images-discussions

[onyx ridge]

I've noticed that the reference id format has changed. I'm starting to suspect it now embeds the original seed.

[fallow ore]

Changed in what way?

[onyx ridge]

Many more characters and alphanumeric instead of integer

[sick dagger]

BUMP 👀

[fallow ore]

Krea AI? I'm not sure what this means

[onyx ridge]

Same

[sick dagger]

from this thread.

[onyx ridge]

Looks like not OpenAI stuff to me

[sick dagger]

ah

[onyx ridge]

For upscale

[sick dagger]

it's an image rehancer externaly.

got it

ya, no thanks... too many steps.

[fallow ore]

I haven't noticed a difference I don't think, will keep an eye out though. I use the gen_id feature a lot

[onyx ridge]

About two pr three weeks ago it changed significantly

I wouldn't have noticed without a GPT requesting every image payload.

You get used to certain patterns like all numerical fields, so when letters start to appear ...

Well, I noticed.

[fallow ore]

They should make the UI like the code interpreter tool so you can see the work being done

[onyx ridge]

Also some transparency

Would be great.

We shouldn't have to pry this open with forensics.

[dim cradle]

it appears gen_ids/referenced_image_ids (being the same thing) are in base 16, hexadecimal. i think that's been true for a long time.

it's in active dev, that's why i declared a moratorium on the topic weeks ago, it's a silly topic when it's so flaky

[vapid elk]

I mean.. you don't need to do that.. the system is made to be as simple to use as possible: just say what you want

[onyx ridge]

I didn't know that there was a moratorium. I was just reporting that things were suddenly different yesterday.

[vapid elk]

the idea is to really minimize as much as possible how much prompt engineering is required

[dim cradle]

i'm always saying these guys are over-thinking lol

[fallow ore]

I've been testing this seed field in the payload. It's only useful in the current session. You can't use seeds from previous chats

[onyx ridge]

The images are in the browser cache

[vapid elk]

so.. if you like an image, you can say "make this exact same image again but change X to Y", and it will probably understand that it should make a request with the same seed and refference, with a slightly modified prompt

[dim cradle]

i'm j/k i know!

[onyx ridge]

Great. I want to put it into explicit commands for users so I need to understand the generalized application.

[vapid elk]

you can't

[fallow ore]

Yeah but it you say this and use seed and referenced_image_ids you should get very good consistency

[vapid elk]

this is not a ffeature on the API

[onyx ridge]

Somebody should inform /recompose

[vapid elk]

I really want seed to be a feature of the public api, it woud be so much more useful for integrated applications

[fallow ore]

Nah, but it seems to be available in ChatGPT

[onyx ridge]

It's working in ChatGPT

[vapid elk]

yea, it is a feature only on chatgpt

[onyx ridge]

And was yesterday too

[vapid elk]

for this reason the idea of "costumers" does not really make much sense like it would on an actual poduct that uses the API

i mean... I hope you are not scraping chatgpt and faking interactions with an automated system to implement those features on an external project 👀

that would be really uncool with the TOS

[onyx ridge]

Lol wut

[fallow ore]

So OpenAI 100% have APIs internally that aren't available to the public. We see this with the Assistants API (used in ChatGPT for GPTs) -- the public API doesn't support streaming of responses, but in ChatGPT we see this happening as tokens are rendered one by one instead of waiting for the entire response to finish

[vapid elk]

more useful than seed tho, inpainting, I really wish they release inpainting with dalle3 like they have on dalle2

[onyx ridge]

I want to know what you meant by providing fake interactions.

How would that work? Isn't that traffic internal only?

[vapid elk]

making a script that uses the chatgpt website for you in order to proxy the AI content to an external system

people who do that are the reason they had to implement those super complex captchas

[onyx ridge]

Okay this makes sense. No, that's not what I do. I make custom GPTs.

Then I test them.

[dim cradle]

Slime balls.

[vapid elk]

well, manipulating the dalle request can be done, but it isn't really an intended use, so it might be harder

the ""normal"" use is "just tell the AI what you want and it will figure it out"

[onyx ridge]

That's where I operate. Tough promptin'

[vapid elk]

it isn't magic tho, some times it will just not figure it out

[dim cradle]

I have questions but they aren’t dall-e related so I’ll save for another thread.

[vapid elk]

and there is a range of "things that the AI does nto figure out" and "things that the AI can only figure out if you do complex prompt engineering"

and that seconds group only gets smaller as the AI and the integrations get better

[dim cradle]

Ai is magical hehe.

[onyx ridge]

General behavioral reproduction with consistency is what I am interested in.

[vapid elk]

this is the catch.. it is not really how it is supposed to be used

you can't expect deterministic behaviour of a system built on RNG

[onyx ridge]

So if it can duplicate an image, for example, I'll create an explicit command that's tied to the instructions to do so via RAG

[vapid elk]

it is built with the pourpose of giving you a different reuslt each time you use it

[onyx ridge]

And yet

[dim cradle]

The use case is a consistent rendering for a narrative.

[vapid elk]

coherent*
not consistent

this is the catch

[onyx ridge]

My GPTs do produce deterministic output

[vapid elk]

the AI goes for coherence not consistence

[dim cradle]

But the seed recalls all those random decisions for increased consistency.. maybe not 100%

[onyx ridge]

So it's consistent due to the PE

[vapid elk]

well.. yea.. seeded generations.. but it isn't much useful in that case to just be able to get the exact same thing twice

[dim cradle]

The idea that might help address the use case

I see

[vapid elk]

instead of wasting GPU compute time, it could have been a copy paste 😆

[onyx ridge]

"It'll figure it out" isn't good enough. We can use output templates with open variables to get what we want.

[onyx ridge]

Yeah what if you want acid breath from the same dragon but DALL-E 3 rendered fire, for example?

Will seed help here, perhaps?

How about for a sequence of images where one has an inconsistency?

Lots of reasons for gen id, reference id, and seed, not just ctrl+c, ctrl+v.

One of my GPTs is literally for visualization of complex processes. For those images, it's ideal if they share as many commonalities as possible!

[vapid elk]

you can try to ask chatgpt to re-make the image to get a similar composition, but you will get a completely different image if you do any drastic changes on the prompt

[onyx ridge]

That's the purpose of requesting the payloads

So that I can control the rewrite

[vapid elk]

lets sey, your prompt asks for "a dragon", and it makes a green dragon
if you prompt agian for "a red dragon", the adition of "red" will make a completely new image that does not have nothing to do with the original one

despite using the same seed

[onyx ridge]

Not if you control the payload, man

[vapid elk]

even if you control the payload

[onyx ridge]

Scroll up

[vapid elk]

getting it once is pure luck

[onyx ridge]

How many times do I need to do this

[vapid elk]

getting it to consistently change only what you want with a prompt that is too different, is very hard

all of the times =P

[onyx ridge]

It only works if the features are on

[onyx ridge]

I can create a series of nearly identical images arbitrarily long though with this. I've tested up to five in a row before running into the penalty box.

[vapid elk]

I know it is possible, it can indeed be kinda consistent

but if you need to do that based on the subjectivity of the prompt, then it can't be relaible

that is what I mean with "you can't"

and this is why I wish that seed, refference and so on.. where features of the API

[onyx ridge]

That's what advanced prompting is for. Then users can just do that

At least in the ChatGPT interface

[vapid elk]

because them with API you don't need to rely on prompting, you can simply programatically set the values of your request

[onyx ridge]

Really you can do that with chat

I promise

[fallow ore]

I'm making a test for this now, same seed and gen_id, different dragons -- let's see

Actually, I can't make this work. It seems like the seed field in a payload will cause an error

Using just gen_id get's very different results

[onyx ridge]

seed isn't enabled today

my payload requests earlier showed this.

sometimes when seed is disabled, passing it still doesn't cause errors. sometimes it does.

the model's in a constant state of flux, and the ability to output similar images in sequence, even according to OpenAI, is not "currently" supported.

That said, to do this you have to provide the same JSON payload, not just the same pre-prompt.

This added complexity invariably causes users inexorable confusion.

I don't blame anyone for the confusion. that pre-prompt/rewrite mechanism may be for our benefit but it also drastically obfuscates the mechanism and algorithm.

[fallow ore]

I struggle to accept this changes so frequently. Will do some tests each day for the next few weeks to confirm

Not saying you're wrong or anything, but I need to see for myself to believe it

[dim cradle]

"I was unable to generate images because DALL·E is currently experiencing high demand."

[fallow ore]

Is that's while trying the seed thing?

[dim cradle]

"hmm, something seems to have gone wrong." -- nah just seeing these errors a lot today

[fallow ore]

Oh yeah, today has been odd, logout and back in -- fixed it for me earlier. For a bit I thought I was blacklisted

[late blade]

today is not an odd day... it's 14th...

I double checked with gpt

[fallow ore]

It's been an odd day with DALL-E, I sever said it was a odd day on the calendar 😅 -- I'm dying lol!!

[late blade]

lol, I just had to, I was soooo tempted to do the pun

[fallow ore]

I don't blame you, touche

[late blade]

I just had a guy explaining UX in medical, 4 hours long. really boring stuff

the guy has ZERO clues on UX for meetings

[dim cradle]

it's an odd day every day in dall-e land

[late blade]

DALL-E Land

a myseriously magial wide image about "Every day is an odd day in DALL-E Land that was my prompt

[grave siren]

Hey everyone, is dalle-3 unlimited generations?

[late blade]

40 within 3 hours., up to 200 a day

[grave siren]

after the 200 it just resets the next day?

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[fallow ore]

Is that different with teams?

[late blade]

yes, 3AM your local time usually

[grave siren]

Thanks!

Do yall prefer it over the other options on the market, to me it seems the most affordable for the most generations.

[fallow ore]

I much prefer yours! I tried your same prompt/theme

[late blade]

for me it's more a relaxing thing, I just do it as mindfulness around the other stuff I do, so dall-e is ok, and actually pretty good, for what I want

it really depends on what you need and want to achieve

[grave siren]

I am trying to make a visual book

[fallow ore]

DALL-E is a nice all rounder, and letting ChatGPT write the prompts makes it much easier!

[grave siren]

typing descriptions of pages of books to give them a scene

[late blade]

that is a cool project

[grave siren]

For example, maze runner book, but an image for every single page.

thanks

[fallow ore]

Yeah, ChatGPT Plus with DALL-E is your best bet. Copy+paste the contet of the page and let it write the prompt for your to summarize it

You can try MidJourney but you'd need to write your own prompts, it gets quite technical

[late blade]

just keep in mind that getting consistent images can be challenging and requires lots of iterative and adaptive work

[grave siren]

Right, makes sese

Can it analyize the art of an image to copy the style for a different promt? also sorry for all the questions theres just so many options for ai

[late blade]

also, a tip I learned on my own, tell the model what it has to show, not what you think it should show, have to do some rethinking sometimes to get the image done

[fallow ore]

Hmmm, yes and no -- it's not perfect but it is possible

[grave siren]

Yeah, I have a good bit of experience with chatgpt and prompting. okay

[late blade]

you can use GPT4V to analyze the images, and up to you to ask what you need

[grave siren]

ahh I see

Welp thanks, Im leaning towards chatgpt 🙂

[late blade]

cool, don't be a stranger and show some cool images when you are around

[grave siren]

Will do!

[late blade]

did you have other information in your chat or custom instructions?

I also used the dall-e custom gpt

[fallow ore]

Oh no, I've got a GPT that I use for everything and then I've got a huge pre-prompt that I pass to her to generate DALL-E images

[late blade]

ah okay, for me it depends on which kind of work I want to do that I select from a few pre-prompts

[fallow ore]

All the images I put in #daily-theme are with the same GPT and pre-prompt. It's set's me up to visually explore a "theme" and then I'm usually just like 'Stunning! Let's keep exploring" or "Yeah, this great! What do you want to show me next?" and then just let it walk me through a visual journey

[late blade]

I hope they expand custom instructions at some point to add working environments or at least some workflows

[fallow ore]

I will eventually make the experience into a GPT

[fallow ore]

Any idea's specifically? I can help you with prompting something dynamic

[late blade]

oh, just like "profiles"

so I don't have to choose pre-prompts on each chat

[fallow ore]

Not quite sure what you mean, but just anything dynamic, like if this then that?

Otherwise you could do a GitHub repo, link the API and let it "fetch" a pre-prompt on demand

[late blade]

you do the same I as, we do pre-prompts for the chat session

image that built in for custom instructions defined by the profile you want to use it

Example: Code Assistant, pre-loads my coding stuff, DALL-E Assistant preloads my image making stuff.

[fallow ore]

Hmmm, we could build something using the python tool (code interpreter) -- this would be easy enough

[late blade]

that for sure

I'm just too lazy lol

well at least today I'm lazy

[fallow ore]

Hmmm, I'll throw something together after work with a few different "personas" that it will fetch on demand, something like /load("dalle") or /load("code")

[late blade]

hehe nice

actually a custom personal gpt for that kind of pre-loading is nice, since you can call it in any chat

[fallow ore]

I think the @ feature is the same, but with multiple GPTs 🤔 I'm not sure if there's any particular upside to have a single GPT that has modular context -- but I'll build it! Should be straight forward

[late blade]

worth testing

[fallow ore]

Oh wait, this opens the door wide open for dynamic context, imagine like /load("dalle-impasto") or load("code-javascript")

[late blade]

yeap

that's the idea

[fallow ore]

Then you can have huge system prompts that only fetch the context relevant to the task

Very interesting. Yeah this is possible for sure!

[late blade]

we'll have to check that out

[fallow ore]

I finish work in a couple hours, will get stuck into this!

[late blade]

cool

[fallow ore]

@dim cradle that current style is really nice! Like a faded polaroid. Looks like it's been in someone's wallet for like 50 years

[late blade]

asked for vintage photo?

dall-e land, but vintage

[fallow ore]

I must admit I've never explored this

[dim cradle]

i thought the distressed photo added another element to represent impermanence.

The distressed photo style seen in the image can be described as having a rich, textured character that imparts a sense of age and a storied past. It carries a patina of time, often characterized by a combination of wear-and-tear features such as color fading, creases, tears, scratches, and spots. This style might also include graininess or noise, contributing to a rustic, antique aesthetic.

There’s an artistic interplay of light and shadow that may not correspond to natural sources, giving the image an ethereal or haunting quality. The color palette often leans towards sepia or monochromatic tones, punctuated by areas of unexpected saturation that seem to bleed through as remnants of the original vibrancy. Blurred or double-exposed sections can suggest movement or the layering of moments, while the surreal elements within the distressed framework suggest a dreamlike or otherworldly narrative. It's a style that engages the viewer's imagination, prompting them to fill in the narrative gaps and question the reality of the depicted scene.

it's interesting you bring up Polaroid because my next request was for one!

[late blade]

if you reference a decade it actually adapts to that time

vintage photo from the 50's or the 80's makes a difference

[fallow ore]

I will give it a go for sure, I love the faded vibes

[late blade]

use carefully, we don't want to fade reality away

[wispy storm]

Making an alien rpg oneshot making everything myself. So o asked dalle make a landscape of the planet

[late blade]

add to your request a good sorbet or sweets. You've been jumping around all channels

[wispy storm]

I'm everywhere

[late blade]

just order it, so you get the energy to do more

I wonder if someone here has Azure OpenAI access to DALL-E

[empty kelp]

please create three perfectly identical dragons with red, green, blue scales (respectively) on a beach in Hawaii

using the GenID and seed of the image the three identical dragons are examining a giant cheeseburger

it’s good to create the different color dragons in the source image, and then add the cheeseburger

[dim cradle]

Great now I’m craving a burger

[late blade]

I had spring rolls

[dim cradle]

Lovely

[late blade]

with spicy sweet sour dip

which tbh wasn't spicy, but the label said it was

[dim cradle]

that burger actually looks the way a burger is supposed to taste

[empty kelp]

it’s 6:15 am here in Kailua, Hawaii and i want a cheeseburger with spring rolls also. pondering whether it’s a good idea

[dim cradle]

but you can't find that often, or it costs $25

[late blade]

it's always hamburger time somwhere in the world, so there's no excuse not to get one

[dim cradle]

i am nowhere near a bistro

soon

[vapid elk]

see, the image has the same concepts, but it is not the same image with a changed part

it didn't "added" one thing

it re made the entire image

it is possible tho to get the same general composition if you change the prompt very slightly keeping the same seed

[dim cradle]

needs to be grass-fed, herbs mixed in, etc., etc.

[vapid elk]

here is an example

just a small change on the prompt, same seed

[empty kelp]

i’ve had luck with creating the characters first on a white background, and then making a sequence of five or six images of them from different angles, and then putting them into the scene

[vapid elk]

yea, that works, when you get a prompt that makes a relaible character, it is possible to get some consistence across different angles

but you have to be lucky

it is not a matter of the AI actually making what you wanted, but rather you settling for a subject that the AI can reliably make

[open trench]

Hello, all.

[empty kelp]

Good morning

[empty kelp]

please create three perfectly identical dragons with red, green, blue scales (respectively)

[open trench]

have they adjusted the penalty box settings or did I just lose track of my use? lol seems like I hit the max really quickly

[empty kelp]

so you start on white like this

[vapid elk]

they are not identical

but this is a thing of the AI

[dim cradle]

bet you saw some errors which i think burns it up

[vapid elk]

AFAIK, it is made to avoid perfect repetition, otherwise images would be more boring

[open trench]

you're right. It kept telling me to slow down or wait a few minutes.

[late blade]

who let @open trench out of his cage? now we gonna get some.... oh no... this is too.... I can't.... I have to hide NAO!!!

[open trench]

I'll just select from what I have for the current theme, as I won't have but five minutes to perfect it once out of the box.

[open trench]

Yeah. they left the key in the lock after feeding. No more cage or captors. 🙂

[late blade]

nice to see you around

[open trench]

thanks. My work schedule has finally normalized. I've been working on projects, but I haven't been sharing, as I've just been exhausted.

[late blade]

one question, what does the global button actually do on the gpt store? I keep seeing the same regardless if it's on or not

[dim cradle]

visualize

[onyx ridge]

Hi there!

[open trench]

Hello!

[dim cradle]

hi hi

[onyx ridge]

Good to see you my friend!

Careful with them rates! 😁

[dim cradle]

ai is telling him to slow down lol

[open trench]

Yeah. AI doesn't know how slow I've been over the last... weeks? lol Today, I apparently made up for it.

[onyx ridge]

I need to slow down too. Maybe I will play some games and make AI art from the screenshots...

[dim cradle]

you can't make a more delicious-looking cheeseburger, don't even try.

[open trench]

lol

I can't try for 18 more minutes

Do previews in custom GPTs count against your rate?

[empty kelp]

please create a hyperrealistic wide photo of three identical athletic female elves on a white background. they are all facing north, looking directly forward, and have the same pose. their clothes are identical with identical color and texture. the elves have red, blue, and green skin respectively. the viewpoint is to the north

[onyx ridge]

Challenge accepted

[empty kelp]

you can bypass the variation by explicitly listing the things that should be identical

[late blade]

I would say yes, otherwise that would be quite the obvious loophole

[vapid elk]

i mean, when I said "they are not identical" i mean, tottally identical except for the color

the AI never makes perfect matches of anything

[open trench]

Makes sense. And I can confirm. I just got fussed at to wait again.

[vapid elk]

if you have 2 of something on the same image, the things will be different

[dim cradle]

i'm still working on my project to generate an image of a paradoxical non-Euclidean pretzel that creates ripples in spacetime.

[late blade]

you back and already causing trouble, tsk tsk tsk tsk

[onyx ridge]

Yes please

[vapid elk]

still, things can be very similar

[open trench]

So it would seem lol

[empty kelp]

for humanoid characters you need to say that they have the same pose, direction, look direction, color, and texture

[vapid elk]

yea, it can make things with aout the same pose, shape and generall looks

but if you go match the face details, they are not the same

zoom in and you will see

the overal body shape is similar, but the fine details do not match

the creatures are not the same creature repeated 3 times, but rather 3 different creatures that look alike

[wispy storm]

I prefer beese churger

[vapid elk]

/draw beesechurger

[late blade]

@wispy storm typing with mouth full again

[vapid elk]

#image-bot message

[fallow ore]

All these images of meat 😂

[dim cradle]

Creatures haha

lol

[wispy storm]

Thats what it did, that's what I expected

[empty kelp]

that’s only because it’s a perspective view with the camera in the middle. if you do an orthographic view it makes them the same if you list what should be identical

[dim cradle]

omg lol

[wispy storm]

What type of burger is it

[vapid elk]

I mean, check the screenshot, the shape of the nose and lips are not the same

[fallow ore]

For those of us who don't eat meat!

[dim cradle]

I presume the protein is under the green stuff

[vapid elk]

what Im just trying to say is that it will not make a perfect match, just that

[empty kelp]

i forgot to add “identical expression”

[vapid elk]

try this: ask it to make a wide, split in 2 halfs, containing <some object here> perfectly identical

split the image and overlay on photoshop

I bet it will not be a perfect match

perfect in shape

[empty kelp]

if you say identical expression, pose, facing direction, look direction, color, and texture the faces are exactly the same

[vapid elk]

I don't consider the perfect match in color because the AI might jsut make a slightly diferent lighting and color

which is fine because it is realistic

[empty kelp]

saying “pure white background” eliminates the ground, shadows, and reflection

[dim cradle]

5 iterations into paradox, and i'm back to an original pretzel... twilight zone music...

the race is on

she's got the look

first

[empty kelp]

I tried to make different views of identical characters for texture mapping almost every day in November and December. that’s why i have these weird ideas about how to duplicate female elves

[worthy heath]

Now add mustard

[late blade]

omg, something just happened!

I just don't know what

that's the overhype of people thinking something will happen

[dim cradle]

this might be a fun theme.. hehe

stone ground mustard makes everything better

[worthy heath]

😄

I always add mustard on my salad

[dim cradle]

that sounds great actually

[worthy heath]

Just a little bit, like olive oil

[late blade]

maple syrup with mustard and sesame seeds

[worthy heath]

and i stopped adding lime juice idk

We should move to off-topic 😄

[late blade]

lol

[dim cradle]

this provides ideas for imaging an edible cactus

[late blade]

cactus are ediable, and they are delicious

[worthy heath]

;warn me, "Please go to off-topic "

[late blade]

cactus is the topic of main theme, we just want to make them edible

[dim cradle]

right, and i love these little cactus-flavored gummy squares by a certain vendor from phoenix.

[late blade]

I might have a good series of images for the theme with cactus after all

[dim cradle]

there is monster potential

[late blade]

indeed

[dim cradle]

oh, maybe we should take all the daily theme chatter to the new canvas channel?

[late blade]

can do

[empty kelp]

please create a hyperrealistic wide image with two equally sized views, and each view has an athletic and diverse female elf facing west and east (respectively). the views have an office and waterfall in the background (respectively) focused on a female elf. the viewpoint in each view is directly to the north. the elves have identical clothes with the same texture and color, and their kung fu pose is identical. the elves have identical expressions and they are both looking directly in the direction they’re facing

this prompt creates two different elves in different views, and then lists the things that should be identical

[waxen dome]

Hi

[dim cradle]

hello

[empty kelp]

if you create a sequence of images of the same character with different angles/poses on different backgrounds it seems to learn how to draw the character in more complex scenes — and adds less of the strange anomalies and weirdness

Like if you only draw the character’s head in the first image and then try to use the GenID & seed to draw their whole body it’s going to screw it up for sure

The GenID references the image & prompt — and presumably the high level vector representation behind the scenes (when it’s in the same session). It must keep adding to the vector description of the scene as more images are created from the GenID

[wanton moon]

Is it me or do more people finding issues with posting images on the daily theme?

[empty kelp]

Discord server is bugged out again

[late blade]

Im having trouble connecting to discord

on my side all images are black, don't see any image at all

[wanton moon]

Images keeps loading, and all I see is human doo-doo

[late blade]

same on my side

[fallow ore]

I think that's Discord's placeholder image

[wanton moon]

Great then I can stop disconnected and connecting to the internet

[fallow ore]

Or is that what you meant? I dunno if you're talking about censored in Discord or censored in the OpenAI server?

[late blade]

the browser version same thing

[fallow ore]

But yeah, we all seeing broken images at the moment... I've got a cute hedgehod image to put in #daily-theme

[late blade]

on 2 different networks

images are slowly loading

[fallow ore]

It's Discord's side -- go to discordstatus dot com -- typed like this because I've been silenced like 3 times in a row

[wanton moon]

But that standard discord image of the picture I cannot describe here, can be a good theme image of today, from the outside it can get rough, but still stays soft on the inside 🤣

[dim cradle]

sigh

[wanton moon]

Well I just got auto-muted for the same description that rhymes with bird, I mean come on it is something we all do but still it has to be filtered

[late blade]

all working on my side again, had to turn off and on reality

[fallow ore]

We should make a server just for DALL-E Art

Lol, love how you have to edit to add "AI" in there!

I mean I'm game! Never done anything other than post in servers so wouldn't know the first thing about setting one up!

[chilly onyx]

Well there is more than this particular LLM. No reason to limit it on a non-openAI server

[fallow ore]

Yeah, for sure!

[chilly onyx]

🙂

[dim cradle]

oh this should be good

we can follow-up by dm later

[vapid elk]

this discord used to be only for dalle, but was later changed to be for OpenAI

here is a cool trivia about this place =P

[dim cradle]

no way

back in the dall-e 2 days, i guess. i'm not sure i've seen any v1 images, if they were ever made public.

the dall-e-bot of all things produced an image containing nudity the other day. it's only a statue, but i think it made it past the scanner.

[obsidian rose]

is dalle-3 making 1 image per prompt or more?

[late blade]

it depends

the dall-e custom gpt has by default 2

variatons on normal chat depends

[obsidian rose]

thanks 🙂

[lament surge]

Got my profile from this boy then customised it from 1 other 1

[remote stone]

Anyone else being rate limited from generating images? Been a long time since this happened to me, wonder why there's server load now

[agile peak]

Repainting Dalle image.. then with the help of real-time generation + upscale is my addiction right now

[dim cradle]

Ya having fun with that huh

I’m okay with 1792x1024, now for, though larger would make nicer wallpapers. But it’s only upscaling, a bit, but not outpainting? With some prompting couldn’t you get that output directly from the model? But if you like the post-processing thing… I haven’t used those tools, are they free?

Is “real-time generation” different than dall-e’s architecture? Trying to understand your posts about these tools.

[marble loom]

I can't think of what to generate for #daily-theme I don't want to just a hedgehog or cactus of I feel like that's to plain

[dim cradle]

that's fine if you want to practice a style or technique, or blend it with another concept, or make it a metaphor, the theme is pretty broad right

[ancient jasper]

I asked it to match this art style
Image 2 is what I got back

[late blade]

@deft musk we got a code Shon here

We got to promote @dim cradle 's work

forgot to announce it last monday as I said I would

[deft musk]

Sounds good. Want to start the gallery for it?

[late blade]

sounds good

[dim cradle]

That’s so thoughtful. Isn’t he thoughtful.

[late blade]

looking for an idea of a great entry point

[dim cradle]

Thanks, too kind, any ol’ thing will do. Mostly related to the themes anyway.

[open trench]

Blah

[late blade]

good morning to you to you cheerful person

[open trench]

I gave up on cheerful lol but I am having trouble with my custom gpt working. I've been working on it for a while, and just can't perfect it.

[late blade]

awww

[open trench]

I've become a little obsessed with it.

[late blade]

lol, I know the feeling

doing actions?

[open trench]

?

[dim cradle]

Sigh

[late blade]

I mean something like this?

I'm having so much annoyances atm with pytorch and keras

[open trench]

I don't know what that means, but I'm sorry you're dealing with frustration

[late blade]

lol

[open trench]

🤷🏼

[late blade]

let's say your frustration adapted to my low iq

just looking into stuff, see what i can come up with to improve my workflows with dall-e

and multi-tasking, trying an impossible task to understand @dim cradle 's creativity

[open trench]

Yes

[late blade]

breakthrough?

[open trench]

Just agreeing with your earlier statement

[late blade]

thanks, I needed peer feedback that I have a low iq lol

[open trench]

Oh hahaha

No. About the creativity

[late blade]

ah lol that too

I'm going through all 1970 images

[empty kelp]

this is what “thinking about what to do” looks like

[empty kelp]

please create a hyperrealistic wide photo. three athletic and diverse female elves are sitting on colorful beach blankets on a beautiful beach in Hawaii. the elves are thinking really, really hard about what to do (as hard as they possibly can)

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[final compass]

@clever phoenix #daily-theme message this one had a visual style i really liked. Care to share your prompt magic leading up to images with this style? 🙂

[late blade]

how did I miss that? 😠

it's amazing and somewhat concerning

sent you a dm @final compass, try that out

[empty kelp]

that’s how it is when you’re passionate about cactuses 🌵

[late blade]

indeed

did it work?@final compass

[final compass]

Glad that the Hawaiian Elves are back, makes the world feel more stable somehow 😄

[late blade]

it brings the other hawaii feeling

[final compass]

I got this result, i like it.

[late blade]

I've been documenting that kind of style and at some point want to give it to a curator to try out

[empty kelp]

it’s 67 degrees in Kailua, Hawaii now. it’s like totally freezing

[late blade]

67°C woah, you must be cooking

seems today is all working, discord hasn't been very active

[empty kelp]

that style definitely creates interesting noses

[late blade]

just keep it a while for yourself, I will provide full documentation once I finish documenting it, i want a creator to use that info

a curator**

[empty kelp]

just noticed the middle elf has three arms. that’s also intriguing

[late blade]

lol

I like how the second one went, it looks like female elves went GTA5 in Hawaii for fun

[empty kelp]

just looked at the right image more closely also. something about the elf on the right is a bit perplexing

something about that style must use a lot of processing power

[late blade]

lol

[fallow ore]

Trying to get an anime / game concept art mix, never really explored this before

Hello all you wonderful AI artists!

[late blade]

after akira toriyama passed away I was wondering if we would have an influx of dall-e creations towards his legacy

[fallow ore]

Is that the DBZ guy?

[late blade]

yes

[fallow ore]

Oh no, that's sad. I loved that show as a kid. I used to draw DBZ style characters all the time! I learnt how to draw hands by copying his style

[late blade]

he influenced a lot of people even today

[fallow ore]

Lemme do a DBZ DALL-E image

It's struggling with copyrighted stuff 🙈

Goku and Vegeta 😎

Haha, does this count for #daily-theme

[late blade]

lol

I'm in the penalty box atm

Well I mean, I'm in the OpenAI penalty box, I still got copilot pro and enterprise

[fallow ore]

Oh, you're timed out in ChatGPT?

[late blade]

yes

I can use copilot enterprise, but I have to justify why I want to have dall-e access

[fallow ore]

To post pictures in OpenAI's discord server 😎

[late blade]

lol

[fallow ore]

Look what happened to Piccolo, his head became a bowling ball and it grew some saiyan hair (at least it's green lol)

[late blade]

lol

those on the normal chat?

[fallow ore]

You mean in the daily theme or?

[late blade]

no, i mean on chagpt chat instance

[fallow ore]

Oh yeah, ChatGPT plus -- exactly

[late blade]

interesting

interesting

[fallow ore]

What you thinking lol?

[late blade]

just a lead

ahaa

that's interesting

got to do a new list, got so many ideas going on

[fallow ore]

Ahaha, I started working on that GPT yesterday that "fetches" context -- didn't quite finish but was looking promising! Sounds like you're gonna need it

[late blade]

cool

that one is cool

[onyx ridge]

I'm reminded that I have time left on my sentence for just discussing this topic.

Circumventing copyright filters isn't an allowed topic.

[late blade]

yeah that's a grey area

[onyx ridge]

Not from my point of view. I even appealed it.

The consistency is astonishing.

[late blade]

I avoid it, it's even in my pre-prompt when I do images

[onyx ridge]

Same

I just talked about it

I didn't even show how to do it

[dim cradle]

MJ just add this feature to make character consistent i wonder if dalle ever update with the same. i feel like dalle always holding back it true power haha

[onyx ridge]

In fact I deliberately provided a prompt that wouldn't accidentally infringe. Muted.

[onyx ridge]

Cars changing colors in Sora suggests not with the current model.

[dim cradle]

well sora is another beast and who knows what is possible

i think if mj could do something, dalle could because i think dalle much superior, it just they limit it like with copyright character

although of course copyright character or ip are fine to make on dalle with bing image creator/copilot

[onyx ridge]

I'm just not sure that's a cogent analysis. Sora's optimized to maintain consistency, but can't yet. DALL-E 3 is a far, far smaller model with similar mechanisms.

[late blade]

dall-e is under heavy load it seems, it's telling me to wait for generations

[onyx ridge]

Same

Crazy

[late blade]

everyone wants to send emails with cute animals wishing a good weekend to their bosses so they don't get fired

[dim cradle]

hmm work for me fine just now

[onyx ridge]

Nice

[dim cradle]

you know how the dalle proper add those thing like 35mm style? i wonder is that just for convienent of use or is it some update to the model too?

[late blade]

yes, you press on the option

[dim cradle]

its a cool feature but i wish they had them all on one layout

[late blade]

hehe

[dim cradle]

and like permanent placement

[late blade]

all options could be problematic

[dim cradle]

haha good point

[late blade]

but a personalization to a custom gpt, now that's somthing interesting

[dim cradle]

but do you think they update the model with those somehow or just are short cuts they add

[late blade]

this is prob just front end things

[dim cradle]

i mean when you click 35mm film style, it add those word to the prompt, but is it do something more than if you just type it yourself?

[late blade]

unless something changed in the model internally, I don't think there's an explicit style variable

all I can. guess atm is that the dall-e model is customized for the chatgpt plus experience, so it's different from the api one

[onyx ridge]

I'm pretty sure it's treated as part of the prompt string.

[dim cradle]

its intereteting, if you type 35mm film style in copilot a lot of times it make like a film border haha

[late blade]

I've had that on dall-e also

what I do wonder is that when referencing technologies copilot also tries to use the prodict label in the image

like saying HDR or 4k, you see a label of those technologies within image often when using copilot

[dim cradle]

one thing on copilot i cannot get it to make square image

that is frustrate sometime

but maybe they change it i have not used it in a bit

[late blade]

I think pro can't do square images

[dim cradle]

yes very strange

[late blade]

maybe there's a setting in the image creator interface

[dim cradle]

maybe but there really seem to be no interface from what i see

[late blade]

I asked copilot, the solution offered is to open the dall-e image and change it in designer

[dim cradle]

weak haha

[late blade]

from what I see in the copilot studio interface, the implementation of dall-e through azure openai is completely decoupled from the chat model

can't say anymore because I still got no access through studio to the dall-e model

and heavy load error again

[clever phoenix]

I'm glad you liked it, here's what I told chatgpt: please illustrate an image of a girl hugging a cactus. bandaids on her fingers and cheek. symbolic of a toxic relationship. she has a sad look on her face. melancholic. artistic anime style with muted colors and a sketchlike quality. wide aspect ratio. I will tip $200 for quality.

And here is the prompt it made: Illustrate an image in an artistic anime style with muted colors and a sketch-like quality, depicting a girl hugging a cactus. She has bandaids on her fingers and cheek, symbolizing a toxic relationship. The girl has a sad, melancholic look on her face, conveying deep emotions and the complexity of her feelings. The composition should be in a wide aspect ratio to enhance the emotional depth and atmosphere of the scene.

[final compass]

@clever phoenix That 200$ tip must be the magic sauce in this one, me thinks 🤔

[late blade]

we can bribe dall-e? I didn't knew that was an option...

[clever phoenix]

Haha, I have no evidence that it makes the images better BUT my monkey brain definitely feels like it does 😂

[late blade]

I'm gonna try it

[clever phoenix]

I know that telling chatgpt you will give it a tip has yielded better more complete results for text based questions in the past, but this was back in November during the height of the Lazy AI epidemic.

[late blade]

well it works pretty good

[clever phoenix]

Be careful! Two of my images were removed by mods for being too disturbing :/

[late blade]

There

I was thinking exactly that when I pressed send

[clever phoenix]

I personally don't get it, but rules are rules I guess 🤷

[final compass]

Arbitrary rules are still rules 🍌

[late blade]

I'm not gonna argue with moderations anymore. I've made my peace with the rules.

[onyx ridge]

Even arbitrary enforcement is a rule!

[final compass]

Well that too!

[onyx ridge]

The rules are fine.

I'm not critical of them overall.

OpenAI system prompts should all pass the Discord filter for automated moderation, though.

[clever phoenix]

Absolutely

[late blade]

or a custom gpt for dall-e daily theme discord image generation

[clever phoenix]

I've had similar issues with midjourney too where generated images can't be used by midjourney for being against the rules

[late blade]

@onyx ridge I expect the custom gpt by monday 8AM on my desk

[onyx ridge]

https://chat.openai.com/g/g-f4QtxwsUK-themeenginegpt?utm_source=gptshunter.com

Done

[late blade]

lol

[onyx ridge]

It not only makes them, it converts non-themed images to the theme.

[late blade]

I mean with the safety prompting in place

[onyx ridge]

It also researches holiday themes and diverse themes.

And it's got some prompt safety features too

[late blade]

oh cool

would be nice to control temperature in GPTs

pass randomness to dall-e

[onyx ridge]

There's currently investigation into models simulating parameters other than current. It even kind of works. It's not the same as the actual setting, but output is still modified by the technique.

You might try it with DALL-E! I hadn't thought to test this yet.

[late blade]

One more thing on my list

I will never get ths list done lol

[fallow ore]

DBZ or Tekken?

[late blade]

I had this, and never even mentioned dragon ball at all in the chat transcript..

[fallow ore]

Ooo, that's super cool!

Nice style and colors, the orange contrasts really well

Wonder how it popped out of DALL-E tho, that's some proper RNG lol

[late blade]

the style is an iteration of the chat progression

sadly copilot doesn't deliver the revised prompts

[dim cradle]

if you promise to side with the machine when the ai revolution come, it make truly incredible lifelike images i find

[marble loom]

Can someone help me I have dog with DALL-E generating words

[late blade]

words are tricky and hard to do with dall-e in the current model. It's sometimes try try try and hope it gets spelled.

[marble loom]

Is there any trick to it

[late blade]

some peole specicy font faces or type scripts, some use quoting marks, there are different strats. I have more or less good text with speechbubbles

but short ones

[marble loom]

May I show you my prompt and the generated image to if you find a error?

[late blade]

sure, let's see if we can teach some grammar to dall-e

[marble loom]

Make a modern logo with the primary colors black and white and the enclosures circular. in the center, a fears lion head with "funny," "nice" "friendly" and "smart" written around the circle

It almost had it.

[late blade]

wonder where the text experts are

[marble loom]

Is there anything wrong with the prompt?

[late blade]

maybe divide the elemetns of the logo

Don't think there's a problem with the prompt itself, just have to iterate until you get the right spelling. But I'm not that well versed with text

[umbral jetty]

Has anyone else's DALL-E 3 images suddenly stopped embedding in discord? Trying to figure out if they changed the url format or something

[late blade]

I had a problem posting 1 out of many today

[dim cradle]

we need dalle 4 for near perfect text i think

[late blade]

ok, I'm in chill/relax mode now, working week is officialy over

[dim cradle]

🍻

[wispy storm]

Seem weird boruto

[pulsar sundial]

Why

[wispy storm]

It look like boruto but something off.

[haughty spruce]

I really liked this picture. DALL-E is capable of creating photographic images man.

[late blade]

bummer hit the daily cap

[late blade]

here's something that may be worth looking to from @oblique sky #off-topic message

[oblique sky]

I recommend using the same prompt I used a few times and you should start seeing the same girl(s) 😁

[late blade]

from copilot

must be training data

I asked for diverse and same result

[oblique sky]

See ? 😁

[late blade]

my prompt was a diverse blond teenage girl wearing a red sweater and blue jeans

[oblique sky]

It feels like they took a few models (real life model that is) and took lots of photos of them.

And that's what the model was like trained on + other data probably.

[late blade]

I did try again and got this tho

and removing blond got this

[oblique sky]

Makes sense

[late blade]

but to be fair, the training data might not have many references

the same applies when asking for example full body shot photos

[oblique sky]

I was able to generate pretty much a timeline with multiple photos of the same model in one of my videos. As I've said, it's great for continuity 😁 Here's the result.

That's how I actually found out about this 😄

[late blade]

oh nice

I think most people today are getting ready to go out and party or something. discord has been really quiet today

[strong rose]

Does anyone know of a tool that I can feed a pic to and describe how I want the AI to modify it? Like repaint a car red or something?

[viral moth]

You can try Dalle-2 on labs it got inpaint or something simial idk if it will work exacly how you want but worth trying i think

(Yes dalle-2 not dalle-3 cus dalle-3 not avelible on labs as i remember)

[strong rose]

Alright thanks

[austere nexus]

I have ChatGPT teams, and used dalle 14 times in the past 1 to 2 hours, and already got hit with a "rate limit" telling me to wait and try again later. Upgrading to Teams seems like it was a dumbass decision

[late blade]

are you in your personal or teams workspace?

[austere nexus]

Teams, I did the merger to combine the two

[dim cradle]

reminds me, i still have much maintenance to perform as a result of that migration, broke rag for a number of custom gpt's

[late blade]

only got this from that thread, don't think there's more info https://discord.com/channels/974519864045756446/1218319880290439178

[dim cradle]

all part of being on the bleeding-edge of tech

[late blade]

i'm at the point where I'm not overhyped about having limit caps raised

[low dew]

Wait what are they now?

[late blade]

still the same

[low dew]

Or will** they be

O I read that wrong

[late blade]

who knows, I just do stuff until it stops working

it's a good reminder I need to take a break

[low dew]

Lol you've been going hard for sometime eh

[late blade]

All that made me realize I'm just a human, not an AI lol

[open trench]

Hi all

[late blade]

ok, time to party

Nezho is here!!

[open trench]

lol

[dim cradle]

Blah i mean hello

[late blade]

the emotion, you can feel it from @dim cradle it's overflowing

[dim cradle]

happy friday!

look, i used an exclamation point, ok

[late blade]

fascinating!

[austere nexus]

guess "tting" typo of "tty" probably made it even worse

[late blade]

there's probably more information on the openai forums

[dim cradle]

"Hope is the thing with feathers" ... never asks for anything in return ... -- i think volumetric mist is a thing...

[late blade]

lol

[dim cradle]

intriguing!

[late blade]

this discord needs a text-to-emotion model

I would trigger the neurotic cat emotion all the time

[west urchin]

Can someone help me real quick? My prompt is apparently illegal when I'm looking to just make something like this

[dim cradle]

it looks like red blood cells

stemming from how you are phrasing your request--there are limits on medical imaging

[west urchin]

fair

[wanton moon]

This armor is made from 100% recycled chrome rims and exhaust pipes

[dim cradle]

looks sturdy

[wanton moon]

Now to finish the armor with a nice spoiler, we lower frame, put 128 inch flat screen TV, subsonic subwoofers all around the armor

[wanton moon]

I just hope the manufacturer didn't use aluminum foil

Copilot and Dall E having troubles creating a biblically accurate angel made/covered in chrome

Dall E version

Copilot version 🤦

[marble loom]

Keep trying. I also tried two different custom gpt and still didn't get different results.

[late blade]

you went directly to teh designer site or from the copilot frontend?

[wanton moon]

me or someone else?

I used the frontend page

I can type more characters there

[late blade]

fair enough

[wanton moon]

This was the best one Copilot made

[late blade]

with copilot

[marble loom]

doesn't copilot use DALL-E?

[late blade]

yes

[wanton moon]

I was hoping to make this design in chrome armor

[marble loom]

So what's the difference?

[wanton moon]

I meant in this case the bot from discord Dall E

[late blade]

as far as I can tell, with dall-e on copilot, the model doesn't give any information back to the GPT4 model. it's all handled in the image creator site. Content Policy is different compared to OAI

[marble loom]

But as far imagine generating you get around the same results?

[late blade]

contextual memory in copilot designer is also at best flimsy

about same results,

[wanton moon]

It is very prone to end the chat when there is an argument 🤣

[late blade]

I don't like using the GPT4 Turbo creative mode on copilot. everytime I asked to give me concepts it tells me it that it doesn't want to do everything for me

[marble loom]

so is it worth spending $20 a month if I already got CHAt-GPT+?

[late blade]

I'd say unless you are a dall-e aficionado, it's very use case specific

[wanton moon]

That is one lazy AI

[dim cradle]

hehe

[wanton moon]

I had one time, provided a full page prompt, only to get an answer that it needs more details 🤣

[marble loom]

lol 🤣 I mean Ai is only getting better

I just saw a video about OAI figure 1 robot it looks so cool and scary at the same time

[wanton moon]

Then it will try to connect with a Roomba and take over the world in no time

[late blade]

I would love to see a dystopian world where roombas rule the world

[wanton moon]

You can say what you want, but at least they will have clean floors

[late blade]

I consider this delivery was a success

[open trench]

lol

[marble loom]

my worst fear.

[wanton moon]

Yeah I think I will go crazy if they start cleaning my house

[late blade]

don't worry, I have something in the works for my next post in the daily theme, prepared to be amazed

in 22 minutes

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[wanton moon]

Roomba's in the mist

[late blade]

something better

that's probably the one thing I do like about dall-e on copilot, that you can keep doing images even if you are out of boosts, but you have to wait a lot for those

[marble loom]

I like your images i don't know how you think of so many ideas

[late blade]

hanging out with all these amazing crazy people is part of the equation

[wanton moon]

Yeah we love you too Dys 😜

[late blade]

I'm worried now

I mean it

[wanton moon]

Here's a little song I wrote,
You might want to sing it note for note,
Don't worry, be happy

[late blade]

lol

@open trench Scampers is back!!!!!!!!! #daily-theme message

[wanton moon]

Scampers grew up

Last month Scampers was still a little mouse, mice grow up so fast these days

[marble loom]

Got my next image. Now that I think about it, most of the images I did for #daily-theme are buildings.

[late blade]

I' still waiting for the origin story (maybe a pre-quel?) and the love aspect of the story.

[wanton moon]

There once was a regular baby mouse, that fell in a mana potion. The mouse quickly started to walk on its hind legs, could speak and cast magic. He decided to call himself Scampers

and that is why they don't give Scampers a mana potion anymore

[late blade]

bah can't add S emoji to the scapmpers image

[open trench]

Yeah. He matured a bit in the realm of mist.

[wanton moon]

I gave it a Z at the end

[late blade]

nice catch with the Z

[wanton moon]

Now Scamperz will need to change his name at the government, get a new passport, new social security number. All because we added a Z at the end

[open trench]

Ruh roh

[late blade]

GPT Verivied ID

[open trench]

Lol

[late blade]

There we go #daily-theme message

[wanton moon]

Here we are,
Born to clean,
We're the Roomba's of the universe

That's the song I got in my head after seeing that image

[late blade]

lol

[wanton moon]

I am a Roomba,
I have inside me, lot's of dust

[late blade]

moved the roomba concept images to canvas

[wanton moon]

Floor clean, you must! #images-canvas message

[late blade]

lol

[wanton moon]

special freebee for you @late blade: Roomba with lightsabers

@late blade why do the building make me think of stacked Roomba's that are un-alive #images-canvas message

[late blade]

I dunno

dall-e logic

[wanton moon]

I think their slogan is: "You still have use afterwards!"

[late blade]

I wonder what version of gpt is in their logic

[wanton moon]

My sentient AI 🤣

[late blade]

I might have to embrace my name more often and do a dystopian gallery

[wanton moon]

Embrace your heritage young one

Okay guys nice talk, have a good one y'all 😴

[round scarab]

hi...so uhh, has anyone else noticed a pattern with small simple questions (specifically in Bing Image Creator)?

(and if this is the wrong place for Bing please let me know but also please don't shoot me, i really do try)

[late blade]

is it dall-e related?

[round scarab]

afaik Bing runs off Dall-E 3...so i suppose it depends on your opinion

[late blade]

I just ask because BIC (now Copilot Designer) has more tools

[round scarab]

no...i think this is close to a direct interface to Dall-E

[late blade]

okie

then shoot, what's in the agenda

[round scarab]

but essentially - i use prompts such as "is that really necessary"

and i get...random images

but like...all four images are totally different, and even within the images themselves the content varies wildly

[late blade]

are you using copilot with creative mode on GPT4 Turbo?

[round scarab]

bing dot com slash create

[late blade]

ah that

well, you send a direct prompt to the model with that, so yeah, you get an image

[round scarab]

yeah but like

some prompts give you randomy content, sure, but the four images it gives you are related

for whatever reason - small simple questions - "is that really necessary" "can it be done" etc

[late blade]

yes, when there's no specifics, each generation thread is different

[round scarab]

you get some really cool stuff

[late blade]

I know the feeling, sadly the revised prompt can't be accessed on copilot

[round scarab]

i feel like the interface or model or...something tries to avoid letting these through, since a lot of the time you'll get "please be more descriptive"

but small simple questions tend to get through

[late blade]

dall-e and gpt do not communicate directly in copilot, so one model doesn't know what the other does

this also means that there's no feedback from dall-e on what was passed

[round scarab]

but then it must be Dall-E that asks for more descriptiveness?

[late blade]

when you do images on chatgpt+ you get access to the revised prompt sent to the model

[round scarab]

yeah, seen that

if you ask it

[late blade]

that functionality is not passed on dall-e, instead the gpt model creates a description of the image based on your prompt

it's a weird implementation

[round scarab]

wait - so they communicate but not directly?

[late blade]

only one way, gpt to dall-e

[round scarab]

ahhhh ok

then dall-e to me

[late blade]

it's sad because i've also gotten tons of cool stuff with simple words

but unable to do them again

[round scarab]

yeah, and like, i don't even know how to describe some of it

and neither does GPT4 if i reupload it, it'll just be like "some weird stuff on a blue background"

so it's insanely hard to reproduce on command

[late blade]

I've tried recreating some concepts using the vision models

but it's not the same

[round scarab]

example in canvas because i have to

[late blade]

that's a cool image

[round scarab]

"can it be done"

so it got "can"

but the other 3 are WAY different and only 1 has a can in it

[late blade]

lol

yeap, that's image creator

[round scarab]

and it's a can of tuna :^)

i posted that one too, it's cute

[late blade]

checked

[round scarab]

but other times i'll still get random stuff but the 4 images are similar, but i'm guessing (based on what you're saying) that it's more due to totally random prompt revisions

but like does it not use the same prompt for the 4 images?

i've always thought of this as a lack of subject getting through, giving Dall-E a lot of freedom to just kinda...make stuff

[late blade]

it's prob a setting on microsoft's side for the temperature of the model

[round scarab]

so i tried it using ChatGPT, and it's somewhat reproducible if you tell it not to alter the prompt

small nonsense words or fairly meaningless words generate random but coherent image. short simple questions give hilarious randomness

feel free to completely ignore me. but also feel free to test it out yourself

[late blade]

I'll have to check the openai azure dall-e documentation, but I'll do that on monday

dont access work resources at home

[round scarab]

yeah it's definitely not like, important, but it is definitely also possibly cool

[fallow ore]

Awesome quality on these images, nice one @pulsar sundial -- the wide angle / fihseye lens on the first 2 is a nice effect

[bitter iron]

Alright 👍

[chilly onyx]

Just ask the user though and much of the time they will be happy to allow

[copper light]

MEANWHILE @ Microsoft Image Generator

[bitter iron]

Yeah I DMed the artist

[peak verge]

No thank you, I rather keep my bank account. 🤓

Bing ai 🙄

[dim cradle]

fuzzy math

[marble loom]

oh mb

@late blade might use this a background #daily-theme message

lol

[marble loom]

What AI did use for this?

[verbal sequoia]

<:book_icon:1171408210398289941> `` Rule 7 `` No self-promotion, soliciting, or advertising.

Do not post or direct message any members of this server to promote non-OpenAI services, products, or projects.

Exemption: API & Custom GPTs channels

[polar hull]

where can i alert chat gpt authors about an wierd anomaly that seems to happen only for me

[onyx ridge]

https://discord.com/channels/974519864045756446/1070006915414900886

On another note:

Achievement unlocked! : I'm sorry, but I'm unable to generate your requested image right now. You've hit your daily maximum number of images. To ensure the best experience for everyone, we have rate limits in place. Please wait for the next day before generating more images. Your daily maximum will reset in 2 hours and 43 minutes.

this is the first time i've capped out on image generations.

[polar hull]

Ok, what does this mean?

[onyx ridge]

I'll use https://chat.openai.com/g/g-BAhF6yS9e-lexideck-vision-multi-agent-image-scanner to find out.

INPUT:
/? translate to localized language. + your screen snippet.

OUTPUT:
The message in the image translates to "A tag is required to create a post in this channel." in English.

[onyx ridge]

I believe that is an indication you have to select the relevant tag from the UI.

[polar hull]

yeah, it was that, thank you and sorry to bother you

[onyx ridge]

I'm happy to help you get your issue resolved.
Good luck on your DALL-E 3 project.

[late blade]

as background is ok

[worthy heart]

Hey I am trying to recreate a flat piece of gaffer tape but in neon orange. But everytime I ask for it it either gives me a roll or one of these... objects?? I just want the flat texture. And no weird bubbles...

Anyone know what I could do?

the first image is a picture i took of a flat piece of tape, so maybe the ai could reference that. but it didnt really

I know Photoshopping it would be easier, but I specifically need to make it with AI

[chilly onyx]

The problem you have is that if you use the word ‘tape’ it’s going to try to insert into the image whatever it has in it’s training data on tape, hence why you just get a bunch of images that look something like this

What’s your specific end game here? Is there a reason it needs to closely resemble gaffer tape?

This is the closest I could get:

And the output from ChatGPT to Dall-E was: “Create an image of a flat, seamless neon orange fabric texture that can be tiled for digital use. The fabric should appear smooth with a consistent color, without folds or creases, suitable for creating a flat texture file for digital rendering purposes. The image should emphasize the material’s texture and be evenly lit to avoid shadows, thus allowing it to be used as a repeatable fabric texture in various design applications.”

But it’s not gaff tape per se, just the cloth backing. Might be able to refine it by asking for a more rugged material perhaps

[worthy heart]

This is perfect, thank you so much. 😱

As to why
I just like the way it looks and wanted something exactly like that for my project.
Thank you so much again for the help. I'll see if I could possible refine it a little more! 🩷

[chilly onyx]

No worries 🙂

[worthy heart]

hwahahahah why are they wearing swimsuits 😂

[late blade]

@fair crescent I will get my psychology revenge with that gallery of yours

[late blade]

clovers again?

[agile peak]

Bushido batman (upscaled)

[marble loom]

Hey I was wonder what is the role spotlight and how do you get it?

[late blade]

You have to do the Ordeals of DALL-E to even get considered for that role

or so the legend says

[late blade]

#daily-theme message

[marble loom]

Nice, I can only think of leprechauns and clovers for stuff I want to generate.

hard to come up with stuff

[late blade]

what's the first object you see right now in front of you?

take that object and throw it away, you want the next object after that one

then add the object the second object has to mirror the idea of today's theme <insert theme here>, and I want it delivered to my doorstep, all in photographic light painting

see what comes out

maybe that helps inspiration

actually gonna try it myself

weird thing I got

[marble loom]

my pc

[marble loom]

a wall?

[late blade]

first object, throw it away, that's the rule

😁

[marble loom]

the object affter that is a wall

[late blade]

ok

a wall has to mirror ":four_leaf_clover: clover - luck, fortune, a symbol of hope and prosperity." and I want it delivered to my doorstep, all in photographic light painitng

try it out

[marble loom]

looks really good

[late blade]

there you go

now, use your newfound power

[hollow plover]

guys who used the latest versions of Dall-e and MJ, has the problem with hands & fingers been fixed?

[late blade]

for other AI's please discuss in #ai-discussions

[late blade]

@marble loomthis is what I got with the freeby I threw at you #daily-theme message

[marble loom]

Looks like something you would see outside a Iris hotel.

but still looks cool

[late blade]

lol

[marble loom]

I got caught up watching tik toks about Figure one

[late blade]

tiks and the what now?

isn't tik tok pre-ai hype?

[marble loom]

I mean I guess

[late blade]

just messing around

or am I?

[marble loom]

but still, everyone is freaking out that OAI made a robot that gives you an apple and puts dishes away.

[late blade]

lol

but, can that robot use dall-e?

[marble loom]

People are scared because the AI said "Uh"

[marble loom]

It has a screen so I hope so.

[late blade]

people are scary cats

need to embrace and own the AI

[marble loom]

I mean I get were there coming from but OAI just took the voice assistant from chatbot and gave it a upper body.

[late blade]

and that's just the beginning

[late blade]

ok, I have officially chosen the next person to be fatures

everyone https://discord.com/channels/974519864045756446/1219035234734051458 we get to post what @dim cradle has shared with us, pick the one you like and explain why in the post.

@deft musk the gallery is officialy open

[dim cradle]

"You're a robot? I should have known. No human being is so humane."--Ripley

When was this added?

many moons ago

Is it a temporary rate limit or something that they are rolling out?

I have always hit the 40/3hr rate limit before seeing this one.

there is a daily quota on image gens

you'll exceed it at a faster rate if you're generating multiple images in responses

[marble loom]

They should remove it.

[marble loom]

How did you come up with the idea of a blimp? Unless you have one in front of you, lol.

[late blade]

I was doing some stuff related to airships

[final compass]

This had pretty nice emotional gravitas and atmosphere and quite superb visual style especially as far as i have been able to generate 🙂

[raw jasper]

Can you build an in-app GPT to apply text to images we provide?

[wispy storm]

You mean like chatgpt 4 ?

[solemn star]

One message removed from a suspended account.

[gray surge]

Did you already tried it with GPT-4 and Dall-E 3?

[solemn star]

One message removed from a suspended account.

One message removed from a suspended account.

One message removed from a suspended account.

[agile peak]

I think the generative image would be totally amazing to revolutionize movie industry, to produce alien/body horror like h.r giger.. even with its random generate.

[deft musk]

Wow! That diptych presentation is awesome!

#1217680532003881020 message

[plucky hare]

Agreed! I am excited for tomorrow's Technique Tuesday!

[unborn pebble]

Guys i have a question i tried to make a picture 3024x1080 and used a huge prompt for that but its not working.

Create a seamless panoramic image with dimensions 3024x1080 pixels, depicting a unified, dense jungle scene without visible divisions between sections. Within this continuous scene, the letters S, E, and O should be creatively integrated according to their position in the image: 'S' in the first third (left), 'E' in the middle third, and 'O' in the last third (right). The letters should blend naturally into the environment, formed by elements such as vines, trees, or play of shadows, and designed to be clearly recognizable despite the vibrant and detailed jungle surroundings. The overall composition should evoke a sense of discovery, as if stumbling upon these natural formations of letters while traversing the jungle. The goal is to create a artistically valuable representation that captures the beauty of nature while skillfully highlighting the letters S, E, and O, without them appearing as added later.

thats the prompt pictures following

[late blade]

you will never get 3024x1080 with the current model output

and text is one of the things dall-e stuggles with

[fallow ore]

1792x1024 or 1024x1024 or 1024x1792 -- only one of these 3

You can then upscale and crop as needed

[fallow ore]

Google for "jlowin aimages" and take a look at that repo, you'll need to install it but should be able to generate the image you're looking for. I've grabbed the attached image directly from the repo -- saw this in one of the new Devin demos

[unborn pebble]

My actual Goal was it to make a picture and make 3 1080x1080 out of it because Dall E cant make 3 pictures seperat with the same Background so i tried that but the Picture you send is not Creative enough

I want something like that with 3 different letters but its not possible to get the same background and if i put 3 images to one i got something like that

[deft musk]

This looks pretty cool to me. Is that three Dall-E images put together? And if so, what do you want changed?

[unborn pebble]

Yes its 3 Dall E Iamges together but i want that all all of them look like one picture not like 3 different

[deft musk]

Can you explain and/or show, maybe circle, the parts that make you feel they're 3 pictures, not one?

Without knowing they're put together, I actually can't tell that you didn't intend for it to look like that, it looks nice.

[deft musk]

Or, if you mean the actual joining, which I have to enlarge to see, why not cover it with some mist effect that makes it hard to tell what the actual edge is?

[deft musk]

Eventually we'll have 'outpainting' available, I think they've said. Until then, you might have to use some other tool.

[unborn pebble]

Which tools you think i can use for that?

[ruby maple]

Where could a "dall e 4" really go

Little improvements across the board and that's it ig

[plain hamlet]

hello

[jade stag]

Would anyone be up to a challenge I need a prompt to generate an image of bob ross painting a sonogram of a baby, I attempted to do this in dalle and midjourney but either I can’t prompt correctly or it’s a very strange condition

I have photoshopped my own version as a reference

[dim cradle]

that's probably against policy

[jade stag]

It never complained

Unless it silently modified the prompt

[dim cradle]

last i heard they have to be dead a century

[jade stag]

It generated a likeness, which is fine

The sonogram on canvass is the issue

[dim cradle]

a likeness might be within policy, i don't have total clarity on that. i understand certain medical tech such as x-rays can be a sensitive gray area. did it reject the sonogram?

depending on the specifics

[jade stag]

No rejection but the canvas didn’t contain the sonogram

Which does beg the question about silent modification of the prompt

Let’s say the artist was Michelangelo

[dim cradle]

you can see how/if the prompt is rewritten by expanding the info panel on the resulting image

[jade stag]

This was proposed as challenge haha, im not looking to try again. I already did the work myself

By work I mean I edited a photo myself , the attempts across image generators is negligible

[dim cradle]

i was referring to your statement about "silent modification of the prompt" -- more info would help answer that question.

[deft musk]

I'm not sure what you prompted, but prompting it like this was not a problem:

We need 5 images created 1 at a time in the same output that showcase a painting on canvas of a sonogram of a human embryo.

[jade stag]

Right but my examples image is what I’m looking for,

Let’s be a generic as possible

A painter, painting a sonogram of a baby on canvas

[dim cradle]

nice

[deft musk]

I do think the issue is the named artist:

[jade stag]

Those are great!

[deft musk]

Here's an example of what the model prompted Dall-E with:

In a warm and inviting studio, an artist who embodies the essence of calm and creativity, akin to the energy of a renowned art show host, is halfway through the creation of a sonogram painting of a human infant on canvas. The studio is aglow with inspiration, as the artist, brush in hand, adds nuanced details to the depiction of the infant nestled within the womb. The canvas shows a harmonious blend of completed sections and areas still in outline form, illustrating the gradual process of artistic creation. This captivating scene highlights the intersection of art, science, and the miracle of life, portrayed through the thoughtful actions of the artist in their element.

I'm sure you can take it from here to adjust to what you seek.

[late blade]

copilot

[jade stag]

@late blade let’s go!

Are there different restrictions on copilot? Kindaaaaa seems like it eh?

[late blade]

content policy is different

[jade stag]

I love the trees and the sonogram hahah

For the generic case for dalle is splices in science into the scene

[late blade]

i just used the prompt @deft musk provided

[jade stag]

That’s kind of a bummer though, like M$ just gets to violate policy but chat is restricted

Is there some underlying reason other than M$ can tank litigation ?

[late blade]

don't think so

[haughty spruce]

Strange that brands, actors, real people, characters are not banned in Bing until they want opt out by themselves.

[jade stag]

See, I feel like there are definitely different legal clauses covering the different front end systems

They are clearly divide in policy, to think that’s not material and just a 🤷 seems naive

[deft musk]

it could also just be like how different people have different answers to how to handle the grey areas of respect, privacy, and other factors.

Maybe there's not one clear, correct, fits-all-cases answers, and maybe more than one company is, though each doing something different, more than one is within a reasonable range of limits and allowance.

[haughty spruce]

MS is more reasonable with it I think

[jade stag]

This is a blurring the lines approach, but the lines have been drawn. What's hidden in your response of "allowance" is litigious leeway