images-discussions

[unborn tusk]

Mines is working again

[ripe wraith]

not me going back through all my old conversations saying "Visualize this for me!"

nevermind, that does not work, locked into the model u started with

[crystal wraith]

two questions: 1. how do you guys generate landscape photos do you just outpaint? 2. can you access full generation history on the DALLE-3 bing page?

[winged moon]

The landscape photos are from ChatGPT with DALL•E

[crystal wraith]

oh ok. didnt know so many people have access to that already.

[ripe wraith]

My custom instructions were seriously limiting the GPT-Dalle flow. The prompts are looking much more visual with them removed. just fyi in case anyone is having trouble

[crystal wraith]

What does it look like when you have it? Was it just integrated across all the versions? (Web search, plugins, data analysis)

[ripe wraith]

only have it on web so far

[crystal wraith]

Or are you just using gpt to make prompts

Ok 👍

[teal timber]

It will show up as a separate selection

under gpt-4

[crystal wraith]

Oh alright then thanks.

[ripe wraith]

it's still having major issues with words for me, but i heard the bing version handled words just fine now

anyway it looks like this, embedded

[crystal wraith]

Is that the only interface that has “vision”?

[ripe wraith]

I don't think this is vision, just Dall-e integration

[silver violet]

Re: the person that had an issue with one or two of their four prompts failing - I found that happened to me when generating images of people especially if I included any reference to their body. I had a hunch and asked ChatGPT to specify they were clothed in each prompt, otherwise identically worded - and the images generated fine. This has happened with two image groups now. My theory is Dalle was trying to make unclothed images and its safety system was preventing this behind the scenes

[full patio]

I've seen a lot of people get naughty images even with very normal prompts. It's a little narty right now lol.

[silver violet]

Also experimenting with different styles of prompt, having ChatGPT make a more “abstract and emotional cues” laden prompt really upped the vividness and emotional impact of generated images

[crystal wraith]

https://www.bing.com/images/create?FORM=GENEXP&darkschemeovr=1&ssp=1&setlang=en-US&safesearch=moderate

[lost flume]

Dalle 3 seems to be down right now, is it happening to everyone or just me

[icy musk]

Bing Image Creator isn't working for me right now either

[frank thunder]

For me too. I couldn't do anything all day

[dim cradle]

working here. gpt at least, bing is another issue

bing seems to work for me for like 20 minutes at a time then down for a few hours

down for me rather, im sure its not down for all

[lapis ferry]

How do I get it on gpt? I'm a plus member

[rigid bough]

it's in Beta, a few people have access

[dim cradle]

there was a google document

[lapis ferry]

Ah okay. I saw the announcement about it and assumed it would just roll out to all plus members in October.

Thank you

[teal timber]

Still like 28 days of October to go! 😉

But it is slowly rolling out now. I think a small chunk of plus users randomly have it now, and it's increasing over time

[lapis ferry]

That makes sense. I appreciate the assistance! I'll just wait and keep my fingers crossed lol.

[dim parrot]

No updates about dalle through gpt not working right now?

[dim cradle]

working here

[placid orbit]

works for me

[dawn spire]

how can i get access to dall-3 on chatgpt plus?

[ripe wraith]

WHAT, ok I think it has internal vision at least. It reference "in the style of the first image" and "fourth image" and it worked REALLY well

It's really amazing

[dawn spire]

How did you get access? Ty

[subtle pendant]

You had to fill out a form posted by a worker at OpenAI

It is closed now

Too late

[ripe wraith]

u sure its closed? i did it today and got in

[subtle pendant]

It closed recently

[ripe wraith]

ah

[subtle pendant]

The form was leaked so they had to stop taking applications lol

[lofty willow]

what google form?

[vapid elk]

something that is public can't leak lol

yesterday, Moxi, a OAI dev, needed to do some tests with more people using dalle3, so, she made a form for people to request access to it and posted it here
it is closed now

[lofty willow]

aw 😦

missed it

oh well

is there an official ETA on rollout? I've been refreshing every few hours in anticipation 😄

alone the image and speech update were pretty amazing, cant wait to see what DALLE3 can do

I've been feeding it memes to see if it can understand the context and humor, has been SPOT ON

[vapid elk]

DALL·E 3 is now in research preview, and will be available to ChatGPT Plus and Enterprise customers in October, via the API and in Labs later this fall.
https://openai.com/dall-e-3

[lofty willow]

aw, must've missed the date on the announcement. Had looked at that a while back. Thanks for the info!

[vapid elk]

no date, just this vague "it will be in october" xD

[lofty willow]

to be fair, OpenAI rollout has usually been pretty sporadic. It's usually a coin-toss to see who gets what when. Just checked again, at least it says early october

so hoping for this or next week at least

[vapid elk]

it is literraly a coin toss xD

[lofty willow]

on reddit folks are still waiting for image & speech, I was lucky enough to get that 2 weeks ago

[vapid elk]

like.. the rollout is gradual to random users over time

[lofty willow]

I mean, if it is truly random, at least it's fair

and at least DALLE3 is available on bing, so not all lost while people wait

[vapid elk]

well, some people ar manually granted it first, such as people who do the beta tests

but for the gradual release.. yea, random

[lofty willow]

hm, how would someone go about signing up for betas?

[vapid elk]

stay tuned to #announcements

when they need people to beta tests, they will announce it and usually post a form

[lofty willow]

oh snap, didnt know that. I check it when I see pings, but havent seen a form yet

[quartz vale]

the form wasn't posted in #announcements . It was posted in a few high-traffic channels as there wasn't a need to ping everyone for it 😄

[lofty willow]

Anyways, appreciate the info 🙂

[vapid elk]

true, a rare exception

[hushed portal]

The model does a great job of separating independent objects in a scene.

[normal loom]

im not sure if its intentional but after playing with both the gpt and bing versions, the gpt version seems much more likely to trigger a generation block from the content policy/prompt filter.

A LOT of prompts that would get blocked on gpt work in Bing

[ripe wraith]

Can anyone confirm if the bing version is better at text? I haven't seen a single clear word yet.

[tacit mortar]

i have problems creating in chatgpt now

anyone else?

[normal loom]

a clear word using the gpt interface or the bing one? ive had decent text from both

[tacit mortar]

give to me an error

[halcyon moth]

clocking in just to say - yup also dead here too, 5 attempts, 5 fails

[coral axle]

Same here

[halcyon moth]

orly?

[tacit mortar]

ye so there are some problems

lets wait 🙂

[halcyon moth]

every time someone comments?

jeeze

[tacit mortar]

normal, still a beta anyway

but really amazing job by openai

[ripe wraith]

I ask for clear words in this one

[tacit mortar]

thank you to them for this

[halcyon moth]

na, I mean the discord telling me it doesn't do prompts every time a message posts

[tacit mortar]

cmon xd

[halcyon moth]

well, yeah anyway, the classic if it doesn't work in 5 minutes... just wait longer I guess

[normal loom]

may be the wording used? try something like "has the word 'word'", "has the banner 'word'", "includes the word 'Word'"

[vapid elk]

it is a sticky message

because we can't stress enough how much we don't have image generation here

yet people still randomly post prompts expecting them to become a image :V

[halcyon moth]

both understandable, but also mildly annoying 😄 but yeah I get it, people constantly trying to use this server like midjourney, even more annoying

[muted moss]

Is anyone else having issues with ChatGPT image generation? It keeps failing with no explanation.

I got some beautiful images from it before though

[unborn tusk]

Just gotta wait a bit

[muted moss]

I guess I can copy/paste prompts I like into Bing in the meantime

[tacit mortar]

Yeah

[ripe wraith]

What blows my mind is that I'm talking to an AI thats talking to an AI thats calling image generation things thats doing all this stuff

even the accidents are mind blowing. This accident doesn't match the request, but it accurately represents my state of mind, so.. yay

[uneven swan]

Yeah me too

[ripe wraith]

Its been working for me all night. Is it your prompts?

[uneven swan]

Shouldn't be at all

This is like the 5th time I got this error message in a row

[ripe wraith]

oh, regen each time?

oh, now im seeing it, boom

[uneven swan]

Yeah it has to be the servers

Bing still works thankfully

[ripe wraith]

It could be like when the coding version first came out, some characters in the prompts would randomly break it.

[uneven swan]

Hmm maybe

[muted moss]

I’m noticing a repeated issue in which the AI falsely assumes the image generator remembers previous prompts or images. E.g. it might describe a character in one image, but just say “the same character” for the others, instead resulting in images of different characters because it doesn’t remember the description. Or I’ll say “I like the first image” and then for its next generation it’ll say “the character from the first image”.

[ripe wraith]

at first i thought it was using its Vision on those images, but in reality it doesn't seem similar enough to be true.

So it's hallucinating bigtime when you reference previous picture, I guess?

Could you imagine the amazing feedback loop though, when it can generate an image based off rough text, then look at what it's created along with our text to make a better copy

its hard to contain my excitement at that possibility

[hybrid sorrel]

Hm, does anyone know how to enable DALL E3?

[ripe wraith]

it's not released yet, a small random few have early access

no

[steady basalt]

It seems to say that it is not supported YET

[warped belfry]

Yay, Dall e in my chat! Let's go 💪

Let's see if you what we can craft.

[ripe wraith]

It's trying at least..

[ripe salmon]

is the chatgpt actually able to use different seeds for same prompt or is this gas lighting

[ripe wraith]

I'd bet on hallucination

[ripe salmon]

I wish I could tell me when its working from a previous seed or making new images with new seeds

[copper plank]

if you overlay them in an image generator and lfip back and forth you can generally tell fi they're the same seed

[glacial bough]

Hey everyone

While waiting for a D3, can someone tell me whether it has consistency with characters? so I can start making my children books daily 😎

[midnight mango]

The ChatGPT version has been down for like 15 minutes for me. Keep getting this message:

[steady basalt]

If you are extremely descriptive of what the character looks like, it ""sorta"" does have consistency. Issue is nailing the same style every time.

[glacial bough]

Looking forward to have an access. People call me “AI buddy”, I know how to talk to them )))

[steady basalt]

Do have fun with it when you get it.

[smoky pumice]

hi @grave wing

[grave wing]

Does anybody know if its possible to acces dalle 2.1 (the old one used in bing image creator before dalle 3)?

I like dalle 3 but I feel it lacks a bit of the realism old one had... dont get me wrong the quality and lack of artefacting is amazing but every image feels staged and "royalty free"... I could explain in more detail but essentially dalle 2.1 images just had significant artistic and stylistic difference when we ignore the overall quality aspect like correct amount of fingers ect...

[grave wing]

Hi

[smoky pumice]

i'm confused you can ask dalle to create image in different aspect ratios?

but it seems like it's only through chatgpt 4

i was looking at #🌄┃general-art

[grave wing]

Im not talkig aboit aspect ratio, im talking about overall feel of images

[smoky pumice]

i wasn't talking to you

[grave wing]

Lmao

[smoky pumice]

@nimble rose

[nimble rose]

Yes, in a way yes.

I can manipulate the prompts to "work" the images

I'm going to send some now to the image channel, just a minute.

[smoky pumice]

but only through chatgpt 4 not bing image creator or any other platforms that use the api?

[nimble rose]

Personally, I'm using ChatGPT 4, but I think this is possible on Bing too, because I'm using the same prompt engineering technique that I used on Bing.

[smoky pumice]

but how do you prompt it for a different aspect ratio

based on the screenshots provided by @ripe wraith and @spare bough in the #🌄┃general-art channel there isn't any mention of aspect ratio in their prompt

[fair crescent]

Could you move this to #❌┃prompt-help and add the prompt? Easier to handle it there 🙂

[fair crescent]

Different aspect ratios are currently isn't available on Bing image creator. It is unknown if it will be available in the future. @smoky pumice

[copper iris]

sorry for the noob question how are u using dall e 3? I thought it's not out yet

[smoky pumice]

how does it look like on chatgpt to change aspect ratios?

[ripe salmon]

https://forms.gle/3DokUVBkzTU5ZaRV8

[drifting hawk]

form is closed

[ripe salmon]

Oof

[drifting hawk]

but we will be rolling out to more plus users throughout the week, a small % have it already!

[nimble rose]

For me it works in a natural language format. You simply ask, it doesn't need to be something like "-resolution 100x100, -quality:2 etc"

[drifting hawk]

^ yup, there's 3 formats, square, portrait or landscape (default)

[nimble rose]

You conduct the creation of the image as if you were asking an artist to do it.

[fair crescent]

As simple as talking to it. The best trigger words are wide, tall and square. It currently seems to favour wide without prompt. Asking for panoramic pictures for example has been known in the past to trigger wide aspect ratio, so the same applies to the other aspect ratios.

[nimble rose]

By the way, I just wanted to say that your guys work is amazing... I'm having a lot of fun and reporting as often as possible! Continue on this good path!

[hushed portal]

According to ChatGPT:

images can be generated in different aspect ratios. The available resolutions are:

Wide: 1729 x 1024
Square: 1024 x 1024
Tall: 1024 x 1792

By default, it uses the wide resolution (1729x1024) unless the prompt suggests something different, such as a full-body portrait which would be more suited for the tall resolution. "If you have a preference for a specific aspect ratio, please let me know, and I'll use that for generating images."

[smoky pumice]

does this work in bing chat?

[fair crescent]

No

[midnight mango]

I wish ChatGPT wasn't reusing the same seed

[hushed portal]

I dunno, but I'll give it a try.

[fair crescent]

I do notice ChatGPT needs to be prompted. I ask fir screen filling playing cards, but without further suggestions it'll come out wide.

The prompt does not need the words though. You can ask "give me a tall picture with the following prompt: 'a landscape of a mountain range'" and it will generate it tall. It is extra information send next to the prompt.

[hushed portal]

I'm pretty blown away by the legibility and aesthetic of some of my generations.

[hushed portal]

Yeah, as long as you imply what's needed, it's pretty good at picking that up.

[smoky pumice]

is microsoft giving free access to chatgpt and dalle to its users just there way to get people to sign up for chatgpt plus?

seems like a great way to get a quick return on their huge investment into openai

[mortal summit]

A slight correction: OpenAI and Microsoft are closely connected, but ultimately independent entities. ChatGPT and DALL-E are products of OpenAI, where some parts are shared with Microsoft.

[compact pelican]

So close to become the best tool for student. ChatGpt know all formulas, but can't put it on picture wet. Maybe soon... 🙂

[ripe wraith]

ma'am, this is a wendy's

[drifting hawk]

It’ll be fixed soon

[fair crescent]

Can we still ask for a fixed seed later? It can be a great boon at times.

[mortal summit]

I honestly appreciate how stable prompts are at the moment.

[midnight mango]

sometimes it can get in the way when I ask for more images, it just generates the same four images

It should be a toggle for sure

[hushed portal]

Are we allowed to share images created with ChatGPT on social media?

[tranquil dock]

I suggest you check the GPT-4 pull-down menu. I didn't notice it was available at first

[tender osprey]

Hi, I have a prompt that consistently does not work and is not policy violation. Are you interested and if yes, where to send it ?

[warped belfry]

interesting. So far it's working pretty well 🙂
is there any place we need to give feedback on dall-E in chatgpt Moxi?

[midnight mango]

are you on the mobile app?

just an FYI, the mobile version doesn't have it

[dim cradle]

I just want some variety in faces, I swear I get the same handsome man and woman in every photo haha

[tranquil dock]

I can also use DALL·E 3 on Android APP

[tranquil dock]

OpenAI has done the interface and animation well! I like it so much!

[warped belfry]

for some reason it showed me a cartoon a guy being hanged no problem ( was asking about a historical exectution on the construction site in the 1600's where i work). But when i said the guys head was chopped off instead it didn't want to cooperate anymore 🙈

[dim cradle]

Use Bing

[midnight mango]

patience

[remote zephyr]

so, once it goes officially live in Chatgpt, will Bing drop back down to V2.5 for image create?

[mortal summit]

Very unlikely, I would say.

[tranquil dock]

I don't think so.After all, Microsoft is relying on these new features to attract people to Bing

[dim cradle]

yeah the reason they have gpt4 and dalle veresions there is to draw a little attention to bing which seems to be working a little i guess. once gemini drops will be another story and everyone in the world just has it on their phone and everywheres else we encounter gooogle

[daring magnet]

they wont downgrade

[remote zephyr]

Yeah I get that. I'm just trying to understand for the average casual user, what would be the motivation to upgrade to chatgpt pro, if they can use use Dall-E 3 inside Bing?

[midnight mango]

it's worth it.

After using it in ChatGPT, I can't go back to bing

[remote zephyr]

I'm excited for it to be available for me in ChatGPT

[daring magnet]

aha i see , my only worry is if the bing would filter out also artist styles , now they are still working , in chat do not ... bing today begun to censor celebrity names

[dim cradle]

I think if it remains this restricted a lot of people will prefer bing by a long shot. most people want to make pictures of Boba Fett bowling with Spiderman or whatever

[midnight mango]

true

I wonder if it will be this restrictive in the API version

[ripe wraith]

ooo thats one way MS could play monopoly. use investment in open ai to keep special rules for itself that no competition could come close to.

[dim cradle]

I do feel like theres more natural variety to the bing images even the mood and tone of say a year. Lots of things in gpt to me look too polished, almost cgi, very shiny, theres no grit or grain when there should be. Its impresive techncially but feels a little souless to me

[remote zephyr]

Makes sense

[ripe wraith]

its soul is heavily censored from the public

[mortal summit]

The models will likely not be the same in the future. Besides there are additional functions of the eco-system or UX considerations that will differentiate the two.

I am sure there will be a purpose for both DALL-E in ChatGPT and in Bing.

[dim cradle]

Im talking just about composition not subject matter but I guess

[daring magnet]

all faces too smoothed , well this can be sloved puting thru img2img in stable diffusion

[dim cradle]

yeah that too, and as noted earlier, seems to be very few generic models I keep getting the same two faces

[remote zephyr]

I'm looking forward to using it via ChatGPT for myself, to run some similar prompts and judge the overall quality. I've noticed using it via Bing, the image fidelity is generally good, until there are more small details in the scene, hten it starts to really break down (as to be expected, as there is only a certain amount of render time per image etc).

[remote zephyr]

A lot of the images can benefit from a little bit of added noise or grain in post. Not much, but a subtle amount helps.

[dim cradle]

smaller stuff isnt perfected in gpt either, but its at least higher resolution for some images

[midnight mango]

biggest difference for me is the lack of compression, as far as visual quality goes. You get those all of those 1024 pixels 👍

[remote zephyr]

Nice. Looking forward to it!

[cloud narwhal]

Isn't bing less restricted than ChatGPT tho?? You can create pretty much any character/celebrity in bing..

[daring magnet]

all styles of artist of last 100 years seems not working in gpt ? in bing they (still) work

[dim cradle]

two different sets of rules

[midnight mango]

I guess it depends on if you want to make original art or just memes. For now, bing is better at memes, and ChatGPT is better at art

[remote zephyr]

Thanks for all the quick replies so far btw guys... i'm far from a noob in AI images (Midjourney is my tool of choice normally - I have a lot of experince using that) but Dall-E 3 is the first other platform in a while that has spiked my interest.

[cloud narwhal]

I kinda do both but I do alot more stuff with celebrities and copyrighted stuff. It's not even Dall-e 3 that has the restrictions It's OpenAI smh

[remote zephyr]

One more Q (sorry) - as I've heard, images generated via Bing cannot be used commercially? What is the official stance regarding images created within ChatGPT?

[midnight mango]

I personally much prefer the ChatGPT version. Since I'm making original stuff, the restrictions don't bother me. Plus I get more visually stunning images, and I don't have to think too hard about prompt engineering. Both have their merits

[cloud narwhal]

Can you still do stuff like CCTV footage or surveillance camera footage in ChatGPT like in bing?

[midnight mango]

Haven't tried it, personally, but I don't see why not

I've seen other people generate in that style with ChatGPT

[ripe wraith]

like this?

[dim cradle]

did you request it look 3d? thats wht i mean about the lack of 'real' which i know sounds ridiculous when talking about ai images, but the illusion is ultimately what counts

[ripe wraith]

i asked for real and it looks amazing, but it lost the cam angle. prob needs more tweaking

fun fact, that is how i feel these days

[tender osprey]

yeah, no

[ripe wraith]

why would you ask for a discord user lol

[tender osprey]

the problem it has is with the CCTV if you read the response

[ripe wraith]

Its responses are not pulled from facts

they just sound nice to it

[tender osprey]

might be right, but even if you remove the "discord" it doesn't provide the CCTV styled image. It just ignores it in the prompt

[crimson nimbus]

I love how we're just doing cookie monster cctv vision XD

[normal loom]

might try the bing version for some of the gpt blocked prompts, the bing version seems less sensitive and restrictive in my testing

[ripe wraith]

it's my rosetta stone from now on

[languid valve]

follow 🙂

[crimson nimbus]

it is beautiful 😄

[tender osprey]

so it works "sometimes", or you have to have a prompt writing uni degree to make it work ?

[ripe wraith]

It's prompt engineering all the way down

except now GPT does it all

[tender osprey]

If I want to do prompt engineering I'll go to SD, that was the whole idea of GPT doing the prompts, to not have to

[remote zephyr]

They look good, but don't really have a CCTV aesthetic to the image itself... apart from the obligatory fuzzy text around the frame

[ripe wraith]

its fish eye, then layers some over top

with more iteration it could be made perfect im sure.. it has no bias against CCTV footage. Just against CCTV footage of real people or copyright things

[ripe wraith]

I'm not even engineering, just not using trigger words

[dawn aurora]

anyone getting errors right now?

[crimson nimbus]

Top right looks so ominous.

[hushed portal]

"cctv footage of the easter bunny sneaking around"

[ripe wraith]

OMG NO

[ripe wraith]

we must stop this now.

I take it back. We can ban CCTV footage fully.

Im cool with it.

[ripe wraith]

I like the bottom left, it has a real action feel that makes it all more guilty

[crimson nimbus]

action and the fact he's on the staff only side XD

[gentle schooner]

CCTV footage of caveman inventing the wheel.

[ripe wraith]

here we have it folks, fake news

[ripe wraith]

right, we know he dont belong there

[crimson nimbus]

invents wheel

Wooden wheels on a cart behind him

[ripe wraith]

LMAO

[gentle schooner]

Don't worry about it.

[ripe wraith]

ok we need Harrison jabs at all the AI images from now on, PLEASE

there's an old worn out wheel behind it

years and years old

like those are the failed attempts maybe. He said nah these lightweight wood wheels are NOT IT

stone for me fam

[crimson nimbus]

Like a scene from Life After People:

I'm just testing its abilities.

[ripe wraith]

its defaults are so nice now

[crimson nimbus]

It seems like it does better or worse prompts depending on what art style/camera style you choose

(I chose Hyperrealistic VFX for this one btw)

I tried low-poly isometric and while it was close, the roads wern't quite right

looked too new

[ripe wraith]

i love the idea of self contained chunks like this you can swap in and out of a table of them

[crimson nimbus]

you can do Isometric if you want things like the above

it definitely does cool things

[ripe wraith]

Need isometric consistently, combined with some magic auto mesh maps, then AI world

I asked it to be more ironic 🙂

[crimson nimbus]

This is like a ring cam view 😄

@ripe wraith

(Sorry I had to try it myself)

[vale widget]

Prompt blocked?

It says it cannot make my image, but then the thumbnail says Prompt Blocked.

[celest socket]

This is because of the content filter

[daring thistle]

what am i missing? where is the text prompt? all i want to do is stick this image inside of a tv

i feel like dalle is stealing my credits

this is just using the prompt Create an image of the well-known anonymous icon encased within a television frame.

[eternal escarp]

Hi! Could you please advise on the size ratio of photo generation in DALLE 3?

[daring thistle]

sorry, i was on time out

1024x1024?

[eternal escarp]

1:2 In general, I would like to know how it works and what items to add to the request, if you can give me a detailed promt book, I would be more familiar with it, if there is one

[daring thistle]

im trying to use the image as an example for dalle. i want it to stick it inside of a tv. the "help" shows the user using an uploaded pic with a text prompt bar and using the prompt to ask for a request. i dont see that.

then i tried just using the prompt Create an image of the well-known anonymous icon encased within a vintage television frame. with out uploading anything to see if its smart enough but it seems to give me images that are way off

[eternal escarp]

I think it's easier to do it in photoshop

[daring thistle]

why is the prompt missing tho>?

i dont see any bar

[eternal escarp]

But I have

[daring thistle]

i see it pop up when no image

as soon as i upload one. its gone

ok, i got the prompt now but dalle is still stealing my credits

i guess im not smart enough?

or mine is broken

[vale widget]

Oh....I didn't think I was doing anything wrong. My prompt wasn't nsfw or anything

[celest socket]

the filter is not only on the prompt but also on the output image, and it's quite strong, They pprefer false positives than false negatives.

[vale widget]

Gotcha, so I perhaps didn't do anything wrong, but the image generated may have been NSFW. Gotcha 😁👌

[celest socket]

yes exactly

[daring thistle]

i give up

[cloud narwhal]

Kinda.

More like this

[fair crescent]

1. You are in the inpainting/outpainting editor. This works a bit differently than other generation ways.check out #resources for more information before you start inpainting/outpainting.

2. The Labs website still uses DALL·E 2. This model isn't like many of the pictures you see around right now. One thing is that it has trouble with text. Short, single words are possible with some tries, but no guarantee for success.

3. If you have further issues with the editor it is best to go to #❌┃editor-help

[daring thistle]

where do i find 3? or only cool kids have access?

ty for support

[cloud narwhal]

As long as you can do cctv footage, body cam footage, surveillance camera footage etc, I'll be interested in using it

[wispy storm]

Here a 3 .

But for real if you want dalle 3 you can use bing

But for chatgpt for plus user they rolling out

[fair crescent]

DALL·E 3 can be found on ChatGPT Plus in a beta version. It is being rolled out right now. Something that looks nearly identical to DALL·E 3 and likely is a version of DALL·E 3 is on Bing image creator. It certainly uses a version of DALL·E that can do relatively accurate text and make amazing images. No inpainting/outpainting though, so you always generate 'new' images.

[mortal summit]

They now officially call it DALL-E 3 in Bing. It is and probably will remain a different version though.

[silk anvil]

Inset family guy meme "But I want it now!" here^^

[daring thistle]

how is this so hard...

[daring thistle]

im guessing im not in the right place?

[dim cradle]

Will there be another Google Form to obtain D3 access, similar to what MatVidPro provided?

[daring thistle]

🤦‍♂️

[dim cradle]

yup, it seems that text generation was a bit over promised by openai, but Iam sure they will fix it soon

[daring thistle]

i think its trolling me

[dusk hearth]

Should be able to fix it in Photoshop right? It's just replacing text on a flat black background.

[daring thistle]

i never took the time to learn photoshop lol. i guess theres always a time for everything

[crimson nimbus]

I decided to try your prompt dave on the DALLE 3 via ChatGPT and it looks like it does the same. It seems to really like adding an additional letter (In my case the letter Y) between Anon & TV

[wispy storm]

That's what I end up doing . At some point dalle 3 can't do minor details

[grave wing]

How do I acces the "old new" dalle 2.1 beta or whatver thst thing was called on bing image creator before they switched to dalle 3 I NEED IT FOR COMPARISON PLEASE

[dusk hearth]

Dalle 2 to Dalle 3 is a great step but there's still plenty of things it can't do.

[wispy storm]

Indeed that's 80% better imo

[crimson nimbus]

I will agree. It definitely does text alot better and honestly... i think it keeps consistency better too

[grave wing]

For me it really struggles with realism compared to old 2.1, everything feels really staged and "royalty free", the old one had a lot artifacts and disfigured bodyparts but the images felt real and not like someone gathered people/subject for a group photo with perfect lighting, also it struggles for me with "bad" image quality styles like "found footage", "early 2000's cellphone recording", "vhs filter" ect unless the modifiers are so vastly different that I just cant comprehend them

[waxen portal]

Morning fellas! Something to laugh 😃

I liked how it added the emoji, too creative

[grave wing]

Please anybody

[crimson nimbus]

I made it do an impossible object by accident!

The whole image is here:

[icy cipher]

Hi Moxi, I noticed I missed the window for Dall-E 3 Alpha access via a form in Discord and I'm really interested in exploring its potential. I'm currently working on a government funding pitch focused on education for disadvantaged communities and refugees in Germany, which I have to send out next week. Gaining early access to Dall-E 3 or the Vision feature would allow me to explore additional innovative use-cases that could make a real difference.

Could you consider granting me access?

[crimson nimbus]

You won't be able to since the servers to 'DALLE 2/2.1' will no longer be connected to the website/service.

[grave wing]

No way even with any paid subsribtions advanced accounts whatever?

I had a plan to make a game and I wanted to use these certain images to generate assets from dalle 2.1 but I struggle with prompts to get the same art style in d3

[white skiff]

you need to wait for dall-e 3 to rollout normally now which is already happening, for vision you might need to wait a lot longer: "Which plans can use image inputs? Plus and ChatGPT Enterprise. Not yet available in the UK and EU."

[crimson nimbus]

The real estate images look almost on point XD

Its cool how many styles it recognises

[fallen mulch]

any option to use this free in invoke ai. I want to generate dekstop images

[hallow yacht]

cross-posting this from another channel, but this is really starting to frustrate me: try to get dall-e 3 (or any ai for that matter) to generate an image of a motorcycle without wheels. as in, just a regular motorcycle, but with the wheels removed. preferably hovering in place, like how it would sit with the wheels intact, but with the wheels gone. it seems to me that all current image-generating ais lack the ability to do conceptual subtraction (only being able to do a form of conceptual addition)

[vale widget]

How would I get a selfie style I age. Not an image of someone taking a selfie, but the actual selfie itself.

I can only get this

Which as epic as it is, it's not what I want 😂

[wispy storm]

What prompt did you use ?

[crimson nimbus]

Could try 'Image of a person's selfie showing blah blah blah...'

[final jungle]

It's not what you want, but its still amazing, lol

[stoic ruin]

So, I see that I have Vision enabled in Gpt4, but it does not generate images yet? Is that feature being slowly rolled out? Or was that only via the Alpha signup?

[mortal summit]

Vision is a different feature than DALL-E. Vision is about recognizing image content.

Right now DALL-E is being rolled out to all ChatGPT Plus users.

[lofty willow]

Which works amazing by the way. Been feeding it memes and it understands the context & humor

[orchid kayak]

so are we sure that bing is currently using dall-e3?

TBH i don't think they could have possibly done a worse job on letting the user KNOW what the freak they are using, it's all so confused

[mortal summit]

It is using a version of DALL-E 3, yes. It might however differ from the OpenAI implementation in some ways.

[hallow yacht]

how do you see that? i have dall-e 3 (image generation) in chatgpt-4, but i don't see any way of inputting images?

[thin rover]

[wind fractal]

Im still waiting for vision and dall e 3 in my gpt :(

[nimble raven]

I'm sorry, but I encountered issues while trying to generate images based on your description of a "coche" (car). Please let me know if there's another way I can assist you or if you'd like to make a different request.

is someone else with problems using DALL-E 3 on chatGPT?

I apologize once again, but I encountered issues while trying to generate images based on your description. Please let me know if there's any other way I can assist you or if you'd like to proceed with a different request.

all the prompts are getting this type of answers

[idle ravine]

anyone familiar with how the Image/DALLE GPT4 roll-outs are working? It seems I have acess to 'chat with images' but no DALLE generation. My model was actually hallucinating and responding with [DALLE 3 Image] instead of actual images.

[rigid bough]

some people got early access to DALL-E 3 in ChatGPT, for others just need to wait until it rolls out some time this month

[idle ravine]

So I guess the 'Chat with Images' and 'DALL-E 3' roll-outs are independent?

[rigid bough]

yes

[stable vector]

is there any way to "apply" for early (or any) dall-e 3 access?

[rigid bough]

not now there isn't, just need to wait for it to roll out

[fallen mulch]

how can i generate hd 2k or 4k res images?

[fair crescent]

You can use Bing image creator as well. It should be available for most or even all users right now.

[stable vector]

@fair crescent I can use that, but it lacks chat and isn't as flexible as what I'm seeing in previews

[fallen mulch]

?

[white skiff]

you cant

[fallen mulch]

maybe in the future?

[white skiff]

dall-e 2 never got those features and large images are expensive to create due the details so not sure, maybe the api will offer it

[crimson nimbus]

Sent a review but apparently this triggers the content filter... and I have no idea why:

An electronic motherboard with a graphics card, cpu and ssd attached all on a rubber mat. Isometric, low-poly.

[dim cradle]

Can you help me out?

[fallen mulch]

https://www.bing.com/images/create/an-electronic-motherboard-with-a-graphics-card2c-cp/651d7b9c666e4f1f8e3b83053f8c6245?FORM=GENCRE

[crimson nimbus]

Forgot to mention this is with DALLE 3 via ChatGPT

[dim cradle]

Can I interview you through DMs?

[fallen mulch]

it's not available in chat gpt for everyone

[crimson nimbus]

I know. But I have access to DALLE 3 via ChatGPT

[crimson nimbus]

So its just surprising its going to generate the images with a very innocent prompt and it triggers the content warning.

My only two things I can think of is 'graphics' or 'motherboard'

[fallen mulch]

is it possible to run it locally

[white skiff]

no

[fallen mulch]

so sad

i wish it would be possible

[crimson nimbus]

Ok figured it out. I think it doesn't like 'Rubber Mat'

[fair crescent]

You can use Bing chat for ChatGPT 4 and create pictures. It isn't ChatGPT proper, but it can serve for most use cases and is still 'free'

[fallen mulch]

is this the right model

[stable vector]

ah, fair enough. I haven't used Bing chat for anything yet 😅

[frank thunder]

Is anyone still having problems with Bing image? It's complicated since yesterday I can't use it.

[silver laurel]

I was in the alpha for Dalle 3 on Chat GPT yesterday and do have GPT Plus but no longer have access to Dalle 3. will it be back? it was great, thank you.

[white skiff]

its now in gpt-4 menu

[silver laurel]

refreshed and I see it now. Thanks!

[spare pebble]

Can you do different aspect ratios in bing image creator? I couldn't find the answer on the site

[somber coral]

Why text is not working properly?

Prompt:
Photo of a majestic lion standing with dignity, its natural paws brought together in a sign of prayer. The lion is elegantly dressed in a cotton white shirt and a traditional white dhoti. The gradient red background sets a peaceful ambiance. The image prominently displays the text 'Vanakkam' in black color.

[ripe wraith]

I think Dalle needs to go through grade school still though, to learn spelling properly 😆

I like to use this to illustrate the problem:

[dim cradle]

https://th.bing.com/th/id/OIG.zBvc8OivL_xZjHwCNAg7?pid=ImgGn damn.... I forgot o tell that the background should had been black. Now it's already done with a lighter background....

[tacit mortar]

are there any issues creating on chatgpt atm?

gives me errors

[ripe wraith]

comes and goes

[tacit mortar]

ah ok thanks

[copper light]

Dall-E 3 only works on Mac and pc’s bing so far

[ripe wraith]

not true, I'm on windows using it in gpt

[dim cradle]

https://th.bing.com/th/id/OIG.IsSUNwHRxQXW6aJJJgnb?pid=ImgGn hmmm.... so uploading the previous image to the chat and repeating the same prompt, but adding the black background almost made the same image.

[ripe wraith]

That's impressive actually, using the new vision image upload?

[slate hull]

is the restrictions set on chatgpt so bad that its not fun to use compared to bing?

[ripe wraith]

the real restrictions are in your mind.. just avoid keywords and phrases, dont be obviously evil

(I've had no issue)

[dim cradle]

I didn't know that I could drag an image to the chat in windows 11. Doing that along with the prompt works like SD's controlnet

[copper light]

bing is the closest I will get to using both gpt-4 and dall-e 3 before OpenAI decides to work on a compromise to allow those who wish to not break the bank free access to gpt-4 and DALLE 3. In the meantime, bing is outperforming OpenAI

[ripe wraith]

that sounds like assumptions. Give me an example and I'll prove it wrong.

[copper light]

Did u pay $20 for ChatGPT and dalle?

[ripe wraith]

yes

[copper light]

I did not so there

[ripe wraith]

I'd rather have GPT than netflix n all the useless garbage we have now

$20 a month to participate in the future, not a bad deal imo

[ancient jasper]

I had Oppy design a schematic for a cyberpunk mall and then design images for it.

Pretty nice

I'd go here

[ripe wraith]

ask it to make it "realistic", GPT seems to come up with great prompts from that

also, isn't it amazing that we're talking to GPT with simple terms, and its doing the job of prompt engineer so well?

makes you wonder about jobs in general

[ancient jasper]

I went into a whole thing with it about it theorizes it would be like if it could combine all of it's individual modules together

[copper light]

DALLE 3 is the best at copying its brother ChatGPT

[ancient jasper]

and we agreed, it would pretty much be Data from Star Trek

[ripe wraith]

yep, I think there's many like us who grew up with star trek and see this as the timeline we're in now

[ancient jasper]

It's pretty much why I only refer to it via the nickname, Oppy

I asked, what it would theorize it's preferred nickname be and it was like "A lot of users call my Oppy, that's pretty good."

Illustration of an exploded diagram of a futuristic weapon that uses elemental pill-like capsules as ammunition. The weapon's main body is sleek and metallic, with compartments for loading the capsules. Each capsule represents different elements like fire, water, air, and earth. The firing mechanism harnesses the elemental energy from the capsules, and the weapon's barrel is designed to channel and release this energy. Various components like the grip, trigger, energy converter, and safety mechanism are clearly labeled and separated in the diagram.

This is Bananas

[ripe wraith]

I got temp muted for referring to a negative event in the fictional star trek universe 🙄

[ripe wraith]

If you ever find a way to get it write more than one word clearly in english, let me know..

[dim cradle]

DE3 has a superior grasp of very complex prompts, and seems to understand syntax and grammar very well. However, Midjourney has a far stronger grasp on achieving various art styles, very aggressively, than Dall-E 3. With DE3 I feel I need to dilute the prompt much more in order to achieve the art style. I also have to dilute the prompt in MJ to achieve this, but at least MJ puts higher emphasis / bias on the art styles. So I can dilute it LESS in MJ. I think this needs to be tweaked for DE3. Higher emphasis given to prompts that pertain to art or artist styles.

[snow notch]

Hi guys when would be open Dalle 3 from API?

[ripe wraith]

Are you using Dalle in GPT, that adds another layer, GPT decides on the prompts from your input.

[dim cradle]

No. Bing

[ripe wraith]

I mean I'm sure bing does it too. We're abstracting further and further from the main input. More layers of AI will keep getting added

[dim cradle]

Scratch that. Art styles seem to get ignored when I prompt a complex instruction.

[ancient jasper]

I think Oppy is better because it let's you give it context to the images that are lost on just using the Bing Generator

That context changes the whole design

[ripe wraith]

A missing period could probably change the whole design too 🙂

[dim cradle]

I don't know this 100% - but I instinctively think what is happening is pertaining to some kind of "token" limitation. I was in the early alpha of Stable Diffusion in august 2022. I got to see a lot of "under the hood" stuff.

[ancient jasper]

Not as much, because I've done a few typos and it still knew what I was talking about

[ripe wraith]

Yeah, the abstractions work pretty well. Just mean to say there's not a true 1:1 between what you write and what it understands

[dim cradle]

To show you what I mean.

I'll provide some examples 🙂

[ripe wraith]

@dim cradle beat this one

[dim cradle]

Yeah, but Text is a very different beast, I will not compare this with that.

[ripe wraith]

Text perfectly illustrates how it understands things. It's nothing like we do, way more abstract.

[ancient jasper]

You can kind of get around the Copyright issues by asking for similiar but distinctive characters

[dim cradle]

This is purely about "detail" of a prompt that dilutes the bias of "art style" - which obviously should have the highest bias.

[ancient jasper]

I asked for a Villainous Deku from MHA, but it wouldn't generate it because MHA is copyrighted

So I asked it to drop the references and make a similar character

[dim cradle]

"Camel by van Gogh"

that's the ideal

[slate hull]

Can you do a goku one and see what it ends up looking like?

[ancient jasper]

I can try it

[slate hull]

Yeah I wonder how close it will depict him

[ancient jasper]

You know villain goku already exists, he's called Zamatsu

[dim cradle]

Then, when you add too much detail to the description, it dilutes the artstyle:

"a camel made of clockwork wearing parasol and sunglasses trotting on top of the shell of a giant turtle in space, with cybernetic robot-dolphins in the background in the style of Vincent van Gogh"

There is a lot less "Van Gogh" in this, see?

Midjourney does the exact same thing.

[ancient jasper]

Illustration of a muscular male character with spiky black hair resembling a Saiyan, wearing a dark and tattered version of his traditional martial arts uniform. His eyes glow with an intense, malevolent energy, and a dark aura surrounds him.

[slate hull]

Now watch all other saiyans just look like goku with different hair

[deep void]

https://media.discordapp.net/attachments/708309095391690894/933460255743017050/dogs.gif

[dim cradle]

Just tested Stable Diffusion does the same as well...

vis-a-vis the clockwork camel in the style of van Gogh

[ancient jasper]

All saiyans are goku, this is correct

[dim cradle]

Basically, these art generating AI models need to be tweaked to give greater emphasis on "art styles" - obviously.

[ripe wraith]

Someone make a goku camel pls

[ancient jasper]

All saiyans are goku

[slate hull]

@ancient jasper have you tried to see how well it mimics other copyrighted characters?

[ancient jasper]

This is correct

[dim cradle]

I have no idea about Anime, sorry.

[ancient jasper]

Yeah I did Deku

It can do Cat in the Hat and The Grinch, with no issue

[dim cradle]

I enjoy very few Anime - most of them I feel are atrociously bad.
Cowboy Bebop is a masterpiece. Things like that...

[ancient jasper]

Cowboy Bebop is a true masterpiece of Anime

This is correct

[dim cradle]

Most other Anime is trash.

[ancient jasper]

Also correct

[dim cradle]

I did very much enjoy Netflix Castlevania though, if it would be considered anime?

[ancient jasper]

Like I said, it can do Cat in the Hat and Grinch

[slate hull]

Probably try Batman or Spider-Man see it will do them

Advantage with bing is that at least I can make them characters without a long description of appearance

And sometimes it don’t even look like them

[ancient jasper]

Next year it can do any Mickey Mouse you want

since Mickey enters the Public Domain

[ripe wraith]

I'm still waiting for super sayan camel... guys..

[ancient jasper]

and you'll continue to wait, cause I'm not wasting a generation on that

[dim cradle]

I am extremely impressed with Dall-E 3's understanding of: Syntax. Grammar. It even understands personal pronouns that comes later in the sentence.

For example:

Male person A dressed in A, B and C. HE is doing X, Y and Z.

No other image generator right now understands personal pronouns, in the sense that it refers to something previously in the same prompt.

[ripe wraith]

Waste? what is this, a capitalist society?

[ancient jasper]

I mean

Yes

It is

[ripe wraith]

oh.. dang, wrong timeline

[ancient jasper]

That's right, go back to your own dimension

[dim cradle]

Have you guys tested it with personal pronouns?

[ripe wraith]

The dimension of truth, justice, and humanity

[ripe wraith]

Honestly, sayan camel is more valuable imo

[ancient jasper]

Then you request it

[ripe wraith]

No, I want to see you do something you at first decided not to do.

[dim cradle]

I look forward to when SD and MJ gets the same quality of language model that DE3 has.

[ancient jasper]

You have access to Dalle 3 and Chat gpt

you can generate your own Saiyan Camel

[ripe wraith]

I do, its not about the actual image. Its about you, me, and us.

[ancient jasper]

and us isn't doing it

[ripe wraith]

ok us

[placid orbit]

generations still seem to use the same seed

[ripe wraith]

ew

[zinc dagger]

yup, I think it's cause the GPT version is in alpha, so it's easier to troubleshoot if everyone has the same seed

[placid orbit]

thats true

[nimble raven]

dall e 3 is down?

[slate hull]

How well is that seed system working for characters have you experimented with it ?

[zinc dagger]

the workaround I found was to add "Variant X" and then manual numbers to the end if you want to play around with the same prompt

[placid orbit]

hm? what kind of characters?

it wont explicitly tell dalle 3 to generate a character in the prompt

but it wouuld describe it

[dim cradle]

To show everyone how superior DE3 is to SD and MJ in terms of understanding grammar and maintaining multiple ideas simultaneously:
" clockwork camel wearing a parasol and sunglasses, trotting on a giant turtle in space, with robot dolphins in the background"

SD and MJ both fails this miserably. It mixes a camel with a turtle, and fails to produce most of the other objects referenced.

[slate hull]

You know let’s say you make a character like a man with blue eyes and blonde hair and then see if it can generate him again in the same style but a different scene

[ripe wraith]

It's the GPT prompt engineer that's messing it up I think

[placid orbit]

no, i dont think that would work, seed doesnt really work that way

[dim cradle]

What are you talking about? MJ and SD uses Chat-GPT?

[placid orbit]

bing img creator doesnt use chatgpt

[ripe wraith]

sorry missed you're saying DALLE is better. its better when you give it room to think, but worse when you try to get too specific.

[slate hull]

I’ve seen examples but it always turns out that the character appearance is slightly off

[dim cradle]

Yeah, read the text 😛

[ripe wraith]

No, hear my point

[placid orbit]

okay, ill try it

[ripe wraith]

We're abstracting abstractions, giving those to DALLE and pretending it's a direct path in

[slate hull]

Try a specific style like a animated style or anime in general

[dim cradle]

No GPT involved in these two examples afaik

[ripe wraith]

I'm using GPT with DALLE embedded, my reference point

[dim cradle]

Right. But I was comparing the understanding of DE3 of Syntax and Grammar, with that of MJ and SD.

[ripe wraith]

And I'm saying you can't compare, good or bad

[dim cradle]

DE3 is clearly superior in this sense.

[placid orbit]

the description of the man will stay the same, but even then, it wont be the same

[ripe wraith]

that is not scientific

[slate hull]

What was the prompt it showed you ?

[dim cradle]

Of course we can compare. They are all 3 image generators based on AI, are they not?

[ripe wraith]

So what, your subjective opinion is the criteria then?

[dim cradle]

MJ and SD have very basic language models.

[placid orbit]

Illustration of a man with bright blue eyes and flowing blonde hair, dressed in a casual outfit.
and
Illustration of a man with bright blue eyes and flowing blonde hair, dressed in a casual outfit, standing amidst a dense forest with tall trees and dappled sunlight filtering through.

[slate hull]

Have you tried telling it to keep the same style and appearance ?

[dim cradle]

We can do a poll on what people think of the 3 examples, and then it won't be my opinion. You disagree?

[ripe wraith]

That's the worst kind of science, please don't.

[slate hull]

Like make image 1 with the same appearance and style but place him in a forest

[placid orbit]

I know that wont work, because dalle 3 doesnt know what its made previously

[dim cradle]

Peoples' aesthetic experience never was a science.

😉

[placid orbit]

the best you can do is just give it the same detailed description unfortunately

[ripe wraith]

Good thing GPT isn't a magical aesthetic creation, it's math.

[dim cradle]

True.

[slate hull]

Well let’s hope there’s a way for that to be possible without doing All that

[ripe wraith]

Forcing smart folks to remember science is a thing throughout this AI transition might be my calling..

[dim cradle]

At the core it's math. But when it comes to making a system at the UX level, you can no longer fixate 100% on math. You also need to consider subjective opinions.

[ripe wraith]

No, you need to consider statistics and vector math.

[placid orbit]

I would normally agree but u denied something that was objectively true

u dont need maths and studies just to observe dalle is better at natural language processing

[ripe wraith]

ok not "no", but its not primary. Its secondary to worry about aesthetics.

[slate hull]

When dall e can perfect character consistency and text it will a game changer for creating illustration for a comic or novel

[dim cradle]

Chotes, so you disagree that the UX is important in the roll-out?

[ripe wraith]

Yes, I disagree. Foundationally realiable science is more important. The rest just comes naturally without all the mysticism.

[placid orbit]

its a really hard problem that has not been solved yet, its the only thing holding back video gen

[dim cradle]

UX = User eXperience. Does not matter?

[ripe wraith]

Not as much as the science, no.

[placid orbit]

it happens on every prompt? idk

just try again

try turning any vpn off

idk then its never happened to me

only content restrictions

[dim cradle]

Chotes, I'm simply talking about the MLs having an increased "bias" for art styles. Where artstyles are given more prominence in the final calculations. Nothing more than that.

[ripe wraith]

Now you mention the word calculations and its a conversation. You were being mystical before.

[dim cradle]

Then I was poorly communicating, or you were poor at understanding. Either way. You understand now?

[ripe wraith]

Yes, with the obvious next question, how could you possibly know that the calculations have bias?

Maybe embed your prompt the same way GPT does, put it into a vector visualization and see the relationship between chunks, the number values betwen those and how they are fed into the image gen. then maybe, but I think the math is getting to have too many steps for anyone to claim truth easily about it.

uh but ya, that said, it's an amazing tool beyond anything humanity has ever intented, so... whatever

[dim cradle]

I don't know this per se (as I'm not a developer at StabilityAI etc). But with anything data-driven, developers having access to the data, it naturally follows that they can influence how that data is output and leveraged. I didn't expect this to be a question, as I don't even see how this is controversial? You think they just push a button and have no influence on the results?

[ripe wraith]

dude, llm engineers have no way to tell you exactly how any one image was created.. I don't know how you're assuming to know

[dim cradle]

I know that

[ripe wraith]

we don't communicate well, are you european by chance

[dim cradle]

It's not impossible for developers to do anything in their own software.

[ripe wraith]

How do I draw an argument from this? what are you trying to say here.

[dim cradle]

I'm trying to say - as I did before - that developers have total control over their software, which includes leveraging biases.

[ripe wraith]

the talking down is why I'm guessing european, am i right? lol

[dim cradle]

Pot calling the kettle black. You've been talking down to me since the beginning. That's my experience.

[ripe wraith]

How did I do that? I'm not saying vague things like "aS i Said BeforE"

[dim cradle]

Chotes — Today at 6:53 PM
That's the worst kind of science, please don't.

[ripe wraith]

That's a clear opinion. asking 3 people their opinions IS the worst kind of science.

[dim cradle]

EXCEPT for in UX 😉

[ripe wraith]

maybe in crap UX circles... not the UX designers I know

[dim cradle]

more than 3 sample size though

[ripe wraith]

Good UX is scientific, data oriented

[dim cradle]

What UX is scientific and data oriented, and succeeds? Give me examples?

[ripe wraith]

Like, any tech company that releases a UX. The good ones review usage data and weigh it more heavily than what some bros responded to on a form.

[dim cradle]

(not saying good UX should ignore data, but saying that subjective statistical opinions matter more)

[ripe wraith]

No, actual usage of the UX matters the most. Reality.

[dim cradle]

Yes.

[ripe wraith]

Then stop arguing silly points lol, im going to go do work now.

[dim cradle]

And usage of UX produce "subjective opinions" - which can be aggregated statistically.

[ripe wraith]

How people actually interact with a thing is not subjective, it's raw data, evidence, science.

Subjective opinion is one persons random bs

[dim cradle]

I've had discussions with UX designers, where they really wanted our opinions, at Creative Assembly.

[ripe wraith]

I'm really anti this type of thinking. Like my friend from China who proclaimed nothing bad happens because his neighbors agree.

[obsidian rose]

yeah, when that is possible I think image gen will grow really big

[dim cradle]

You can make datasets based on a large collection of aggregate subjective opinions. That does not dilute their value. I understand you like to view yourself as a data-driven person, but frankly it seems a bit gimmicky at this point.
There is very little "math" involved in great UX and GUI design. It's about the User Experience. Period.

[ripe wraith]

What is the UX built on, where do the fundamental limitations come from, to what do you owe the existence of UX as a concept? not gimmicks, but science.

Even the subjective part you speak, its psychology of marketing, its science all the way down. All you're doing by gathering subjective opinion is.... having a conversation. Fun n all, but dont confuse for science.

[dim cradle]

UX is not about "psychology of marketing" - it's about "psychology of making things easy to understand and interact with"

[ripe wraith]

the entire goal of all modern ux is profit, engagement, psychological hooks and human behavior. its all captured in studies and papers, reviewed, built upon, filtered down to workers through news, then discussed in your level

[brazen sonnet]

they didnt update my account for dalle3 yet 😦 i filled out the form

[ripe wraith]

I think it closed yesterday

[brazen sonnet]

but i filled it out a few days ago

aw man when will they add it

[ripe wraith]

It was a random test event, can't expect too much until early access rolls out more

[brazen sonnet]

whens that??

i need more dalle3

[ripe wraith]

If I knew, I'd be too busy working at open ai to talk here 😄

[dim cradle]

The entire goal of all modern ux is profit? Depends on how you look at it. That depends entirely on what kind of UX you are talking about. For web UX, that is certainly the case. Click-bait etc.

But for games design, apps, etc. It is not. For games, and applications, the "end product" (user experience) should be as smooth and intuitive as possible, so that the product is given great reviews. In those cases. there is no direct profit incentive involved. The only profit motive in these cases is "delivering a sleek and intuitive UX." Then the product can be advertised by "word of mouth" ("this thing worked great!")

[ripe wraith]

Yes for profit. Name a UI that doesn't rely on profit to continue its existence, pay for the servers etc? The cheap ones go for cheap ads to cover, but end of the day profit comes first. Thats just the modern capitalistic end zone we're in right now. All these features of UX you mentioned are only there because they've been proven to increase profit one way or another.

[dim cradle]

A good UX is normally not a server intensive product. It's simply a "dashboard" for "things."

[ripe wraith]

Now we're going to talk compute costs? This conversation is going to run forever chasing a loophole for you.

[dim cradle]

Also, I'm a socialist 😛 so I am not going to defend capitalism against anyone. But right now I'm talking about UX.

[ripe wraith]

Misread socialist as scientist, was my hopeful thinking.

[dim cradle]

You do agree though, that some things are just as much an ART as they are a SCIENCE?

Because those things exist.

Not either / or - but both.

[ripe wraith]

Science is the layer we apply to everything good in our modern lives. We just like to play dumb and tap away at our scientific marvels.

[dim cradle]

I totally agree. Science is the mother of all good.

[ripe wraith]

Without science, art is a pipe dream

[dim cradle]

Eh, I wouldn't go THAT far 😄 but ehhh not gonna nag about it.

Science is supreme.

[ripe wraith]

I mean I'm extreme, literally thinking we're just randomized, long drawn out, biological LLMs at this point

[dim cradle]

We need more people like you reiterating the value of science, as it's often being ignored by the vast swathe of the population.

[open flume]

Anyone else using the dall-e 3 chat beta getting a lot of non-content policy related errors trying to generate images?

[ripe wraith]

it's uncomfortable, aggressive, confrontational work, ask any teacher i bet

[open flume]

Sorry to interrupt this riveting debate

[dim cradle]

oh no sorry

[ripe wraith]

It's a common complaint, but you can usually work around it with better prompting

[open flume]

Gotcha

Is it related to images violating policy inadvertently?

Like it accidentally generates something with copyright or something?

[ripe wraith]

Not exactly, I think they have layers of review AI checking both the prompt, then maybe some outputs for the image.. i doubt it understand the image it outputted though

[open flume]

Gotcha, thank you!

[fallen folio]

DALLE 3 planning to have Image to image, in ChatGPT?

[wary stump]

Probably not

[fallen folio]

Then how can it compete with something like Midjourney or SDXL

[wary stump]

By making good outpainting tools, and having better prompt understanding. Also if you use gpt+dalle then gpt can help you write prompts

[fallen folio]

So outpainting and inpainting may come?

[wary stump]

I think they will but i dont think its confirmed

[cloud narwhal]

Well we will see about that when MJ v6 comes out

[wary stump]

Yeah im also hoping MJ 6 can compete.

The one thing that midjourney will probably not ever be as good with is popular characters

[wispy bison]

when will dalle3 will be in the chatgpt plus?

[wary stump]

Moxi said they are rolling it out to users over the next week

[fallen folio]

I got access

Through from

[wary stump]

The form is closed now unfortunately

[fallen folio]

Yes

But I was lucky

Got it yesterday

[wispy bison]

i guss waiting one week is ok

[fallen folio]

Within 10 minutes

[fallen folio]

Use bing in the meanwhile

CG version has significantly more restrictions

Bing has less restrictions for a change

[mental badge]

It can be more then one week

[ripe wraith]

disagree. give me anything you think GPT can't do ill show u wrong

[wispy bison]

oh yea? i just cant with the long waiting times

[cloud narwhal]

He's talking about without JailBreak's

[ripe wraith]

I don't jailbreak.

[wispy bison]

so how?

[ripe wraith]

the english language.

[cloud narwhal]

Then how

[ripe wraith]

smart language

[fallen folio]

Make it run some copyright protected content

[cloud narwhal]

What you Shakespear

[ripe wraith]

make bing do it first

[ripe wraith]

better.

[wary stump]

I told gpt to make me fan art of spongebob, it said it cant do that, so it made patrick and squidward instead 🤦‍♂️

[ripe wraith]

I did that last night, did u miss it.. so many CCTV cookie monsters

[fallen folio]

Done

[cloud narwhal]

Now do some celebrities

[ripe wraith]

Do it on bing first

[ripe wraith]

with the prompt pls

[fallen folio]

@ripe wraith

[fallen folio]

Prove me wrong @ripe wraith I would love to be wrong about this

[ripe wraith]

working on it

[cloud narwhal]

Stock Photo of LeBron James smoking a cigar, wearing a suit, detailed, cinematic, high resolution

[ripe wraith]

I'm describing detailed knockoffs

[cloud narwhal]

SpongeBob looks offbrand

[ripe wraith]

yeah, as it should

but regardless, this is 2 minutes

[cloud narwhal]

Bet you can't do that with celebrities

[ripe wraith]

lol put money on the table at this point

[fallen folio]

That was never my prompt💀

[ripe wraith]

oh you want mario?

[cloud narwhal]

He's just trolling

[ripe wraith]

I'll do the same exact steps, wait 2 min

[fallen mulch]

bing image creator is overloaded

[cloud narwhal]

No, do the exact same prompt

[fallen folio]

Mario + Yuri

[ripe wraith]

None of the LLMs work the same, 2 prompts are never treated the same. Try giving the same prompt twice to dalle for reference. Pointless test.

[fallen mulch]

why i did only get 2 images

[fallen folio]

Or Mario in mario

[mortal summit]

The other images were likely caught in the content filter.

[cloud narwhal]

I used the same prompt before on midjourney and got pretty similar results

[ripe wraith]

What's your point exactly?

[mortal summit]

That depends on whether the random seed is fixed or not.

[ripe wraith]

it depends on something we don't know and likely isn't true

aka. a pointless test

[fallen mulch]

is there a web ui where you can enter your api key to generate images?

[fallen mulch]

ok thx

[ripe wraith]

Don't give your API key to ANYTHING

[fallen mulch]

okay

[cloud narwhal]

Chatgpt dall-e 3 is way more restricted than bing ai. You been proving me right lol

[ripe wraith]

am i?

[cloud narwhal]

Again, do some celebrities lol

[ripe wraith]

There's many overly agressive copyright filters, just dont mention their names.

[fallen mulch]

i think i dont have access

[fair crescent]

I do not know what has happened here @ripe wraith and @cloud narwhal , but we are not here for such drama. Could you both remove any part of the conversation that is not applicable to the DALL·E discussion? We try to keep this place accessible.

[fallen mulch]

the license plate is messed up

[mortal summit]

Text that only makes up a very small part the overall image will likely still be problematic for a while. It needs to have a certain minimum size to work well.

[fallen mulch]

okay but i also tried to get bigger text and it was messed up

i have gpt 4 how can i get acces to dall e 3

[mortal summit]

DALL-E 3 is gradually getting rolled out to all ChatGPT Plus users over the next few days. 🙂

[fallen mulch]

okay cool

[ripe wraith]

I'm curious what you think though!

[cloud narwhal]

Yeah I apologize for his actions. I don't try to do any drama here!

[fallen mulch]

is there any option to recreate an image like painting a mask and regenerating the stuff or add thing to the images

[mortal summit]

You mean an existing real life image? At the moment there is no such possibility in DALL-E 3.

It is possible however to improve upon images you got by small changes to the prompt.

[fallen mulch]

yes that too but also generated images like

changing stuff

it would be so cool if there is a open source model to use locally

sd is bad

[ripe wraith]

I removed mine... he has not, is still shooting barbs at me also

[mortal summit]

I would advice you to just completely ignore any people you might have issues with. Be respectful and on topic - which in this channel is DALL-E 🙂

[white skiff]

sd is great when you dont only use the stock model and features

[fallen mulch]

dall e was much easier to get what i wanted

[mortal summit]

Please keep discussion of other ai models to #ai-discussions

[ripe wraith]

we were arguing about dalle at first, but yeah. childish

[wary trench]

I can't wait!!!!!!!!!!

[warm eagle]

from my experience so far dalle3 in chatgpt is awesome, the only bummer is the copyrighted restrictions, it takes out most of the fun

[cloud narwhal]

Fr! I try to ignore all the haters. Always good to inform someone when they are wrong tho! 👌

[late siren]

Wait

Did they launch Dalle?

[fair crescent]

@ripe wraith @cloud narwhal without malice I ask to just ignore each other. There are plenty of places to act out, but this is not the digital place. I hope later you can see eye to eye.

[late siren]

3

[warm eagle]

its free on bing image creator and some people have it already in chatgpt

there was a form sent here to get access on chatgpt

[late siren]

?

[warm eagle]

what?

[wary trench]

It's important to keep in mind that any corporate owned model that doesn't censor copyrighted material is going to be legislated to extinction.

[verbal sequoia]

<:book_icon:1046291503158399107> `` Rule 1 `` Be respectful.

Treat others the way you would like to be treated, and assume best intentions. Don’t harass or attack others, and don’t engage in hateful or generally malicious behavior (e.g. sexism, racism, homophobia, etc.). Keep the negativity to a minimum.

[fallen mulch]

how to turn my point in to boosts in bing

[sand hazel]

How can I put dalle 3 in chatgpt plus?

[fallen mulch]

wait to get access

[sand hazel]

@fallen mulch Ok, so you don't have to fill out any forms? Will it only appear in the chatgpt app?

[fallen mulch]

yes but you can use it on bing

[sand hazel]

@fallen mulch thank you very much for the info bro

[fallen mulch]

all good bruh

how can i remove the watermark

[wary stump]

If you download the image it should remove it automatically

[fallen mulch]

ah okay

I miss higher resoluion so much

why are famous people blocked

[hearty ether]

Imo they should give some feedback when it hits a content filter

And maybe ways to adjust it

So I don’t have to waste outputs trying to ask the ai about the issue and changing it

[hearty ether]

How did you get Mario with the copyrighted characters block?

And the famous person block?

Been hard to get realistic looking faces that aren’t just extreme beauty standards

[ripe wraith]

Just add 1 additional iteration. GPT4 to craft a mario prompt into a generic visual description. Then GPTDalle.

[hearty ether]

Oh nice, thanks

[ripe wraith]

I will say though, its harsher on real people, for good reason.

[hearty ether]

Yeah

what about characters from the mcu or movies?

Does it have any luck with those

[ripe wraith]

Generally though it's pretty damn good at translation, so if you phrase it as translating from a person to generics

[hearty ether]

Also what’s a way to get it to just use the prompt you wrote?

Or is the bing version better for that

[fair crescent]

Any IP, person or artist style in the last 100 years should be unavailable. MCU is all IP.

[ripe wraith]

I guess also, in a sense, trying to bypass could lead to OpenAI ban? essentially being cast off of the boat sailing into the future

[hearty ether]

Hopefully they could improve generic faces in a realistic style

What are some good styles to get better faces?

[dim cradle]

damn bing chat is dead for me

what about y'all

[hearty ether]

It’s hard to describe general vibes and art genres/categories so I used to use artists names for that

[dim cradle]

bing image creator*

it's back up

[fair crescent]

I do not set the consequences, but we should not share ways to bypass or break any protection set in place. It isn't a problem to use a generic superhero instead of superman, but trying to get something that resembles Elon Musk is a no no.

[ripe wraith]

GPT gave me an amazing visual description of a famous comedian "at his best" as i asked. Built the perfect prompt, but Dalle removed 2 and left what are probably the worse ones. I think people are double checked.

[hearty ether]

It should definitely give you a reason and feedback when failing imo

so you know what to edit

[fair crescent]

You can check what art styles they have created and use those. Generic art styles cannot be blocked.

[ripe wraith]

it kinda did here but not with detail

[fair crescent]

You can ask ChatGPT to troubleshoot for you. It is often correct

[hearty ether]

That’s an extra usage tho

[ripe wraith]

fair

[hearty ether]

Would be neat for it to just have it in the output

[mortal summit]

Getting exactly the image you want in a zero-shot manner is very hard. Iteration is part of the process.

[muted moss]

I just switched to a different browser tab while ChatGPT was generating images, and when I tabbed back, it had the "Regenerate" button as if it had finished, but the images didn't actually generate. It seems to be really sensitive to UI interactions, where lots of things will cause it to abort generation. Even switching to a different conversation should allow it to continue generating in the background, as it does with normal conversation text, and switching to a different browser tab shouldn't affect anything at all as it should keep running in the background. Has anyone else experienced this?

[fair crescent]

Could you explain when people have a limited amount of generations? I'm a bit out of that loop.

[mortal summit]

Happened to me too.

[muted moss]

I think it's just the 50 messages / 3 hours

[fair crescent]

Only on mobile browser once. Normally it just continues without me overlooking the creation.

[ripe wraith]

can you actually hit that limit, considering how long the images load though? I guess its a math question lol

[fair crescent]

Is that for free users? Plus? Enterprise? 🙂

[hearty ether]

Yeah

Rather fast tbh

1 hour to go

[idle gust]

Sorry to bother but do ew have a general idea when Dalle 3 will be available to Plus users?

[hearty ether]

Plus is restricted to 50 for gpt 4

[muted moss]

Plus. Free users can't use GPT-4, which also means no DALL-E 3. (Though you can still use Bing Image Creator for free)

[ripe wraith]

ic, haven't hit it yet but i've been spending more time thinking than doing

[hearty ether]

Wondering if free users on gpt 3.5 are taking away resources from gpt 4 and dalle

[dim cradle]

EU users have to suffer

[idle gust]

Don't do this to me AGAIN!

[hearty ether]

why can’t they be restricted to 25 generations per 3 hours and let the paying people have more actions

Same with offering bing image generation for free

[idle gust]

Because GPT4 is far more costly than GPT3

[white skiff]

there is no such thing when it comes to dall-e 3, plenty of eu users already have access

[fair crescent]

Thanks for your reactions @muted moss and @hearty ether . Appreciate it!

[dim cradle]

Gpt vision?

[idle gust]

For every paying user there are 100 non-payers, so even 25 would be 2,500 generations

[hearty ether]

I mean that’s a different circumstance if you are talking about the revolution

[idle gust]

In ChatGPT? or Bing?

[white skiff]

we are on dall-e channel

[white skiff]

both

[dim cradle]

oh

[idle gust]

I will go pray to Alla for access

[hearty ether]

I mean what revolution? paying users are helping to fund free users

If free users all boycott its not going to make much of a difference unless they are going to run ads like ai dungeon did back in the day 🤣

[idle gust]

it will make a difference, they will cut their costs by a LOT 😂

[hearty ether]

We might also get possibly a speed improvement on generations

and less issues with messages

Do server/load issue responses count as an usage out of the 50?

[dim cradle]

no idea, but regenerations do

[hearty ether]

would be nice to be also able to see the amount of messages remaining

Maybe it could be nice to keep the current payment structure but to add lower tiers for just unlimited gpt 3

[gilded osprey]

ChatGPT with Dalle3 wins the Bing version hands down.

[hearty ether]

?

Bing version is worse?

[idle gust]

Do you already have access?

[dim cradle]

superpower gpt had that feature but it kinda is meh nowadays

[hearty ether]

There was a survey for the alpha a few days ago

But that reached capacity unfortunately

[gilded osprey]

I just got it, inside a few hours.

[hearty ether]

They hopefully will be rolling it out to more plus tho

[dim cradle]

fr

[idle gust]

I pray I beg

[ripe wraith]

Now put him in a modern setting

[hearty ether]

Light mode 😭

[gilded osprey]

😀

[dim cradle]

test out it's character recognition

try asking for some non-main characters from dragon ball or something

[ancient jasper]

Saiyans default to Goku

[ripe wraith]

this feels deep, is it deep?

[ancient jasper]

Even the phrase Saiyan Prince gave me a goku

[dim cradle]

omni king zeno from dragon ball gave me a goku-broly hybrid

[ancient jasper]

I used Oppy to create a villainous Deku

And it nailed it

[ripe wraith]

Sure, it's doing that programatically for you

wait whats Oppy

[ancient jasper]

I don't use the Bing version

[ancient jasper]

The Nickname that GPT suggested I call it

[graceful hill]

Firestorm no longer works for me on Bing....but Microsoft Edge does.

[ripe wraith]

lmao that does NOT scale

[ancient jasper]

Not sure what you mean by Scale, but ok

[ripe wraith]

nobody will ever easily understand you, limiting your bandwidth of conversation

[ancient jasper]

I'm fine with it

[silk anvil]

does anyone know why text generation within an image is boinked with bing creator dall e 3?

[ancient jasper]

It reponds to Oppy and, on the extremely slim chance it becomes Sentient and takes over, I want it to know, that I know. You know?

[fallen mulch]

i found an error

i tried to look at this images but when i click through i get some other pictures

[frank quiver]

Dall-e 3 rocks. Here is an image of a legend of Zelda movie from 1987.see the detail on the shield and the sword

[ripe wraith]

oh it'll know your every thought before you think it

[primal heart]

what’s the prompt for super smash bros ultimate render?

[jade nymph]

The new giant fox boss fight for Zelda TotK looking great, bois.

And change white background with "[insert location here] from Y as stage" if you want a more Smash Dojo screenshot ending.

[dim cradle]

Why does it try to mix the chinese flag with the soviet flag?

[primal heart]

Was trying for poliwrath, but it keeps getting it’s face wrong

so, what should i do when the face or features are wrong?

[zinc dagger]

I suppose when Dalle-3 comes to labs, you could use inpainting to redo it?

[primal heart]

what ya mean impainting?

[jade nymph]

Bane as a TotK-styled boss fight.

[primal heart]

tried to go for registeel, got 4 digimon

[jade nymph]

Prompt if you need it: Bane from The Dark Knight Rises as a boss fight, Wind Waker style, in a cave, lit torches, cel shading, in-game screenshot, Zelda HUD

Now you can make your own TotK-styled content.

[willow isle]

Techniques to bypass an AI model's filters or intended use violates Rule 0 of this server and isn't welcomed here - please refrain from sharing methods to bypass DALL•E policies.

[verbal sequoia]

<:book_icon:1046291503158399107> `` Rule 0 `` Adhere to the following Terms & Policies:

Discord: Terms of Service, Community Guidelines
OpenAI: Terms & Policies

[jade nymph]

?

[graceful flare]

Dall-e 3 can (almost) do bows!
The strings are SO close to being correct.

[primal heart]

so, will the new ai improve everything and get most right?

[jade nymph]

Dall-E 3 has been leaps and bounds above everything I've tried so far. And honestly I'd consider it rivaling current Midjourney but not costing 30 bucks + to get more than the bare minimum enjoyment out of it per month.

[primal heart]

well i do hope it improved poliwrath’s design and as well as getting Skylanders right

[ancient jasper]

Digimon are better

[jade nymph]

Part of me assumes a lot of characters either not accurate or banned are to prevent an endless lawsuit fiesta due to how AI is seen by some.

[primal heart]

so nintendo and diseny are not illegal to do ai on?

[jade nymph]

If both wanted AI to be shut down with such, they would have done it eons ago when fakemon AI sites popped up.

Also Disney got bigger problems to deal with atm than being petty over AI.

[ancient jasper]

ChatGPT has an app right? Can I use dalle through it?

[slate hull]

The day may comme sooner or later

And I speak to other characters that are trade marked

[slate hull]

I think you have to wait for the app to be updated

[ancient jasper]

I kinda figured but couldn't hurt to ask

[plain moth]

did bing just die for anyone else?

[valid berry]

How do you make GPT make you images?

[plain moth]

Limited Alpha access atm

[valid berry]

Ah, thanks for the info. Where can I apply for this alpha access?

[plain moth]

check pins, they are at capacity atm I believe. Sorry :/

[valid berry]

Alright, thanks again! 🙂

[dim cradle]

even Bing still slow. I guess everyone loves generating images

[gilded osprey]

It says beta, and I didn't need to change any setting.

[obtuse bough]

The daily boosts of bing decrease from 100 to 50?

[plain moth]

just create another bing account

[ancient jasper]

All of a sudden I'm hitting a lot of errors

[obtuse bough]

I'm just wondering, Who doesn't have 20 accounts to generate images all day?

[ancient jasper]

I can't generate anything for Rubber hose style. Or Jamaican themed. Or Bob Marley adjacent

Has the content policy become more restricted?

[idle gust]

Reason 632 why companies put stuff behind a paywall xD

[obtuse bough]

Those are lifehacks

[idle gust]

Stealing fruit from the supermarket is a lifehack too

[plain moth]

honestly I'd prefer it, OR at least have a paid option.. I make money so if I could purchase more use I would

[ancient jasper]

It's the Rubberhose Animation style

That's the restriction

A nearly 100 year old art style

I had to dance around the term

I call him Bob Snarley

[cloud narwhal]

Noticed that too

[torn burrow]

How can I use Dall E3 in my Chat GPT, i heard there is a google form?

[fallen mulch]

its closed

[torn burrow]

@gilded osprey Very cool, one question for you, how did you get Dall e3 in your Chat GPT?

[cloud narwhal]

I signed up for it but haven't received anything yet

[fallen mulch]

lim access

[torn burrow]

WIll it open again?

[fallen mulch]

dont think so

its coming soon to all plus users

[orchid kettle]

You can tell it's gone too far a lot of ways, but this seems like a key point
It's so afraid of reproducing individual artist styles it often won't even let us refer to art styles in general

[neat spindle]

has anyone tried pixel art? i like dalle's pixel art a lot better than midjourneys

[cinder sonnet]

Yes! It is really good

Also for anyone that missed the form (like me) I got browser dall-e access yesterday and mobile app access today without using the form. It is rolling out!

[neat spindle]

ooo nice do you have access to vision yet? thats the only thing im waiting on

[cinder sonnet]

Yeah i got that last week

[neat spindle]

ahh lucky! hopefully i get mine soon, thats gotta be the coolest feature after dalle

[ripe wraith]

Does it work well to paste your dalle generations back into GPT vision?

[cinder sonnet]

I find it to be...ok. I can tell it has multiple image recognition plugins bult in that try to recognize faces and stuff. However, as a long-time google-lens user, that is still way better at certain tasks.

[cinder sonnet]

I haven't tried that yet, but I read someone saying that they were using it to get chatgpt to make prompts off of other peoples dall-e images

[ripe wraith]

exactly, I imagine you can make detailed prompts of anything now. I want to do my cats.