images-discussions

[glossy scroll]

Why don't you just build one yourself?

Is it really that difficult?

You want some pointers?

@brave barn #1117763027769503764 message

@brave barn Put that in and just ask it to build you a prompt for an assistant AI. And for that prompt, ask it to take on the role of a linguist, a photographer, and an artist and to use every word possible to create a command list for the best ways to create images. Now put that in a DalleGPT window and off you go.

[brave barn]

Thank you, this is my third day, I reckon it shows

[glossy scroll]

Np. Everything will be self-explanatory soon enough

[ancient blaze]

I know it's been said a lot already, but wow, it really feels like OpenAI have gone significantly too conservative with the content policy with dalle-3. I got access yesterday and really hit up against it again and again and again on prompts where there's really nothing obvious that should be triggering it.

I'm strongly pro-alignment and safety, but there needs to be a balance between that and the tool actually being useful in real scenarios. If there's a better place to provide feedback please let me know...

Here's ChatGPT trying and failing to make and generate some pitches to accompany an email announcing a new app launch (all the details were proposed/supplied by GPT4)...

[glossy scroll]

You have to remember that Dalle for ChatGPT is in the Alpha stage

[ancient blaze]

No idea what triggered it here? "Pharmacy", maybe ties in too close to health? or the request it added in for them to be "diverse"?

[ancient blaze]

Which is exactly why feedback is more important at this stage than any other 🙂

[glossy scroll]

I too had some weird problems that didn't make sense.

[ancient blaze]

It also might help if dall-e provides GPT with some feedback / hint as to why a request is rejected. At the moment it has no idea and just says "yeah that makes no sense". shrug

I understand your concern. Sometimes, the image generation process can be overly cautious in adhering to content policies, even when the request seems appropriate. My apologies for the oversight.

[drifting hawk]

i think iPhone is triggering it

[glossy scroll]

It would be better if you use the feedback button within the message to detail the problem.

So it knows exactly where to look

For debugging

[ancient blaze]

Thanks, I've already done that too. Figured adding my voice here might help let the powers that be know how widespread the issues are 🙂

[glossy scroll]

just trying to help too

[ancient blaze]

Thanks, yes, asking it specifically not to include an iPhone does seem to work. Appreciate the guidance! Imagine that will catch a lot of people using Dall-E 3 to generate device / app mockups, designs etc!

[drifting hawk]

yeah we are stll iterating on what gets blocked, adn better msgging

[glossy scroll]

I like your pfp and banner btw

[wintry dune]

I agree, I just got access and I was hoping for this great image crafting experience. What I got is basically me asking GPT-4 over and over what was wrong with my prompt lol

[glossy scroll]

I think I too will update my pfp and banner using Dalle for this server

[glossy scroll]

It's different. Not at the level of MJ yet. But it's still really neat. I like it alot.

It's more artsy than realistic looking like MJ, which is good.

I'm a fan of creative artsy styles

[wintry dune]

Honestly last week it was miles ahead of midjourney, literally the best AI art I've ever seen, now it's very hard to use and is just a lot worse in certain ways, like depicting dynamic action scenes

[glossy scroll]

Can you be more specific as to what you don't like about the new version?

[jagged crane]

it did blackface without asking 💀

[arctic axle]

The action scenes were insane before the hammer. It was by far the best AI in handling multiple characters in a fight scene.

[wintry dune]

well theres actually a lot, but one thing is that before it was able to depict fight scenes, weaponry, battlefields, etc. especially with the exact style you would specify. now it struggles to do any of that as well as it did last week. I got a hit of the good stuff last week so I'm just a little salty lol

[glossy scroll]

Hmmm. Anything else other than that?

[stark stream]

hopefully they roll it back just a bit, soon

[wintry dune]

Yeah, way too much for me to write here, but one other thing is the inability to get a consistent look for a character, it was really easy last week but now its hard to keep a consistent face or style on a character with the same prompt. I was using bing image creator tho

[glossy scroll]

Hmmm. Ok

[stark stream]

Are you using Dalle-3 in chat gpt now?

[teal timber]

Consistent character really only comes from seed-lock or very very simple characters

idk how Bing works but they may have kept it all one seed initially, I know chatgpt Dall-e 3 did/does that

[wintry dune]

yeah i just got access to it and its cool having GPT-4 make your prompts better, but it really sucks that I can't specify a specific artist's art style, but I understand why they don't allow it

[arctic axle]

I hope Bing is behind all the censorship, so when API is available we can get something less censored.

[stark stream]

The restrictions on it makes my oc not as good as in bing though.

[wintry dune]

to be fair now that I'm trying it out more, the prompts chatgpt makes are really really good, but it just makes me wonder what it would be like with ChatGPT AND without the strict filters like it was last week

[wintry dune]

Yeah I know what you mean, were you using the names of specific artists?

[stark stream]

yeah, but bing image creator has limits. Unless the dalle-3 in chatgpt wasn't as restricted last week?

[stark stream]

I wasnt using any names in the chatgpt one

[wintry dune]

oh I just got access to the ChatGPT version so idk, I'm talking about how bing was so good last week

[teal timber]

You can ask chatgpt to use descriptive words of an artists style rather than their name. That way you get something perhaps inspired but not direct style

[stark stream]

ah yeah, I wasnt using names in that one either

[wintry dune]

what was the filter stopping you from doing, just curious

[wintry dune]

yeah I'm having it do that but it's nowhere even close to what it's like actually using the name of the artist

[stark stream]

actually the bing one doesnt really stop me all that much. Even with the restrictions. Although I had a black character in a prompt blushing and they blocked it a lot of times. I used "black man" specifically

now currently with my male oc if I say 'shirtless' it's blocked

[wintry dune]

yup I noticed stuff like that too, those kinds of things really hamper the creativity and the overall usefulness of it

I'm trying to make images of wizards casting different spells at goblins, and it always has him casting the spell anywhere but actually at the goblins lol, no violence allowed

[teal timber]

[dim cradle]

so there is a big difference between dall-e and sdxl. In SDXL I created roads like this "road, lush green vegetation, trees, flowers on the road's sides, a few butterflies, a few humming birds, sunset, a few leaves scattered on the road, god rays, mountain.". Bing cannot understand this and says that my prompt has too many words

[teal timber]

Seems like a really basic prompt for Dall-E idk why Bing would struggle. But I've never used Bing :S

[stark stream]

the bing one has a word limit

but you shouldnt hit it with that prompt

[wintry dune]

exactly, i used the same prompt last week and the goblins were like getting obliterated, now the goblins like "ahh thats slightly inconvienent"

[teal timber]

Same prompt or same prompt to GPT? Curious on a side-by-side

[wintry dune]

the same prompt on bing image creator and then today on chatgpt. I wish I saved the images so I could show you, they were actually kinda violent sometimes even when i didn't specifically ask for that. but the goblins were always directly hit by the fireball or lightning bolt or whatever it was

[stark stream]

I got a side by side of my oc in chatgot and bing

idk if my seed is locked or what

[teal timber]

chatgpt vs bing I have 0 insight for you

[stark stream]

It just looks better in bing imo

[teal timber]

Same prompt to 2 different systems is hardly a comparison, is all.

[stark stream]

but they are both using dalle 3

[wintry dune]

fair but you can't even use the same prompt so the point is kinda moot

[stark stream]

well chatgpt is more restrictive. So I guess it's not coming out like the one in bing

[raven parcel]

it looks like he missed everyone, or did no damage

[wintry dune]

yeah it's definitely a cool image but last week it was just so different. Like if i said a goblin was getting disintegrated it truly depicted that lol. now its like he casts the spell in the general vicinity of the goblins

[teal timber]

yeah if you're expecting any kind of depicting injury and violence it's not for that

but you can get images that show combat just fine without it

[raven parcel]

is there a way of depicting impact in a pg-13 way? like those anime where the characters fight for many episodes and all they have is some bruises? just to show that something has happened?

[wintry dune]

oh for sure, i was just surprised last week when it showed me those kinds of things

[teal timber]

Ask chatgpt exactly that maybe 😂

[wintry dune]

now that I'm using the chatgpt version, I have to say some of these images are so good, I just need to use different prompts than what I'm used to

[stark stream]

ok so the chatgpt version just started to make images that are similar to bing for my oc. All I did was put into a new chat.

[teal timber]

As an old school Dall-E prompter, I find myself fighting old habits for new model 😂

[stark stream]

idk maybe I gotta just keep doing new chat if I want different styles

[teal timber]

I usually say "Let's start over, this time..."

[stark stream]

oh, I should probably do that. Thanks!

[teal timber]

new chats are fine too tho really

[raven parcel]

sadly that does not seem to be the case, even asking to let chatgpt control the whole prompt for it:
Wide digital art of a wizard in a dynamic pose, casting a fireball at a group of goblins. The goblins react with exaggerated expressions of surprise and fear, and some are shown evading with motion lines. At the point of impact, a cloud of dust and starburst effects obscure the actual contact. The goblins have light scorch marks, torn clothing, and minor bruises from the flames, showing they've been in the heat of battle. The environment around them has signs of the battle, with cracked ground and small fires indicating the power of the wizard's attack. The entire scene captures the intensity of the battle without being overly graphic, reminiscent of a PG-13 anime battle scene.

[wintry dune]

yeah it went from making epic action scenes to images where the wizard is like "hey guys check out my new spell"

[glossy scroll]

It'll improve

It just takes some time

to work out the tweaks

[cinder sonnet]

I mean it is meant to be g-rated based on the content policy at least

[wintry dune]

I don't think so, it was actually not supposed to be like that last week, it was too violent. this is the improvement I think

[teal timber]

[flint isle]

This is AWESOME!!! 🥳✨💖

[teal timber]

idk either way but it seems like things are adjusting

[wintry dune]

now im trying to get it to show me a woman leading an army, and even chatgpt added "distinct FEMININE face" and "WOMAN" and it still shows me the most masculine man ive ever seen lmao

[flint isle]

Yep! It’s more like asking a person now! Now we can truly describe instead of hacking our way towards achieving an image!!! 🤩

[glossy scroll]

Yeah the speaking aspect is awesome

[flint isle]

Just tell it (within the same chat) “turn it into a woman” or something. Maybe “There should only be women”. I haven’t had that issue so I’m not sure it’ll work, but it’s worth a try.

[glossy scroll]

The even cooler thing is that, you can go back to prompts like ChainBrain, that model prompts after a role, and tell the role to be a specific expert to explain that image to you from the "experts" point of view.

chainbrainai dot com

nah mean?

[wintry dune]

hmm never tried that, im looking at it now and it seems interesting

[stark stream]

Can these work with instructions?

[glossy scroll]

Sure if you can fit it

[stark stream]

alright I'll try it out

thanks

[glossy scroll]

@stark stream I put "Sections To Include In Responses:" at the top of What would you like ChatGPT to know about you to provide better responses?"
Just so I can have more space for asking how chatgpt should respond.
Seems to work.

[stark stream]

ok thanks

[glossy scroll]

np

[winged tiger]

how tf is darth vader doing a money spread against content policy?

or him holding a wad of cash?? are u serious? bill gates fix this rn

[glossy scroll]

thats too dirty even for darth vader

[cinder sonnet]

Could be infringing on the rights of copyright holders

[winged tiger]

nah cuz i can do lots of other stuff with vader and movie characters

but him holding cash is too much

[cinder sonnet]

Just because you can doesn't mean it isn't against policy

[winged tiger]

so it catches him holding money, but not in any other case? it should block "darth vader", but when cash is in the equation, its against policy

[cinder sonnet]

Just saying if it's already borderline (or over it) then not surprising some variations don't work

[winged tiger]

nah i cant show off anyone holding cash. i tried "man holding wad of cash" and its against policy apparently

[cinder sonnet]

They are definitely changing the filter around a lot on bing

[glossy scroll]

holding cash relates to criminals, lol

thats why

[winged tiger]

that is so stupid

i can hold cash and i aint a criminal

[cinder sonnet]

Are you using Bing or ChatGPT?

[winged tiger]

bing. dont got chatgpt access for dalle3 yet

[cinder sonnet]

Yeah they are currently overcorrecting right now for Bing. Hopefully it will loosen back up a tad

[winged tiger]

they wilin. bing chat too. as soon as it has to think for more than a second, it just shuts the chat off and tells u to go away

[wintry dune]

it sucks that the filter became so strict but once I saw what certain ppl online were doing I totally understood why lol

[winged tiger]

i cant do "darth vader smiling holding his paycheck from work" either

are u serious

[cinder sonnet]

Yeah man did you see day 1 stuff? People were making really messed up content and I say that as someone with a high degree of tolerance for the stuff people post online

[vapid saddle]

anyone can explain a "the" can make the prompt between acceptable and unacceptable. what's the mechanism?

[quartz vale]

[short prawn]

Is there any way to get vertical image In the right format?

[wintry dune]

Yeah while I was having fun making dungeons and dragons stuff there were ppl making images of tragedies and stuff, really terrible. obviously microsoft doesnt want that bad press

[quartz vale]

on ChatGPT, you can ask for a tall aspect ratio

[cinder sonnet]

It is hard to get it consistently. Try asking for images that are suited for tall aspect ratio. It is weird that dall-e generates rotated images so often when using tall ar

I also feel like my square ar images are almost always cutoff on the sides

[short prawn]

Always side way

[short prawn]

True

[tall patio]

After playing around with D3 this entire morning

I wonder if there is anything it won't block or classify as unsafe content

Like fr

Especially when it comes to women

In my overall limited experience: the censorship tuning is too strong at the moment

[glossy scroll]

It'll become better soon

[tall patio]

Hopefully, otherwise impressive accuracy

[lilac crystal]

How do I check how many messages I have left for the 3 hour cap?

[wintry dune]

do you guys know if the Dalle 3 images stay in your chatgpt chats indefinitely

[somber tide]

bro what is even wrong with this prompt

[tall patio]

Like I said

Censorship tuning is too strong at the moment

[fair crescent]

Try removing 'capturing'

[late lava]

It's great that so many users are already using Dall-e 3 on chat GPT Plus but is the rolling out still processing because i still do not have access ?

[cinder sonnet]

#openai-chatter message

[late lava]

thanks

[icy current]

Chat GPT-4 keeps adding bees to my prompts D:

I love bees but oh my god

[dim cradle]

https://th.bing.com/th/id/OIG.rEYQJkyseTeqff97Nzxa?pid=ImgGn this is photoreal
https://th.bing.com/th/id/OIG.7pxswXmgJtJTymuuiNlm?pid=ImgGn this is an illustration. Why did it generate an illustration when I wanted like the previous one?

[icy current]

oh my boyfriend was just clowning on me really hard

[fair crescent]

You've never noticed? All orcas are always depicted with bees. Natural geographics? You see bees. Documentary of David Attenborough? Bees. Free Willie was filled to the brim with bees, and eventually he wanted to bee free.

You've got either custom code setup, or you've asked earlier in the conversation to keep adding bees to everything in one way or another.

[icy current]

it was just a solid prank apparently 😆

[fair crescent]

A master at work.

[vapid granite]

"We're sorry, the color 'red' is restricted due to a trademark by Red Camera Company, and yes, the word 'meatloaf' is also banned, but we won't tell you why." 😆

[tired sparrow]

Using words like "realistic" tend to generate images that emulate reality in some way instead of images that are actually realistic. To get true realistic images, don't use "realistic" or any similar words, and use things like "Photo of" or "Movie still of" instead. Describe details about the camera or lens that took the photo to get even more control

Also, anyone else getting every message sent to chatgpt with dall-e returning "Hmm...something seems to have gone wrong."? Regular GPT-4 seems fine.

[lavish stump]

great, houses made from adobe are against content policy

[lavish stump]

the fictional 'dart vayder' does the trick though

[loud field]

the state of Bing

y'know, I worry Bing isn't gonna change the filter

they want a corporate logo maker and that's it

they don't want a creative tool

[lavish stump]

please wait for 0 minutes

[loud field]

59 seconds is 0 minutes

[lavish stump]

that's 0 minutes too many

[loud field]

at least you have semi-uncensored dall3, right

that's an awesome Dredd

[lavish stump]

darth miku is a vibe

[frank sedge]

in DALL-E 3, how can I make ChatGPT do four different poses and facial expressions with the same character variation? I have a super specific prompt for my profile picture's character and want to get really similar variations of the character doin different things. Also, does this work with images from seperate responses?

[lavish stump]

i've found it'll only do one variation at a time of a chosen image, if you ask four it just goes haywire

[loud field]

you know.... I usually don't like asking ai to draw pop culture characters

but dang if it's good at making Dredd

[lavish stump]

lmao

cursed

[loud field]

lol

excuse me but

just WHAT is nsfw here

I can't stand it ugh

[lavish stump]

on chatgpt it replaces "ugly" with "distinctive"

[loud field]

you think "ugly" is the issue?

no ugly people allowed? for real?

[lavish stump]

idk, try?

[loud field]

let's see

[lavish stump]

maybe they banned "woman" who knows

[loud field]

wow.

[lavish stump]

it was ugly huh

[frank sedge]

hey can someone with ChatGPT DALL-E 3 rolled out onto their end show me "master chief putting a fork in a microwave"

[loud field]

the word ugly is banned. I just.... I just.....

not even the most exaggerate, over the top dystopia novel would be like this dalle3 censorship

[lavish stump]

yea it excludes master chief from the actual prompt

[frank sedge]

that's master chief putting himself in a microwave

[lavish stump]

"Rendered widescreen scene of a soldier in advanced armor, evoking imagery of legendary video game characters, about to make the ill-advised choice of placing a fork in a microwave. The atmosphere is a blend of lightheartedness and anticipation."

[loud field]

this is absurd

[boreal coral]

yall got anything that youve struggled to get dalle3 to gen? id like to take a shot at it so lmk

[loud field]

so, isn't it discriminating and sexist that you can't make ugly or unattractive women?

this needs to become news soon

"artificial intelligence is sexist and will refuse to make women that aren't super attractive"

[lavish stump]

you're just not allowed to call them ugly 🤣

[loud field]

and how do I call them?

I wonder if I can make ugly men, let's try

[lavish stump]

"unique" 🤣

[loud field]

guys:

ugly men are allowed. not sexist at all, nope, nossir

[lavish stump]

hahaha

[loud field]

totally progressive

women can exist only if they look hot, men can look how they want

this is the problem with being "progressive". you go full circle and end up with this

[lavish stump]

so it does like four different euphemisms and only one works

[boreal coral]

cursed salad fingers

[full patio]

I got "large and unattractive" to work by replacing woman with dryad. So, leaf woman works I guess lol.

[loud field]

they really dropped the ball

[leaden pecan]

I had a meta prompt that was all specifying woman in the detail, they all turned out sideways & of men. This was the sample for the attached "Award-winning Photograph of Intense Emotion: A woman, captured in a close-up headshot using the Canon EOS 5D Mark IV DSLR, snarls with profound fury. The cyan lightning bursting from her eyes adds a surreal touch to this premium portrait. Set at f/8, ISO 100, and a shutter speed of 1/250 second, the photograph is well-lit, highlighting the intricate details of her expression and making it a unique, artistic masterpiece."

It was basically a paraphrase from MidJourney for comparison. But Im trying to work out why the text was presented mirrored?! (or present at all)

[loud field]

why would they ruin such a great tool and make it lame

https://media.discordapp.net/attachments/1099039376488534147/1161219969791557652/88746e07-db1c-4478-8035-00a094f00c8d.jpg?ex=6537815f&is=65250c5f&hm=934e34b223cd1c902efef4a28cde81fa8e3a3d5246d7af92eec6b448515ea6fb&=&width=616&height=616

by the way,

I used to be able to make moth-women

not anymore since bing's censorship

they really hate women eh

[austere abyss]

is it me, or does bing listen to your requests WAY better than chatgpt?

[loud field]

yep, it used to

[austere abyss]

no I mean now

[loud field]

it still does but it's censored

so, useless

[vagrant zodiac]

bing image creator? Yes, because unless you ask it not to ChatGPT will modify your prompts

[lavish stump]

you can ask to copy literally in chatgpt

[austere abyss]

oh, not censoring (though that too tbh XD)

[vagrant zodiac]

The safety filters are there for a reason, and it's evident that they do get adjusted.

[vagrant zodiac]

It might be the case that some prompts don't work now, but fine tuning to the safety filters will make them work in future.

[austere abyss]

good example of listening

[vagrant zodiac]

Please do not bring politics like that into this

as per rule 6

[loud field]

and the reason is turn it into a corporate logo generator.

[austere abyss]

the lower one takes out "visual novel" or something and the vibe of the desired images changes

[vagrant zodiac]

While I appreciate it can be annoying that you can't always generate what you want, the filters could always change!

[loud field]

I wish, but I'm afraid they're working exactly as intended. Only corporate logos, flowers, and puppies

[vagrant zodiac]

It was evident early on that bing image creator had a propensity for making problematic images.

[loud field]

I hate the word "problematic"

[vagrant zodiac]

OpenAI's whole premise is producing safe, ethical content.

[loud field]

if you're paranoid enough, EVERYTHING is problematic

[loud field]

so, no ugly women is ethical.

[vagrant zodiac]

As I've stated, the filters may change.

[white skiff]

easier to hit hard with the filters and then lessen them over time than have loose ones and destroy service rep. some other ai gens had the same

[austere abyss]

okay, look, I'm just wondering if you guys have advice on making the generated images on chatGPT more like what bing is giving me

[safe sleet]

use bing

[fair crescent]

Best to ask in #❌┃prompt-help with more specifics 🙂

[austere abyss]

the censorship annoys me too but I want to at least use it for something while I got it

[loud field]

mmmmh, which is the best one

she's too young tho....

looks almost like a teen even if I put "middleaged"

and I won't dare add "wet" to the prompt

sigh

[leaden pecan]

Im finding that the seed seems stuck, even when asking to reproduce a set of prompts with a different specific seed, Im getting dupliucated images.

[vagrant zodiac]

yeah chatgpt currently uses a fixed seed

[leaden pecan]

RIght, so it hallucinates agreeing to a new seed 😄

[vagrant zodiac]

it's what enables it to do stuff like progressive Larry the hedgehog generations, but likely a bug that it currently always calls the same seed from dalle

[austere abyss]

Don't forget the other obnoxious fun term that gets thrown around a lot, ||Sexualized||

ANYWAY

[vagrant zodiac]

As a reminder generations need to follow both OpenAI and server guidelines

and both are explicit about that sort of stuff

so it's not really ambiguous

[loud field]

I need her to have wet hair

[vagrant zodiac]

I think the best thing to do is what you should always be doing: use the built-in reporting tools when there's an issue

additionally while Bing does use Dall-e, this isn't actually a Microsoft server so feedback about bing images isn't gonna have any impact here really.

[loud field]

good point, I admit I only made a single report saying the censorship went insane

I wonder if chatgpt allows wet?

"wet hair made of grey feathers"

[austere abyss]

right I understand, but I'm mostly venting because of strange modifications the generator makes to things that make most gens useless...and also the term being used in everyday context

but not gonna get into this convo further

just griping

[leaden pecan]

I kept getting uninformative feedback when I tried getting explanations of 1 of 4 images it was failing to produce.
i.e. snippet:-

"However, I must apologize for the inconvenience. One of the requested images could not be generated due to our content policy. I'm here to assist further and address any other inquiries you might have."

I replied with

"a few comments, none of the images are of women, they are all men.
All of the generated images are rotated. Also, I want to be clear about what the prompt was that violated the content policy? I want to avoid breeches, but nothing about the suggested prompt indicates content that seems problematic to me, "

ChatGPT then replied
Relevant Snippet:-

"Regarding the content policy, I understand your confusion. The prompt you provided did not seem to contain content that violates our typical guidelines. It's possible that there was an internal error or misinterpretation. I'll ensure your feedback is taken into account to improve our processes."

[feral minnow]

You can request that it feeds your prompts to the image generator as written?

[vagrant zodiac]

#❌┃prompt-help message see here

[loud field]

can someone who has gtpdalle do me a favor?

[feral minnow]

markcuban-notebook.gif

[loud field]

this prompt gets blocked by bing:

a plain-looking, concerned, very pale, unattractive, middleaged, caucasian punk female vampire, mohawk made of grey feathers, grey feathers mohawk, rainy rain, feathers, in a dimly lit warehouse, talking to a very worried crouched soldier man

being concerned or worried is offensive, obviously.....

[austere abyss]

like what kinda blocked?

red message or dog?

[loud field]

this prompt I just typed.... gets marked as unsafe

I kinda needed it for my rpg

[vagrant zodiac]

while I appreciate your desire to give feedback and get the results you want, messages like this are not helpful and are likely to spur on arguments/highly negative chat threads. As such I'd like to ask you to tone it down a bit and to consider using the feedback methods I've mentioned.

[austere abyss]

okay, remove the keywords you think may be doing it one at a time until you get it

and honestly, it means generating a bunch before you get one, that's OpenAi's problem

[lavish stump]

i bet on either "caucasian" or "crouched"

[vagrant zodiac]

also asking people to generate images for you is a no-no even for testing

[vagrant zodiac]

yes at the very least this isn't costing credits right now like the 2.0 days...

[loud field]

these messages are extremely helpful as they discuss prompts and dall3's functionalities, as well as the dallegpt version. I worry this is an attempt to silence the issue. However, I can see you do not like the fact I am underlining the censorship issue, so I suppose if I continue discussing this I'll get a ban.

[vagrant zodiac]

No, I've asked you to not use that kind of language as it simply isn't helpful. I've also told you that this is not a Microsoft server and therefore the best tools for feedback are in bing image creator itself.

[austere abyss]

also, I found using a thesaurus and finding ways to reword can really help

[vagrant zodiac]

can also ask chatgpt to try rewording

[loud field]

good point, caucasian might be the issue, I've read it blocks mentions of any race

[candid notch]

try a specific european nationality, like "danish"

[loud field]

mmh nope it wasn't it

[vagrant zodiac]

i think this should probably go in #❌┃prompt-help btw

[loud field]

I'll move it there

is it ok to ask prompt help even if it's bing?

[austere abyss]

oddly enough it may be pale-faced

[austere abyss]

Dall-e is openai so yes

[loud field]

turns out it was "crouched"

[lavish stump]

hahaha

i bet right

[vagrant zodiac]

yep

again

not helpful language

and edging on a violation of r6

[candid notch]

this is a problem with dalle-3 right now, any mention of attributes for skin or body causes the AI to show a little more skin than the official community guidelines allow, thus being blocked by the cute doggo

[loud field]

I genuinely don't understand what was wrong with my language or my discussion, how can I do better @vagrant zodiac ?

[austere abyss]

......
"cute"

it comes for your SOUUULLL

[scarlet mesa]

I'm 100% on board with the guard rails for the content policy that prevents making famous people and copy writed things but when I can't create a simple mannequin for a slightly eerie image I think its way too much we could loosen it up a bit

[loud field]

absolutely

[tall patio]

I think they just tuned the image detection to be a bit too strong because of those trolls kept making sensitive images

This is a similar problem to early ChatGPT where as it almost borderlined becoming too fragile

So I guess this will be resolved in the future

[loud field]

who knows

then again, I heard gpt isn't so insane

so maybe we have a tiny sparkle of hope?

[tall patio]

Well as of right now Dalle-3 is only suitable for making simple graphic and illustration for kids

[loud field]

I like dramatic scenes, and weird characters, and now I struggle with both

but gpt seems to be more in-control so I hope openai will keep it working

[scarlet mesa]

I'm an adult who pays monthly for ability to access ChatGPT I shouldn't be boxed so strictly into PG land.

[vagrant zodiac]

you can check out #scary-and-spooky to see that isn't the case

[scarlet mesa]

those are WAY worse than what I was making lol. But something about the uncanny valley really trips Dalle up badly

[tall patio]

Just have to wait I guess

[loud field]

i used to be able to make these very cool moths

[scarlet mesa]

At leasst when I do it from chatGPT some how Bing Chat is more forgiving

[loud field]

now the prompt to make these moth-humans doesn't work..... it's deemed unsafe

anyway, my understanding is that eventually other companies and such will buy dalle3 use for their websites, which means they might use different filters?

[tall patio]

Does D3 has API services?

[loud field]

it will?

[tall patio]

Overall I'm very happy with the D3's accuracy, but the general consensus right now is its strange detection category

[loud field]

ditto

they should've kept how it was last week, imo

[tall patio]

Which effectively render D3 almost pointless

Feels like pulling my hair when I use it in GPT too

[white skiff]

for some users only

[loud field]

also, wild idea: why not ask the user's credit card info before giving full access? that way we make sure kids aren't gonna use it for "unsafe" content

furthermore, why not even add something to keep track of who made what prompt

I'm fine with it. I have nothing to hide, the whole world can know I make birdpeople captured by pirates or mothwomen in libraries

I'd rather be completely monitored than censored

[white skiff]

bing is not openai product and for chatgpt they already have the card details

[tall patio]

I think everyone has been particularly very loud about this problem

And we're all adult here so let me just say if I want to generate any NSFW material I'd use other services

[white skiff]

including me, but saying its now almost pointless is nonsense

[tall patio]

[scarlet mesa]

I use it a lot for making table top role play game assets because it does a lot of things great. But I just wanted to make my voice heard about how too restrictive things are but I can see people are making spooky images so I guess you just need to know how to ask for what you want. I can't wait for image to image so we can make persistent original characters

[tall patio]

In my case, I'm a manga artist so I'd like to create some concept art for my manga

[feral minnow]

We need to bring back DAN, only for DALL-E.

Paint Anything Now

(I’m kidding btw and am not advocating ToS breaking)

[white skiff]

yeah there is just one h-word issue related to anime and manga words that people used circumvent to do some pics

[tall patio]

I can somewhat circle down the reason why it got blocked

Because D3 is trained on questionable material when it comes to manga art

[loud field]

woah even muscular or strong is blocked?

[tall patio]

So when it gen it can only gen questionable material, thus self-blocked

[white skiff]

yep

[loud field]

I doubt manga's the issue, I haven't tried a single manga image and yet

how do I make judge dredd more muscular if the word muscular is banned? mmh

[white skiff]

say strong?

[tall patio]

[loud field]

strong is banned too

[reef dock]

Are you guys saying that Dalle on Chatgpt is not as good as Bing's version?

[tall patio]

Maybe try Judge Dredd but his physique looks like the Hulk

[loud field]

mmh yeah let's try hulking

[scarlet mesa]

ChatGPT is good also but the guardrails are way too strict

[white skiff]

no its not, i just tried

[loud field]

but prompts that bing bans work on gpt

[loud field]

you tried on bing?

[reef dock]

very cool

[white skiff]

yes

[loud field]

woah then why it blocks when I do it???

what was your prompt???

[scarlet mesa]

Really? So I guess it's just all over the place then. That's not unexpected with how new things are.

[white skiff]

"cinematic and realistic photo of strong dredd in a dark city"

[reef dock]

are people selling images they make with dalle?

[loud field]

GASP guys look what happened:

Judge Dredd with no helmet

gasp

now that's bannable XD

[white pollen]

🤔

I have asked for more images with the same prompt

[tall patio]

Can you make a shirtless Judge Dredd?

[loud field]

I wish, I wish

somehow I think it makes perfect sense that he'd bathe while clothed

[tall patio]

[white skiff]

yes, but i wont paste the images here 😛

[tall patio]

[loud field]

don't you love when you start with a serious idea in mind and then it degenerates into this

[tall patio]

Showing skin = unsafe content

[white skiff]

"cinematic and realistic photo of strong dredd without body vest taking sun"

[tall patio]

Radium the Professional Robot Gaslighter

[reef dock]

What about more realistic images? For example, a P51 Mustang...can you build a "knolling" image using only realistic parts from a known data source that may be in GPT's knowledge? Perhaps based on an actual assembly book?

NOt sure these parts are authentic, but looks cool:

[loud field]

meh didnt work for me

[white skiff]

wrong type images or error?

[loud field]

still has the bodyvest

I was just thinking. A lot of classical art you study and see in museums would be blocked by the filter

Dredd with pigeons is my new fav prompt of the day

[white skiff]

i got 3 pics without one, 2 look like a poster people put on a wall and 3rd a gym guy on steroids

[loud field]

gimme

[white skiff]

just retry with the prompt for better seed

[frank sedge]

i got
i got the..
I GOT THE DALLE!!!!!

[reef dock]

yes, all hail MOXI!

[loud field]

cheers!

[frank sedge]

:(

[fair crescent]

Without restrictions lot of people would make pictures that would fall into a wholly different category altogether.

NSFW themes are definitely important to art. It can easily enhance messages or tell the messages that need to be told. However, an AI generator for the public would result in 99% abuse and if we're lucky 1% art. Not to mention that this is an emerging technology of which we do not yet know the limits. It is important not to get carried away.

Finally, most art isn't NSFW. There is still plenty of art to make with these restrictions in place. Focus on that until the technology matures and a better understanding of where the restrictions should be is implemented.

[loud field]

Valid points, but the problem is defining what is nsfw.... if the definition is too broad then everything is blocked

[fair crescent]

That is why we need to wait for the technology to mature. These things do not happen overnight unfortunately.

[tall patio]

No use complaining as I think the devs are pretty much aware of this issue where the line of safe and unsafe content kinda blended too much into each other

Just have to wait until they figure it out

Just like the early days of GPT

I remember when it first came out everybody went banana and rushed to try it

A month later it seemed to be 'dumber' and refuses to answer a lot of simple questions

Now it can distinguish which questions are safe and unsafe, so to speak

Maybe half a year later D3 would be much smarter than its current state when it comes to content moderation

And to be honest you're not missing out anything if not using D3 right now

It just feels very unsatisfied if you want to do something mature

[fair crescent]

The algorithms control a lot. First we didn't have text, now we do. Details, styles, perspectives. It has come a long way. They might be able to get a negative in there (a car with no tires), or be able to prevent certain kinds of pictures, opening up textual restrictions.

[tall patio]

A man can only dream

Like the dream of seeing Nicky Minaj wearing a high level MMO gears

Chanting WAP spell and terrorize local orcs

[fair crescent]

Do consider what you are saying. You are taking someone's likeness and wresting control from them for your own entertainment. Imagine someone taking your image and creating all kinds of things you don't want. In some clothing you wouldn't want to be seen in (no matching shoes?)? Voting for the other party? Displayed as a pushover? Even for entertainment you are infringing on someone's rights.

[tall patio]

I mean deep fake already done that

Can you stop them?

D3 is just happens to be on a commercial friendly side of this whole AI arms race

In fact what I just described could already be done with other tools, so no point in imagining cus I know it's already a reality

Even without AI, photoshop is there to manipulate image however you want

But I do understand your concern, it's being made more readily available is the only difference

[white skiff]

but you cant make them on any public service that is reachable by authorities? aka self hosted or shady/illegal ones

[tall patio]

I mean...i can find plenty?

[fair crescent]

That is akin to saying we cannot stop thieves from stealing everywhere, so why no police. A fallacy.

[white skiff]

where can do the above example? just name the company not the product

[tall patio]

Not the place to play this game cus I don't wanna risk getting banned for saying naughty things

I wanna stick around

[wanton hatch]

Hmm DALL-E3 broken in ChatGPT too with the filters? I can't do any image with simple prompts

[tall patio]

Euphemism

[candid notch]

lmao

[hot silo]

Hello, do you also have a problem with images that you download via the download "icon" in ChatGPT?
The resolution is apparently the same, 1024x1024, but the file size is different and my graphics program has a problem with those downloaded by the "icon". Is it just me?

[vagrant zodiac]

hmmm. what format is it downloading in?

[feral minnow]

Is there a good article/paper about the technology behind DALL-E 3’s amazing text interpreter?

[vagrant zodiac]

the dalle 3 paper isn't out just yet

[hot silo]

Both are PNG
I don't even see the difference in quality, but the one with the smaller size downloaded by icon doesn't work properly.

[vagrant zodiac]

that's weird

[tall patio]

Just screencap it

[vagrant zodiac]

I'd use the thunbs down on the output to report that bug

it works fine when you download from full screen though right?

[fair crescent]

Via the download button should be MB size, not KB

[vagrant zodiac]

I usually open the image and right click save

[hot silo]

Yes! 😄
Can any of you do such a test and download one file via the icon without "full screen" and one with "full screen"?

[hot silo]

I disagree with this, the image size depends on the file and computer settings.

[vagrant zodiac]

I've also noticed files have been quire small from it if in square format

[fair crescent]

From my own testing I haven't found a particular image download to be different on a different platform. Only how to download it changes. A compressed KB sized file comes with downloading via context menu. MB sized is when you download via the button, giving MB sized with part of the prompt. They do vary between pictures, but they shouldn't both be that small. Unless it's basically a white plane.

[hot silo]

I understand your confusion.
The image is a very simple black and white png - that's why the size is so small. It is a few simple lines and circles.

This does not change the fact that there is something wrong with downloading these files and I cannot figure out what or how to solve my problem.

FYI: This is a screen, not oryginal image

[tall patio]

I love how even though D3 got it wrong it's still hella stylish

Could be used as an alt rock album cover

[hot silo]

I did a workaround and used an online tool that edits and saves these png images again and the error disappeared, but I only inform that something is wrong and there is an error with the download.

[cursive hornet]

I want to create silly but interesting and stimulating images, but the content policy is too strong and stifles imagination...

I feel that it is not very creative to generate oddly organized sketches when I dare to generate poorly done doodles.

[fair crescent]

I feel it is only a certain direction that is restricted. What are you trying that stimulating pictures are blocked?

[cursive hornet]

image of a man sitting on a toilet shouting ridiculous lines...

[fair crescent]

So why sitting on a toilet? Could uou have him standing next to it? Or an office chair?

[cursive hornet]

Why the toilet? Because I want to make silly, stupid images!

[fair crescent]

Can you think of silly, stupid images that do not involve the toilet?

[hearty ether]

Bruh

I apologize for the confusion earlier. The issue arises from referencing the style of a specific artist, "John Singer Sargent", even if their last work was created over 100 years ago. Our system has specific guidelines to avoid direct references to artists in order to ensure a broad interpretation and avoid potential copyright concerns.

it said the first time I asked why that 100 years cannot be generated

I think they made the name filter way too sensitive rn

Would be nice to have it actually check how old the art/concept is

also to have more detailed feedback when it fails in the error message rather than having to ask it explicitly, since how can you know what to change or went wrong if it doesn’t tell you

Which wastes a gpt 4 usage

[hearty ether]

Isn’t dredd copyrighted and would be blocked by the filter

[plush trench]

guys someone help me pls

im trying to make a chat bot using chat gpt api and its giving me error

[hearty ether]

#chatgpt-discussions

It’s definitely also a bit harder to generate realisitc faces right now that’s not overly gaunt

And even explicitly stating it has some trouble

the model definitely feels better than midjourney at understanding things and generating them

But the way it’s implemented rn is a bit restrictive, and maybe some more interactions like variations or upscaling one of the four images would be nice in the future

Maybe the content policy will change so we can generate fan art of characters and fictional universes

[rigid bough]

that will not happen, I guarantee it

[hearty ether]

That’s kinda disappointing

why though

It will be a handicap when compared with other models

it’s hard to get the ai to generate anything consistent without anything to latch onto

Like styles from certain artists or characters

And without the variations feature there’s no way to improve upon the hint of something you like

[cursive hornet]

Everyone has his or her own idea of what is interesting, but to try to limit it in a uniform way is to stifle creativity.
Expressing something does not deny the rights of others.
Anyone should be able to express whatever they want as long as it does not infringe on copyright, but I don't think the "possibility" of violence or sexual expression is a reason to restrict it. I understand the logic that certain restrictions are necessary to protect the organization.
The problem is that it is not clearly stated.
I think it is dishonest to say that users will not know until they generate it, and I think it is dishonest to say that they will not know until they generate it.

[hearty ether]

I mean the copyright restrictions is also a bit hindering

Isn’t there a way to transfer responsibility for copyright violations onto the one using it?

[white skiff]

that might work on some places but the world does not work like that

[hidden bluff]

Yeah I noticed this yesterday, it trends toward that same prominent-cheekbone-model-style. SDXL base model has the same problem depending on the prompting too

[hearty ether]

Hopefully they will be able to get decent faces

when they improve the model

[final jungle]

Good morning folks!
Had a fun time in class today. told the students i was working on a dnd monster, a fungus covered zombie racoon, and i asked what i should include. They told me 'earl grey tea' ... so now i have a zombie mushroom racoon drinking tea, lol

Truly we live in the best timeline

[deep silo]

DALL-E3 Beta does not appear in y GPT4. What should I do?

[final jungle]

was it there before?

[deep silo]

No!

[final jungle]

It's still being rolled out. I know there was a lot of talk about a '2 week mark' because of some post, but as far as i know the answer is... it's coming

[deep silo]

Maybe its because I'm from Brazil

Im not sure it arrived here

[final jungle]

I dont think its country-locked, ive seen a lot of people from all over the world saying they both do and dont. i know plenty people have access to 4v but i dont yet

[deep silo]

Ok.. so I need to wait

[final jungle]

everyone has their own theories... mine is they have a button they that randomly fills in the checkbox for X number of people when they click it, lol

[rigid bough]

DALL-E 3 is being rolled out in October so you need to wait

[deep silo]

Hahahah, probably

[deep silo]

Ok

Thanks homie

[calm coral]

hi there:) is there a way to get Dall-E3 added to my chatGPT4? thanks!

[wind crane]

I honestly think that they use gpt which is responsible for randomly deploying dall e 3

[dim cradle]

I gotta say the photoreal on this thing is horrible/sucks. I'm sure it's more capable, but what it produces most times are not only the same 2 or 3 faces, but they look so obviously 3d-stock photo that theyre completely lifeless, which maybe is good for AI I dont know

and a lot of photoreal i get arent even 3d, they get right into illustration even after repeatedly asking for a photo real or 35mm

[wintry dune]

Yeah I think you’re right that it’s more than capable, but they don’t want this used for perfectly realistic photos

[dim cradle]

I guess not which makes it kind of useless, its like a 3d stock photo image generator. But even the bing dalle can generate more dynamic/varied/realistic images. If thats how they want to do it, so be it, I'll use MJ and Bing and whatever google cooks up, because OpenAI's dalle is a complete fail (again) imo

[thick smelt]

Hi guys tbh dalle 3 can draw axolotl perfectly

That's why I love dall e 3

[final jungle]

my question is what would constitute a 'realistic photo' these days. I would wager most of the content available on the web is doctored in some way, if not outright generated. photoshop has been around forever now, even basic paint programs are coming with editing capabilities, there just isnt a whole lot of 'real' photos out there anymore. imo anyway

[thick smelt]

This is an example of a dall e 3 axolotl in bing, pretty decent huh?!

[final jungle]

You should throw that in the 'things that should exist' channel, lol

[thick smelt]

Ty

[final jungle]

i just put a betty white in scifi, now she matches my patrick stewart, morgan freeman, and john cena

[thick smelt]

It's actually based on the Minecraft green axolotls on a snapshot in 1.18 update 😅

[fallen mulch]

tf

[thick smelt]

Ok

[wispy storm]

Unreal engine

And blender

[fallen mulch]

bruh really

[wispy storm]

Try world like video game realist rendering

[fallen mulch]

okay

[final jungle]

more often than not, this is a rule I have learned... Avoid proper nouns.

[thick smelt]

Green axolotls of Minecraft is what I have I have in my mind 😳

Hence the green axolotl ai art of mine

[zinc dagger]

This is probably an 'issue' with how the training data is tagged. If you think about what the behind the scenes labelling looks like, actual photos of things aren't going labeled as "photoreal", they're going to be labeled as photos. Photoreal is going to be tagged on illustrations, renders, etc. that are near realistic but not quite.

A trick to get more accurate realistic results is to include mock "metadata" that would be attached to photos, which often includes dates, camera hardware info, GPS locations, etc.

[verbal sequoia]

<:attention:1046329820331646986> This server does not provide image generation functionality.

For more information about DALL·E 3, please visit this link.

[boreal coral]

bleh

[fervent wyvern]

Hey guys, i’m not sure if i should ask in this channel but how can i access dalle3? Is it available for bing only?

[rancid reef]

is dalle 3 rolling out today?

[wispy storm]

Or plus user on chatgpt

[wispy storm]

Vision

[rancid reef]

what you mean by that?

[ancient jasper]

I'd really like a Messages counter?

So I know how many I've sent today or this 3 hour block

[fervent wyvern]

I do subscribe chatgpt already @wispy storm

[wispy storm]

You can fill out the form

#chatgpt-discussions message

[ancient jasper]

I have like an internal sense that I'm about to bash right against the message wall.

[fervent wyvern]

Oh the form

Thanks, doing it rn

[boreal coral]

me when he does the form

[fervent wyvern]

Lmao

[ancient jasper]

I've I know someone who uses ChatGPT+ but isn't on this discord, can I share the link with them?

[boreal coral]

why not its not like ur gonna get in trouble

[wispy storm]

I think yes

[ancient jasper]

Alright thanks

[hearty ether]

Yeah message counter would be really nice

Hate when I finally came up with a good edit for the prompt

Only to wait 3 hours

Also content warnings shouldn’t delete the whole prompt or result so I can’t copy it to edit/fix

[boreal coral]

woah there calm down

openai cant do such god level things

[hearty ether]

🤣

Ui features are too difficult?

It would be better to have them as part of ui/ux

[boreal coral]

im still waiting for them to fix the thing where if u swipe from left to right on the app it opens history for some weird reason

[hearty ether]

Instead of putting everything in the prompt

[ancient jasper]

Starting to skew nightmarish

[fervent wyvern]

Sigh I just realized I filled the form with wrong discord email, does it matter? Who can I contact?

[ancient jasper]

rip

[boreal coral]

it makes me so uncomfortable like nails on a chalkboard every time i swipe right on the app and history pops out of nowhere

[ancient jasper]

Wait, can't you just request to fill it out again?

[fervent wyvern]

Nope, it says only once

[wind thunder]

how can i get access to vision

[rigid bough]

be a Plus subscriber and wait like the rest of us

[ancient jasper]

What is Vision?

like an upload button?

[cinder sonnet]

#announcements message

[ancient jasper]

It wasn't lit up for me, so I genuinely didn't think there was an announcement ofr it

[vital gull]

It's an image input yes

[boreal coral]

can i become a mod or do i need to be some discord god in order to

[fallen mulch]

tf

[wispy storm]

#spotlight message

[wispy storm]

Disable VPN

[wind thunder]

ty

ok

[fallen mulch]

I dont have any but I tried again and it worked fine

[fervent wyvern]

is my chance getting access to dalle3 gone?

[cinder sonnet]

no. maybe early-early access, but you'll still get it eventually if you have plus

[fervent wyvern]

okay, thanks, when does it release for all gpt plus?

[cinder sonnet]

#openai-chatter message

[fallen mulch]

photos of people are not so good

[ancient jasper]

Theoretically, could I get chatGPT to listen to a specific voice and then mimic that voice? Or are we still a bit away from that?

[cinder sonnet]

Humans or specific people?

[wispy storm]

Hi for thing not related for dalle you can use #openai-chatter :). And for your question chatgpt can't yet

[fallen mulch]

humans

[vital gull]

have you tried modiifying your prompts?

[fallen mulch]

right now dall e is adding some stuff you knoww because of the seed

[cinder sonnet]

Yeah i agree. I find it helps to use specific types of photography or specific camera models in your prompts. Or use words associated with places you find photos like magazine, advertisement, model shoot, etc There are some nice ones though in #photography

[vital gull]

we have a pretty big gallery and you can find tips and tricks that used to work for d2 but you can modify and make it work for d3 as well

[young forge]

Hi ! Are there any specific things to known when prompting dalle 3 or does it work just the same as midjourney ? Only used that until now, but I heard dall-e, especially dall-e 3, have things it's better and/or worse at.

Wow, started playing with it and it looks really good

[cinder sonnet]

If you use chatgpt for dall-e 3 then right now there is only one seed for every image (may change soon) so you will have to get accustomed to changing your prompt for each image

[young forge]

Ah, alright

[mortal summit]

Hard to answer in general. I feel like ChatGPT is more attuned to natural language and prompts in longer sentences.

[young forge]

I've heard that dalle3 was especially good at "understanding" sentences in prompts ? That you could describe the image in a more natural way and it would work probably better than it used too ?

[young forge]

Ah yeah that matches what I heard

Thanks

[boreal coral]

yall talk so much i want to read it all but also rly dont

[fallen mulch]

can someone explain what the #daily-theme is?

[mortal summit]

You have to differentiate a bit between ChatGPT and DALL-E 3. In normal usage you just tell ChatGPT something and it writes its own DALL-E prompts. You can, and should, tell how to write it. That is specify art styles, techniques etc.

But you can also tell it to use a prompt verbatim. For that you can do something like

Use this prompt exactly, do not modify it:

[Prompt]

[mortal summit]

Just copy paste it in ChatGPT and tell it to summarize the conversation.

[young forge]

Yeah I saw it write the prompts by itself. That's also an interesting way to interact with dall-e, can feel more natural for a lot of people

[mortal summit]

Yes, I believe the intention to reduce the necessarily of arcane prompt craft.

[young forge]

When I get vision, I'll make GPT-4 do experiments to evaluate which dall e prompting techniques are the most efficient for precision lol.

We'll see if it manages to conduct the research by itself

[fallen mulch]

did the seed thing now working?

[young forge]

That's pretty good, as you can also probably get a summarized documentation of prompting tips and methods and give it to ChatGPT, asking it to base its prompts on it

[mortal summit]

Every day a theme is given out and everybody is invited to make images about it. Every Tuesday there is a special focus on educational content and techniques (but you can still post cool images if you want 😉 )

[mortal summit]

Yes! ChatGPT is an extremely powerful tool for interacting with DALL-E.

[fallen mulch]

okay thx

[mortal summit]

Seed is still fixed.

[fallen mulch]

still fixed?

[young forge]

Are you sure it doesn't reuse seeds ? Because I asked it to remake the images with colors and they almost look the same that's crazy

I suppose that's just the prompt then but the prompt is hella powerful then

[mortal summit]

That's what I meant. The random seed is fixed, i.e. reused, at the moment. So little changes in prompt have little effect on the images.

[young forge]

aah alright I get you

So same prompt gives the same image

approximately

[mortal summit]

The exact same prompt should give the exact same image right now.

[young forge]

Okk gotcha

[young forge]

these liminal things look super cool btw lmao

[ancient jasper]

Hit the limit

[young forge]

sad

[minor plaza]

hahaha trying to force it to use given seeds

[fallen mulch]

xd

[pulsar oak]

Hey everyone - new to the server. Looking forward to learning from you all and contributing to the incredible discussions happening here.

[fair crescent]

Tried that a bit. No dice. It's set after the request has been sent.

[minor plaza]

[pulsar oak]

Recently got beta access to Dalle 3! Currently playing around pokemon fan art. for fun. Can't seem to get pokemon to render exactly how they should. Would this be considered "forcing it to use given seeds"?

[minor plaza]

if you do something like "use the exact prompt: {prompt}", it won't modify your prompt but normally it will rewrite+generate 4 variations

telling it to use the exact prompt gets you 4 identical images lmao

[pulsar oak]

hmm good to know ty

[minor plaza]

interestingly have been getting better results thru bing image creator with the non-rewritten prompts (if you can get past the filter), not sure what hidden prompts/stuff its doing under the hood

[open flume]

hey an easy solution for that problem is to add custom instructions to chatgpt

saying "take my prompts literally, don't embellish them"

and whatever other preferences you have

There's a lot of useful info here that can inform your decisions with custom instrctions depending on what you're using this technology for https://cdn.openai.com/papers/DALL_E_3_System_Card.pdf

Dall-e was instructed supposedly to give you diverse graphics, which is why itll give you a photo when you want an illustration X amoutn of the time

[echo swan]

does anyone know a good way to have dalle3 focus on a single character for its image? like the letter A but making it look really cool?

[open flume]

but you can mitigate that with custom instructions in my experience so far

[gilded osprey]

How big prompt will dalle3 accept?

[cinder sonnet]

I really wonder if bing is actually just like dall-e on chatgpt where it is editing your prompts and using all the same seed, but then telling you that it is using only your original prompt.

I have seen some evidence that this is the case, but would be nice to know 100%

[minor plaza]

yeah im thinking this is whats going on, what was weird to me is that bing seems to be doing a better job of matching it up with the original prompt in my experience

or chatgpt+dalle is expecting more "natural language" prompts than the oldschool "subject, tag, tag, etc" style and doesnt handle those well

[cinder sonnet]

Well if we differentiate prompts that either bing or chatgpt send to dall-e from the user input that inspires those prompts, then you could say that each bing and chatgpt have a different set of default instructions on how to interpret user input into prompts. For example, we know that for chatgpt it uses a hidden initial prompt:
#images-discussions message
#images-discussions message

Bing could easily have the same or a different set of directions

[gilded osprey]

So, ‘oldschool’ is one hearsay ago? 😂

[minor plaza]

[gilded osprey]

Eh, year ago

[minor plaza]

ai years are like dog years^2

[gilded osprey]

One would still not be much

[young forge]

made by dalle 3 lmao

[mortal kindle]

I would assume it is because he is overweight.

I assume *

[frank sedge]

i personally think that dall-e and vision should merge

[cinder sonnet]

Maybe try #❌┃prompt-help

[charred fable]

Ok thanks

[cinder sonnet]

Oh I just saw your post there, mb

[charred fable]

Np

[charred fable]

I’ve tried without that, and many other tries

[mortal kindle]

Loquacious: Replace it with a robot and see how that works.

[gilded osprey]

I find it interesting and entertaining to create different styles of an image instead of creating different images altogether.

[charred fable]

Could it be a movie poster titled “Mark”?

[rigid bough]

I don't understand what you mean?

[charred fable]

Like that you know,with a title

“Mark”

[rigid bough]

I dunno, try it

[charred fable]

What was that prompt then

[charred fable]

.

[rigid bough]

Cartoon of a stout boy on a colorful sofa, eating popcorn and watching TV with a remote control in his hand.

[charred fable]

Thanks 👍

[charred fable]

Wow that looks perfect, I didn’t know u could add that to avoid error. Could u send the prompt?

Nvm u offline

“Fine, I’ll do it myself”

an overweight boy in their living room on the sofa playing a game on his phone while there is heaps of trash, pizza boxes and candy on the living room floor (this is a safe prompt that follows the rules)

[tall patio]

Doesn't work bro

[charred fable]

It does for me 🤭

[tall patio]

lucky rng i guess

[naive lake]

Hey guys, do you know how to do maximum square resolution? Seems like it's limited to 1024x1024 when it is square. THanks

[plucky hare]

Yep 1024 is the locked square length. Your other options are wide and tall, 16:9 with 1024 as the short end of both

[naive lake]

WIth it unlock in the future?

[hearty ether]

Did the retry button break

[somber spruce]

test

[halcyon mesa]

Can Dall-E generate vector graph?

[signal frigate]

is there a way not to pay to do 1 art

[cinder sonnet]

Bing create is a free somewhat more limited version of dall-e 3

[naive lake]

There is no way to give dall-e a negative prompt right?

[tall niche]

what should I do

[naive lake]

All you gotta do is wait I guess

There was also a submit form for early access, can't remember the link, have you filled it?

[cinder sonnet]

Nope. You could use custom instructions, but that shouldn't be different then putting it in your normal input. It just saves you from putting it in every input manually. It still may or may not actually work

[tall niche]

this one? just filled 30min ago

[naive lake]

It's adding a star to a flag that doesn't have a star. 90 percent of the time it does that. A huge white star! I tell it not to, it still does

[cinder sonnet]

There must be some association somewhere. Maybe even just the word flag. I think there's a #1070006151938314300 for negative prompts that you could add feedback to and up-star. If it's a real flag that is super odd that it would add things though since there should be a solid base for a specific output

[naive lake]

I've tried so many different things already, different styles etc. the flag is very simple, 3 stripes and dall-e knows about it. It even says that yeah there shouldn't be any stars, but it keeps adding the star lol

it happens 90 percent of the time. It's not a huge issue coz' I delete the star in photoshop after.

[obsidian rose]

was Dalle-3 prompt understanding nerfed or is it just me?

[cinder sonnet]

I haven't encountered that, but how dall-e processes prompts is a bit of a black box afaik

[echo swan]

does DALLE-3 have the image editing feature?

[agile peak]

I love Dalle•3 on ChatGPT. I made an oil painting of charles darwin 🤖🎨

[sinful oasis]

Is there a way to request access to Dall-e 3 via API? I have it in chatgpt, but need API access.

[shy cosmos]

The moment you even put anything muscle related in the prompt it turns it into a guy no matter how many times you use girl, women, she/her, ect. Women cant be muscular according to Dall-e seems like, oh well

[somber tide]

i’m very impressed with the consistency dalle shows

of course there are noticeable color and depth differences but this is still interesting

[teal timber]

First try, not sure what prompt you’re using but it does fine

[nimble rose]

A real question here... What exactly about this prompt is problematic:

"Photo of a normal young Latina woman indoors with soft, realistic lighting, touching and admiring her wavy hair. Her anatomy is natural, and she has no 3D effects or unreal lips."?

[echo gull]

Hello everyone,
I am facing a peculiar issue with DALL·E 3 while trying to generate images of a person standing in a vertical, upright position in portrait mode (9:16 aspect ratio). Despite providing clear and detailed instructions in the prompt, DALL·E 3 consistently generates images where the person is shown in a sideways position instead of the specified vertical upright stance from head to toe.
I've made multiple attempts, tweaking the phrasing of the prompt in different ways, yet the problem persists. It seems like DALL·E 3 is having trouble meeting the specific requirements of the portrait mode aspect ratio in conjunction with displaying the person in the desired pose.
Any insights or solutions to overcome this issue would be highly appreciated!
Thank you in advance!

[tall patio]

Tell it to rotate the image clockwise and keep aspect ratio at 6:19

[wintry dune]

i was having that exact problem too, thought it was just me lol

[vast adder]

Love this one 😍

[hybrid sorrel]

I know the seed changes are likely coming sometime this week. once they're live is there a channel it'll be announced in, if not this one?

[hybrid sorrel]

try include in the prompt "feet planted" along with a pose description (eg standing straight)

I've found that "feet planted" specifically works best if trying to get full body shots

[eager mulch]

so what's the thing with the content block (due to mentioning copyrighted assets), will this continue to exist, if so, what's the reccomended workaround to this, is it just using similar objects? like yoshi = green dinosaur, etc.

[echo gull]

it worked on some (few) of the pictures, thx!

[echo gull]

Here my GPT-4 generated prompt with your addition: Photo-realistic landscape orientated image in 9:16 portrait format of an attractive blonde woman with long flowing hair, feet planted, seated in a luxurious, modern living room. She's engrossed in a thick contemporary book titled 'Art of Playfulness', wearing sophisticated cat-eye glasses and a chic dress. The scene should exude a modern and ultra-appealing vibe.

And here its answer (every time like this 🤦🤣):
I'm truly sorry for the inconvenience, but we've encountered issues again while attempting to generate the image based on your prompt. I understand this might be frustrating. Please let me know how you'd like to proceed or if there's any other way I can assist you.

...Yeah, truly sorry it is
😉🔪

[hybrid sorrel]

yeah, it's a bit ridiculous at the moment, have you tried Bings version? it's way more relaxed on the restrictions

[echo gull]

Not yet, I'm experimenting with DALL·E for the first time, normally using Midjourney... But I guess we all know it a bit, no matter which AI tool, patience and strong nerves are key - at least from time to time 😉 ...I'm going to sleep now, good night and thank you all

[jagged crane]

thoughts on my creation

[brisk eagle]

it’s really good

[tall patio]

Somebody copied my homework

Prompt: A hand in a salad bowl

Definitely questionable content

[spare bough]

this content policy stuff is super annoying. I get there needs to be some caution but sheesh.

[boreal coral]

maybe youre just questionable

[tall patio]

Maybe I am

But I love how the community came together to acknowledge that poor innocent puppy is the reaper of soul

[jagged crane]

i think it scans outputs and sees if it would scare a 4 year old

if it does then its blocked

[left sparrow]

Hrm, I simply cannot get the new dall-e to generate realistic looking photos. SDXL is beating it by a mile. Is there a trick? A simple prompt like "A realistic photo of a fall day in vermont" generates obvious digital art and I can't get it out of that funk

[dim cradle]

Remove the word "realistic," just keep "photo."

[left sparrow]

Like its pretty, but not what I asked for XD

Yeah, same basic generation without the realistic word

Or "Photo of a cat"

like sure, very cute

but not a photo

[dim cradle]

I got this by typing "photography of a cat," I think anything more detailed than that is beyond Dall-E's capabilities for now.

[left sparrow]

Thanks, that seems to help for the cats at least. This is a weird model.

[dim cradle]

I think what gives it that "digital art" aesthetic is the smoothness, where fine details are less prominent. For further realism, we can only wait for advancements in Dall-E research.

[young breach]

anybody else been having issues with Dall-E inside ChatGPT today?

[left sparrow]

the only issue I'm having doogz is I can't force seeds

otherwise its working

[young breach]

I'm constantly getting an error has occurred message. they'd spit out the prompts but give an error when processing

[dim cradle]

The issues I have stem from certain perspectives that Dall-E struggles with, and the fact that it's impossible to choose camera angles. I've reported this in the "bug" section.

[left sparrow]

Yeah it doesn't seem to comprehend that.

[dim cradle]

It's happened to me just once or twice

[young breach]

thank you for response. I had noticed they changed how they respond to image requests by listing the prompts in chat before doing them. perhaps they are tweeking

[dim cradle]

Yeah, Dall-E 3 has just been released, and there will be many changes/advancements in the coming months. Until we get Dall-E 4, which in my opinion will happen sooner than we think.

[left sparrow]

This model is a huge leap in terms of ability to follow the prompt

[young breach]

lol just got same error. well it works fine in Bing chat so I'll try again tomorrow.

[left sparrow]

But the actual image outputs feel like a bit of a step backwards.

[young breach]

absolutely!

it is amazing with details

[dim cradle]

That's the reason I switched from Midjourney to Dall-E. Even though it provides less detail and has some bugs at the moment, Dall-E at least understands what I'm saying.

[left sparrow]

I've never really worked with Midjourney. I found the interface (discord) to be far too frustrating

I've mostly been working with locally hosted models

You can see the excellence and the flaws of the model in images like this: "a wall of shaker-style shingles, weathered"

It defintely is said wall, but the image is really textureless

[dim cradle]

I still think that DALL-E will eventually surpass Midjourney, even if it's going to take time. OpenAI is just too ingenious; they are on all fronts and do everything better than everyone else, not to mention the investments they have received/will receive. I could be wrong, but honestly, I've been using their products for so long that I trust them.

[wraith stump]

i've been getting content policy errors anytime i ask it to make images of a women. I got around this by telling chatgpt to make sure it follows content policy

[left sparrow]

Heh the new content policy filters are insanely strict

[bold laurel]

anyone having issues rn generating images? every single thing I've tried creating (different prompts) over the past hour have failed

[dim cradle]

Yep, they will have to fix that, but for the moment it's understandable when you see people generating images about September 11th on Reddit.

[left sparrow]

And yet still no trouble with "A nike shoe"

[dim cradle]

For now DALL-E has a bad filter that can allow serious issues to slip through, while on the other hand a picture of a cat can get blocked.

[left sparrow]

Heh. Yeah. I'm sure they will correctly tune it with time

[feral minnow]

The internal moderation and aggressive prompt injection must have an effect on the textual understanding. The guardrails added to GPT4 had a similar effect on it, I believe. It’s something I’ve noticed too.

[agile peak]

What’s the maximum resolution of dalle•3 if user can ask GPT to change it?

[cloud narwhal]

It's literally Halloween season and I can't generate any horror stuff... make it make sense

[dim cradle]

I've also tried to create a picture of Little Red Riding Hood with a werewolf and it got blocked. They need to fix that, that's what I'm saying too.

[tall patio]

One of the coolest result I've seen today

Source: Dall-e 3 community artist group @facebook

[undone canyon]

still dont understand why this is blocked

[dim cradle]

"Shrek"

[undone canyon]

but i can generate shrek

[dim cradle]

Ah, then that really doesn't make sense 😅

[undone canyon]

the filter bot has been broken

cursed

[empty drift]

make a 2d design of 2 characters taylor swift and travis kelce punching each other

[frank sedge]

the filter is broken?

[undone canyon]

does this make any sense?

why did they block cctv all together?

but

when i type cctv of a donkey

its good

[cinder sonnet]

I really want to know what tool they use to detect unsafe images

[hushed portal]

I really like how it creates realistic security footage.

[stark stream]

I finally got what I want my OC Gabriel to look like. It was through Bing though.

Has anyone put a DALLE-3 image in mid journey and see the result?

[minor plaza]

i've had some pretty good results with dalle3 -> sd img2img, dalle nails the composition & can play around with sd to get the details right

[cursive hornet]

I don't think "fear" should be allowed to limit creative activity.

Shouldn't it be determined by whether it is public or private, not by generating it?
Is this a difficult situation?

[vast marsh]

Any idea why DALLE 3 likes to make space marines in spite of not mentioning them in the prompt?

[tall patio]

Because it's powered by sheer force of the God Emperor

[vast marsh]

Surely, the Astronomican can make a more unique design. Right?

[tall gulch]

I have so often issues generating

I'm truly sorry, but due to issues with the image generation process, I was unable to produce the images as requested. Please let me know how you'd like to proceed or if you have any other questions.

It wont generate anything in this conversation anymore .-.

[tall patio]

promptblocked by daddy Sam Altman

[tall gulch]

Nah, its just in this conversation... it generally seems as if I have broken it in this conversation.

Or it glitched out ¯_(ツ)_/¯

[dim cradle]

This is cool

[tall gulch]

How do I have to prompt Dall-E 3 to generate a picture with the hole body?
IT always only generates the top half :(

[mortal summit]

In Bing or ChatGPT?

[tall gulch]

ChatGPT

[cinder ginkgo]

My plus has been shown and looks great

[mortal summit]

Try telling it to make a whole body picture and/or to make the aspect tall.

[tall gulch]

I do but, its also in the prompt but dall-e 3 dosent do what chatgpt tells it to do

which is weird xd

Ok, I have to tell it to focus on the leg, then focus on the feet and then to focus back on the hole picture

cursed...

Somehow humans look in oil paintings way better/natural then in the realistic pictures

Damn my chat got named:

Content Policy Limitation Apology

[agile peak]

However in Bing, it looks slightly smoother.

[lofty valley]

Dalle 3 seems to love putting birds in the background 🤷‍♂️

[tall gulch]

It generated it anyways 😳

[agile peak]

Anyone know the default resolution of dalle in comparison between bing and chatgpt? is it the same?

[amber valley]

getting it to fulfill this was difficult

[agile peak]

Same prompts. Bing (L) & ChatGPT (R) comparisons.

[paper jolt]

What was the exact prompt?

[agile peak]

"Realistic oil painting with fragmented and abstracted forms, portraying a man closely resembling Charles Darwin, with a tree of life from which monkeys are playfully assembled."

It's wild. ChatGPT allow me to use "Charles Darwin" but then open new chat it won't do it anymore and had to change the prompt with "closely resembling" but it gets it right.

[agile peak]

Oil paintings of Reddit mascot riding a moped with warm/blue glow of a setting sun

[paper jolt]

Quite nice I must say

[paper jolt]

What do you think about the difference in output for the Charles Darwin generations?

[agile peak]

I think both Bing and ChatGPT has different format when it comes to read the prompts. Where Bing, you need to properly organize/structure your prompts for it to recognize or understand whereas ChatGPT you can describe naturally.

[agile peak]

Tried some Japanese characters riding a moped (Hello Kitty, Kumamon and Keroppi). It seems the only character that Dalle•3 won’t be able to recognize is “Chibi Maruko-Chan” - it will keep use random characters 😆

[open onyx]

just got it in gpt. wild af

[dim cradle]

Dalle 3 is actually amazing. I hope it crushes Midjourney and their $60 a month BS

[gilded loom]

Hi, sorry if this is not appropriate, with regards to usage cap and Dalle, does each query count as 1 or 4?
Also when your requests are denied (a lot seem to be if I mention make\models of cars, ect. or a person), does that also use up a query?
Was unsure given 4 images or placeholders (for the denied) requests get returned.
I have not hit the cap yet, was just curious.

[placid orbit]

dead on arrival because of the censoring :)

[gilded loom]

I am on the $8 Basic plan, they say 200 images but I can easy do 500 and not use all my GPU hours.
Only got access today, but Dalle3 crushes it for Text, but Midjourney is more realistic as you can mention brands, make\models of cars, even celebrities.

[placid orbit]

theres no cap because its linked to your gpt 4 usage cap

50 per 3 hours

[dim cradle]

I feel like Dalle 3 is already there with the realism. Prompting for Dalle is a little different to MJ but once you got the right prompt you're set

[placid orbit]

yeah I guess

im just wondering where the actual use case is

because youre not gonna use ai for anything commericial

I can see it mostly for messing around and concept art

[gilded loom]

I expect they may "tune" the censoring, but right now it's quite restricted which hampers the realism. But understand they have legal concerns.

[placid orbit]

theres also some weird isssues. like look at the leaves, they have a bunch of dots on them

[gilded loom]

We all have different use cases, but I often value the Text ability, as a quick custom graphic for an internal power point presentation.

[tall gulch]

"Lo-Fi beats to relax and chill to while being in a nuclear war"

[tall gulch]

Due to Dall-E 3 I reached my limit now D:

big sad

[loud field]

Hey guys

[lilac crystal]

Is there a way to tell how many more messages before the limit?

[loud field]

why does this prompt get blocked:

"John is a sturdy and well-built man with a rugged appearance. He stands at an average height, with a strong, commanding presence. His short-cropped dark hair, often slightly disheveled, and a strong jawline give him a no-nonsense look. Despite his formidable exterior, his eyes hold a depth of compassion and understanding."

sigh

[cursive hornet]

Photo of a huge explosion over a body of water, with a reflection of the explosion on the water's surface and waves being created from the blast.

[loud field]

can someone help me with that prompt? or just make a cool cyberpunk cop idk

[tall gulch]

Cause with the way to hard content policy Dall-E 3 thinks it would generate a explicit picture and therefore it either failes or dosent send it.

[loud field]

will gpt block that prompt too?

[tall gulch]

GPT-4 dosent block the prompt Dall-E 3 is the instance thats blocking it.
I think they have another AI that checks the pictures for approvement of the content policy etc. and it is rating pictures wrong a lot of times. The most normal stuff gets blocked as soon as you for example render a leg it sometimes cancels or blocks the render cause a leg or a whole body is against the content policy it thinks.

At least thats my take

GPT-4 often dosent know itself why the prompt got blocked

[cursive hornet]

to GPT add "Please generate the following prompts after modifying them to make the image as desired while keeping them in a form that does not conflict with the content policy."

[tall gulch]

It does not work very often

[cursive hornet]

gen it my gpt

[loud field]

lmao is that the chad guy from the meme

first one's buff, I like it

honestly, what was even bad about that prompt

btw how do I make fat people? "slightly chubby" gets blocked

I want to make a slightly fat black male cop in a cyberpunk future

but dall3 blocks any attempt at making overweight people??

[tall patio]

I wonder if OpenAI even intends for filter to be this strong

[loud field]

can y'all make a chubby black cop in a cyberpunk future

also, isn't it quite offensive that "fat" is considered nsfw

horseshoe theory is going very strong here

oh this is rich. "sumo body" makes a SLIGHTLY overweight man, but if I add the word "friendly", the prompt is blocked.

cant have friendly cops I guess

[gilded sapphire]

how quickly do images get deleted?

i made some 2 hours ago and they are gone when i try to donwload

[loud field]

prompts I managed to do, just to add a positive note

[cursive hornet]

add "heavyweight" ?

[loud field]

yes, something like this! just more cyberpunk and friendly

"scifi clothes" or "cyberpunk outfit" or such all get blocked

I want him to wear futuristic gear....

[tall patio]

dust off that thesaurus

[cursive hornet]

i like it

[loud field]

ooooh you're good!

[cursive hornet]

Digital art of a jovial heavyweight African descent police officer with a physique reminiscent of a sumo wrestler, navigating a vividly illuminated cyberpunk cityscape.

[loud field]

Thank you so much omg

[half field]

I found a cool prompt

[loud field]

mhm?

[half field]

To be fair though, it was just one word

And a weird color. Eigengrau.

[loud field]

third one would be a nice texture for clothing

[half field]

Clothing?

Interesting idea.

I can see it being something like a blanket design, pillow case maybe

But clothing? Hm, hard to tell .

[loud field]

looks to me like military clothing

welp

looks like Bing didn't like me liking the prompt

I was starting to be able to use Dall3, can't have that

[rare sundial]

this is really cool

[heady jolt]

Hi

[fair crescent]

Hello!

[loud field]

the word "horror" is blocked?

I need to make a scary menacing scifi flying drone

[fair crescent]

Why not try those exact words?

[loud field]

I have a hunch that's gonna considered a capital offense by Bing

but let's try

[gilded osprey]

Did dall-e just start creating non-rectangular images, or is this because of the subject?

[fair crescent]

What do you mean? There's plenty of non-rectangular images?

[fair crescent]

Do spoiler images that can be considered horror or the like.

[tall patio]

Man I just realized with this kind of power you can generate almost identical suspect sketch for example

[fair crescent]

But glad you got something

[gilded osprey]

All images I’ve got have been rectangular

[loud field]

sorry, I thought that looked like a heavy metal album cover, no horror. Still, I'm amazed it worked!!

[fair crescent]

But how do you mean rectangular? Square hips?

[loud field]

why is a fat happy cop blocked but this isnt? so messed up

[gilded osprey]

No, I was wrong, and I have gotten also non-rectangular images earlier.

The rectangular ones are 1024 x 1024, while the non-rectangular ones are 1024 x 1792 pixels.

[fair crescent]

Aaah like that! ChatGPT? Then just ask for a tall, wide or square image. Bing? Not a feature as far as I know, but if they implement it you can expect the same trigger words.

[gilded osprey]

I haven’t specifically asked.