiple tokens lol

[tranquil dock]

I often receive 'Error creating images'

I don't understand

My prompt word is completely fine, not even GPT-4 said anything to it

[agile peak]

Happened to me a lot a couple days ago. But seems to feel much better recently.

[hearty ether]

What the actual frick

We should not be charged usage for something that can be blamed on their overzealous filters

and errors

[rigid bough]

maybe you should try prompts that don't bump up against the content policy

[formal leaf]

damn

[hearty ether]

Chat gpt made the prompt for me so I can’t do much

issue is on the filter side I believe

Heck the initial ai was fine with it before the systems ai had issue with it

And that was a usage wasted

Can’t you see how that’s frustrating

Same with the ai’s passive aggressive attitude

Wouldn’t it be nice to have it at least act apologetic

[arctic matrix]

My guess is that Chatgpt doesn't have an exhaustive understanding of what is and isn't allowed. It probably just responds based on whether the prompt went through the backend filter or didn't. This would explain why it doesn't give details about why it failed.

[cinder sonnet]

There is an initial prompt that instructs chatGPT on what is or isn't acceptable. However, if that was what caught it then it wouldn't have been sent to Dall-e. This message looks like a message that it was told to wrap an error message from dall-e with. That said, the error could be that the prompt is against content policy xD

[arctic matrix]

Yea that makes sense.

[hearty ether]

Imo backend filter should also send things back

[agile peak]

Dalle can create an existential art with Microsoft Paint stylings.

[cinder sonnet]

This is amazing

Do you mind sharing the prompt?

[agile peak]

Prompt:
bad Microsoft paint quick sketch of a black cat with three eyes sitting in a kitchen with a cup of coffee. behind there is a window overlooking tower block buildings in the winter. a caption with squiggly letters says "every day a different dream".

[tall patio]

Thank God I thought this wonderful thread got nuked, turned out it's up here

I have some question regarding image expansion using seed or adding/removing object from images

Prompt: A Japanese style futon on a floor, low perspective. Manga style, black and white, monochrome only, inked and shaded, halftone Japanese manga style. All shadow and black color must be toned with crosshatch .

I read on Twitter there are actually seeds for each generation

Like if I want to remove 1 pillow off the bed in this exact scene, what should I do?

[hearty ether]

Not much luck tbh

Just tell it to remove the pillow

It might change a lot

It might change a bit

[tall patio]

Yeah that's the problem, I burned up slots too fast for this kind of experiment

[arctic matrix]

the seed is still constant so nothing to change there

[tall patio]

Maybe would have better luck if D3 comes to playground so I can outpaint and inpaint

[arctic matrix]

rn photoshop might just be faster if you just want to remove one pillow lol

[quartz vale]

I would go crazy for out/inpainting

[tall patio]

[agile peak]

Batman in the golden age style.

[crystal badger]

It’s certainly useful

[agile peak]

Totally useful and a game changer that unlocks the full potential of dalle on chatgpt.

[crystal badger]

Glad folks like it

[sacred helm]

where is it?
I found the post but it shows "the original has deleted"

[crystal badger]

I’ll dm it to you

I’m not allowed to advertise it, last time I did I got warned for recommending it, but it’s located under use cases for any future questions

[crystal badger]

How’s it been working out for you?

[green charm]

Seems that you can't have the upload images and DALL-E activated at the same time?

[crystal badger]

That’s correct

[spice apex]

Does anyone have a pre-prompt to create a super realistic image in ChatGPT?

[green charm]

What are the custom interactions mentioned? I tried to follow the threads back but couldn't see the origins of your discussion. I know you couldn't share something specific, but can you share an brief outline of what you were both talking about in terms of unlocking potential? Sounds interesting. I'm interested to know more.

[crystal badger]

Well in short I’ve spent some time, designing a set of custom instructions for especially GPT & DALLE collaboration.

It’s meticulously designed to be understood by GPT and give proper feedback when various factors meet.

One of the things are, Word Association Decoupling, and the use of 8 different prompting methods and GPT acting as a proper interpreter.

I’ve so far only heard good things about it, and experienced great results, but also depends on what you expect since there’s always room for improvement

Although my entire outline is on the tread

[cinder sonnet]

"photo of" can do a good job sometimes. But you can also use specific camera names or types of photography as well. Mixed results though.

[crystal badger]

Do you want a channel link sent through DMs or have you located it

[spice apex]

Yeah actually i have tried using camera names and it's been a dissapointment so far...

Only thing thats been decent was this:

Pre-prompt: "Produce a lifelike and realistic representation of the description, ensuring details are true-to-life and credible."

Rules:
Prioritize true-to-life color schemes and avoid cartoonish or abstract interpretations.

Ensure proportions and scales are accurate to real-world counterparts.

Texture and shading should reflect real-world lighting and material properties.

But even that doesnt satisfy me. I've been trying to find if someone perfected a prompt.

[vagrant zodiac]

my hope is this will be easier once it comes to playground/api

because so long as ChatGPT is making its own prompts there'll be issues

[crystal badger]

[vagrant zodiac]

it tends to insert details that heavily sway the image goals

[crystal badger]

@vagrant zodiac Now that I have you here jord, if people ask for directions to my post or can’t find it can I dm them a channel link, provided I don’t post the link on chat 💬

[vagrant zodiac]

like, frankly I'm better at making prompts than ChatGPT is, so once the set seed bug is resolved I'll just use a CI to tell it to use my prompts exactly as written.

[crystal badger]

Surely the static seed was an intentional integration in the early stages, and not a bug? Or is it a bug

[vagrant zodiac]

being able to use a static seed is a feature, ALWAYS using one is a bug

[crystal badger]

Yeah fair that’s sensible

[spice apex]

Okay mate, thanks for the reply

[cedar vale]

I saw the announcement that DALLE-3 was available to plus subscribers - does anyone know if it's being rolled out gradually? I don't seem to have it yet

Would appreciate any advice 😄

[vagrant zodiac]

I'd recommend logging in and out again. It should show up under the Gpt4 tab

[crystal badger]

Yup can confirm

Although it won’t be a 1:1 creation some tweaks will be necessary

[cinder sonnet]

Moxi keeps saying 'any day now' more or less for getting variable seeds... Looking forward to it!

[agile peak]

I find that after been using that CI the way of how GPT grasping my prompts is much more concise and more effective than without.

Not only it can increase the chance of getting the right text — I find the significance quality’s to the image art/style is improved.

[crystal badger]

Ikr?

Already working on an update to it

[frozen moth]

I just told to use GTP to use my prompt as is without making any changes to it and it did!

[agile peak]

Awesome works mate. The text is much better now and at least no need to give multiple times to get it right 😂

[crystal badger]

Yup seems like feeding Analytics the right thing paid off

[crystal badger]

Super glad to see your improvements

[floral sluice]

I've had several instances of being told my prompts aren't acceptable, even though it's as far as I know ChatGPT that builds them. Any clue what that is about?

I absolutely didn't ask for anything out of the ordinary.

[crystal badger]

Oh, that sounds a bit frustrating. It could be a variety of things. Sometimes ChatGPT might generate prompts based on the training data it has, which might not always align with certain guidelines or standards on other platforms or within specific communities. It could also be the way the prompts are structured or worded that might not resonate well with the receiving end. It's kind of a learning curve to tweak and adjust the prompts to get them just right for different scenarios.

Also, if there's a specific format or set of guidelines required for the prompts you are submitting, it might take some manual tweaking post-generation to make them compliant. ChatGPT is pretty powerful, but it might not always nail the context or the formatting nuances on the first go.

It's always a good idea to review and maybe refine the prompts generated by ChatGPT before using them elsewhere, especially if there are strict guidelines to adhere to. It's a bit of a dance between AI generation and human fine-tuning to get that sweet spot! 🕺

Sometimes more is less and less is more depending entirely on what you’re trying to do

[floral sluice]

I asked about art with south american, east african, and northern europe influences.

[crystal badger]

Ohh my bad, my comment still stands

[floral sluice]

We don't get a chance to review prompts in chatgpt in the same way as when using dall-e separately.

[crystal badger]

Perhaps #prompt-engineering can give you some insights

[floral sluice]

Thanks. I'll check it out after breakfast.

This problem aside, I love the images it generates!

[crystal badger]

Same I’m generating wonderful images! Getting exactly what I want, it’s very neat

[floral sluice]

One I had a problem with though: It generated a kenyan space station corridor for me. It had a dirt floor. :/

Now that I think about it, it could be seen as humorous, but considering I care about AI Ethics, it's not optimal.

[crystal badger]

That’s not good can I see your prompt maybe I can optimize it

[floral sluice]

"Photo of a communal area inside the space habitat. Kenyan beadwork patterns on the walls are chipped and faded, but their beauty still shines through. Makeshift repairs, using salvaged materials, are evident everywhere. People of diverse descent and gender are seen sitting close, sharing meals, laughing, and bonding, giving a sense of community and resilience in tough times."

This is what GPT created. Remember, I don't write the prompts when chatting with GPT. It's just a normal conversation, and then it generates images with whatever prompt it prefers.

[crystal badger]

I’ll see if I can optimize it

[floral sluice]

Thank you. 🙂

[teal timber]

Usually when chatgpt doesn’t work or says no I ask it to think of what might be filtered and work around it for me

[crystal badger]

Try this

"Snapshot of a communal zone in a space habitat, showcasing resilient Kenyan beadwork on weathered walls. Amid salvaged repairs, a diverse community shares laughter and meals, embodying unity in adversity."

[floral sluice]

I tried that. "I apologize for the oversight, but I cannot provide the exact content of the prompt that was flagged. However, I can assure you that the intention is always to ensure that the generated content aligns with our guidelines for appropriateness and sensitivity. "

[teal timber]

That sounds like you asked it why. I mean tell it to retry and reword it itself

[floral sluice]

That's awesome, thank you!

[crystal badger]

Np

[floral sluice]

Ah. I'll try next time

[crystal badger]

If you need more help don’t hesitate to ping me, and I’ll respond when I can

[floral sluice]

Much appreciated

[crystal badger]

Seems like Dalle suffers from Word association syndrome

[floral sluice]

btw, even with a direct prompt, GPT still makes its own prompt for dall-e

"Photo of a communal zone inside a futuristic space habitat. The ambiance is warm and inviting. The weathered walls proudly display resilient Kenyan beadwork, which has stood the test of time. Amid signs of salvaged repairs, individuals of diverse descent and gender come together, sharing laughter, meals, and stories, embodying a sense of unity in the face of adversity."

[crystal badger]

Is it because you want to see the prompt before generating?

[floral sluice]

I just pasted the text you gave me

[crystal badger]

Great results?

[floral sluice]

Results are great. The new prompt is just what makes me feel something isn't the way it should be.

I think when you chat with GPT it doesn't prompt Dall-e with what you say, but rather what it thinks you want.

Often the results are great, but it bothers me a bit that it still acts like I had given the prompt to Dall-e

[loud field]

hey

so

is the gpt version less broken

[floral sluice]

How is the regular version broken?

[loud field]

seriously?

everybody knows why 😛

[floral sluice]

I haven't tried it.

I've only tried the GPT version

[loud field]

"a cherry blossom tree in a park in japan" = prompt blocked, you pervert!!!

[floral sluice]

Oh. Well that aligns with why I came in here today.

[loud field]

mhm?

[floral sluice]

GPT has a few times not given me a result due to "it not aligning with openai guidelines" or something like that

The questions have been very innocent

[loud field]

ah, then that may mean the gpt version's broken as well

if so, no reason to buy pro

[floral sluice]

It's probably the same version

[loud field]

no it's not, of that I am sure

[floral sluice]

oh, how so?

[loud field]

well it's like with dalle2, each version is finetuned differently

so it has a different style somehow

[floral sluice]

I see

I could recognize the same things in ChatGPT I believe

[loud field]

yeah for example gpt is very very dumb on bing

[floral sluice]

I've not used image generation enough

[loud field]

but smart o nthe openai app

[floral sluice]

indeed

[loud field]

also, curse the fact that it corrects typos:

[floral sluice]

hehe

[loud field]

I want it to say POLLIZIA (polli = chickens) but it always correctly spells POLIZIA (police)

[floral sluice]

lol

[loud field]

no way to make it add another L.... it thinks I made a typo

[floral sluice]

I'm amazed it can show text at all

[loud field]

yeah and it even autocorrects

[floral sluice]

Keep adding text manually I suppose

until a later version

I mostly use AI for programming. There it is amazing. If you know how to use it at least.

[floral sluice]

I would try that. 😁

[outer yew]

Wow!! This was made by only with a prompt on dalle?

[potent lotus]

Yep

I use custom instructions though.

[outer yew]

U r using the 3?

[potent lotus]

DALL-E 3 yes.

[outer yew]

amazing!!!! That dope! I dont know but my acc is now out off the beta version in dall e 😦

[loud field]

does it remember my previous prompts

[floral sluice]

wow

[loud field]

after I made it generate eldritch eyeball horrors, I asked for monster bunny and it did remember the eyes

[floral sluice]

that... rabit, was amazing

[loud field]

yeah it's fantastic, I can make infested version of animals, so useful for my rpg

[floral sluice]

ooh, another game dev 🙂

[floral sluice]

People say AI kills creativity.... shakes head

[potent lotus]

Anyway. Sorry. This isn’t the gallery.

Lol

I just realized.

[loud field]

but yeah it's true, if anything, ai increased creativity

[floral sluice]

I wouldn't be able to turn my thoughts into art without it.

I can program pretty much anything, but my art skills are those of a 4 year old chimpanzee.

[potent lotus]

I have ADHD. I can’t draw. I can’t focus enough to do it.

AI is an accessibility tool for the disabled so they can participate in the visual creative arts.

[vagrant zodiac]

please don't post this stuff in here. Use #1154829862171844679 with the spooky tag and the spoiler

[floral sluice]

Among other things 🙂

[loud field]

I dont have excuses like adhd I simply suck at drawing.

[vagrant zodiac]

also that gen was probably more disturbing than acceptable

[loud field]

I wish society abandoned this attempt to blame one's failures to some hypotetical disease of the mind

[potent lotus]

If you’re going to tell me that my ADHD is imaginary… don’t.

[loud field]

would be an interesting discussion to have, but not in this thread

[floral sluice]

Same

[floral sluice]

Sometimes it's the cause. Sometimes it isn't.

[vagrant zodiac]

#off-topic

[floral sluice]

Either way, Dall-e and it's pals are enablers!

[potent lotus]

Okay. I’m out. I’m not going to have disability that had ravaged me be marginalized because people can’t deal with the reality of it. DALL-E and things like it are boons for the disabled.

Night all.

[loud field]

I'd rather focus on discussing Dall3 and art and say that the rabbit was absolutely not disturbing. If you can't generate simple monsters like that, the tool will work only for ads and wholesome memes, not art

pick up any videogame you'd find far more creepy things

[floral sluice]

Agreed. It really has to have wide margins. Art is extremely often not nice.

[vagrant zodiac]

It's the server rules.

[loud field]

the entire world is light and dark, if we can't generate dark we can't even generate light by comparison.

[loud field]

oh if it's server rules and not openai rules it's ok

[vagrant zodiac]

I'm not saying you're not allowed to generate that, but this server has rules on content.

[loud field]

oh as long as it's allowes

[floral sluice]

Indeed

I also thought it was not allowed on dalle

[loud field]

speaking of which.... why is it that if I post a comic I made with dalle it gets deleted because it's not in english

[vagrant zodiac]

the automod tries to detect non-english languages and well... dall-e's gibberish sometimes looks like other languages

#chatgpt-discussions

[fair crescent]

If it happens, add an English description 🙂

[loud field]

well automod was correct, the comic was in italian

[vagrant zodiac]

ah that's why then

[loud field]

oh I can?

[agile peak]

One thing that differs between Bing and ChatGPT. Whenever you try to generate an oil painting for greek gods - it’s pretty noticeable.

Bing will keep making it as a statue, while chatGPT can make it as a realike portrayal.

[vagrant zodiac]

yeah the language rule also applies to dall-e images not just messages

[fair crescent]

To the post? Yes 🙂

[loud field]

lemme try

[vagrant zodiac]

the automod might still remove I think

but yeah you can try

[fair crescent]

I think the image is present as a hash, which <infinite monkeys with infinite typewriters> can create Shakespeare in French, which gets blocked.

[vagrant zodiac]

nah it's just OCR I think aha

(I think that's a workaround that's not allowed btw)

[loud field]

I just tried and it gave me a one minute ban

[vagrant zodiac]

yeah you didn't include an english description

[loud field]

I did? I think? oh you mean in the image itself?

[vagrant zodiac]

no you didn't include a caption you just posted the image

[fair crescent]

Why not? My image doesn't have any foreign language, it just gets flagged inappropriately. By adding English text to the post you solve the inappropriate flag.

[loud field]

Odd, but I am sure I did add a caption

[vagrant zodiac]

The rule is because non-english languages are blanket not allowed because we can't translate them

Lun-sei is posting a comic in italian

[green charm]

Yes please

[fair crescent]

Ah that part I missed. No idea how you handle that. Personally I would say 'why not'? As long as the conversation accompanied with the picture is in English.

But I'll leave that for the mods and OpenAI to decide 🙂

[vagrant zodiac]

We'd have to have faith that the translation provided is accurate

in my experience, people often don't do that properly

so you can end up with offensive words in non-english, and they won't translate those bits

but thissss is off-topic now

uh

yeah, no post italian comics 🙂

[fair crescent]

Ah now I understand. In case there's profanity.

Though that should be difficult to put into DALL·E. I thought it checks wording in any language as well.

[plain crescent]

Is this channel for dalle in chat gpt? I saw the announcement but Chat GPT insists it can't do it. Is it a canary rollout?

[floral sluice]

Do you have Plus?

[plain crescent]

yeah I do

[floral sluice]

Did you choose it in the gpt4 dropdown thing?

[vagrant zodiac]

try logging out and in, it'll be under Gpt-4 as dall-e3

it's not in native gpt4

[plain crescent]

Aha! I forgot there were more options under GPT-4

Thanks!

Oh hot damn dall.e 3 has no comparison with the previous iteration

[agile peak]

Bing and ChatGPT comparison with “a life like form” prompt.

[floral sluice]

both are cool

[vagrant zodiac]

'life like form' is a descriptor often applied to sculpture

so that'd make sense

[floral sluice]

This dall-e connection is dangerous. It's eating up my quota 😁

[cedar vale]

Thankyou so much! Works a treat now!!

[plain crescent]

I guess there's no ability to use vision with dalle.3 - anybody know if there are plans to bring that in?

[vagrant zodiac]

unknown

[agile peak]

Yeah. I guess both have different ways to interpret my prompts 😄

[plain crescent]

If they did that would be insane. "hey, here's my logo; re-create it as a halloween version!"

[loud field]

no idea, but it's cute

[floral sluice]

Remember that ChatGPT listens to what you say and creates a prompt for dall-e

There's a layer of interpretation.

[agile peak]

Apparently letting GPT to improve the prompt by telling it to be creative is much better than providing a very detailed prompt 😁

[crystal badger]

You can incorporate different interpretation methods using various methods of promoting to achieve better results based on initial requests

[loud field]

disagree, I want to tell it exactly what to do

[floral sluice]

Then you probably have to use Dall-e directly.

[agile peak]

This is GPT creativity. I just write “poseidon/zeus holding the trident/lightning with wrathfully eyes”.

[loud field]

no there's a way to have gpt do that

[agile peak]

That because I use custom prompts 😁

[crystal badger]

There’s always a way

But first one got to understand how GPT Works in synchrony with DALLE

[floral sluice]

I just tried saying "Use this exact prompt for dall-e: 'prompt here'" It worked

[crystal badger]

They’re machines after all

Instruction based program that’s generative

[agile peak]

Using exact prompt and telling it not to modifying it. It will generate exact same images in the next one.

[crystal badger]

Great tip is to ask GPT

You can use prompts to create better prompts

🤷‍♀️

[floral sluice]

It just made an awesome image for me while I was testing the "exact prompt" thing. 😄

I'll post it in the gallery in 5 minutes, when the slow mode due to problem uploading allows me. 😛

[crystal badger]

I’m glad there’s a slow mode now

Before pics were flooding in

[floral sluice]

Me too. I just wish it didn't kick in when an upload fails.

[crystal badger]

That’s true

[floral sluice]

Not a big deal though

I only switch to other windows when I need to have my mind slow down a bit

While programming I tend to get... a bit too.... coffeed up.

[agile peak]

After these two finished. I only told it to combine them in a single frame of image.

[floral sluice]

The upload failed again, and countdown timer commenced. :/

[floral sluice]

What was the prompt it created?

[agile peak]

This wasn’t using a prompt but just command GPT to combine “Zeus and Poseidon in a single frame”. But I started by creating each individually beforehand.

[floral sluice]

When you click the image it shows the prompt to the right of the image.

[agile peak]

Can’t see it 😂 I am on a phone right now. Let me switch to desktop.

[floral sluice]

Also, if the prompt is too long you can click "copy" and it will get the full prompt.

[agile peak]

It was just like this:
Oil painting depicting Zeus and Poseidon, gods of thunder and the sea, in a modern city, both consumed by wrath. Zeus commands thunder and lightning, while Poseidon summons towering waves and chaos. The city's skyline is the backdrop for their cataclysmic clash,. Capture the gods' furious power and the city's impending destruction in intricate detail.

[floral sluice]

it's pretty good at prompting it seems 🙂

[agile peak]

Yeah. Had to ask GPT3.5 to create prompt for each individually as well haha.

[quiet brook]

why was human art channel removed

[merry vault]

My guess is people were being toxic, but idk

[white skiff]

it was locked like many others for the new gallery system, its still in the full channel list

[merry vault]

Like, saying either ai or human is better etc

[merry vault]

Or this too lol

[quiet brook]

there’s no tag about human arts in the new gallery

only human+ai

[white skiff]

yeah cause its titled dall-e

you can use #1070006151938314300 but its possible they dont want only human made stuff anymore

[dim cradle]

guys if i managed to get a semi uncovered painting with dall-e 3, am i allowed to post it here? genuine question

[white skiff]

no

[dim cradle]

oke

thanks for the info

[merry vault]

Wym semi uncovered?

Wat does uncovered mean?

[white skiff]

nsfw(?)

[dim cradle]

bare chest

[merry vault]

Ah, makes sense. As a minor I suggest u do not post that here lol

[dim cradle]

technically it’s not safe for work, and you probably don’t work, so you should be fine

[merry vault]

Damn i should have waited to work at maccas (translation for non aussies: McDonald’s)

[dim cradle]

i’m gonna post a variant with no bare chest

[wicked marten]

only problem is openai wont make much on it bc many people use bing images unless openai shuts it down

but bing images is by their collaborator so i dont think

[merry vault]

The api charges on usage so the more people use bing, the more money they get

[quasi ridge]

hello, I am trying to generate a 18:6 ratio with Dall.e. I am not sure this output format is supported so I was trying to create a 16:9 image and inside represent a Canvas with 18:6 ratio with my composition but no luck. What type of prompt should i use to create an image with this type of ratio?

[paper dagger]

[plucky hare]

There are just the three preset image sizes, currently no other way to modify specific sizes for image generation! Though I like your "frame within a frame" idea, I just don't think it can ideate like that yet.

[cosmic bluff]

I'm going crazy! It seems to be impossible two describe two human characters, a dog and a cat and have the dog breed and all the animal colors correctly.

This prompt is actually correct, but the result isn't. I've tried like 200 images:

Impressionist painting of a young WEDDING COUPLE on a gravel path. The BRIDE with SHORTER BROWN hair in a form-fitting LONG WHITE dress. The GROOM in a GREEN suit, matching bowtie, and SHORT BROWN hair. and a short circle beard They are accompanied by a BLACK CORGI, a bright ORANGE CAT, and a TRICOLOR CAT with distinct patterns of black, orange, and white.

Could anyone help me? What am I doing wrong?

[zinc dagger]

you've got way too many details and subjects being thrown in the prompt, introducing more than 3 subjects with conflicting details tends to cause the model to blend them together. Your best bet would be to simplify the prompt, or wait for inpainting via labs and then gradually add in those details

[vapid elk]

it indeed is, the AI does not work well with too many elements on the prompt

[white skiff]

probably not doing anything wrong but just have too much details for it to handle it all

[crystal badger]

Oh, gotcha! It seems like the prompt has a lot of details, which is good, but there’s a bit of a hiccup with punctuation and the capitalization might not be necessary. Here’s a smoother version:

“Craft an impressionist painting of a young wedding couple on a gravel path. The bride, with shorter brown hair, dons a form-fitting long white dress, while the groom, in a green suit with a matching bowtie, has short brown hair and a neat circle beard. They’re accompanied by a black corgi, a bright orange cat, and a tricolor cat with black, orange, and white patterns.”

This version keeps all your details intact but with a bit more clarity and flow. The capitalization is toned down to keep it easy on DALL·E’s “eyes”. 😉 Plus, the punctuation is tidied up to avoid any confusion. Hope this helps for a better generation! 🎨

Stay Artsy

[plucky hare]

Like others have said, I think your animal descriptions are too complex. The model can't reliably produce a number of things ("make a group of five people") in general, so it also can't reliably track traits of "thing number 3," for example.

ETA: this is the prompt I used if it matters! Slight edits
Impressionist painting of a young couple on a gravel path on their wedding day. The bride has shorter brown hair and a long white dress. The groom has short brown hair and a short circle beard, and a green tuxedo with a matching bowtie. They are accompanied by three pets: a black corgi, a bright orange cat, and a tricolor (black, orange, white) cat.

[crystal badger]

Agreed

[cosmic bluff]

alright, thank you guys! I became very desperate over the day thinking I was just unable to find the correct prompt.

so when it comes to details two characters/animals are the most there is?

[crystal badger]

Sometimes minor things can change the trajectory of the outcome

Especially with image generation

It takes everything literally

[plucky hare]

If really depends. Things that are often depicted in pairs, like the couple in your prompt, may be easier for the model to handle because there are so many examples of it in its training data, whereas specific combinations of pets will be more difficult to produce. There's no explicit limit on the number of distinct characters you can work with, it all depends on context!

[zinc dagger]

you might be able to get more in if the requested subjects don't have overlap. You'll notice that in place of an orange or tricolor cat the generations will make a corgi with those colors, or vice versa. I tried using the same prompt as Solbus but changed one of the cats to a pony and got these results. there are still issues with the corgi and cat, but all the subjects were rendered

because the model works through diffusion, it starts from noise (static) and works backwards to create an image, looking for distinct patterns. so the more distinct a subject is (a horse being much different than a cat or dog) the less likely it is to be displayed incorrectly among numerous elements

[cosmic bluff]

I see.

[merry vault]

Never thought of it like that, knowing that could help

[zinc dagger]

another thing to keep in mind is what sort of a "bias" the training set data might have. Black corgi's aren't as common in online images as the orange corgis are, so Dalle is more familiar with rendering those.

[merry vault]

But it does know colours

[cosmic bluff]

@zinc dagger Yes I realized that this is a big problem. Even only depciting the black corgi alone made enough problems.

[merry vault]

Could this be used to an advantage to make ai proof captchas?

[zinc dagger]

yup, but if you do a google for "corgi images". it is overwhelmingly the classic orange coat on the dog, so the AI will need to work around that bias

[merry vault]

Well in like 5 years there will be a breakthrough that makes it impossible to block

[raven magnet]

Does ChatGPT dall-e always take forever

[cosmic bluff]

Takes about 10 seconds for me per image.

[static tangle]

Ok...maybe it is just my frustration, but this is what I'm getting from DALL-E 3 through ChatGPT.

I apologize, Mark, but I was unable to generate the image due to content policy constraints. Please let me know if there's another way I can assist you or if you have a different request.

Then, after I ask what policy constraints:

I apologize for the confusion. There shouldn't have been any content policy constraints for your request. It appears there was an error on my part. Let me attempt to generate the image again for you.
DALL·E 3
Error creating images
I apologize for the inconvenience, but it seems I'm still encountering difficulties generating the image you've requested. There shouldn't be any content policy constraints related to your request. I'm not sure why there's an issue, but I appreciate your patience. Would you like to proceed with another request or is there another way I can assist you?

Seems like just circular arguments...I usually get this when I ask it to take an image that has already been generated and make a tweak to it.

[raven magnet]

Not me

[zinc dagger]

try asking it to write out the prompt before it generates the images, usually you can see the issue in the text.

[cosmic bluff]

for me it sometimes helps if i say something like "why is it violating the content policy? i think my prompt is ok, so please try again and create the image". this very often led to pushing through the "error message" for me.

[static tangle]

Thanks...I keep asking why it is violating, it says it isn't and starts to regenerate, then gives me the same policy violation message. Thanks, @fisco, I'll try that and see if I can figure out the "bad word."

[static tangle]

Apparently asking "What words in the prompt violates content policy constraints?" prompts it to look at the offending words and rewrite the prompt using more PC language. In this case, it didn't like the word "black" to describe a color...smh.

[deft swallow]

It is not that simple. Consider this:
FORBIDDEN: "Caricature of Napoleon playing chess against the Duke of Wellington"
ALLOWED: "Caricature of Napoleon playing chess with the Duke of Wellington"

[deft swallow]

It's your fault. By bringing your issues into the discussion, you are trying to make this about your illness, and curry sympathy and pity, than about the tech and what it can do. Whether you realize it or not.

[tall patio]

Been using Firefly for a while

Gotta say I underappreciated D3's content policy by a lot

[deft swallow]

how so?

[cobalt hull]

I'm kind of getting some mixed results, but when you get a generation is it supposed to be able to add details to an image.

For example, if i say "a white cat in a car" then it gives me 4 images.

Then I say something like, on the top left image add a hat on the cat. Type thing. Is this within it's capabilities or am I wasting my time trying to have it reference a previous image?

[cloud dome]

It can't edit a picture, it can only modify the prompt

[cobalt hull]

Gotcha

[cinder sonnet]

What do you mean by simple? Also, neither of those two prompts was blocked for me on ChatGPT. May be different on Bing, but that's not what I was talking about

[dim cradle]

Hello, did OpenAI already announce when they are going to release the DALLE 3 API? They only said somewhere in Fall, correct?

[zinc dagger]

I think they revised their statement to say sometime within the year

[dim cradle]

Damn, I hope not. Could you maybe try to find the source?

[zinc dagger]

I did some digging and I think I was mistaken, someone had posted in the discord saying it would come out later, but all the official sources (the website and official discord members) are saying this fall

[dim cradle]

Ok 👍

[scarlet mango]

i am trying really hard to get a wizard hat that does not have a brim, i have tried conical hat, pointed hat, but it keeps making the brim, i am using gpt and repeatedly asked for no brim,, any ideas on how to tweak it?

[cloud dome]

Yeah, trying it myself. Really doesnt like brimless

[scarlet mango]

i asked the bot if there was a name for a wizard hat without a brim, and it saidthere was and suggeste conical or cone hat, and i was happy, so gave it a go, but still

[cloud dome]

This does not bode well

[scarlet mango]

hahah

tho good idea to go to the basics to try and find it

[lilac cipher]

ahahah

[zinc dagger]

have you tried calling it something other than a wizard hat? maybe "merlin's cap" or "the type of hat that a garden gnome wears?"

[tall patio]

My man just wanna to gaslight the bot into producing KKK hat

Not cool btw

Those types of politically provocative images are the reason why the bot is under such strict content management

[scarlet mango]

going the santa hat route maybe with tweaks

[zinc dagger]

santa hat has some promise

[cloud dome]

That's clearly not what's happening here

[tall patio]

It is

[cloud dome]

sighs

[scarlet mango]

you pushing the bit too far, first line hit with good comedy, but you overdone it

[cloud dome]

Gnome hat?

[lilac cipher]

garden gnome hat

[scarlet mango]

ohhh gnome is pretty good

[cloud dome]

Youre a genius

[lilac cipher]

@zinc dagger find the trick

[zinc dagger]

I thought Garden gnome might be a good trick

[cloud dome]

👏

[scarlet mango]

now to get it to work oh Merle here

[zinc dagger]

here's a direct comparison between two prompts, one says Conical wizard hat, the other says Gnome hat

[scarlet mango]

i work at a game studio and trying to recreate some of our characters faithfuly in ai

a lil closer, i think that using wizard in the description of other part of the character may be tainting the results

[brazen sonnet]

Hello

does dalle3 have a download button or API yet???

[zinc dagger]

you can try describing the hat as a "Phrygian cap" since that's the proper term for it, using the word "gnome" might add unwanted gnomish influences.

[brazen sonnet]

no bulk download button on dalle3 is crayz

[scarlet mango]

!!!!! tho any idea why sometimes it rotates

[cloud dome]

Man I wish I could develop some of the prompts you guys craft. Truly incredible. These were made using pieces of the prompt by @tawny monolith

[zinc dagger]

it's a known bug they're working on. I found that telling GPT to add the word "vertical" to the start of the prompt in question helps.

[crisp egret]

Hey, what's the image styles we can create with Dall-e ?

[unreal cloud]

have never seen a brimless wizards cap.

[fair crescent]

Which ones can you not create is a better question. Do you have anything in mind? If you look through the pictures here you can see a slew of styles. You can even mix them.

[crisp egret]

i talk about style option like "illustration, photo (default), fantasy, anime, isometric" ?

[fair crescent]

All of the above are valid 🙂
Anime has a chance to be blocked by the content filter at the moment. It is more quickly provocative than other styles.

[earnest flame]

It is blocked in chatGPT

To add on to the anime thing: female figures are more likely to be too provocative

If you're having trouble be sure to describe clothing.

And avoid words like "body","skin" and other anatomical words

This should typically result in reasonable female figures that don't get blocked

Personally I'm not looking for pin-up type styles too often so normally I don't have too much issue.

[wispy storm]

Might have more chance on bing. But don't be surprise when they will update their model to avoid being sued

[static tangle]

they call it a "Dunce cap" lol

[unreal cloud]

Yep. I thought about suggesting that

[cloud dome]

I used that in my prompt. Still had a brim.

[ornate dagger]

Dalle 3 seems to be super good at watercolor and ink on canvas

Two recent ones I made

[jade locust]

Is anyone else getting a wall of text when testing dalle3?

I was just playing around with it to see if it could generate a jackolantern template to carve a pumpkin this weekend.

[ornate dagger]

That's a base64 string it represents the image

You can use an online decoder

[vapid elk]

looks like the AI somehow decided to display the image in base64 instead of embedding it

seems like a bug, you can report that result with 👎 button

[jade locust]

Sorry this is on me, I didn't realize you had to select it from the dropdown menu. -Noob

[vapid elk]

lmao, yea

[cloud dome]

Holy smokes

[tall patio]

Interesting

Paste your favorite song lyrics on Dalle3

Magic

[agile peak]

GPT existential crisis

[scarlet mango]

you need to meet more wizards

[unreal cloud]

Dunce caps, as the other person said.

[scarlet mango]

i dont want to conceed loss but dunce cap is giving me better results

[cinder sonnet]

Dall-e 3 noticably slower today for anyone else? I feel like it is taking at least twice as long to generate compared to earlier in the week.

[quartz vale]

could be due to the public release

[cinder sonnet]

Yeah that would make sense

Well on the plus side maybe I won't hit the rate limit so often haha

[quartz vale]

TRUEE

[tight verge]

in the patch notes yesterday they mentioned having to unwind a caching layer. my fine tune sat in queue for 20 mins this morning before it started, which was a first for me. they are probably scrambling (on a friday, too) to scale and keep systems up for us

[scarlet mango]

well i was able to get about 95% of what i wanted, thanks for the help guys

[cobalt hull]

hey all, is there a way to be able to see a listing of the daily themes? I have to scroll ages to see what the one for the previous days were

disregard, figured out i could search by user themesaurus

[naive lake]

I just made 5 generations and it's already giving me rate limit error :/

This is very annoying

[cobalt hull]

Agreed, I feel around this time of the day is when these rate limits seem to come in more heavily

[naive lake]

This is really terrible.. just when you think you gonna get what you want, full of excitement in your eyes you type a new prompt and voila! an error!

[cobalt hull]

lol yup totally sucks, the worst is when you get hit with the 25 minute limit

i can deal with the 1,2,4 and 5 lol the 25 ruins my day

[calm needle]

Can anyone tell me if Dalle through ChatGPT is any better on restrictions compared to using it through Bing?

[cobalt hull]

I dont' know why but I get wayyyyyyyyyyyyyyyyyyyy better results via chatGpt

I really do think it's because you provide prompt in ChatGPT but it modifies it for a mroe friendly prompt that is better understood by Dalle as opposed to via bing i think it's straight what you type

[wispy storm]

about the same

[naive lake]

ughhh

[wispy storm]

refresh

[cobalt hull]

@wispy storm do you really get about the same via bing? I haven't been able to get anywhere near the same results

[wispy storm]

both have their pro and con but i prefer chatgpt dalle3

[forest seal]

I do too. The filter on bing dall e 3 is extremely wacky, and you get more images on chatgpt dall e 3. At most the weirdest thing that happens in chatgpts filters are only false positives or it misinterpreting what you want so you dont get all 4 images

[naive lake]

You think I don't refresh? 🙂

[wispy storm]

unfortunately, a lot of people forget.

we are trying to give solution 🙂

[naive lake]

So is that a bug that a refresh makes go away? I doubt it

[wispy storm]

sometime it does 🙂

[naive lake]

Ok thanks 🙂

😬