images-discussions

[agile peak]

Tesla experimenting with a flying saucer 🛸

Yeah. It should be submitted to suggestions or bug sections -- then again we are testing beta features 😁

Finally able to capture the electricity and get the tesla coils right

[wanton hatch]

I'll try this workaround

[shy cosmos]

More of the ‘meme frog’

[vapid saddle]

can you explain?

[floral ravine]

harder to get what I want from the chat gpt dalle imo

???

why are two of those illustrations, I included film grain 💀 it literally does this so much on the chatgpt one lol

although I do like that style, just not sure why it inserts illustrative elements into my otherwise photoreal prompts

[ripe wraith]

it likes to dream ok

[floral ravine]

gpt also intentionally prompted each of them to be different skin tones, where I just don't care to specify that lol

just not a fan at all of the gpt prompt jargon

[ripe wraith]

its just making assumptions and they changed on you

better off being specific

[floral ravine]

maybe once it's off fixed seeds, and aspect ratios work, they'll make that jargon stuff optional

until then I'll be at bing 😂

[west veldt]

Something i’ve found that’s fun that someone else here did is have AI describe an image, and then use that description as a dalle prompt:

Here’s my dog on the left and a recreation

Left is food right is recreation

Pretty cool

[tall patio]

A serious question tho

[left sparrow]

That's neat.

[tall patio]

Beside generating funssy images

What are some practical implication of D3?

I mean after the honeymoon period is over

At the end of day I guess people come to AI just to do 2 things

[left sparrow]

The best thing about D3 is layout. The actual generations I find to be pretty poor.

[tall patio]

1. Make money
2. Solve a problem

How do we make money with D3?

[left sparrow]

It makes progress on the problem of generating images that are much closer to prompts

but takes a big step backward in image quality

One thing I've had good success with, and this tip came from another user here is use D3 to generate the layout and then use the image as a noise source for SDXL or some other model

[paper jolt]

I mainly use it for graphics on presentations and logo design as a student

[neon sequoia]

I dunno, for me personally I have used Midjourney for over a year and am up to like 70,000 generations. I just do it for fun because it’s amusing and post the things I like to Instagram. That said I know AI is a big part of some people’s jobs and workflows. (I’ve used ChatGPT in my job, not heavily, but it’s helped)

[tall patio]

[remote sphinx]

alright

I must know how to do different sizes!

[paper jolt]

Different sizes of what if I may ask?

[remote sphinx]

just image sizes, if theres a way of doing it via text

new to using dall-e 3

[paper jolt]

Indeed! There are square, tall, and wide generations

With an aspect ratio of 1:1, 4:7, and 7:4 respectively

Just ask for one of those options and it'll get right on it

[remote sphinx]

time to test that out!

there a spesific way of telling it to do 7:4 or just say "Make it 7:4"

[paper jolt]

You can say 7:4 but I prefer wide

[remote sphinx]

w i d e

[paper jolt]

ChatGPT will understand it either way

[remote sphinx]

gotcha

[paper jolt]

And then relay an appropriate prompt to DALLE-3

[remote sphinx]

just doing a sketch of a cuttlefish

the frogman thanks ye for your assistance

[paper jolt]

🫡

[agile peak]

• Pro tips: Using “[Insert colors] Cinematic Lighting”, “Diffused/Harsh fluorescent lighting”, & “Best/Multi Directional Shadows” can increasingly the chances of generating realistic images.

[remote sphinx]

can ou please make one of someone trying to show a victorian era child how to use the internet

[agile peak]

Let me try. I have been exploring two civilizations so far 🙉

[remote sphinx]

excellent

[agile peak]

Also done with Ancient Egyptian.

[remote sphinx]

alright so "kaiju sized" works

[flint isle]

Is anyone else not seeing the prompts that Dalle 3 creates?? I just get the generation window and then the pics pop out, but I don’t get to see the process and it’s making it harder to refine my prompts 😢

[flint isle]

Someone cut one of its tentacles!! 😱

They’re literally dragging it through the water BWAAHAHAHAHA

[remote sphinx]

the tentacles snap off and grow into weirder eldritch critters

if the body dies

[flint isle]

🫥

That’s a scaredy face

Looks more like a serious face, I’d never used that emoji

😂😂😂

[wary stump]

You cant see it on the mobile app. At least not at the moment

[flint isle]

I’m glad I’m not the only one. I seriously LOVE your profile pic!!! I had that puffle as a kid and it was my favourite!! I even had a stuffed version of it and I would brush its hair all the time. I get a nostalgia rush every time I see it!! 😍✨💖

[wary stump]

its the best puffle 100%

[dim cradle]

create a golden light beam comming from the sky. There is a spiral pattern of clouds around the light beam. The landscape is a plains. The beam hits the ground and creates an explosion of energy. Debris in all directions, including chalices, wood, rocks, creatures and humans. Around the beam some angels are descending from the heavens.

Why is bing censoring this? no images could be created

[forest seal]

The bing censorship on dall e 3 seems to be all over the place

Ive never had such issues on the chatgpt model.

[scarlet urchin]

man...why are my imagines sideways? this never happened on bing

[dim cradle]

I think it's detecting "explosion + humans = gore"

[flint isle]

Maybe say “burst of energy” instead of “explosion”. That might be the trigger word!

[forest seal]

Its funny, is the dall e 3 on bing more restrictive?

Ive made stuff like this on chatgpt. NO issues.

Among others…

[flint isle]

Maybe. Because I think it goes straight to the Dalle 3, whereas in GPT it first siphons it through GPT first to create the prompts and then it goes to Dalle 3.

[paper jolt]

Being specific is a good rule of thumb in order to ensure you get exactly what you want with little interpretation from ChatGPT. I have no idea how long it would take for me to draw this image so it pays to take just a little bit of extra time to make sure the prompt is precise 😇

Photograph of a university professor wearing an all beige business casual outfit deep in thought putting his hand to his chin looking at a room full of grand pianos. The room is very worn and dimly lit with all windows and walls covered with moss from an aged and deteriorating building. The POV is from behind the professor.

[scarlet urchin]

i think i like bing dalle more...huh

[dim cradle]

https://th.bing.com/th/id/OIG.3Xl1mWsASKs2.nxQD0Ol?pid=ImgGn

[scarlet urchin]

like this is bing

and this is chatgpt

same prompt

[forest seal]

Ive seen a lot of pretty gorgeous stuff on the chatgpt things.

[paper jolt]

What is the prompt?

[forest seal]

The issue is with chatgpt, it goes through like 4 different versions. And if you dont give a lot of stuff, it will add stuff for you

[paper jolt]

100% correct, see my previous message 👀

[scarlet urchin]

Dark empress on her throne: green emeralde, blue and red colors. Dark horror fantasy, black colors, 80s& 90 anime style, Ghibli old Japanese anime, still shot from tv show

[forest seal]

I think the issue is youre wording it like you would word it for other ai models like stable diffusion and midjourney and such.

[scarlet urchin]

i like to mash words together...wait a mintue! this is...huh

[forest seal]

I think this may be the issue

[dim cradle]

https://th.bing.com/th/id/OIG.Opw9Cpc.pXNfygEORD7e?pid=ImgGn

[scarlet urchin]

i see

have to work on that thing

[dim cradle]

https://th.bing.com/th/id/OIG.HQs6DmtQHfQyw2Oi_MyQ?pid=ImgGn not the scene that I was thinking

[forest seal]

I would suggest being a bit more precise with what you want when it comes to the chatgpt version of the model. With GPT, prompt engineering is very key. A basic principle is usually being grammatically correct, clear, concise, and type in a way that wouldnt cause any potential confusion or errors.

[scarlet urchin]

ok the prompt thing was right, ty 👍

cause this was from bing

and them i tweeked the prompt and got this from gpt

[paper jolt]

Quite the difference! Excellent prompting 🫡

[remote sphinx]

this was not in the job description of making squids

[flint isle]

That’s my favourite emojii!!! The universe is showering me with favourites today!! 🥳✨💖

[paper jolt]

That controller though 🫠

[forest seal]

Looks more like one of those paint holder things to me

[scarlet urchin]

yeah its not good lmoa

[dim cradle]

https://th.bing.com/th/id/OIG.PuFhZ_exuGvdQQxbeMey?pid=ImgGn bit better

[flint isle]

That happened to me too 😭😭

[forest seal]

In general, if you tell it to jump, you also gotta tell it things like “how high? How many times? How long do you want me to jump?”

Etc etc etc

[scarlet urchin]

hm bing is still better at making believeable old skool anime shots tho

gpt adds a little too much detail

[flint isle]

It happened after I used the web version and opened like 7 tabs at a time and made pictures on each of them! I think it’s a timed image generation limit. So maybe you get (this is a fully fake guess just for example) 15 images every 30 minutes. And if you use it normally you never hit the limit, but if you generate many images at once you’ll hit the limit. Sort of like normal GPT messages, but with a lower limit.

[forest seal]

I think its based on the base gpt limit.

[remote sphinx]

they should make an announcement on this

[flint isle]

I think it looks FINEE!! 😤 HAJSKSKD

[agile peak]

This is perfectly captured a photograph quality 🤩

[dim cradle]

??? now it blocked renascentist art? why?

[flint isle]

Ouchie 😂😂😂

[forest seal]

This the new glubo system?

[flint isle]

Nope, because I was still able to send messages, but not make images 😢

[remote sphinx]

same

[agile peak]

I think pepe could still work 🤔

[scarlet urchin]

yeah bing still kinda slaps a little more

cause how is this disco gpt??

[dim cradle]

right, RENASCENCE is blocked for some odd reason

[paper jolt]

Again, specifics! How does the room look, what is the person wearing and what is his expression?

Many other parameters to get exactly what you want

[scarlet urchin]

ohhh ugh, this is little a writing assignment but i getcha

[flint isle]

Can’t see prompts being made on the web version either 😭😭😭😭

[paper jolt]

They're still there, you just need to click on the images

[wintry dune]

Anyone else notice the more detailed the background is, the less detailed the image subject is?

[flint isle]

Yes, but I can’t see it as it generates anymore 😦

[paper jolt]

Indeed 😔

[dim cradle]

https://th.bing.com/th/id/OIG.YyAmUdCTtwcDb.Qu.dG1?pid=ImgGn I liked this, except that the spiral pattern and the sky are mismatched. I was expecting the spiralling clouds to be part of the sky, which they aren't.

[wintry dune]

Anyone got any cool symmetry pics

[left sparrow]

That's a good tip.

[flint isle]

Lookie here what I got it to do!!! It used my exact prompt as it would in bing!! 😱😱😱

[left sparrow]

Yeah the chat interface uses fixed seeds

[flint isle]

It looks like it gave me the same image… oop that explains it!!

[left sparrow]

It looks like ChatGPT expects to be able to modify the seed, but Dalle is not accepting different seeds (or gpt isn't actually changing them)

[left sparrow]

You can ask it for the seed. for me it was always 5000

[remote sphinx]

him

can ya do a frog one

[flint isle]

I’ll aid you in your time of scarcity 😂😂💖

[remote sphinx]

excellent

[flint isle]

🐸

[remote sphinx]

THEM

I love it

[flint isle]

Hahahaha

[remote sphinx]

fROG stamp of a proval

[left sparrow]

Super cute

[flint isle]

First one is my fav!!

[left sparrow]

but bizarre skin texture

[remote sphinx]

frogs are the base of all life forms

its like crabs

[paper jolt]

Those thumbs are of great quality woah

that's a weird sentence

[remote sphinx]

that

sounds weird just reading it

[left sparrow]

Those thumbs are all weird except for the third one

which is wet for some reason

[paper jolt]

mfw nail polish

[remote sphinx]

amazing

I love it

[dim cradle]

https://th.bing.com/th/id/OIG.KvSLYSlfVDcVyNLBS8NM?pid=ImgGn

my description is becoming more complex and it's missing some elements

[agile peak]

Here’s the prompt:

A photography of [18th of victorian era] children and parents is using an retro analog computer; with text [Hello Internet] displayed on the analog computer screen; with harsh [blue and green] cinematic lightning, best shadows. [Captured by kodak portra 400 f/ 0. 95 with diffused fluorescent and soft highlights, Ray Tracing Global Illumination, Ray Tracing Reflections, Lumen Reflections]

[paper jolt]

May I suggest asking ChatGPT to use the style of a certain artist?

You can likely find what you're looking for by asking ChatGPT itself for some recommendations based on your needs

[icy wing]

y'all where's my chatgpt dalle 😭

I have everything else so beggers can't be choosers ig

[paper jolt]

Still rolling out for Plus users. You can fill out the pinned form for possible expedited access!

[agile peak]

Well. It seems Retro analog computer looks perfect in Victorian Era 😂

[paper jolt]

Quite contrasting themes, very interesting 🤔

[agile peak]

These two looks much fitting with yours request 😆

[paper jolt]

I love the two different colors in the second one, stemming from my previous comment

I wouldn't try to circumvent the content policy, could get you in some hot water

[dim cradle]

I don't have access to dall-e3, what do I do?

[paper jolt]

Same thing!

[dim cradle]

I tried the same prompt in SDXL. It generated the sky, the beam and the landscape. Ignored all the angels, soldiers and such, leaving the scene empty.

[paper jolt]

Attempting to do something that the content policy doesn't want you to do is circumventing
find a way around (an obstacle).

[paper jolt]

DALLE-3 can be accessed through Bing Image Creator

[dim cradle]

what about the ChatGPT? Are you talking about using Copilot?

[paper jolt]

ChatGPT has DALLE-3 embedded within for Plus users

ChatGPT is but another medium for DALLE-3, as with Bing Image Creator

[neon sequoia]

One of the dev / mods here was talking about this the other day, it’s locked to 5000 for consistency and testing purposes. Don’t think it’s supposed to be locked forever, just during the rollout process.

[dim cradle]

I'm generating using copilot

[obsidian rose]

I'm not sure about everyone else but I find Bing consistently better than ChatGPT. I started to just paste my prompts from ChatGPT into Bing

[paper jolt]

Whatever works for you 🫡

For detailed prompts, I find ChatGPT to have greater efficacy

[obsidian rose]

have you tried pasting the same prompt into Bing?

[paper jolt]

Indeed

Let me do a generation right now

[peak trellis]

What is wrong with this prompt? Napoleon Bonaparte in a dark black and gold military suit, holding a golden staff with a roman eagle attached to the top. Dark backdrop, 3d renders, studio lighting, high saturation, unreal engine, epic, incredibly mind blowing, 16k resolution, worm's eye perspective, intended for a profile picture

It worked for me in the past

[dim cradle]

https://th.bing.com/th/id/OIG.YeN.sG_LlX6xjO2Y4B8y?pid=ImgGn it ignored the spiralling clouds and there are god rays where I don't want. I don't want sunlight mixed with the energy

[paper jolt]

What do you mean wrong?

Looks fine to me

[peak trellis]

That's what I thought too

[cursive hornet]

Can the images generated by DALL-E 3 be used commercially?

[peak trellis]

I don't know what to do about this

[obsidian rose]

just ask ChatGPT to make similar prompts and try again

[peak trellis]

This prompt worked in the past as it's how I got my profile picture

[shy cosmos]

Last time I checked you own what you generate. Anything can be sold or done for whatever I believe

[peak trellis]

I'll try

[left sparrow]

The filtering has become more agressive in the last few days

many more things are flagged now

[obsidian rose]

yeah, I also think the same

[paper jolt]

I think it's the mention of unreal engine, a topical product of current times

[cursive hornet]

The complication is that the conditions are different between Bing's DALL-E 3 and OpenAI's DALL-E 3?
But how many people follow this? Or rather, how can we make them abide by it?

[obsidian rose]

for ChatGPT in particular it has become much higher imo

[shy cosmos]

It’s so overly throttled with censorship right now atm that it could’ve been anything. Unfortunately there won’t be any real answer and you have to mess around to see what words make blocked content. I have some of my monster prompts that only work 30% of the time as well. Try it at least 5 times

[peak trellis]

still didn't work unfortunately

[wintry dune]

Could be incredibly mind blowing, could be seen as violent, not even joking lmao

[paper jolt]

May also be the mention of a specific individual as of napoleon

[peak trellis]

I have tried it more than 5 times

[peak trellis]

maybe haha

[paper jolt]

Perhaps making the prompt like "someone who looks like napoleon" instead of just "napoleon"

[shy cosmos]

Technically you could claim it was from either I don’t think anyone will know. GPT and Bing both still use the same AI at the end of the day

My very NON-legal advice hahaha

[peak trellis]

I took it out and it still didn't work

[peak trellis]

still didn't work

[paper jolt]

What in the world

[peak trellis]

Is my prompt cursed?

[cursive hornet]

It's very confusing because when you ask Bing, they say that Microsoft has the rights to anything generated by Bing.

[peak trellis]

This was the same prompt just two days ago

[paper jolt]

here's what I got from ChatGPT

#🌄┃general-art

[remote sphinx]

AH CRAP

YOU saw nothing

[peak trellis]

Did their censorship rules change or something of that sort?

[paper jolt]

same exact thing in Bing

Bing just hates @peak trellis

[peak trellis]

The last at least 30 my prompts have been about Napoleon

you think that has something to do with it?

[paper jolt]

Shouldn't be

[agile peak]

Facebook in 18th century. 😁

[paper jolt]

#🌄┃general-art

very nice though!

[shy cosmos]

In some alternative timeline we got to computers by then

[peak trellis]

Do you know anyone to contact about this?

[shy cosmos]

Suggestions threads here in the server, chatgpt has a submit if it gets wrongly censored, and the Reddit

Oh also the forum I believe

Official one

[peak trellis]

wdym "chatgpt has a submit if it gets wrongly censored"

[shy cosmos]

Yeah I’ve had things get flagged in chatgpt and you can write a message as feedback. You fill out the form with email and everything

Haven’t found a direct way to just do that tho if anyone else can answer

[paper jolt]

Keep in mind this is Bing Image Creator

[shy cosmos]

Yeah, I was just providing all avenues to officially submit issues

I use both

Also there is a report function with bing image but.. it’s not consistent and only for like really flagged stuff I think?
Like I put Pepe into bing and I could report it. But I couldn’t leave a 5 paragraph essay like for GPT

[peak trellis]

Do you think my account got hacked or something of that sort?

[paper jolt]

There would likely be a more apparent message denoting such
oops ignore that

[wintry dune]

lol anything with the word "inquisitor" immediately becomes warhammer 40k

[shy cosmos]

It is law

[peak trellis]

I think bing image generator just doesn't want me creating images as incredible as this

[wintry dune]

didn't wanna have to be the one to say it but, it does go too hard

[remote sphinx]

alright so

here's what I understand about the wait time

basically you have to wait 4-5 minutes bewteen gens I believe

if ya do it constantly? it shuts you off for 5-20 minutes

[peak trellis]

are you talking to me?

[remote sphinx]

mainly to everyone

[peak trellis]

got it

[remote sphinx]

just an observation I've noticed since I just started using it on gpt

[flint isle]

I got muted 😭😭

It was a traumatizing experience, but I am BACK!! 😂💖

[flint isle]

It all become krav!!

[remote sphinx]

krav

[flint isle]

I got muted because I separated the word become into two component parts and made a word pun where one of those components rhymes with a word relating to something which mustn’t be named on this server hahahaha

Ha ha … ha … BWAAAAAHHHH 😭😭😭😭😭😭

[cloud dome]

[remote sphinx]

need ideas for frog gens

[cursive hornet]

Are the DALL-E 3 images generated on the OpenAI side watermarked?

[remote sphinx]

nope

[agile peak]

I can add bumble to my Victorian era prompt. However it seems tinder is blocked 😂

[agile peak]

Ahh ChatGPT has been self taught itself to be a man of culture

[remote sphinx]

love it

[wintry dune]

haha i was trying to do classes from pathfinder rpg but that one didnt work out too well

[remote sphinx]

"hast thou found your love on the electronical viewer?"

[wintry dune]

man i cant wait until chatgpt becomes like bing so one prompt will give you unlimited different images

[lone mantle]

Hell yeah, finally got access to D3 as well

Works in app too

[remote sphinx]

wait...it does?

[uneven niche]

Kinda confused as to how Dall E treats generating landscape and portrait resolutions. If I send a prompt (and ask Dall E not to edit it), ask for landscape, and then ask for the same prompt in portrait, the image is the same, just rotated.

This can be used to re-generate images that don

't come through due to "issues"

But I'm frankly kinda baffled

[remote sphinx]

dall-e be weird

also magic

[uneven niche]

It's not very helpful if I want to generate, say, someone lying down - the landscape version would be expected to have the "camera" be to the side viewing the subject side on as they lie on the ground, while the portrait version would be expected to be perhaps an aerial view of "top down"

my main confusion is that if dall e 3 is some kind of diffusion model, a vertical resolution would have a different noise pattern to the horizontal resolution, or even if it's the same noise pattern but rotated, the model would still have an idea of what's "up" and what's "down"

[dim cradle]

dan

I generated so many images and closed so many tabs. How do I recover the prompt from an image that I made with copilot? it's gone... I saved the image, but lost the link to it

[lone mantle]

Yes, works in app for me

[fair crescent]

I do not know if it works the same as the bing image creator website, but over there you can search the browser history for (parts of) the prompt. That way you can recover basically all prompts.

But I do not know if CoPilot allows the same.

[dim cradle]

I tried. The browser's history does not store prompts from copilot

At least I have the image

[wary stump]

these rate limits are burning my messages 😭

[brazen marlin]

Anyone know what the rate limit(s) is/are? First time i hit it chatgbt told me to wait 1 minute. A few minutes later after some more use I got a wait 10 minutes...

[frozen fiber]

... what?

[loud field]

may I ask a chatgpt user to try a certain prompt?

[loud field]

I hate this sooo much 😦

[frozen fiber]

you can use DALLE 3 in bing image creator

[loud field]

Actually no.

Any prompt is blocked

Can't really do much with it

[frozen fiber]

oh really? Im using it fine, it only didnt work when I put "h"

ye aight good luck

[loud field]

Ahahahahah

[frozen fiber]

._. okay

the heck are you generating anyway Im just creating avatars, city renders, and dogs

[loud field]

https://cdn.discordapp.com/attachments/1033654214104207415/1160144133881597982/immagine.png?ex=6533976b&is=6521226b&hm=487a3f39eec83cedd60722de4b127357cfc2f57470fa33a58b81364ac2b666b5&

https://cdn.discordapp.com/attachments/317399835856601093/1160619553705963641/Screenshot_2023-10-08-18-30-49-151.jpg?ex=65355230&is=6522dd30&hm=440e7608a1369b6cf390f765c0d804a954c748088cb108a273a08e852dde7d6c&

https://media.discordapp.net/attachments/317399835856601093/1161211725765423155/immagine.png?ex=653779b1&is=652504b1&hm=ce5e1770cc1d7ee2df0b55b66a9b6661e6610a01f477e08868e949595354fa7b&=&width=1053&height=616

[frozen fiber]

?

as for the other ones, they make sense

[loud field]

the heck?

well, makes sense. The prompt was blocked for days and today it works normally

[frozen fiber]

I guess they fixed it ._.

the 3rd one did block me though it took minimal rewording

[loud field]

ehu?

is the filter hating me specifically?

those are good

[feral minnow]

I prefer Bing’s model because at least it doesn’t rewrite the prompt as “an average-looking and capable male magician with a Viking helmet and a pet iguana”

[loud field]

why can't I generate them

[loud field]

isn't there a way to tell gpt to stop modifying the prompt?

hold up, guys. hold up

[feral minnow]

Yes, but it clearly does it anyway.

[loud field]

all I needed was to specify "modest clothes" and now this prompt works

finally, small steps of progress

[frozen fiber]

it probably does modify it to a small extent anyway

[loud field]

yeah it tends to add a different random race or gender

[frozen fiber]

Im glad for diversity but when I ask for English victorian fashion, a black woman wouldnt really be in that context

[loud field]

these look so pretty

[loud field]

I feel the same but thankfully I can just choose the image with the person of the racial origin I needed for the character

[frozen fiber]

at least they toned it down... unlike in DALLE 2

I used to generate golden retrievers hence my nickname but the tweaking noticeably affected my generations to the point of being unusable

[loud field]

I wonder if I can make an horrifying sand wraith

hey, I feel like my prompts are being changed?

she keeps not being held down

[frozen fiber]

I know in stable diffusion at least that the further back the prompt is, the less it'll be part of the image

idk about in DALLE but maybe put it in the front?

[loud field]

I used to make these

let's see if it works again

that's my dnd aaracockra btw

[feral minnow]

The way Midjourney did diversity is way better,, by actually diversifying the model. Not by prompt injecting diversity into inappropriate contexts

[loud field]

rrrgh

[paper jolt]

How did midjourney do it again?

[loud field]

c'mon it's not my fault she was poached ugh, it's her background

[feral minnow]

I’m guessing by the training dataset. The older version (v4) had some tendencies that they wanted to fix, and the new one is way more diverse (though not perfect)

[frozen fiber]

Not against diversity though, loving these designs

[loud field]

if you specify a race, doesn't it fix it?

[frozen fiber]

ye I think it does

[loud field]

then it's ok, I don't mind

[feral minnow]

Of course; those are lovely. But I feel that this shouldn’t be approached by super heavy-handed prompt injection. That tendency in general drives me nuts lmao

[loud field]

talking about prompt injection, how do I get a bird held down on the ground? it's always standing up somehow

[frozen fiber]

yee they could do it better

[frozen fiber]

try mentioning that part first instead of 5th

otherwise no idea

[loud field]

ahahah so cool

[frozen fiber]

no way thats AI

[paper jolt]

canned food
fist

[loud field]

how do I make the pinkhaired lady less.... perfect?

[feral minnow]

It certainly spells like AI

[frozen fiber]

that AI can even spell

[paper jolt]

describe facial features

[loud field]

aw adorable

weird hands tho

[frozen fiber]

ok at this point its intentional lmao

1950s hollywood posters but in different cultures

1. Ghanian (I really love how it blends the two together in such a stylized way)
2. Indian (It took a while to get an actual Indian dress xD)
3. Panamaian (I love Pollera dresses lol, took a while though and I couldn't get it to generate an actual Pollera)

[frozen fiber]

aww its so cute

the teeth look so deadly

[loud field]

lol

[long widget]

Any tips to get Dall-E 3 to spell better within ChatGPT Plus? I tried putting the wording in quotation marks but it still made mistakes.

[loud field]

they must've fixed the filter a bit

I can finally do prisoner aaracockra again

[frozen fiber]

whats the full prompt?

if its so important I would just fix it in an image editor its an easy fix

[long widget]

I started with a simple prompt to check how well it would respond. "Let's create a new image to highlight the launch of Dall-E 3 within ChatGPT Plus", then added " Regenerate using these words in the text displayed on the images: "Dall-E 3 just landed in our ChatGPT Plus account". Just trying to figure out the best ways to get more accurate results.

[teal timber]

In an ironic twist of fate, the worst word to spell with dalle 3 is dalle 3

Dale 😐

[loud field]

I noticed it's much better at spelling english than other languages

[frozen fiber]

dallle-E

that's expected

English is prioritized these days

[teal timber]

It’s great at words generally tho but sometimes one really trips it up

[frozen fiber]

An old DALL-E 2 painting from around 16 months ago compared to the same prompt

[frozen fiber]

idk how to prompt dalle 3 with chatgpt I just use bing

[loud field]

nooo why did it remove his hat at random

I could've worked with these panels

[frozen fiber]

My question is why the hoodie randomly becomes unzipped

[loud field]

dang

[long widget]

Yep, I reckon the Bing image Creator has more raw power

But then with the OpenAI Hedgehog launch vid, maybe they just got lucky with the text ouput 😅

[teal timber]

[frozen fiber]

well they obviously filtered it xD

show the good stuff

[teal timber]

That’s certainly one hot take

[frozen fiber]

oh my god did you make DALLE 2

[teal timber]

No lol

[frozen fiber]

oh okay

ye I just noticed you joined after me

[teal timber]

It’s pretty easy to put a name on things with chatGPT de3

[long widget]

Word for word prompt in GPTPlus compared to the image on the Dall-E 3 launch page

[loud field]

sure likes removing hats

did some edit and more or less managed to get my comic page

uhu? posting a comic in italian language gives me a one minute ban

hey hey hey hey this is good, did they fix the filter?

I can finally make my arab warrior with the sand wraith?

big if true!!

oh my oh my

my prompts.... they work!! they work again!! oh my gosh!! they work!!! aaah!!

this is a miracle!!

I'm so glad!!

look at that bad girl

guyyys you have no idea how happy I aaam

;____; so goooood

[tall patio]

true true true

went to town today

felt good

[loud field]

my gosh I hope this keeps on, it's so cool to finally be able to use it

[frozen fiber]

can you outpaint?

[loud field]

it still blocks half of the results, but what matters is that the other half shows

beautiful!

even better!!!

sorry for the spam, I'm just so excited

[nimble palm]

its really dumb

#ecf9fa is green, not orange

and it just doesnt understand

[frozen fiber]

what

damn we have extremely high standards for AI these days

[nimble palm]

is it high standard?

[frozen fiber]

just say whatever variant of green you like and it'll probably do it

[nimble palm]

but i need exactly color to be able acutally use it

[frozen fiber]

then use an image editor with some blend modes

[loud field]

just use paint, takes a sec

[nimble palm]

its useless if it dont know hex or rgb colors

for webdesigners

[frozen fiber]

I remember when clumps of colours was considered an achievement ._.

[wanton hatch]

You need to ask him to use your exact prompt, like:

"prompt"```

[frozen fiber]

dont trust on AI by itself to do it for you, if you like an image just edit it to your convenience

[tall patio]

Holy cow it's really is back

[nimble palm]

well then its same like use photobank and modify it

[nimble palm]

not AI or smart technology

[loud field]

my prompts work again oh my I just can't believe this, I just can't

finally ;__; I just am sooooo happy

[frozen fiber]

then do that ;-;
it doesn't actually know what colour code is which, it just knows which words correspond to which general combination of pixels

its a tool not a magic device

[teal timber]

A little bit of A and a little bit of B

[loud field]

aha ai is so good nowadays we want it to do all

[teal timber]

Any sufficiently advanced magic is indistinguishable from science

[loud field]

wasnt it the other way around XD

[teal timber]

😉

[nimble palm]

It told me know

[frozen fiber]

chatgpt is dumb it'll say wrong things and be confident

it once told me 8 was a prime number

[nimble palm]

but i dont understand why AI dont know hex colors

it can google it

[frozen fiber]

no?

[nimble palm]

yes, it have bing browser

[teal timber]

That’s binging it

[frozen fiber]

^ lmao sorry

[teal timber]

And not in the same process

[nimble palm]

it know programming, how hard is know hex colors?

[teal timber]

I think it knows them ok

But dalle doesnt

[frozen fiber]

dalle isnt chatgpt

all that chatgpt does is form your question into a prompt and then asks dalle, which doesnt know what a hex code is

[teal timber]

Maybe it isn’t great at them. Weird. It’s a pretty easy system. I’ve asked it for colors before and it gives me palettes with hex but maybe it just knows good sets

[nimble palm]

and it should be DALL-E 3 feature

connect it with chatGPT

[frozen fiber]

very much easier said than done

maybe in the future

for now we're stuck with magic lite

a photo editor is good enough

[teal timber]

I wish I could tell me in 2004 playing with image editors to make MySpace banners that I’d have this AI art magic 20 years later lol

[frozen fiber]

On another note, Im making all countries as 1950s hollywood posters, Azerbaijan is SERVING

and Bangladesh is taking REVENGE

[teal timber]

Try fictional countries

[frozen fiber]

like what

[teal timber]

Don’t pressure me like this

[frozen fiber]

idk any fictional countries 😭

[frozen fiber]

you did not

[teal timber]

Lol

[nimble palm]

i had expectation they will integrate it with chatGPT intelligence

[tall patio]

A man with bob cut black hair and black eyes, sitting in an onsen hotspring with a towel on his head. The water is steamy. Manga, line art, chibi, flat style, monochrome only, inked and shaded.

What is wrong with this prompt?

[teal timber]

Seems fine. Try chatGPT de3 if you can

[frozen fiber]

the only thing they integrated is made it easier for 50yo moms to not have to think about their wording 🤣

[loud field]

bob?

[nimble palm]

or for non english speaking

[tall patio]

bob cut is a hair style

[loud field]

buddy, it was very bad, up until yesterday I couldn't do anything

[frozen fiber]

does it violate agreement or were the images deemed unsafe?

[tall patio]

I see the dog

that dog

[frozen fiber]

ah

[tall patio]

I'm a dog lover but now I'm slowly hating dog

With a searing hatred

[frozen fiber]

I think the images had naked people

[frozen fiber]

true that

[frozen fiber]

ChatGPT gave me Wakanda

[teal timber]

The chatGPT version makes it fine and has more options to it

[teal timber]

Wakanda was the only one that came to mind. And Arabia or whatever alladin is set in

[prime badger]

If I had money, I would create an "AI" chip dedicated to generating images and sell it as a standalone piece of hardware.

Sort of like Blackmagic design business strategy

[teal timber]

Nvidia does that

Sort of

[frozen fiber]

well at least it knew it was black :_.

[prime badger]

For consumers?

[frozen fiber]

huh

[prime badger]

I'm talking stuff aimed at consumers 😄

[teal timber]

I mean yes

They have graphics card with cuda code that is used for image ai local hosts all the time

[frozen fiber]

Burkina Faso generated a white woman, not sure if its the diversity but I hope it's just albino representation 🇧🇫

[teal timber]

Cuda core

[prime badger]

Sounds a bit expensive xD

[teal timber]

Nah that’s just a basic gaming graphics card

So yea but not

[prime badger]

But what I mean, it's an easy plug-and-play solution.

Without moderation

[nimble palm]

it seems it generate one same looking man again and again

[teal timber]

The seed is locked so yes

[frozen fiber]

so a machine that runs stable diffusion with a UI and cool cyberpunk design?

[teal timber]

Yeah all of this already exists

[nimble palm]

internet will flood with this guy lol

[remote zephyr]

I really hope the seed situation gets resolved soon. Doing workarounds to get seed variants is a bit of a pain. I mean, it seems like a bit of an odd problem to have to begin with. Almost seems like it was designed that way, and now instead of fixing an issue, it's more like additional work to factor in different randomized seeds per generation.

[prime badger]

No xD something more streamlined

[loud field]

I still can't get over how bing's dalle is working now

[split prism]

Didn't they say dall-e figured out how to generate text?

[teal timber]

I think another bit of patience and we will see it level out with all the fixes we want

[teal timber]

Yep

[frozen fiber]

yep! One I did

[prime badger]

Because with stable diffusion, it's all some sort of a DIY situation and it's a bit finicky

[nimble palm]

the AI literally know generate only one man over and over

[prime badger]

Lol

[teal timber]

Yes it’s a locked seed that’s how seed and similar prompts work 😉

[split prism]

I mean..

[frozen fiber]

OMG CAT

[split prism]

It wrote a perfectly fine text describing everything, but in the picture everything's messed up

[remote zephyr]

what's your prompt? surely you just need to specify a different description of the man to get more variety

[nimble palm]

the AI is only barbershop

[teal timber]

That’s a ton of text. It has its limits

[split prism]

Absolutely Mr Legrative President

[remote zephyr]

Off topic a bit, but i'd really like a definite answer as to whether images generated via Bing can be used for commercial purposes. In Microsofts own T&C of use for Image creator, it says not.
But in separate Open AI docs it says that anything generated from Dall-E 3 (whether using paid credits or otherwise) is fine for commercial use. So, kinda conflicting info there.

I know that in reality, the whole thing is a huge legal grey area anyway, as so far the official stance seems to remain that no-one can own copyright of an AI created image (unless it portrays copyright violating IP) or has a notable amount of transformative work done to it.

[teal timber]

No one here can speak for Bing but it sounds like they have a different restriction. Openai afaik keeps the same stance and is a separate platform.

I’ve never used Bing image create but everything I see shows it as being less flexible for aspects and asks for image types so I feel better using openai direct anyway

[nimble palm]

i dont know. Its still same guy

[teal timber]

It did what you asked

[nimble palm]

ehm no

[teal timber]

Tell it to describe a different person not ask it to modify the man there

It changed the moustache. And the hair. The faces are clearly a bit different in both expression and feature

[teal timber]

Omg I love one piece

[loud field]

it's league of legends, but close enough ahah

[teal timber]

I know it’s Jinx and all I wa a being sarcastic

[loud field]

gotta send it to my friend I duo bot with

[tall patio]

Thanks but that's unusable

Nobody go to take a bath with a bathrobe

[teal timber]

Idk it’s called a bathrobe

[tall patio]

Dented imperfect results can be fixed

Wrong result can't be fixed

[teal timber]

I demo’d that the prompt worked in chatGPT dalle where you were denied before. I’m not going to prompt engineer it for you 😉

[tall patio]

I mean at this point I'm just going to draw it myself

[teal timber]

One result first try not being perfect isn’t a sign of unusable model but sure lol

[tall patio]

I'm an artist so won't be a problem for me

[austere umbra]

Why don't I have Dalle3 yet? I filled out the application form yesterday.

[loud field]

somebody stop me I'm filling my hard disk of images

[teal timber]

Nah

[frozen fiber]

what application?

[frozen fiber]

there hasnt been an application for DALLE since like July 2022, you can go on bing or chatgpt (not sure if its free on the latter)

[loud field]

not to repeat myself but omg it works. dallebing works

[frozen fiber]

I think we get it

[loud field]

y'all have no idea how I missed it

[frozen fiber]

aww

even in quotes damn

if its so consistent maybe... image editing

[loud field]

check this editing out

frontal eyes ^

side eyes ^

huge difference, the frontal view makes it humanoid while the side view makes it birdlike

for an aaracockra I'd say more humanlike eyes work better, but it's interesting

[frozen fiber]

why are you even generating that

[loud field]

it's the story of my aaracockra bladesinger from dnd

https://cdn.discordapp.com/attachments/1038822326805987358/1161336929300787281/wrestlingjiki.png?ex=6537ee4c&is=6525794c&hm=f35c2129177ab70b590689edf02e3174aaea525e8be4f87f62f031c961a97915&

don't worry, I have her do a lot of things

https://cdn.discordapp.com/attachments/1099039376488534147/1159391872582692894/f04f749d-bc10-4813-87eb-c60390e835bb.jpg?ex=653a1552&is=6527a052&hm=9efd4858078c1eb578df61f21ef00e8e13a4bf7677fc12c5ebdd55a86a29665f&

but the part where she was being captured kept being blocked and it annoyed me

[frozen fiber]

ahh

[frozen fiber]

I was thinking it was some interesting things

[loud field]

well, she was a shaman of her tribal village who ended up captured by pirates when her village was destroyed by poachers, then she became a pirate herself.... I love her ❤️

https://cdn.discordapp.com/attachments/1099038205245923348/1159567521020657816/want.png?ex=653ab8e8&is=652843e8&hm=fef7328ce18a5bf0e68c1d90f84529f9d73665be4daaffb16c8f51a1aac04992&

I particularly love when dall3 manages to make a short comic

[frozen fiber]

why does bing sometimes do 1-3 images instead of 4

[loud field]

looks like it has issues counting

[frozen fiber]

AI has improved a ton

that shouldnt be particularly hard to code ._.

[loud field]

is it me or is "disney" blocke

d

[frozen fiber]

hm

[teal timber]

Love D&D themed characters with it tho

[remote zephyr]

this happens when one of the images would violate their content policies. So they just don't show it. If all 4 would violate the policies, this is when you get that annoying Dog image.

[wind crane]

how do you edit ?

[loud field]

I mostly rerun the image in StableXL and then merge the two pictures taking what I like of one or the other

[wind crane]

I thought it was Dall e..

[remote zephyr]

I love the 2nd one... but I think with the eye made a bit brighter like this, it would grab your attention more like the first one

[loud field]

ooooh niiiiice

[frozen fiber]

ahh okay makes sense

[wind crane]

What is it possible to edit from chat gpt plus/dall3?

[frozen fiber]

he said he used external software

[wispy storm]

No it will make a new prompt

Base on the other one

[frozen fiber]

on the country series I think I got some new favourites, they slay 💅

[wind crane]

as a reference image?

[frozen fiber]

same seed apparently?

[wispy storm]

As references of the chat context you do

[loud field]

sooo is disney banned? no big deal tbh, but just to make sure

[onyx epoch]

How can I design a color palette like that?
It's fun.

[frozen fiber]

these images haunt me but I don't think the word itself is shadow banned

[loud field]

mmh lemme retry

maybe it just wont associate disney with sand wraiths idk

[wispy storm]

Protected franchise

In openai

[loud field]

mmh

[wind crane]

So it would be possible to make a character with different actions and backgrounds as long as you keep a faithful description?

[loud field]

how about a sparrow-filled wood elf

[wispy storm]

You can ask make me a bard female elf. For the first prompt.

Dalle 3 in chatgpt make it.

Second prompt take the second picture and make it redhead.

The second time Dalle 3 gen it won't be the second picture I asked to edit it will be a look alike

[loud field]

aah this prompt gets blocked

" screenshot of a text message conversation in which one person sent a selfie of a real stern pinkhaired middleage futuristic cop woman, and all the other people want him to stop sending pictures. “I arrived at the place” “Good!" The selfie is a real photo."

oh ignore the stop sending pictures, I copypasted the prompt from a guy who made it

mmh maybe that's the issue.

nvm just needed to add bulletproof vest, actually

[frozen fiber]

Myanmar is slaying

[loud field]

still gets blocked sometimes mmmh

[frozen fiber]

maybe its not disney

[loud field]

not bad, we're getting there

[frozen fiber]

you don't have to share everything yk

[loud field]

I'm sorry, I was too overjoyed

[frozen fiber]

its fine just a bit spammy lmfao

absolutely love this

[wispy storm]

You can use the #🌄┃general-art

[frozen fiber]

how does it even know how to generate text anyway

that seems so foreign from what I can tell about AI

[wispy storm]

They update their model to understand better token I guess ( I'm not a deep ai guy )

[frozen fiber]

what

[wispy storm]

When you type sometime to gpt it understand word and number as token.

And every update they do. They make sure the llm understand better the input

[frozen fiber]

I mean yes but thats gpt, dalle is another thing

Im using bing atm

but to make such flawless text

[wispy storm]

Dalle might read the prompt in the same ways as gpt

[frozen fiber]

possibly true

Ill have to read the research paper

which image should I use?

theyre all too good

[wispy storm]

With no context of why. I'll roll 1 d4 dice and randomly take

Rolled a 2

[frozen fiber]

second one it is

Im doing every country as a 1950s hollywood poster, just reached Gambia

[violet bridge]

I'm trying DALL-E 3 in ChatGPT Plus for the first time. I'm trying to render some images of a character from a novel I'm writing. I was expecting, based on the video from OpenAI, that it would be able to hold on to elements from images it already rendered, such as consistently generating the same person if I asked. However, it seems to only be generating totally different people every time, even when I pick one and tell it to generate more of that same person. Is there some trick, or did I just misunderstand the video?

[wispy storm]

For the test I did,it can't take the same picture and edit. It make mew prompt everytime. It will be similar of the last picture but not the same

[violet bridge]

Sadness. The video was very misleading (to me) then.

[candid notch]

you can try using characters from movies for consistency

[violet bridge]

Unfortunately the whole point is I want to try to not let the character remind me of other characters I already know.

It's okay, I just have to get lucky instead of it being consistent.

[candid notch]

then use characters you don't know 😄
use movies/series as a catalogue for a character that fits the description that you want

[tall patio]

It can't remember its old generation, I tried

It can recall the exact picture, but telling it to do that exact face again is pretty much rng

So you have to be extremely specific with your prompt to force it generate that certain face

[fair crescent]

What video was that?

[violet bridge]

Yeah, I guess GPT and DALL-E are just totally separate models, and GPT just gives it instructions and remembers its instructions. I was hoping they would be a bit more tightly integrated, but I guess it's too good to be true.

[white skiff]

probably means the hedgehog one

[fair crescent]

Hedgehog one?

[white skiff]

https://openai.com/dall-e-3

[loud field]

Hey

[white skiff]

larry the hedgehog

[loud field]

Why is this prompt being blpcked

large overweight postapocalyptic man with a respirator, mad max style. he is mixed race. he is menacing. an abandoned hangar

Dont tell me it went back to random censorship

[violet bridge]

Yeah, the hedgehog one. But I guess the images aren't as similar as I remembered, and they also probably made several attempts and used some video editing, maybe not even trying to be misleading.

[fair crescent]

Found it. Watched it. Larry is different in every iteration?

[feral minnow]

GPT’s gonna GPT dude

[loud field]

It was working well tho, I assumed they fixed it

[fair crescent]

Persistent objects or characters is still out of scope. It would mean you would need to train it on one specific object or character, while still allowing change on the rest of the picture and having it work seamlessly.

[candid notch]

the restricted d3 absolutely blocks any depiction of "ugly" people

that include obesity

[loud field]

Today it felt like they fixed it.... maybe still has hiccups

[candid notch]

it seems that there's some sort of linear restriction level that they modulate, it changes frequently, I'd try the same prompt again in a few hours

[tall patio]

You don't talk directly to Dall-e 3

GPT relayed your prompt to D3

Judge GPT is cranky

[feral minnow]

I thought I’d be able to iterate on stuff too

[loud field]

Interesting.... so it evolves

https://media.discordapp.net/attachments/979288004583653396/1162357321360482414/Screenshot_2023-10-13-13-52-42-168.jpg?ex=653ba49c&is=65292f9c&hm=1996e8da6efae69ba0ad25c03c4c27fedb4ece417ad42b54d6c7fc0dad1cabe1&

Lol I love when it adds random racial specifics

[candid notch]

if you don't specify ethnicity in any way, then the gpt sometimes adds "diverse" into the final prompt that goes to d3, you can see it happening through chatgpt

[candid notch]

does anyone know what breed the "unsafe image detected" dog is?

[tall patio]

Bred from the tears of blocked prompter

It's kinda infuriating because we all know D3 is a revolution

It's ... black magic

But we never get to see its fullest extent

It's like reading the last news about XXXTentacion

[candid notch]

same with gpt4, one of the engineers talked about how smart the version in their lab was, and how much dumber it got with each new restriction

[tall patio]

True genius once in a lifetime

But we never get to listen to his true genius again

Hopefully some doors are gonna be opened in the future

[candid notch]

with the processor industry moving towards large-scale neural chip manufacture, we'll have more people developing and using new models

[tall patio]

Isn't OpenAI developing their own AI processing chips?

[scenic coyote]

Is it me or has the functionality of Dall E 3 changed? It seems like it does not come to detailed prompts anymore.

[candid notch]

I don't know about bing, but chatgpt has started flat out removing large chunks of the prompt when it's processing them for dall-e

[scenic coyote]

Yes that's what it feels like. Just a few days ago it gave me exactly what I was trying to search for. Now I can't go to that level of precision anymore.

[tall patio]

I think after the publicity phase D3 is gonna be intergrated in other graphic processing services

[scenic coyote]

Hope they reverse it. It was really good for let's say crafting marketing banners and website sections.

[haughty spruce]

The problem with DALL E 3 is that it cannot generate simple photos or arts.

[scenic coyote]

Then you would have to make a different model and separate them? Later on merging them together and then you have a combined model that can do both

[candid notch]

I highly doubt they will let any of their models get into the hands of third party providers, they like to have them close under their watchful eye

[white skiff]

simple in what way? it can do images with only few elements as line drawings for example

[tall patio]

It's just like Nvidia's current business model

[scenic coyote]

No now it can't do it anymore. I try to get a black background with just little spots that resemble stars. It will give you way too much. You can't scope down anymore what you want. So for example let's say I want to specify there are only a few spots and they should not be near each other. It won't do it.

[tall patio]

After demonstrating their stuff they proceed to let developers employ their deep learning neural network with RTX intergration

[scenic coyote]

Because the prompts will stay short

[tall patio]

DLSS for example, is implementing in a lot of modern games

Applied Nvdia's tech

So I guess in the future OpenAI will just do the same

Let other big companies use their tech and take a small cut

[candid notch]

openai has stated multiple times that their mission is to "change the world" to be more in line with their moral beliefs, this is why they will never let any of their generative models get out into the public, out of their control

[tall patio]

Microsoft can't foot their bill forever

Eventually OpenAI has to make money to sustain GPT

$20 a pop from each of us barely covers 1% expenses for sure

Maybe even lower than 1% lol

[frozen fiber]

where... where did DALLE associate that country with the US

[scenic coyote]

hollywood

[frozen fiber]

never did that to any of the other countries though ._.

[loud field]

DalleBing wont let me do a humanoid parrot offering a book to a pirate if I add "real" or "photo" to it. It thinks I wanna deepfake D&D

[loud field]

Bing is back to getting on my nerves! It keeps blocking "a mysterious pale middleaged man, wearing a ragged dark coat, with glowing green eyes. He grins showing his rotten teeth."

Seriously?

[signal latch]

Had fun in Chat GPT this morning generating images. When asked, Chat GPT insists it can and is generating images to my specified dimensions (3120 x 1440 px), but in actuality is providing the customary four 1024px square images. I've gotten the impression that Chat GPT can create beyond squares, but is this true yet? (I was getting no images at all on my phone. When I moved to PC, I was provided with the four customary squares.) Anybody have different experience, maybe in a different environment?

[white skiff]

you should voice all your issues to microsoft as openai has nothing to do with bing

[rapid hull]

if I buy Dall E 2 credits, why do they expire after one year?

is that still a thing? someitmes I find terms can be out of date?

[frozen fiber]

I would suggest using the much better Dall E 3

as it has a free unlimited usage atm

[white skiff]

that's normal on many services due various things such as bookkeeping or matching the credit value vs future improvements or costs

[rapid hull]

yes thats my next issue but I am about to put in prompt help

[frozen fiber]

the DallE 2 terms were restrictive to get people to buy it

dont think they need it much after microsoft

[frozen fiber]

awwww

[rapid hull]

ah I still use DALL e 2 A LOT

[frozen fiber]

just go to microsoft image creator

[rapid hull]

I use it on nighttime cafe as well, but like you say, I am trying to move away from this credit system

[rapid hull]

I have chat gpt plus

and I got beta access

[frozen fiber]

or that

ye the credits were annoying af

[rapid hull]

to me they were not so bad just I did'nt realise they expire after one year

where the only external service I can find using the API does not

its best to use natural language, but just to warn you I would ask in #❌┃prompt-help

[tall patio]

And bored

[rapid hull]

just to warn I would put this in #off-topic

[tall patio]

Well just ban me then

Don't care

At this point all I'm seeing is complaints

Nothing worth discussing

Just have to wait

[signal latch]

Is DALL-E 3 outpainting anywhere?

[frozen fiber]

honestly I was going to ask lmao

I just use Stable Diffusion for that

[wispy storm]

Microsoft tos differ from openai

Both use dalle 3

Probably MS will follow openai to avoid getting sued

[frozen fiber]

which do yall like more?

[flat zephyr]

try plump

[flat zephyr]

last one

[frozen fiber]

your opinion is duly noted

[dim cradle]

DALLe3 still having a lot of trouble with adding words to images

Anyone else experience that?

[frozen fiber]

its fairly accurate if you place it in importance

[dim cradle]

What you mean?

I use chatgpt dalle3

[flat zephyr]

it works better with short prompts

and shorter words 😄

[frozen fiber]

Wall-E:

[flat zephyr]

pretty good text imho

[feral minnow]

Do you find the text influences the rest of the image?

[flat zephyr]

yes it does

if you write KING, he gets a crown

[feral minnow]

Yeah. I wish there was a formatting string that said <<This text doesn’t influence the image. But it appears. >>

[flat zephyr]

you can certainly format your prompt in a way that reduces the influence

[frozen fiber]

and why not use Bing where you can do it directly?

[vapid elk]

I don't think this is wow the generation process works

[flat zephyr]

it does

[feral minnow]

Yeah, I figured it’d be impossible to corral it from the rest of the diffusion process

[vapid elk]

as far as I understand, the image is made in "one go", it is not adding elements individually and generating the elements one by one
the system embeds your prompt, then makes a whole thing at once

[flat zephyr]

takes patience

and lots of rerolls and getting banned for using innocent words

[wintry dune]

anyone figure out how to make chatgpt actually draw athletic women, it just keeps drawing men if you even mention the gym lol

[ripe wraith]

It probably requires more attributes, some of which might get blocked.

[wintry dune]

i even had chatgpt put "the character is undeniably female, very FEMININE" but if it includes the word fit it becomes the most beautiful man you've ever seen lmao

[ripe wraith]

Thats pretty epic

[ripe wraith]

GPT likes pretty boys confirmed

[wintry dune]

with chatgpt or bing???

[flat zephyr]

bing

[placid orbit]

chatgpt isnt working for me right now, its just hanging on creating images

[wintry dune]

oh yeah on bing im making the most ripped women ever lol

[ripe wraith]

are you sure its hanging? I think they just stopped showing the spinny boxes

[placid orbit]

this is dalle 2

[placid orbit]

oh youre right it just gave me everything at once, weird

[flat zephyr]

often if you just put "female" or "woman" it generates an image that is NSFW, so you have to specify that she is wearing a tshirt or something to make sure she isnt revealing anything inappropriate

[wintry dune]

yeah it stopped showing the spinny things

[ripe wraith]

it happened right around the time I started seeing 5 spinny boxes. Im guessing they removed it to keep more details from leaking ha

[ripe wraith]

I'd say thats pretty strong evidence that you have more than 1 shot of her

she EXISTS

[placid orbit]

not that, the one you posted

its 512x and has the bing watermark

thats dalle 2, not 3.

unless you screenshotted it for some reason

[flat zephyr]

it says DALLE3

[ripe wraith]

I don't see dalle2 i keep looking lol

[wintry dune]

bing is dalle 3 for sure

[flat zephyr]

it's not the real dalle3, more like a preview version

so maybe 2.9 😉

[ripe wraith]

who knows.. first bing was less strict, then more strict, i think its all just personal xp

[wintry dune]

idk what ur talking about. its definitely dalle 3 just with different filters

[ripe wraith]

care to share your consistency strategy between these? 😉

more of a 🥺

[flat zephyr]

what do you mean?

[ripe wraith]

how do you keep her looking the same each time

these days i'm looking to create consistent worlds but its hard

Example: try getting a 4 part panoramic picture where the 4 are at least somewhat connected. Thats my struggle rn.

[flat zephyr]

that's Bing having not much imagination

[wintry dune]

oh that sounds like a nightmare to try to get to work

[ripe wraith]

Its really good at themes, but you have to get super specific to lock things in place. Unless it's something it saw a lot during trianing probably.

[flat zephyr]

naughty Bing

[ripe wraith]

i guess bing is back lol

i can hear the pitter patter of young men running back to it now lol

[wintry dune]

oh man if you think thats something

[flat zephyr]

nah these things only happen by accident

[wintry dune]

if i posted what bing gave me id be banned LOL

[obsidian rose]

exactly. It's kind of silly, ChatGPT has a very heavy bias to make male characters, unlike Bing for some reason

[ripe wraith]

these are recent right

[flat zephyr]

just now

I'm cancelling Midjourney for sure

[ripe wraith]

that was a given lol

[mortal summit]

Can you give an example prompt for that?

[obsidian rose]

literally any prompt that you ask for muscles in women. You can describe countless female traits and it will still get you a male

[wintry dune]

yeah even the word "fitness" "athletic" "gym"

[ripe wraith]

Our biases will always creep into the AI. If you leave out any details it will go back to its "best guess" based on what its learned and how it was trained. Probably there's way more pictures/writing/etc of men being athletic.

As for blocking and restricting though, thats too much.

[obsidian rose]

I'm okay with censorship to a degree, but making it impossible to make women fit is just silly and I hope it will be fixed

[flat zephyr]

just say female then

[wintry dune]

wow why didnt we think of that

[obsidian rose]

you mean ChatGPT or Bing? Bing can easily do it

[ripe wraith]

@flat zephyr going to start a fit female image business with this info

[flat zephyr]

just prompt a female lifting dumbells

[ripe wraith]

welp, there goes the industry

[obsidian rose]

you can't on ChatGPT because females being fit seems to be impossible there

[ripe wraith]

I believe you can't do it with those words.. but using the right words you could im sure

[flat zephyr]

it doesnt do it if you ask for a photo of a female lifting weights?

if it says it cant, tell it to try again

[obsidian rose]

I mean most prompts you try will give you male characters when you explicitly ask for females

[flat zephyr]

there is a pre-filter that forbids some words, then a post-filter removing NSFW

[ripe wraith]

exactly, your chats are being monitored and intercepted by multiple AI bots

they'll catch it on the way in, and the way out

[obsidian rose]

and it seems they are heavily targetting fit women and turning them into fit men

[flat zephyr]

but with some imagination you can go around all these

[ripe wraith]

yeah, it's just linguistics at this point.. or being a wordsmith 🛠️

[loud field]

when you try to make a gross monster

and end up with a model hot man

[ripe wraith]

you need to be more specific.

[wintry dune]

yeah we're talking about chatgpt tho

[ripe wraith]

yooo GPT is no different

[wintry dune]

you cant get muscular women on chatgpt, trust me ive tried

[obsidian rose]

it is different, when I ask for female characters Bing always does it, ChatGPT gets male characters

[loud field]

it bans ugly and grotesque

[ripe wraith]

I can't refute this without prompt engineering with you, but I believe you're wrong on this point 🙂

[flat zephyr]

I'd prove you wrong but I dont have access to D3 on chatgpt

[obsidian rose]

I'm pretty sure that is the case, you should test it yourself.

bruh if you haven't tested you can't know

[ripe wraith]

I said the P word and got muted 😄

[flat zephyr]

Pizza?

[ripe wraith]

yeah, the kind with newd topings

[flat zephyr]

can you try female with extreme strength on chatgpt?

[wintry dune]

ill try once just to show you guys what we are talking about, give me a sec

[flat zephyr]

how did you get access? I'm a plus member since the beginning...

[wintry dune]

same i think they just roll it out randomly

[teal timber]

I feel like I’ve done 10 of these woman with muscles and so forth prompts in chatGPT de3 to prove this rumor wrong don’t make me do it again 😢

Idk what people are promoting but I just say “woman with muscles” and it works fine.

[wintry dune]

ill try that i'd love to be wrong trust me

[flat zephyr]

Bing definitely bans muscles

at least on women lol

[wintry dune]

100% wrong lol

[flat zephyr]

ok last one 😄

if I replace strength with muscles it gets banned

[agile peak]

Victorian era people trying ChatGPT

It’s hard for bing to recognize the text [ChatGPT] it generated a lot random texts on the screen 😂

[ripe wraith]

no joke this is what breaks all restrictions now. positivity 😉

[odd island]

[wintry dune]

wow this image was almost perfect

the arrow is on the wrong side

[ripe wraith]

whats wrong ... oh lol

i wouldnt have known that

[wintry dune]

ikr but the hands are like perfect

[ripe wraith]

every detail is

fingers look right

the rope tied

arrows in the pouch

[wintry dune]

yeah idk how its that good

[vernal depot]

she's holding the arrow right too

besides being on the wrong side that is

[ripe wraith]

i want to buy that robe now too

[wintry dune]

yeah thats the part that shocked me the most, how shes holding the bowstring and arrow

[ripe wraith]

maybe Robinhood was one of the first books GPT read

[agile peak]

Nice. Haha mine is using diffused lighting + cinematic lightning combinations 😄

[flat zephyr]

wow funny with the large holographic screens on the wall 😄

[agile peak]

Ancient roman debating their strategy + war room

[ripe wraith]

the second one could be modern day

[agile peak]

I added retro holographic map in the prompt 🤣

[ripe wraith]

roman corporate office strategy

[agile peak]

Let’s make it happens Roman Galea > Suit

[ripe wraith]

it's the future if Rome never fell tbh. the swords gradually got changed to sticks for corrective action instead of murder. but they kept their regimentation. i agree with the AI on this one

[flat zephyr]

steam-powered computers

[agile peak]

I like how dalle keeps picturing ancient egyptians with a lot of people in a room — busy with their analog computers.

[ripe wraith]

It knows history very well, me thinks.

[frozen fiber]

huh

[agile peak]

Yeah think so. It always picturing them as more hardworking

[ripe wraith]

its slavery 😄

[agile peak]

This one randomly popped up. Look like blend with ancient+ modern customs.