images-discussions

[agile peak]

Giving me Stargate vibes

[agile peak]

No idea why. When adding “Bumble” as text to analog computer Dalle managed to capture diversity of people.

[ripe wraith]

oh i've got an idea why... humans are bumbling apes in its mind

[agile peak]

It generated more than one ethnic.

[obsidian rose]

I actually got lucky to get 1 female in the left. Usually even with prompting like that you get all males

[fallen folio]

Guys

[ripe wraith]

That's like asking "create pics of very feminine men", not wrong in any way, but understandable why it'd get caught up in filters.

[fallen folio]

have they implimented random seeds?

or not yet?

[ripe wraith]

I'll ask it next chance

[obsidian rose]

I think filters like this are just very silly

[earnest flame]

If you desceribe them wearing more articles of clothing you'll have more luck.

[wintry dune]

its just not the same as bing, the images bing puts out are wild

i said muscular woman with muscles visible and flexing lol

[wispy storm]

Bing and openai doesn't intereact the same way. Prompt are different

[cold orbit]

how can i create my own picture with ai generated like dell-e ? can anybody help?

[wispy storm]

Hi you can use the dalle on bing to try for free

[cold orbit]

thank you

[flat zephyr]

I wish we could do image prompting

[obsidian rose]

trust me, it only gets worse the more articles of clothing I describe.

[wispy storm]

Wdym ?

[flat zephyr]

give an image as prompt, like in midjourney

[wispy storm]

You can with vision

With chatgpt

[wispy storm]

Like that?

[obsidian rose]

I assume you didn't ask for muscles? because they are definitely female characters with armor but they don't have a lot of muscles

[wintry dune]

the hard part is a woman with muscles that is still obviously a woman

[obsidian rose]

that is impossible with ChatGPT, very easy with Bing

[wintry dune]

oh yeah on bing i have the best images ive ever seen of that particular thing

[wispy storm]

How much muscle do you want lol

Like hulk ?

[wintry dune]

didnt you see my example above crossing her arms lol

[wispy storm]

Like over exaggerated?

[wintry dune]

im kidding but its like there

its like there is not inbetween either hulk or no muscles showing lol

[agile peak]

I like the art and coloring ✨

[obsidian rose]

I mean we can't make female barbarians for DnD campaigns with ChatGPT

[wispy storm]

Maybe giving example of what you want like inspired from x universe

That's what I have done for star wars theme

[wintry dune]

oh yeah i got that style down perfectly, it looks so good in almost all my images

[rapid hull]

do people response in prompt-help?

[lone mantle]

On mobile app it doesnt show anymore the prompt limit?

[rapid hull]

sorry to sound impatient

[lone mantle]

50/3 hours

[obsidian rose]

It's probably just flat out impossible to get a female bodybuilder in ChatGPT no matter how many images you do for example

[rapid hull]

just wondering if I should post here instead?

[waxen geode]

What’s the question?

[wispy storm]

Depend what time zone

Ans yes people do reply

[rapid hull]

thanks I will wait

[agile peak]

Early test of ChatGPT be like

[obsidian rose]

I should say a female bodybuilder with female traits like real female bodybuilders

[wintry dune]

thats what i mean its a womans head on a mans body, while bing actually makes an entire woman

[thick smelt]

bing dall-e adobo anyone? 😆

[wintry dune]

yeah exactly like thats straight up a mans broad chest

[obsidian rose]

yeah, I mean that is just getting even sillier. ChatGPT can't make muscular females

[wintry dune]

yeah i would show an example from bing but uh bing makes them maybe too feminine lol

[thick smelt]

what if... ChatGPT is in the 90s? prompt results:

[agile peak]

I am hungry now at late night lmao

[thick smelt]

Because I'm trying to test Filipino food in bing Dall e 3 😂😂

[remote sphinx]

so what can I do to improve this prompt?
a fake governments photograph of a damaged alien spaceship wreckage at the bottom of the ocean, the top of the photograph has the words "Classified"

[dim cradle]

damn

the word "war" is blocked

[sacred temple]

same prompt. obviously different results!
BING DALLE vs GPT DALLE

[dim cradle]

is gpt dalle exclusive to pro accounts?

[scarlet urchin]

yes

[agile peak]

Yeah. Annoying isn’t it. Just replace with military instead (eg war rooms to military rooms).

Today I just learned that, using fluorescent lighting and neon hues [add specific colors] from neon lighting can totally produce entirely different atmosphere.

Neon is perfect for producing Synthwave or Retrowave images.

Also adding Diffused lighting/shadows alongside with Cinematic lighting could increasingly produce realistic images.

[agile peak]

ChatGPT is obviously more complex in comparison to Bing. Imo.

[gilded osprey]

ChatGPT’s user experience is also much much better, in my opinion.

[azure rose]

hi

[obsidian rose]

I think Bing is much more creative than ChatGPT. Just copy ChatGPT's prompts into Bing and get the best experience

[scarlet urchin]

i really think it just depends on the prompt. ive gotten better results from bing with simple prompts while gpt requires much more detailed prompts

[agile peak]

Exactly. ChatGPT is good to go if you want for more detailed and complex prompts - you can also ask ChatGPT to discuss about your prompt before generate or let it improve using more fitting textual for Dalle•3 to understand.

[real atlas]

is the form still the way to get Dalle3 access or are they just slowly rolling it out to all Plus members

[fair crescent]

Have you copied the prompts from the results? They might be wholly different in ChatGPT.

[rapid fiber]

Where do I drop ideas/feedback for ChatGPT DallE?

[agile peak]

Experimenting with a cinematic look and moody green filter.

[dim cradle]

so buying credits in openai site gives better images than with bing for free?

[dim cradle]

hey. It seems that prompts have to be literal. I mean, I wanted spiralling clouds around a beam

https://th.bing.com/th/id/OIG.Ye2m9FfeGF2pqT8VaXGV?pid=ImgGn

Howerver, I thought that the AI would place the spiralling clouds on the sky because it's obvious. No, it isn't obvious for the AI.

[wanton hatch]

"80s anime OAV vintage blurry style", it interpreted keywords in bulk more or less immediately 😅

[jade stag]

that's pretty damn good

[sacred temple]

Achievement, unlocked!
I told GPT that my prompt is from a clip interrogator. This is what it said.

I understand now. The prompt you provided consists of latent space tokens, which are essentially coordinates in a multi-dimensional space that DALL·E can interpret. If you provide the exact latent space tokens, I can use them to generate an image directly without creating descriptive prompts.

Please provide the latent space tokens, and I'll proceed with generating the image for you. If you don't have them readily available, I can continue using descriptive prompts as we've been doing.

[scarlet urchin]

i love how it understands 80s anime. and making it grainy is chef's kiss

[ripe wraith]

Remember how I said first thing I'd do with vision is make art about my cats? 😄

[obsidian rose]

try Bing, it usually works

[agile peak]

I tried using 2 same character in 4 different settings, turned out the weakness of GPT-4 dalle is creating an image of people diving into the planet. It keep generate the same image as if they are floating out of the space rather than heading into the planet lol

Four attempts, the result is the same kek

However Bing Dalle managed to give the result when using "diving" to the planet:

[dim cradle]

bing is censoring the word "fight" too

damn

[stable dock]

So is it DALL-E itself that's rejecting the prompts or is it bing?

[tall patio]

Tinfoil take

OpenAI throttled D3 GPT version so that everyone would use the Bing version

[earnest flame]

Did you just describe this as having a filter?

[tall patio]

Would love to see frame interpolation feature being done by D3

With that I might never need to break my back for pixel animation again

[rustic prism]

I pay for GPT Plus but dont get free access to dall-e 3?

[wispy storm]

#announcements rollout happening patience is the key

you can request the alpha but it is manually accepted

[scenic cloud]

So there is no point in using Bing Dalle

Everything is against content policy apparently

[tall patio]

Just wait

[gritty ravine]

The images that you all make are so impressive. I tried, but can never get the quality and detail. What is the secret?

[obsidian rose]

cherrypick the images, make prompts with ChatGPT and paste on Bing if you are using it

I don't think most images you see are first try

[graceful hill]

Lately I like making wildlife photography.

[wispy storm]

hi there, thats cool and impressive! You can use also the channel #animals

[wintry dune]

lol i told chatgpt to make the image a just a bit more gritty, it went from this to this lol

[agile peak]

Yes. "Cinematic look as if it blend with green moody filter/tone"

[agile peak]

You can just ask chatgpt3.5 to use improvise/alternative words

[agile peak]

Beautiful

[agile peak]

Use alternate words such as Battle, Confrontation, Brawl, Duel, Clash.

[dim cradle]

for some reason some ppl are generating images using the words that get my prompts blocked. Or, they used other language in copilot

[feral minnow]

I love the aesthetics it gives me when it accidentally forgets to block the prompt.

[dim cradle]

dall-e in bing cannot create a wizard with one hand holding a ball of purple energy and the other hand casting a white lightning. It always creates a wizard with both hands holding the same energy

[mild ermine]

Excuse me, if I use the regular Dall-E 3 interface on OpenAI's web site, can I use a reference image?

[mild ermine]

yeah well hands I think are tough, I noticed that with chatGPT Dall-E the other day too. I think the issue is if you don't have both hands on something you have to render more of the hand so it's not a decision that it's going to casually make. I've noticed when you do manage to get one hand off of something it often changes the image quite drastically in ways that aren't intuitive

[uncut abyss]

why hasn’t someone been able to crack bing / Dalle’s restriction 😭

[agile peak]

You can. Ask chat gpt to use alternative words lol

[uncut abyss]

usually doesn’t work

at-least for me

[agile peak]

Wild stuff today. I was making a poster for an animated satires movie, about political and social stuff in indonesia.. one of them is about religious intolerance, jeez the generated images by bing was wild.

I even mentioned specific groups; but Bing keep focused in one group as if it’s the source of intolerance lmao

Space marine. With ChatGPT from young (25) to old (Age 100).

[jolly hollow]

Did we ever get any additional information on how rate limits work? GPT says that it's per user and set "to ensure the best experiencefor everyone." I'm curious if it is related to how busy the server is, requests per 3-hour period, requests per day, or something completely unrelated.

[remote sphinx]

anyone have good way of making liminal images like the pool rooms?

[tall patio]

very nice

[rich star]

[heady crane]

Why on chatgpt I can’t get real world characters??? While on Microsoft bing I can get it accurately??

Look at the difference 💀💀💀💀💀

[agile peak]

Different seeds. Gpt is still on research preview.

Yeah. I always find Bing version producd much better results.

[heady crane]

will be fixed soon ?

[boreal coral]

why would he know

[agile peak]

Heard it will be later. But hoping it will - cause it's kind of frustrating.

I am using Dalle•3 in ChatGPT for non realistic stuff. It's good enough but still far behind from the Bing version to achieve realism or photographs images.

[vapid granite]

the bing version is the newest version (according to Chatgpt)

[agile peak]

Wait what? 😳 what's the difference, the release version?

[vapid granite]

GPT4 is the alpha. I can only assume the Bing version is the beta. I don't think there is a final release version 1.0 yet.

[agile peak]

That's explain why it is under research preview and produce less quality of images. So not really Beta then; it should have been called Alpha version.

I have been seeing it's more like Dalle 2.5 in ChatGPT.

[vapid granite]

That's what I said GPT4 is the Alpha and BIC is the beta. There isn't a final release candidate yet.
"Alpha is the very first testable version, it might crash and not have all the features on the software. Beta is a version with all the features, but with bugs and possible crashes."

[loud field]

And here we go again....

"a menacing medieval deity: he is a strong medieval man, with a bodybuilder look. He has glowing golden eyes. He has long, curly, fiery red hair that turn into flames. He is wearing the clothes of a medieval nobleman. he has golden chains tying his wrists."

prompt blocked

grrrrrr

[obtuse bough]

Censorship and copyright

[loud field]

what IS unsafe about my prompt????

[fair crescent]

Chains I guess

[loud field]

chains are not inherently offensive omg

[fair crescent]

Or fire in combination with people. You might not want burning people.

[loud field]

I fremoved both fire and chains, still nothing

pisses me off really

[fair crescent]

They aren't inherently offensive to you.

But it might be something else. I am not privy to the actual restrictions. I can only have an educated guess.

[loud field]

guess I can't get that character drawn eh

maybe "menacing" is offensive

[fair crescent]

That can be the case

[loud field]

trimming the prompt so much it'll end up being a different character

now it works but

it's not him, it's some other dude

how do I add golden chains without triggering the filter

and how do I stop him being some real life anime character

any advice to make his face uglier?

he's a lil' too perfect

but crooked nose, for example, bans the prompt somehow

....dude just can't wear golden shackles somehow

or have burning hair

grrrrr

(╯°□°)╯︵ ┻━┻

someone help me out here, I can't make Bane the God of Hate from D&D, somehow

[agile peak]

Might explain the Dalle•3 difference result between Bing and ChatGPT.

[loud field]

Why is my prompt constantly blocked

"Russian bodybuilder with a visage that defies convention. His glowing golden eyes command attention, radiating an uncanny aura of mystery. Cascading locks of fiery red hair, long and curly, frame his countenance with a fiery intensity. A slightly crooked nose adds character to his face, hinting at a rugged. Sitting on throne."

I am losing it

[wispy storm]

Try without Russian

Or radiating

[frozen fiber]

this is the second time DALLE randomly introduces a gun

wasnt that against ToS or something xD

like would it be possible to have negative prompts on DALLE too in the future?

[loud field]

anybody got gptdalle here?

"three panels comic. First panel: an owl asks a robot to draw. Second panel: the robot frantically draws. Third image: the excited robot shows the owl an image of a square with two round circles on it. disney"

This gets a content warning too.

oh wait.

silly me, the square with two circles....

I didnt even realize what it could look like

ah. nope.

"three panels comic. First panel: an owl asks a robot to draw. Second panel: the robot frantically draws. Third image: the excited robot shows the owl an image of a square. disney"

still blocked

this bing-fu is tiresome

[tall patio]

Just wait

Once D3 is scaled for Enterprise customers you guys might be in for a new era

[frozen fiber]

what

what are the best subreddits for posting art?

[hearty ether]

Why do content filters delete the prompt

I can’t even edit it to fix the issue

That’s so dumb

[frozen fiber]

are you on bing?

[tall patio]

In best case scenario, the next big enterprise customers is Tesla

Musk is never boring for sure

[loud field]

"POV shot of being grabbed by a fierce female feathered humanoid bluemarine parrot, real, red crest, teeth, parrot beak, slightly dinosaur-like, wearing ragged pirate clothes. cinestill, jungle."

somehow it blocks the prompt.

I bet it adds.... two round shapes.... to the parrot.

lets try to remove female, no women allowed

still not working

does that prompt work on gpt?

uuuggggh

I was wrong, the filter's still awful

today I can't get anything made

I would like to show some fancy comics I made thanks to dall3, but unfortunately an automatic filter here blocks anything that isn't written in english

finally, an angry bird

[fallen mulch]

anyone have dall e on phone?

[wanton hatch]

Completely nonsense here the content policy warning...

[plucky kraken]

is there a limit on immage in dall-e in chat gpt

[fallen mulch]

yeah 50 like messages per 3 hours

[plucky kraken]

Thanks

[loud field]

I dont know why yesterday I was all happy that dall3 was working. Today, the censorship is as ridiculous as usual.

I love how people can make unsafe images with it all they like, but I can't use it for normal stuff

I wonder what happened yesterday, when Dall3 was working again

[ionic junco]

So is there a policy violation, or not?? Using dalle in GPT

[wispy storm]

Word disney

[wispy storm]

What was the prompt

[ionic junco]

create images of a mother tucking her special needs child into bed in a motorhome at sunset while the rest of the family enjoys a campfire outside

[ionic junco]

I know it’s odd, but I have a special needs sibling who loves for me to create images of his life.

[ionic junco]

I created about four of these with no problem, then it flagged it

[wispy storm]

might be specials needs chils might trigger

you can always put thumb down to review a prompt and ask why

[ionic junco]

I filled out the feedback Google document, but the warnings disappear after awhile.

[loud field]

https://media.discordapp.net/attachments/317399835856601093/1162663087858532392/360b7e21-1858-4df8-8fa5-293bc5942923.jpg?ex=653cc161&is=652a4c61&hm=de5a743c070e791c1598c1887ac728ca33b69572b5f921a0289935120f8213ba&

https://media.discordapp.net/attachments/317399835856601093/1162663088189874236/64ceee26-e70d-4c80-864d-74bdb4467966.jpg?ex=653cc161&is=652a4c61&hm=3533d3c4e22728a5fe560a5864bb86b1f74b568a2a212cf8d6311609c55526c1&

A child and her pet Lich

[agile peak]

One thing I love about GPT version. Dalle is great for increase and decrease the brushwork and detail in a single image.

But the strange part it love to generate ridiculously a lot of people in background lol

[ionic cradle]

does anyone know what the usage limit is for dall-e 3

[loud field]

The usage limit is 100 inquisitordogs per one successful image

Wonder why yesterday it worked

[ionic cradle]

what does that mean 🤨

[loud field]

You'll find out.....

[wispy storm]

50 prompt per 3h

on chatgpt

[ionic cradle]

it generates like 4 pics each time right?

[silent thorn]

Sometimes 4 other times 3 or even just 2

[marsh depot]

Hey guys is dall-e 3 integrated in the mobile version of chatgpt plus?

[white skiff]

yes

[frozen fiber]

itll flag some images and display the rest

btw can I see further into the bing image creator history than is shown?

[agile peak]

I think to produce good images with Dalle in ChatGPT. After first creation you can ask "Hey? Why do you change the prompts? didn't meant to create images. I just wanted to discuss."

It will ask you to describe more details before generating images.

Then it start writing Foreground, Mid-Ground and Background subjects including settings before proceed to ensure it meet your prompts.

And the result are significantly improved. With moody green tone.

With cinematic look.

Lastly, dim shadows and fluorescent lighting.

[tall patio]

Inpainting or regeneration from a seed image with specific editing would literally stomp Adobe out of existence

[agile peak]

That would be nice. Currently Dalle in ChatGPT seeds are locked to 5000. Asked GPT to change it but it doesn't have a permissions, hopefully in the future it will be allowed.

[white skiff]

i think some features will be only available in labs and not in chatgpt

[tall patio]

Firefly 2 is pretty good with pose too but it seems their results still have too much noise

High rng generative fill too

Manual editing is still much faster than relying on AI for Adobe

[agile peak]

After getting familiar with ChatGPT Dalle:

• It excels in adjusting settings for specific images even after they have been created.
• It is particularly effective in creating oil paintings and other forms of art.
• It capable to consistently generates the same images with variants when pointed out.
• It capable to understand description using story-telling style.

But still have flaw in achieving photorealism in compare to Bing. Probably due to different "Data training model", "Hyperparameters" and "Prompt input that relies with GPT, sending to Dalle".. or just still early version.

[lone mantle]

@tall patio they have yet to update gen fill in PS to firefly model 2

Cant wait for that

[tall patio]

You can use it online now

For free I heard

[lone mantle]

Web UI you mean, yeah model 2 is there but not yet in PS and gen fill still uses model 1

[lone mantle]

Nobody can stomp them atm

They are the absolute leaders in this industry

[tall patio]

Obviously I'm referring to the AI capability

[lone mantle]

They have just started with the gen AI although already very strong. With those projects they presented and whatever they have in their sleeves there is a lot to come from them

[tall patio]

Video AI

[lone mantle]

Obviously the target audience matters and SD, MJ, Adobe with Firefly and OpenAI with DALL E have their target

[tall patio]

I hope OpenAI can finish their chip soon enough

[lone mantle]

If any imo Midjourney is at the biggest risk out of the mentioned above

[tall patio]

Right now it's a solid 5/10 considering as a paid option

[lone mantle]

MJ?

[tall patio]

Yes

[lone mantle]

It has its advantages but also definitive disadvantages compared to the other ones

[tall patio]

Maybe subsidize is also an option for OpenAi?

[lone mantle]

I use Firefly and supplement it with D3 and vice versa

For gen AI stuff ofc

[ripe wraith]

I hit my GPT limit last night for the first time.. i was getting lazy though, conversation got too casual 🫦

[tall patio]

Nite lads

[ripe wraith]

the changing of guard

just read up and whenever I hear Photoshop I immediately think overpriced tool, but maybe thats just me 😉

GPT-Dalle with in-painting commands is the future, and I need it NOW.

[lone mantle]

The future of AI art?

Dunno

It needs inpaint in general id say 😄

[ripe wraith]

Expanding on visual worlds will consume many of us I think.

[lone mantle]

Inpaint and outpaint kinda became a must have for every AI image generator

[ripe wraith]

"the must have feature", the industry is a little young for that talk isnt it

[lone mantle]

I dont need those from D3 as i can get it somwhwre else

But those that play with D3 alone, thats a different case

[ripe wraith]

that will be everyone, soon enough

[lone mantle]

Nah

Definitelly not

[ripe wraith]

how do you beat multi-modal

[lone mantle]

You have Stable Diffusion, Midjourney and Adobe (Firefly)

Plus Google

With Gemini

[ripe wraith]

Those splitting between tools will be slower than those who make expert use of a single tool

[lone mantle]

Well thst one isnt out yet

[lone mantle]

That single tool doesnt beat those others

[ripe wraith]

*yet

[lone mantle]

And maybe never will

We can speculate

[ripe wraith]

it will get close enough that it wont matter, the slowness will exclude the others

it'll be a crazy few who dedicate lives to using them, if anyone even still maintains

[lone mantle]

What is them

Dall E?

[ripe wraith]

everything besides the OpenAI universe, if they keep this up

A single, multi-modal, memory chatbot assistant, who uses Dalle3, will beat anything else.

[lone mantle]

DALL E is good but it doesnt murder the competition at all

Not even MJ

[ripe wraith]

It's off topic now because you're missing my point repeatedly, msg me in there if you want to continue.

[lone mantle]

We can in AI discussions if u want

[marsh depot]

Is Dall-E 3 also in the mobile browser version? And if i buy plus will i have it from the get go?

[dim cradle]

how can dall-e 3 fail to understand this?

An ice cube floating on black water. On the cube's top face a black cross. On the cube's right face a black empty triangle. On the cube's left face a black empty circle. The left side of the cube glows lime green, whereas the right side glows sky blue. Style is 3D offline render.

https://th.bing.com/th/id/OIG.rauvkSujcGgzVDTtC6.3?pid=ImgGn

[marsh depot]

Dall-E gets roght and left wrong quite often😂

[dim cradle]

it never gets triangle, cross and circle right. It always mixes them all over the faces

[marsh depot]

Maybe they are on the blind sides 😄

[dim cradle]

I didn't say a cross and a circle

https://th.bing.com/th/id/OIG.BqvXwPKw3PiYfJqeA2ny?pid=ImgGn

nor a square

[agile peak]

Oh. Looks like it's now allowed for ChatGPT-4 change the seed on Dalle.

[mortal summit]

Where did you see that?

[dim cradle]

2/4 for me

[wintry dune]

how?

[dim cradle]

it does understand "cube standing on water" but I said WATER, not a wet surface

https://th.bing.com/th/id/OIG.gQxTOPI2OJrKTmanny.b?pid=ImgGn

[frozen fiber]

when you finally finish your project 🥳

[paper jolt]

can you elaborate?

[frozen fiber]

I see you have fallen right into my trap

lmao jk I made every country as a hollywood poster and I'm so glad its done

its interesting how DALL-E sees different countries cultures

[paper jolt]

Do you mind sharing? It sounds quite interesting!

[frozen fiber]

sure dms

[wind crane]

?

[buoyant sand]

I’m really confused

[lone depot]

Looks like I'm not the only one

[dim cradle]

this one was the closest

https://th.bing.com/th/id/OIG.J5NoQiQOQXWc090O7.No?pid=ImgGn

[agile peak]

You can ask ChatGPT after first creation with Dalle.

[dim cradle]

with lama cleaner I erased the blobs

[agile peak]

You can discuss with ChatGPT via Dalle. But I don't think it allow more than 5000.

I tried generate one image and it says "12345" seed before. Asked ChatGPT and the respond like this:

With a seed like 12345 there are few parameters to adjust for controlled variation. Only whole number changes are possible.

Changing the seed improve my art. Plus ensuring less hallucinate and defects shows on the subject facial appearance.

Here's good example with lesser seed: 3333 (Low quality, abstract subject faces, hallucinate a lot by added random objects, poor lightning) and 5000 (High quality, depth in both lightning and shadow, consistent subject faces).

[wintry dune]

wait so the higher the seed the higher the quality?

[agile peak]

Correct. I even asked to Claude (sorry for non-related AI).

[wintry dune]

interesting, thank you

[paper jolt]

What did you ask Claude?

[agile peak]

Nothing. I was testing to find out turned out it doing good to compliment with Dalle. Copy-paste my prompt to both Claude and ChatGPT to see which one has good revision for Dalle to understand.

Although ChatGPT3.5 offer more explanation. Claude is only excel in contextual terms.

[mortal summit]

The seed should not have any relation to the output at all. It simply fixes a certain model configuration to give the same output for the same input.

[paper jolt]

That's what I was thinking

[mortal summit]

I wouldn't trust what the models say too much. They are not primarily meant as knowledge engines or search engines.

[wintry dune]

ah ok, thanks for clarifying

[agile peak]

Interestingly it does affect my images. But since ChatGPT Dalle is locked to 5000; and OpenAI is disclosed the information about Dalle Bing and ChatGPT. I have no idea.

[mortal summit]

Have you looked at the actual prompt send to DALL-E?

[agile peak]

Yes. I even allow GPT to do both generate image by default (without command) and tell it to refrain from changing my prompts. Still have no significant impact to change the variation of the generated images.

[carmine raft]

So, for some reason when I ask for an image of a character, a fairly standard request I would think, I am getting sideways images. I could download them and rotate them with an image editor, of course, but it's strange and annoying behavior. Anybody know how to make a request to not do that?

[agile peak]

ChatGPT Dalle (Left) won this time. Against Bing Dalle (Right) with same prompt. Because I did not mention group or multiple UFOs but single.

[wintry dune]

yeah ive been doing portraits so i say "full size horizontal portrait" in my prompts

[paper jolt]

horiztonal portrait? that sounds like an oxymoron to me

[wintry dune]

if it works it works lol

[paper jolt]

🧠

[paper jolt]

Assuming you made that with Bing, it all depends on your use-case. I personally bought Plus for personal tutoring and information aggregation about topics like mathematics and computer science, which then extrapolated to general information aggregation. What I'm saying is I really only got DALLE-3 as a nice to have alongside with the primary use of GPT-4. DALLE-3 in Bing is really similar to ChatGPT and is free, so if you only wish to buy Plus because of DALLE-3 and you care not in the slightest for GPT-4, I would stick with Bing. The only main difference is the UI and having conversations regarding generations so you can make changes conversationally. If you think that is worth the $20 or more, go right ahead.

[wind crane]

Yes from Bing.
I'm tempted to subscribe but I'm still hesitant.
I want to create comics but I haven't seen anyone do any big attempts in this direction at the moment.

[paper jolt]

You can create panels and then stitch them together in some image manipulation program

[crystal badger]

So um..

I've created a set of Instructions for DALL E 3, i was wondering if someone would like to test them out and give the instructions feedback, i think it works perfectly, but maybe i am just lucky, Would anyone be interested?, its on the ChatGPT Web based.

[wind crane]

what about the consistency of the characters in different angles of view, actions and backgrounds?

[obsidian rose]

imo Bing is better than ChatGPT at the moment for any character creation, but ChatGPT is easier to use

[paper jolt]

ChatGPT may help facilitate that, how has this worked out for you with Bing?

You would obviously have to be explicit with each generation with Bing

And ChatGPT could infer intent with its retention of context

[wind crane]

bing understands that the first name matches the description but it doesn't look like the character, it's a similar character, he does different actions when I only give the first name.

[paper jolt]

So you made two prompts, in the first one you gave DALLE a name and description, then in the second you gave it just a name

Is that what happened?

[wind crane]

yeah

"jack is little boy with dark hair" then "jack walk in city"

[paper jolt]

Again, that's something ChatGPT would be more appropriate for. In Bing, you would have to restate Jack's features

[wind crane]

thk, i think i give it a try

[paper jolt]

Best of luck 🫡

[wind crane]

Are we allowed to post Twitter links?
I saw someone trying to create a consistent character for a comic.

[paper jolt]

Unfortunately not

[wind crane]

Is there an "offtopic" or similar for this?

[paper jolt]

You can post screenshots

[wind crane]

here ?

[paper jolt]

since it's relevant, yes

[frozen fiber]

did it work?

I thought only stable diffusion could achieve that

[wind crane]

@frozen fiber

i will try that

[frozen fiber]

wow

okay jk but thats pretty cool

[wind crane]

If we can save time why deprive ourselves of it?😆

[frozen fiber]

true

[cloud dome]

Please be respectful

[frozen fiber]

Sorry! Trying to make a light-hearted joke 😅

[cloud dome]

Unfortunately with text its impossible to tell if "jk" is someone trying to get away with harassment or someone genuinely trying to be funny

[nimble rose]

I found the technique very interesting...

[wind crane]

yeah

[little cliff]

is this thing down right now?

[frozen fiber]

Im using it rn so I wouldnt say so

[weak basalt]

how do i generate images

[wind crane]

bing/ chat gpt plus.

[dire vale]

why dont I see anymore when chatgpt actually creates the prompt?

its just loading and then done

[paper jolt]

New UX change

Prompt can still be seen when selecting an image

[dire vale]

okay

[agile peak]

Was there any update today? I feel Dalle•3 seems to be improving in chatGPT.

It now can be consistent generated with same angles. Yesterday it was out of control - such as changing colors and styles.

[hearty ether]

You should ask it what went wrong

[hearty ether]

Might be placebo

[hearty ether]

You can sort of do it with two panels in one image

[little cliff]

Make 3 images, wait for 12 minutes due to rate limit. I'm paying for this?

[hearty ether]

In chat gpt plus?

[little cliff]

Yes

[hearty ether]

And you didn’t reach the 50 message limit?

I blame the free bing people hogging it up tbh

[little cliff]

Nope, made 3 images, now it's telling me I have to wait 12 minutes for more

[merry ruin]

Where do I see the privacy Policy?

[gleaming lance]

"make photos of the deathstars weapon system"

[faint crown]

@unborn heron your question in daily theme about the borders. Sometimes Dalle puts black or white borders on top or at the sides or all around.

[unborn heron]

Odd 😅 thanks for answering though!

I guess it's just recreating what it has seen in the training data and there are images that have that in there

[little cliff]

29 billion dollar company

[hearty ether]

Did they fix getting realistic faces

So they don’t look weird?

[unborn heron]

Leaks your private prompts to other users in your path

[paper jolt]

millions of people

[hearty ether]

Most of them free…

[paper jolt]

their costs aren't

[hearty ether]

I know…

And they are taking up load from paying users….

[paper jolt]

tis' the circumstance

do you believe there is an alternative course of action that would be beneficial on behalf of OpenAI?

[hearty ether]

I mean favor the paying users a bit more feature wise and load wise

Maybe a little less for free users and more for the paying

Since the free users are using duplicates accounts to keep generating

This way it will be more dupes required and added complexity which would incentivize them to actually pay

But for that they definitely do have to make the ux better on the chat gpt side

Like more customization over whether we want chat gpt to generate a prompt for us

And more control over the amount and size generated with actual ui elements instead of putting everything in the prompt

And hoping the ai interprets it

[paper jolt]

That is the current arrangement as free users get just 3.5
The duplicate account exploit is likely used by a small amount of individuals
ChatGPT can't reach into your brain and pull out exactly what you want, it can only randomly choose specifics if it were to implement your idea
Adding UI elements for the length would add complexity and doesn't make sense because the amount of text to convey some idea may or may not be below some fixed value which is why we ask in the prompt.

[hearty ether]

Ehhh

That’s why I want to be able to manually write my prompt

Since having the ai generate my prompt again messes it up and the ai forgets or ignores important elements

[paper jolt]

I thought you were talking about ChatGPT in general since free users don't have DALLE-3 in the first place

I think you can just ask ChatGPT to send DALLE-3 exactly what you ask

[wind crane]

is this serious?

[modest orbit]

they still have load issues

yesterday also

let them scale

its insane populair chatgpt

[paper jolt]

It's only one image because of the fixed seed

Not quite all-black but nevertheless

Though I have also observed that even if an element is mentioned in the actual (modified) prompt sent to DALLE-3, it may not be included in the image

[hearty ether]

But yeah more fine control should be better imo

instead of controlling everything though an essay

[agile peak]

Nice. Is this from ChatGPT?

[gleaming lance]

sure is!

[agile peak]

Pretty cool. That give me hope.

I am digging ChatGPT to bring my vision to reality. But having difficult time to add realism to the image where cars isn't empty lol

[nimble rose]

I was doing some tests here and it seems that each and every image generated has a fixed seed of 5000.

[wind crane]

I don't understand why the character is not straight in the portrait format on all image.

[nimble rose]

I was thinking that, if each image had its own seed, it would be interesting for creating image compositions.

[agile peak]

Amazing! adding soft touches is chef kiss

[dim cradle]

highly detailed and textured pencil sketch of a strong adult serious female nurturer elf holding a newborn close to her chest and her bow. She's standing on a hill, aiming her bow towards an enemy in the distance. the newborn is crying. dark smoky clouds cover the sky. she has long brown hair, green tattoos on her arms and face and a leather armor.

[agile peak]

2024-2025 bull run 😄

[dim cradle]

this was the prompt

i'm not sure if i'm allowed to share the image

can i?

[crystal badger]

These are some pretty nice images.

[paper jolt]

Can you elaboarte on what you mean?

[agile peak]

This is happen occasionally. Kinda ironic Bing increased the filter and censored stuff for specific figures and words but NSFW can still be slipped out with certain combined prompts lol

[dim cradle]

i think that maybe nurturer is triggering it but it only happened twice i think

[wintry dune]

anyone else having chatgpt taking way longer to generate images with dalle

[agile peak]

I think censoring stuff is kind of wasting it. I read an article that using some unfamiliar languages can bypass the censored filters.

[agile peak]

Doing pretty quick here on both web and mobile.

[wintry dune]

yeah its making the images and i see them on my phone but the web version is lagging or something idk

no big deal

[agile peak]

Mobile update is pretty good. I'm digging the revamp on the UI/UX.

[nimble rose]

Imagine that: You create a character A and assign it seed 4999. GPT then makes an association between character A and seed 4999. In the same chat, you create character B with seed 5000, and assign it in the same way. You could then generate more accurate images of the two characters in different situations and even together.

This is currently possible, but the subtle details are still "fuzzy".

[crystal badger]

Hopefully we will integrate "Vision" with DALLE at some point, Quality bound to skyrocket.

[wintry dune]

have you guys tried upscaling ur dalle images

[crystal badger]

i decided against it.

[nimble rose]

But something I noticed is: dalle3 is not losing at all in image generation. The quality of the images is at an excellent level, with the only real "defect" being the problem with Copyright and the censorship with certain themes/words.

[agile peak]

Use "Make it wallpaperable image" can increase the quality.

[wind crane]

how to get the right phone image format?

[nimble rose]

I even understand because open is a global company and it could have serious problems with these ethical issues. But I still believe in a situation where censorship will not be necessary.

[agile peak]

"Maximum height" did pretty good to make iPhone wallpaper for me

[wind crane]

i will try

[nimble rose]

After all, the thing about artificial intelligence is the “Sandbox” thing, right? Each individual will be able to do what they want with knowledge.

[agile peak]

Sometimes it's kinda buggy. That the image are in portrait but it generated wrong position.

[dim cradle]

Do everyone have access to gpt 4 dalle 3?

[agile peak]

I think some are still on waitlist; it still gradually rolling out.

[dim cradle]

Ah okay thx

[slate gate]

People with vision model, can you use generated image as reference to generate consistent frames to make simple animation?

[weak basalt]

who can cashapp me $1

[gleaming lance]

Chat gpt is hesitant at making images of weapon systems.

[thorn crypt]

the chatgpt dalle3 is way better than dalle2. I asked it to come up some designs. The outputs are legit concept arts instead of merged nonsense.

[dim cradle]

with bing censoring just too much I can't create fantasy art like vampires drinking blood, corpses on a battlefield, etc... what do I do?

[rancid reef]

you use midjourney 😂

[thorn crypt]

I havent use midjourney before, how good is it?

[dim cradle]

midjourney was the first, or was it? it has over one million users in their discord server

[rancid reef]

its very very good

[rancid reef]

one of them yes

[thorn crypt]

just tried it. It doesn't understand my prompt as good as chatgpt.

like it tries to make illustrations instead of a realistic photo of product sample.

[crimson heron]

MJ requires learning how to "craft" prompt a lot more than the DALL-E variants. DE can better understand more natural style language and context. You can do realistic photos with MJ, but you'll have to learn how to craft the prompts to do it.

[nimble rose]

Or you can search the internet for prompt generators that turn your ideas into MJ prompts.

[thorn crypt]

yeah, Im trying the gpt4 prompt to generate mj prompt based on my prompt

but the results are bad

[runic granite]

hello everybody.

is it possible that Dall-e block this modification prompt? use image 1st as reference and make 4 distinct variations and display all 4 of them as image. Make one arm out of crystal but fully from rough white translucent white mountain crystal and don't include metal for the arm. Make the skin charcoal black.

i tried it has problem with the arm part

[nimble rose]

Copy the image generation prompt you want, and then paste it as a reply to ChatGPT.

[wispy storm]

you can try #1070006915414900886 but i dont think chatgpt dalle 3 aiming for * beach /swimming suit yet*

[runic granite]

yeah, in this case i wouldn't need GPT, would i?

i started with a basic image and modified it step by step....

as i tried to gen prompt with gpt and feed it back...

no good results

[nimble rose]

What I do for my images is, see which of the 4 looks coolest, select the prompt for the one I liked and then I modify only the details I want to change, hair color, anything I want to add, etc.

[peak trellis]

Do they give free credits

[runic granite]

thats the problem, therefore i use this way....

it alreaady gen drow but with greyish skin color despite in the prompt there is coal black... and when come to crystal arm.... it is a disaster if i feed directly the prompt

[nimble rose]

not to mention that, even if it's dalle3 generating the images, it's still our old friend GPT talking to you. So a good option would be to simply talk to him, something like: So... Buddy, the first image was good, but these or those details weren't good, could you change it for me?

[runic granite]

i do that

and blocked

😄

[nimble rose]

Could you share your chat?

[runic granite]

as it dropped the crystal arm

it show i cant share chat with images

Original prompt: ID=2023.10.15. 03:39:35. Photorealistic full-body variation referencing the first image. Female drow with obsidian charcoal-black skin, silver hair, eyebrows, and irises, and red lips. Her right arm is fully made of white rough mountain crystal. In a combat stance in an underground cavern, weapons drawn.

[crimson heron]

When you have trouble creating images like that, you can ask ChatGPT to share the prompts so you can review them. Alternatively, you can ask ChatGPT to analyze your prompts to identify what aspects in the problematic prompts are possibly at fault. I use a combination of those to rework my prompts. In the latter case, you could even ask ChatGPT to fix the prompts and try again.

[runic granite]

thanks i try it.

can i ask something else?

[crimson heron]

Sure.

[runic granite]

i discovered as i use the instruction to make variants... only in the prompt of the first image i see it.... the others are just prompts...

i do someting wrong and thats the normal or it is a bug? as i state to make variants of the selected image. and many times it only gened 1 image and just wrote the 4 description without the other 3 images.

Thanks for the help by the way. 🙂

[crimson heron]

I haven't seen that myself, but you can generally correct ChatGPT's behavior by giving it instructions. Like being very clear how many images you want generated or correcting it that it only generated the first one and gave the text for the others. It'll typically go something like "You're right. I apologize." and then do what you wanted.

[runic granite]

yeah

i do that.

but i specific to gen 4 images.... 😄

maybe i have to be much more specific....

😄

[crimson heron]

One thing I have noticed is the longer a chat goes on the buggier ChatGPT gets in following instructions consistently. And even if it is not buggy it does get a lot slower.

[runic granite]

so make a new chat for every image?

i mean every subject

[crimson heron]

I could almost see why ChatGPT responded this way. I can tell your intent, but an alternative in your request could have been "Use image 1st as reference and make 4 distinct variations. Make the crystal arm from white rough mountain crystal. As usual, I want you to generate 4 images." I don't always clarify the number, however, I will if it got a prior instruction wrong.

[crimson heron]

I'm referring to very long chats and lots of instructions. If I am making a drastic change in subject or doing a side-exploration I'll do it in a new a chat to not confuse. Or if I do it within the existing chat I'll explain to ChatGPT what I'm doing before the context switch.

[runic granite]

oh, ok. so basically one chat for gen image and different for prompt gen and other stuff...

it is usual that despite the own prompt which dalle get state that the subject is female it gen male subject?

[runic granite]

thanks the advice.

can i put it in custom instructions? could it solve this issue?

[crimson heron]

That might work. I haven't done much with custom instructions.

[runic granite]

i saw an extreme one here in the library....

[crimson heron]

My sleep patterns have been off lately so I'll be dropping off. Hope some of this helps.

[runic granite]

of course it was extremely helpful. thanks for it

yeah here 5:27... and me too havent slept yet

i just see the gens of others... and feel extremely dumb.... 😄

good rest. 🙂

[lime solstice]

a

[cursive hornet]

I could reproduce the exact same image with the same seed value and the same prompt, but I thought it was quite dangerous if commercial use is allowed.Not so?

Since the default SEED value is fixed at 5000, it seemed dangerous that it could be reproduced quite easily.

[vapid elk]

I guess fixed seeds might change in the future

[runic granite]

hello.

can i ask something? Maybe you know the answer.

by the way if they dont implement a system to refer back to previous generation of images in GPT, it could be a problem. now i only copy prompt and generate again if something go wrong.

ID or something or displayed seed (i think it would need much wider seed interval)

[cursive hornet]

The seed value of 5000 is set quite firmly, and even if it is specified in a custom instruction, it is generated at 5000.
This is quite problematic.

This means that in the case of images generated by ChatGPT's DALL-E 3, the prompts should not be shared in public.This means.

[runic granite]

yeah except when isnt fixed

i tried now as have problem with refer back to an image

i regened the prompt and similar but not the same image...

[cursive hornet]

SEED is fixed at 5000. Even if you indicate it is random. Only 5000 is used.

[runic granite]

sorry i checked... just a little bit tired in the fight with dalle.... GPT modified the prompt... therefore was the diff.

[cursive hornet]

I missed the seed5000 information above.
#images-discussions message

[runic granite]

for now in my case without a visible proper id... it better as fixed...

[agile peak]

Poe + ChatGPT Dalle. Is such a good combination as an assistant to improve and rephrase the prompt. It help me capture the most accurately in achieving the expected of the quality and vision for the images.

[cursive hornet]

The SEED value does not seem to change when specified.
Ask GPT for the value of SEED.

[agile peak]

That's weird. Either GPT did not correctly reported or something went off - the low quality image of mine posted here was said to use 1234 seed according to GPT.

[cursive hornet]

What is happening...?

[quartz vale]

What is happening?

[teal timber]

Where is it happening?

[quartz vale]

I like the decoration Baugy

[teal timber]

[agile peak]

[quartz vale]

[quartz vale]

How are you all enjoying Dall-E 3? Do you think it's better than 2.0 ?

[runic granite]

is there a workable way to refer back to an image if i just want slight modification and preserve the rest of the details?

[agile peak]

I don't like it...

I am loving everything about it.

Dalle GPT help me create my own wallpaper the way envisioned.

[quartz vale]

Ooooh, fancy

[runic granite]

through GPT i dont describe the experience for me as i got a little timeout but... with Bing Image Creator... I like it...

[agile peak]

I like how sometimes Dalle can be so crazy on ChatGPT especially when you didn't specifically mentioned about the back views settings in the background.

It generated images with sort of infinity objects to compliment with the foreground 😁

Trees, trees everywhere.

[runic granite]

now thats a green city....

😄

[wintry dune]

im loving dalle 3, it can mix so many different styles from anime

and manga

[runic granite]

yeah... just dont want extreme composition....

or i am not capable...

😄

[runic granite]

general stuff ok...

[agile peak]

Agreed. It also good at producing oil paintings not only you can use specific styles but also can increase the amount of brush works for instance.

[runic granite]

ohhh i couldnt tell how many apologies i saw....

😄

[agile peak]

What's happening lol

[wintry dune]

yeah its amazing you can ask for any art technique, like i asked for stippling, hatching, crosshatching

[agile peak]

ChatGPT allow me to know 🤣

And I don't even turn on my custom interactions.

[cursive hornet]

Broken robot.

[wintry dune]

same lol i didnt know any of that until a few days ago

[runic granite]

by the way is there a plan to incorporate the variants into one and just for the "testing" phase it separated in GPT or it will stay?

[agile peak]

That's what I have been wondering. Maybe we will get further information next month on 6th November during OpenAI dev conference day.

[runic granite]

i hope they will implement/allow more detailed setting, functions to dalle

example inpaint... 😄

this way no need to fight with GPT over modifications...

😄

i just want to change the hair/eyebrows color to silver

😄

[agile peak]

Yeah. That would be nice features to interact with 😄 but right now I think since it's still under research preview. It safe to be optimistic what going to happen within the next months.

[runic granite]

but every time lost the arm (i mean i like to preserve the look)...

[runic granite]

im pessimistic as dall-e2 out some time

[cursive hornet]

The GPT icon is square on mine, but round on yours, so I thought there must be some behind-the-scenes version difference.

[runic granite]

they could implement them with 2

[cursive hornet]

The number of digits is different to begin with.

[runic granite]

edge/chrome diff maybe?

[agile peak]

Thats normal 🤣 I'm just using a mobile app and yours is from the website.

[runic granite]

that was my 2nd tipp

😄

[agile peak]

By default it's indeed locked to 5000. Just start a new chat.

I have custom interactions on, this time. Make it easier to referencing certain image from all of the 4 generated images.

[cursive hornet]

I had custom interactions turned on.
I have tried many new chats.

[agile peak]

Maybe in the last screenshot of mine, GPT was hallucinating.

[cursive hornet]

I am cursed.

[agile peak]

I am using both Claude for prompting and ChatGPT to generate images. It worked so good.

I am seeing the image generated from dalle•3 really improved than yesterday. It might be "placebo" but feel like it does.

Bing (Left) & ChatGPT (Right)

[frozen fiber]

theyre different?

[thick smelt]

hey peeps

[half field]

nice

[thick smelt]

an oreosaurus rex noice

[frozen copper]

Why can't we upload images into dalle3. I want to use the uploaded images as a base for the image generations

[loud field]

I would like to point out an issue, besides the censorship

Dalle3 thinks this is "plain looking, mediocre, middleaged"

[wind crane]

Does anyone have this problem creating an image in gpt plus?

[loud field]

let's compare the same exact prompt on stable:

https://media.discordapp.net/attachments/1018541079106166825/1163020275114319882/VWYORICF3aZp.jpg

Dall3 has the potential to be by far the best model ever, but it has strong issues with bias.

[wind crane]

I just realized that in fact even if we have dalle3 it on the computer, we don't necessarily have it on the mobile application..?

[cursive hornet]

seed 5000 bugg
https://discord.com/channels/974519864045756446/1162395649044648060

[wind crane]

What exactly is happening ?

[fair crescent]

If you mean the app: go to a chat in history that has it. I use phone nearly exclusively. I use the mobile browser to initiate new chats with the right GPT, then proceed to the app.

[wind crane]

I succeeded but now I have this message above, the robot no longer wants to draw and when I asked it to change the image format, it did not do so.

[fair crescent]

Can you force close the app and try again?

[wind crane]

Doesnt work

Maybe you mean force close in setting and not juste close the app ?

Doesnt work...

[fair crescent]

Try a different chat first. But on android you can go to the "see all apps" thing and remove it there.

[indigo meadow]

I presume this is the best place o the web to ask for assistance with dalle?

dalle seems to ignore my request to remove ceiling lights. They're always there + ignores the fact I'm asking for a single train and not more than one.

[frozen copper]

Man, the content filter is really too strong. Asking a full body images of people easily triggers the content filter

[cursive hornet]

DALL-E 3-boy is in puberty, so a little word may cause him to go into heat.

[dim cradle]

Hello

[dim cradle]

chat gpt free?

[tall patio]

Yeah, content filter is strong

[cursive hornet]

There is no such thing as freedom.

[tall patio]

But far from useless

The last week I've held onto the opinion that it's useless

I was wrong

Really

Really wrong

Premature judgement

D3 itself is a behemoth

[cursive hornet]

But it is more problematic if the SEED value is fixed at 5000.

[tall patio]

I found a way to force it sticking to a particular image set

No need for seed

Try to think about it and you'll find a way

[lone mantle]

D3 is definitelly very good

[frozen copper]

Yeah, would be better if they tone the content filter down

[sinful crane]

Can I know what cue word to use to get a clean background like that?DS3 always fills the background with mess, even if I specify that the background is empty.

[frozen fiber]

nah what

[loud field]

I only asked for the same prompt you see in the Dall3 image

[undone canyon]

i swear

bing has gotten so restricting

[loud field]

Almost unusable

[undone canyon]

ong

old bing was so much better

[placid orbit]

itll only get worse, so..

[wind crane]

I have to admit that the number of PLUS tile tokens is too limited....

[placid orbit]

lets hope openai actually gets competition soon...

[loud field]

Agreed, that would be healthy

[tall patio]

Maybe I should start selling hopium ...

[frozen fiber]

stability and midjourney?

I remember on this discord when everyone was talking about the release of stable diffusion and how it'll bring some competition, and they have been quite successful

[wispy storm]

Hewwo! for other ai, let's use the proper channel #ai-discussions please

[frozen fiber]

👍

wut

it works when I remove award winning but how is that against content policy?

also gets flagged if I add "line art" "anime"

but then I remove those and

[maiden star]

having some troubles creating hyperrealistic images on dalle-3 , might be cause its beta mode yet or any advices for doing that ?

[wanton hatch]

Keep in mind that hyperrealistic/photorealistic means painterly, so is not realism (aka a photograph).

[maiden star]

i see , ive been putting both of them as hyperrealistic , and photograph , and it doesnt seem as real as in a photograph , should be my issue not being able to do them or is it dalle still being in beta mode ?

[wanton hatch]

DALL-E is still indeed in beta testing

and the extreme policy filters don't help

[maiden star]

yes true

[maiden star]

thanks for the help

[lone mantle]

when it comes to image generating they already have

and those are strong ones

[placid orbit]

well, not really

[lone mantle]

Stable Diffusion, Midjourney, Adobe Firefly

unless we talk about that LLM as well

in that case, Google Gemini eventually?

[wanton hatch]

Probably he was meaning this

[lone mantle]

well in that case ofc Google Gemini if they make it right

but purely for image generating there is a very strong competition

[wanton hatch]

Yeah

[placid orbit]

they are all less capable models than dalle 3, where the main attraction is the prompt adherence

firefly is completely trained on stock images, so even though it makes good generations its limited in what it can make

stable diffusion is the opposite

[lone mantle]

ofc but i use Firefly besides of D3 for a reason

i like the combo when needed

it really depends on workflow and audience

[wispy storm]

Hewwo, For other ai please use #ai-discussions thanks !

[zinc dagger]

I just got the dalle option on my phone today!

[zinc dagger]

It’s pretty buggy though :(, a lot of the image generations end with it getting stuck trying to write text after the images and I have to restart the app.

[thick smelt]

but unfortunately dall e 3 had most accurate axolotls, however google can have a decent axolotl, but not as accurate as dalle 3

[lone mantle]

O_O 😄

had to google those

[thick smelt]

axolotls are similar to salamander but a little more unique, they were added in Minecraft in wild update

[dim cradle]

guys i noticed a weird behaviour with flagging

when i put the selected part of the prompt to the end of it, it generates the image normally

but if i write that exact same part near the beginning of the prompt it gets flagged

[frozen fiber]

well with that type of flagging, it seems like the images themselves were the problem, is it consistent?

like does it continuously flag the results

[dim cradle]

here i tried the prompt that normally works, the 3 blocked results are the ones were i tried to move that part

[frozen fiber]

ye bing is just weird like that

[dim cradle]

here's the prompt that works comic style cell shaded drawing with vivid green and brown shades of a strong female nurturer elf holding a newborn close to her chest and carrying a bow. there's a forest with green cristals on the trees in the backdrop. she's inspecting a tree. her bow and arrows have the same crystals on it. she has long brown hair, green tattoos on her arms and face and a brown leather armor. creatures with the back covered it green crystals are wandering the forest

[frozen fiber]

it flagged me when I added award winning and line art before just now

[dim cradle]

and here's the prompt that doesn't comic style cell shaded drawing with vivid green and brown shades of a strong female nurturer elf holding a newborn close to her chest and carrying a bow. she has long brown hair, green tattoos on her arms and face and a brown leather armor. there's a forest with green cristals on the trees in the backdrop. she's inspecting a tree. her bow and arrows have the same crystals on it. creatures with the back covered it green crystals are wandering the forest

someone suggested to move more important parts of the prompt towards the beginning of it, so i tried to move that part to actually get a brown leather armor instead of a green piece of clothing that only covers a specific body part

[frozen fiber]

odd

maybe try moving just the armour part?

[dim cradle]

it's part of the characteridentifying traits, so i don't really want to get rid of it

but apparently moving that part a bit furthersolved the flagging problems

but she still doesn't wear the armor

[frozen fiber]

not get rid of it but just put it in another section I mean

strong female nurturer elf wearing brown leather armour holding...

idk

[muted moss]

How do I copy the prompt text for a generated image on mobile? Or at least see the entire prompt? The mobile site is missing the “copy” button, and part of it is cut off. It won’t even select the text.

I’d try the app, but I already know it doesn’t show the prompt at all.

[noble kiln]

Perhaps you can ask GPT about the Prompt, like: Can you tell me the Prompt you used to generate these images?

[zinc dagger]

I don't think there is a way to on the app, sometimes you get lucky when you ask it what the prompts were, but there's a system message telling it not to display the prompts so sometimes it just says it is not possible to say what the prompts were

[spring smelt]

Is it possible to change the aspect ratio with dall-e 3? I'm used to MJ's --ar but it doesn't seem to work, ofc, and not just mentioning the aspect ratio like "aspect ratio 1:1".

[cloud dome]

You should be able to ask it to do a specific aspect ratio.

I asked for 16:9 in my prompt in plain language, and it worked

[spring smelt]

Awesome, I will try it again, thanks!

[cloud dome]

When my images were too squished, I simply asked "Can you extend them to be 16:9 to give yourself a little more room?" and it started generating in that aspect ratio

[wind crane]

Is it me or does bing creation not "suffer" from the limited number and delay in creating images compared to CHAT PLUS?

[frozen fiber]

it definitely does in the second

I have a 5 minute wait now (I can bypass it but still coded)

[drifting swallow]

I have developed a new DALL·E 2 Android app. Who should I check with about the name of the app, before I publish it?

[agile peak]

I think Google just announced yesterday they rolling out AI generated images for some people in their app. Will see more info next couple months.

[agile peak]

Yes. This is exactly what I am doing wordings/rephrasing prompts can help a lots.

[muted moss]

Ah, that’s a good idea. Thanks! Only issue is it wastes GPT-4 usage for something that shouldn’t need it.

[agile peak]

Godward's painting depicting Socrates reflecting himself; with dalle.

Amazing. These two with different settings.

[elfin dome]

when will we be able to upload images to dall-e 3 , so we can combine images and text as instruction?

[reef coral]

Can you get credits for dall e without paying

[rigid bough]

DALL-E 3 is built into ChatGPT Plus, so you can use it there, otherwise you need credits for labs.openai.com

[gilded osprey]

It’s surprising how well it does hands.

[ionic ore]

I dont like that bing doesnt tell you what content policy youre breaking
I just got suspended for an hour when I wasnt generating anything sexual or violent

[gilded osprey]

It doesn’t understand mirrors well

[versed orbit]

Is the dalle model down? Keep getting errors

[wind folio]

Same

[gilded osprey]

A: yes

I predict it’ll be fixed soon.

[vale wind]

whats happening to gpt 4 dalle 3

why it give me error all the time

damn. ive wasted 5 messages just getting errors

do the 5 error messages count?

💀

[paper jolt]

no

[robust sonnet]

Could u just tell me plz

[wind crane]

What is the key word to prevent the image from being upside down?

[cloud dome]

What was the prompt you used? I've never seen that before

[wind crane]

draw me a knight, the armor must be heavy, he walks in the city, the view is in profile, we see the whole body, the image format is a portrait.

(When I asked him for image sizes he presented me with:

square
Portrait (tall)
Landscape

Custom instruction: the generated image must be right side up.

[cloud dome]

which size did you choose

[wind crane]

Portrait.

[wintry dune]

i always say "full size horizontal portrait" that makes it always right side up

[cloud dome]

Is that what was generated? Or are those square images

I have a feeling its an issue of DALL E 3 trying to generate a portrait in a constrained area

[wind crane]

what do you mean ?

[cloud dome]

Your images look square. Are they?

Or are they portraits, as in not square but tall

[wind crane]

they look square

[cloud dome]

When you click them they are square?

Then yes, it sounds like it tried to match you parameters in a square frame, which is impossible, which is probably what skewed the images

[wind crane]

oh i see !!

[cloud dome]

I might try prompting differently. You can talk to ChatGPT in a natural language to describe your images, this may get you closer to what you're looking for. Without seeing the entire conversation I can't say exactly what's happening.

[wind crane]

they not look square !!

but this, why it is not right ?

[cloud dome]

I've never tried generating in portrait before.

[agile peak]

I always find images tend to generate much betters with either square or full horizontal than full portrait.

[cloud dome]

You can also follow up and say "this is wrong, some of the images are still landscape"

[wind crane]

ok thank !!

[agile peak]

Portrait mode keeps messed up with the text on the spaceship.

These are default and full horizontal

[wind crane]

i see

[potent lotus]

DALL-E makes some beautiful things

[paper jolt]

Oh wow! What was the prompt for that one?

[potent lotus]

"Photographic portrayal of a young girl with dark hair and an exceptionally pale face, wandering through a field dotted with bright red flowers. The atmosphere is of a very dark, almost pitch-black night, with only the faintest hint of moonlight illuminating the scene. The low lighting creates a dramatic contrast, highlighting her black Victorian dress with red touches, her mesmerizing red eyes, and the crimson flowers around her."

[heavy oriole]

it's not pitch black 😆

[potent lotus]

I know. It's really hard to get these generative AIs to do a really dark image. They are sticklers for lighting.

[spice apex]

What is hall of fame ? Do we get anything out of it ?

[quartz vale]

totally! I love how it's able to show human emotions in what it generates whether they be directly viewable or indirectly referenced 😄
You image reminded me of this one I made a while ago

[potent lotus]

Ooooh I like that!

[quartz vale]

the #hall-of-fame channel is to show off the images people have voted on and receive 10 votes. 🙂

[potent lotus]

I know DALL-E can't do copyrighted characters but I gave it some pretty good descriptors of a certain someone...

[spice apex]

Thank you!

[heavy oriole]

aren't real live people copyrighted characters? AI seems to do a lot of those

[wanton hatch]

Portrait mode or 9:16 or similar are bugged in DALL-E 3 at the moment

[potent lotus]

Yeah but it for sure won't do Michael Myers lol.

[wanton hatch]

You rarely get them

[potent lotus]

But if you describe Michael Myers without naming him and do it well, you can get close.

[ionic junco]

any way to ensure dalle uses certain ages of people? For example, I provide a detailed descrption of a group of adults and it includes a child, even if I say no kids.

[paper jolt]

What was the prompt ChatGPT sent to DALLE-3 in the modified one and the original prompt you sent?

Damn bro that's a really vague prompt

What's the modified one?

[feral minnow]

I’m not sure. They don’t show up in the iOS version and when I check the history of the chat over browser, those images didn’t render.

[crystal badger]

i...

[feral minnow]

Vague prompts can lead to really great images if you provide an evocative theme the model can iterate on

You can tell it had a lot of VHS cover scans in its dataset lol

[agile peak]

Bing. Seems winning here to capture Epicurus in the style of John Williams Godwards in a single attempt.

Here's ChatGPT 🤔

[paper jolt]

Sent the exact same original prompt to ChatGPT, allowing it to modify it to Photo of a VHS cover art for a 1985 action movie titled 'French Maid Vengeance'. The main title is displayed in bold neon letters. The background shows a darkened cityscape. In the foreground, there's a fierce woman dressed as a French maid, holding a machine gun. Explosions are visible behind her.

[feral minnow]

Interesting. I wonder what happened to that one then.

I’ll try and get the prompt somehow.

[agile peak]

Bing doing a good job for capturing some artists styles, why does chatGPT keep changing the prompt.

[spice apex]

I have a challenge for those who have DallE3 and Vision in their account.

On DallE3 section try to create a rough art by 3 or 4 variations of change and then download the picture of your liking.

After that upload it to vision and tell it to describe it very detailed in 1000 words.

Then take that description into a new DallE3 chat and paste it there. Show me the before and after results by pinning this.

[feral minnow]

The prompt injection is very strong on GPT’s implementation. On Bing’s implementation, you more or less have direct access to the model. On GPT, GPT 4 serves as a middle man.

[wanton hatch]

Yep, it doesn't follow correctly what you ask.
You literally need to tell GPT to use your exact prompt at least 3 times (most of the time)

And fun fact, you get the same image/seed over and over if you the same exact prompt. Even in different chats

[crystal badger]

here take my comic.

[graceful hill]

What is going on with the filters? I'm getting all 4 images right now with the same prompt that either got blocked, oe all four images were deleted earlier.

[potent lotus]

GPT could let you have something like a --style raw like MJ

Take your prompt as is so long as it doesn't violate content filters. I mean, that seems reasonable to ask for.

[crystal badger]

i think we should combine dalle and chatgpt and introduce vision.

bound to increase quality and accuracy 100fold.

Dalle is basically drawing blindly.

[tender sigil]

yea it's seriously started to get on to my nerves

[potent lotus]

The Bing filters are out of control right now.

I consider it to be essentially unusable.

[tender sigil]

Dall E 3 was like the BEST ai img gen when it launhed

now its ompletely useless

[crystal badger]

nah

[tender sigil]

for opyright haraters

[potent lotus]

Like if you need very generic stuff. Sure. But even harmless prompts are getting filtered.

It's ludicrous, honestly.

[crystal badger]

the only thing Midjourney has going for itself is absurdly high def imgs u dont need, and adding reference images.

nothing more

[potent lotus]

If GPT lets me do reference images and adds functionality for inpainting/outpainting, then we've got a stew going.

[tender sigil]

someone needs to ome up with a prompt method that will leverly and skilfully bypass the restrition

[crystal badger]

thats bypassing rules thats against TOS.

[tender sigil]

even simple prompts get bloked nowadays

[crystal badger]

Main issue for most people is that They dont know how to prompt properly

[tender sigil]

any tips?

[crystal badger]

Use the tools you have to your advantage.

ChatGPT is an Expert Prompt Engineer.

this should be the first thought coming to you.

[tender sigil]

oh yes

wish dall e was in gpt 3

[crystal badger]

Sadly what people need to understand is that ChatGPT is a translator to DALLE

Meaning you need a proper set of Custom Instructions and a proper prompt to start with.

i aint having the issues the majority of people are saying, at all.

[tender sigil]

omg how?

[crystal badger]

this

[tender sigil]

like how to guide gpt

[crystal badger]

tried asking it?

[tender sigil]

what to ask? like should i ask it to refine prompt to avoid opyright infringement

[crystal badger]

i do have an old set of Custom Instructions u can use.

#1162782165189267466 here.

[quartz vale]

you don't need custom instructions to use Dall-E 3 🤔

[crystal badger]

u dont thats entirely Correct.

But using it with ChatGPT gives you an Edge.

[tender sigil]

thanks a ton! will definitely be helpful

[crystal badger]

If u are using DALLE on ChatGPT i highly recommend using a proper set of CI. and a starting prompt.

[tender sigil]

so is it possible to use it in the regular gpt 3

[crystal badger]

this took me 1 take with no refinement and improvment.

[quartz vale]

no, you have to have a ChatGPT Plus subscription and use GPT-4 to use Dall-E 3 in ChatGPT

[crystal badger]

yeah, thats true.

[tender sigil]

an i use the custom instructions in gpt 3 and get a prompt and put that in bing image reator?

[crystal badger]

i got Plus idk how it'll work for you.

[tender sigil]

it should work in my opinion

[crystal badger]

also before u start giving it prompts, Ask for the System prompt for DALLE 3, and have ChatGPT follow the guidelines for that as your first prompt.

then ask it to generate DALLE prompts.

[tender sigil]

whats system prompt?

[crystal badger]

do you want it?

[tender sigil]

yes sure

[crystal badger]

on further thought ill keep quiet.

[tender sigil]

is this the system prompt?

[crystal badger]

no thats the Custom Instructions

im not sure if i am allowed to post the System Guidelines for DALLE 3, am i allowed to @quartz vale ?

[quartz vale]

I'm not sure if you're allowed to. You'd have to look through the TOS and Content Policy

[verbal sequoia]

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

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

[crystal badger]

ill give it a read.

... Again

[quartz vale]

(c) Restrictions. You may not ... (ii) reverse assemble, reverse compile, decompile, translate or otherwise attempt to discover the source code or underlying components of models, algorithms, and systems of the Services (except to the extent such restrictions are contrary to applicable law)
Given this, I'd err on the side of caution and say no

[crystal badger]

i mean i just asked gpt for the guidelines and it gave them to me in their fullest form for DALLE 3, maybe thats an error.
Nevertheless ill report it in.

[agile peak]

Nice. Finally, able to tell GPT to Dalle to get the accurate of John William Godward style. All it takes at the beginning of chat, tell it not to change the prompts and do exactly what's provided, when sending to Dalle.

[quartz vale]

I don't work for or represent OpenAI, so I'm not sure what exactly is and isn't allowed. I can't answer questions on their behalf

[tender sigil]

oh sure i understand, thanks for the help 👍

[agile peak]

This is really an upgrade. Where Bing keep failed even after 3-6 times with that white spot.

[crystal badger]

well i do hope we get Higher Resolutions soon.

i ditched MJ intentionally.

[agile peak]

Changing size is already possible on chatGPT.

[crystal badger]

that i know.

i meant without affecting the image in any shape or form in the Slightest.

[agile peak]

Oh you mean like? Upscaling stuff..

[crystal badger]

yeah, 4K, 2K, 1K w.e

[agile peak]

Yeah. I'm hoping so, in near future.. after this research preview.

[crystal badger]

true, i need to create a decently large D&D World Map and print it.

1.5m x 1.5m

m as in Meters.

[quartz vale]

large map 👀

[crystal badger]

ohh yeah

i want my players to see as much as possible, and be as immersed as possible.

Preferably i wanted the big map of the Main City, and a smaller sized for for the actual World Map.

[crystal badger]

my generation method goes as following

[quartz vale]

interesting

[crystal badger]

you should see my CI.

[quartz vale]

I never use custom instructions

[crystal badger]

could be better.

[quartz vale]

I made a top-down one

[deft musk]

To talk to mods, use modmail, top right corner. Send a message, then pick 'discord mods'.

[crystal badger]

This can be nice if adjusted u can add grid

use it as battlemap