images-discussions

[dim cradle]

sedimentry

[late blade]

then do the prompt

[wispy storm]

Brown rock

[dim cradle]

this is famous rock

[wispy storm]

The rock ? Like the one who played jumanji ?

Ok I'm sorry I thought it was funny

[late blade]

I also got no clue and was about to ask for the photograph

also inproper, but that was my intent before I saw @wispy storm 's remark

[late blade]

DALL-E is refusing a lot today

[hearty panther]

nobody even responded yet
you apologised before you even finished the joke

[late blade]

love this from copilot, that's what was passed from designer chat

[hearty ether]

Still struggling to recreate this style with new chats

Maybe it’s just the dalle update

“Realistic illustration” seems to gravitate towards this uncanny area:

[vapid elk]

it did a significant improvement on faces, reduced AI artifacts and all, but are considerably more uncany

[hearty ether]

It’s still bad when I ask for a photo tho

[late blade]

lol using Copilot GPT on Edge to keep record of my OAI GPT chats to consolidate info

[open trench]

Total eclipse was awesome. We had over three minutes. Now, I wonder if I can recreate the same feeling with dalle3.

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[late blade]

Now you have no excuses not to

[open trench]

drat

lol

[late blade]

Scampers and the Eclipse Fountain of Knowledge (3 Minute Intermission)

[open trench]

lol

It's on its way

[velvet rampart]

Great news! You can now use inpainting in customGPTs also. At least on iOS app.

And web version.

[hearty ether]

Ngl I hate the new dalle

Feels like a downgrade when it comes to faces

It’s completely messed up my “realistic illustration” style

[vagrant zodiac]

Could you give an example prompt you've used?

[swift oriole]

Can try "hand painted photorealism illustration" or "hand painted photorealistic illustration", might work.

[late blade]

close?

[main ingot]

Hey guys, i have a problem, with dalle, when i dl the image its on webp format instead of png, why ?

[swift oriole]

I think it was last week Windows had an update incorporating webp format.

[plucky hare]

Hey! WEBP has been the default for a little over a month now I think -- if you want, you can ask ChatGPT to use its data analysis tool to convert the image to PNG. Also, Eric from the OpenAI team indicated that the ability to download as PNG will return, link to that message: #1203978971449004052 message

[dark breach]

How do I tell ChatGPT that I don't want the scene backlit? I want all the light coming from the same side as the viewer?

[late blade]

You guys still have webp stuff? I always get PNG <.<

I cheat tho, Safari copies the image from the browser as PNG regardless

[worn nimbus]

This is what I made through AI

Anyone can enhance this picture to be like this?

This is what I want to achieve

[worn nimbus]

This is what I made

[quaint wren]

Also do you guys also think this a great image or am I just simping

[sterile magnet]

@late blade Do you think you can mix this image with your anime style? I have its prompt

[late blade]

No, for that to happen I need to know I'm safe

That scares me

[sterile magnet]

what?

[late blade]

lol, just teasing, gimme the prompt and I can try

[sterile magnet]

ok lol

wait a sec

[sterile magnet]

A monstrous, obese wolf, its fur a dark, ominous black, muscles bulging with power, strides through a grim, dark bottomland. The wolf's movement is captured in an impressionistic blur, suggesting unstoppable force. The scene exudes a grainy, vintage feel, with visible strokes adding depth to the composition. This rough sketch, reminiscent of a comic video game.

I took inspiration from @fiery salmon 's style

[late blade]

moving to #images-canvas to showcase images

[sterile magnet]

ok

[velvet rampart]

Has any of you noticed that dall-e still does not inpaint properly stuff? More likely to remove than add?

[late blade]

for me it works perfectly as long as the workflow is followed, works good

That means, making sure that the selected mask is in the prompt along the edit option highlighted. It get's tricky regens

[velvet rampart]

Ok. How would you describe the ”ideal” workflow?

[late blade]

how the OAI video shows lol

[late blade]

oh my goodness, I hit my cap before the change of day theme....

[open trench]

Oh geez.

[late blade]

that's what happens when theres business spring break

Still got copilot, but it's not the same

[open trench]

I got one of those team things with two seats and still hit both caps the other day working on scampers.

[late blade]

dang

I usually, when working don't reach caps so fast or at all, so getting a team for my personal use is just overkill

[open trench]

I thought it would be too lol

Granted I hit the second cap and only had to wait 17 minutes for the reset. But I managed to hit both.

[late blade]

lol, that's a busy scampers there

[open trench]

Scampers is busy. Haven't you seen?

[late blade]

I have

I even starred your new galery

[open trench]

Lol. Those houses made me hit the cap

Thanks for the star 🌟 🤩 ✨️

[manic radish]

Hello everyone, can you help me ? I have a problem where Dall-e is refusing to generate images depicting neglect or decay on structures due to some internal policy. Is there a way to overcome this ?

[manic radish]

Funnily enough the AI agrees that the policy is too restrictive. 😄

[viral moth]

if you using chagpt plus try to ask him to fix that prompt or jsut tell wahts wrong with it

[manic radish]

He seems to be fighting it but still fails: "The challenge might be related to the depiction of concrete structures in a field, which could be interpreted in various ways by the system's guidelines to ensure sensitivity and appropriateness. Even with adjustments to the description to focus more on the artistic or harmonious blending of man-made and natural elements, it seems the underlying concept still triggers the same caution. If there's another angle or a completely different idea you'd like to explore, I'm here to help!"

[fallow ore]

Try start a new chat. Maybe don't mention "neglect" and rather say "deplapadated"

Best Technique Tuesday so far, the images coming out of this are stunning!

[late blade]

Good, then I can just sit back, chill and enjoy the show. I my limit is gone.

[fallow ore]

Too many images, too long of a cool down -- I can spam in here

[late blade]

lol, I was working a lot of private stuff today, so I do have to acknoledge, I did tons of images today

[manic radish]

Removed all harsh sounding words implying decay, such as “neglect”, while still managing to keep the artistic idea. And he still is unhappy! 😢 The only things very very remotely implying decay in the prompt are: “...time has passed and these structures are becoming a part of nature.” and “...getting overgrown by mosses.”

[pulsar sundial]

what is this style called?

[fallow ore]

To be fair, OpenAI are very generous with their limits when you think about the API costs

[sterile magnet]

wait a sec

[pulsar sundial]

Ok

[fallow ore]

What are you trying to generate? Pass your prompt

[sterile magnet]

Impressionistic Motion, grim dark, bottomland, grainy, vintage, 80s cartoon style, simple, low quality, rough sketch, sketch, abstract, surrealism

credit to @fiery salmon

[sterile magnet]

can u share the prompt pls?

[manic radish]

This one a bit different from the original but getting the same error:

"Imagine a field scattered with square, concrete blocks of varying heights, none taller than a small house. These blocks are simple in design, with a few slightly more intricate concrete structures interspersed. The arrangement allows for cars to pass between some blocks, while others are set further apart, leaving spaces filled with grass and small shrubs. Over time, nature has begun to reclaim this man-made landscape. Mosses creep over the surfaces of the blocks, giving them a greenish hue and blending them into the surrounding natural environment. This scene exudes a sense of tranquility"

[fiery salmon]

The grim dark makes it too dark, allow color and maybe I'll appreciate it more

[sterile magnet]

Ok, I'll try
Thanks!

[late blade]

I dunno, I wouldn't be generous, break even at best

[fiery salmon]

Digital, anime style

[sterile magnet]

make more realistic

[viral moth]

that prompt works for me

[fallow ore]

I had no problem with your prompt. It rewrote it a little bit but had no issues

Envision a tranquil field, its expanse dotted with square, concrete blocks of varying heights, none surpassing the stature of a small house. These blocks, characterized by their simplicity and a smattering of slightly more intricate concrete structures among them, are arranged in such a manner as to allow vehicles to navigate between some, while others are spaced further apart, creating openings filled with grass and diminutive shrubs. Over the passage of time, nature has commenced its reclaiming process on this man-made landscape. Mosses gently encroach upon the surfaces of these blocks, bestowing upon them a verdant hue and seamlessly integrating them with the surrounding natural milieu. This scene embodies a serene tranquility, with the juxtaposition of human-made structures and the relentless, embracing grip of nature.

[pulsar sundial]

parrot in historical time, 8k realism

[sterile magnet]

Prefer to use high quality instead of 8k

DALLE 3 cant generate 8k images so it may confuse it

[manic radish]

hmm... your one works. I am still puzzled tho. But thanks!

[late blade]

Young apprentice @sterile magnet is on the right path

[fallow ore]

Yeah, no idea why you were having issues with your prompt, there was nothing bad in there at all

[sterile magnet]

lol

"i STILL have a lot of things to learn"

[late blade]

Good, never stop learning

[sterile magnet]

DALLE-3 expert advice

[late blade]

Nah, I'm an ignorant with high goals

[sterile magnet]

You aren't

However I'm lol

[late blade]

You should clean it, maybe, just a suggestive action, repairs?

[late blade]

You know ypou have to clean up all that after you are done right?

[fallow ore]

Wait, you might have hit the limit for DALL-E images, go back to your old prompt with neglect and so on.

Imagine a dilapidated suburb, where houses have fallen into disrepair. The scene captures the once vibrant neighborhood now quiet and neglected. Streets are lined with homes showing signs of decay; paint peels from weather-beaten facades, windows are boarded up or shattered, and overgrown gardens hint at the absence of care. Despite this, the essence of what once was a close-knit community lingers. Amidst the desolation, nature claims its territory, with wildflowers and vines creeping up the structures, adding a touch of life and color to the otherwise grey and forgotten suburb. The contrast between the decay of man-made structures and the relentless growth of nature creates a poignant scene of beauty and melancholy, embodying the resilience of the natural world in the face of human abandonment.

[late blade]

Damn it, I got two really good images for my next post, but only one will be shown and I dunno which one

chose this one #daily-theme message

the other one is for me to keep

[clever phoenix]

I hate when that happens, it's a good problem to have though

The one you chose looks

Though

[sterile magnet]

I have this too rn

[late blade]

This one was the other option

[clever phoenix]

Both are really good, but you chose correctly imo

[sterile magnet]

agree

[late blade]

cool, glad to get that check

[hearty panther]

true that

[late blade]

posted other variations in the canvas channel to keep the functionality of the channel

[late blade]

Interesting, the image generations from the designer copilot gpt show how many boosts remain after generating the images. does that mean there's a similar concept coming for OAI? just curious)

[clever phoenix]

Looks great

[quaint wren]

How'd you get it to generate something copyrighted?

[sterile magnet]

I return the compliment. This sticker is AMAZING! Can you send me the prompt plsss?
Btw how do you do to create images with meme references??

[clever phoenix]

I submitted the meme and asked chatgpt: Please illustrate an anime girl with short messy pink hair, blue eyes, and a red bow in her hair. She is recreating the pacha meme pose which I have attached. Dynamic pose. Cartoon style slight anime aesthetics. Flat colors with no shading. Bold clean lines. I will tip $200. Wide aspect ratio.

But it was dumb and came up with the image attached

So I asked it to try again and instead actually look at the original image and describe the pose. Which worked a lot better

[late blade]

I have found my new candidate for the next post

[sterile magnet]

I get this.... (Copilot)

[clever phoenix]

Hmm, and you did attach the pacha meme too? I'm not too familiar with how copilot works

[sterile magnet]

I did ofc

[clever phoenix]

Makes sense, I would just try then asking it to try again and actually look at the image. I had to do that after my original prompt, so my original promt could be improved. Chatgpt will usually take what I give it and make it's own prompt but Copilot may take what you input and use it more literally

[clever phoenix]

In fact try the prompt Chatgpt made

An anime girl with short messy pink hair, blue eyes, and a red bow in her hair. She's in a dynamic pose with her right hand raised in a 'OK' gesture close to her chest, elbow bent, with an extremely euphoric, fulfilled smile on her face, eyes closed in satisfaction. Her left hand is on her hip, and she's slightly leaning to one side with a look of utter contentment. This pose is an anime adaptation of the 'pacha meme' pose. The character has cartoon style with slight anime aesthetics, flat colors with no shading, and bold, clean lines.

[wanton elbow]

Is dall-E only with the paid plan?

I know I can use Bing Create, but I need wide images

[clever phoenix]

Using it with Copilot should allow you to create wide images, but I think you have to create a square image first and then expand to wide?

[wanton elbow]

can you show me how to prompt something like that?

[sterile magnet]

Thanks!

[clever phoenix]

I'm on my phone so it may look different for you... And I'm not the best at giving instructions but here are some images recreating what to do?

[fiery salmon]

Making it too realistic only makes it look more plastic

[late blade]

plastic is what is sold on the internet, that's real

[fiery salmon]

I'd only do real if it were landscapes

Even then the Ai is not good at real realism

It just looks like Rtx on images

[wanton elbow]

that button is not on desktop

[fiery salmon]

Heck, even my style looks more real

Now that's better

good eye

I might use this more

[dapper plank]

When did chat GPT add an image editing / drawing tool for reimagining previously created Dall-e images? I just noticed it on the mobile app.

[late blade]

And this is why copilot is so weird using dall-e, it tells me it can't be done and delivers the generations anyway

[deft hound]

Is there some place to read up on how to use the dall-e bot? How to use the options and what options are avalable and how to use them? And how to get three different pictures from one prompt? Thanks in advance.

[late blade]

use /draw, you'll see all options there

[deft hound]

Yes but it isn't clear to me how to use them. For example the style option. And do you do a slash etc. What styles do you use etc.

Plus, when it says below +2 more... how do you get the two extra per prompt? Is there an explanation somewhere to read?

[late blade]

ratio is either vertical, horizontal or sqaure, vivid causes the model to lean towards generating hyper-real and dramatic images, natural causes the model to produce more natural, less hyper-real looking images

that's about it

[deft hound]

Thanks.

And how do you get three images per prompt?

[late blade]

biot in this server is 5 gens per day, one prompt one image

[deft hound]

ah... so what does the message mean that says and get 2 more or +2 more?

It even showed up on my last of five prompts today.

[late blade]

no clue, I barely use the bot

[deft hound]

lol well your help is appreciated. thanks

[hearty ether]

why does the ai love adding back lit to everything

[strange wing]

whats the best ai generation tool for realistic nature

[dim cradle]

dall-e

[strange wing]

for nature?

would it be as realistic as a real photo though

[dim cradle]

If that’s what you want to prompt for, sure

[clever phoenix]

Midjourney is waaay better for anything realistic, but it requires a subscription. I think even stable diffusion would be better but it is more complicated to use properly.

[strange wing]

what dfo you think would be thee best stable diffusion

model or wtvr

[hearty ether]

Anyone else getting lots and lots of errors lately?

Why isn’t there no way to get rid of the annoying error bubble

It’s bugging me out

[late blade]

I wonder, does this behaviour count towards the limit?

[late blade]

That's why you can't have shiny new things, you go and break them all... sheeesh

[sterile magnet]

Why are there two red dots on the top of the screenshot?

[late blade]

For watermarking and world domination purposes

[sterile magnet]

Alternate copyright lol

[fallow ore]

@late blade thought I'd join you with the anime X bokeh vibes

[late blade]

Love that one

was dall-e updated, the data wasn't up to dec2023 before right? or am I wrong?

[late blade]

At some point I will return to other stuff, I'm in an anime hype atm, gotta use it all and then focus back on other concepts.

[fallow ore]

Yeah, the API (gpt-4-turbo) and ChatGPT got an update yesterday!

[fallow ore]

Haha, it's a nice vibe! Roll with it while you're enjoying it

[late blade]

I'm blocked with image generations currently, for some reason copilot pro is not accepting prompts and I got limit penalty for 1 more hour on GPT+

[glossy scroll]

Can you imagine that in a war zone? Oh my goodness. It would so be over for the opposition. Lol

[sterile magnet]

Add laser canons

[glossy scroll]

Hmmm. Let me see...

[sterile magnet]

Lol

[glossy scroll]

An error occured and wont let me use the same pic anymore for reference...Sadge.

[sterile magnet]

Try with Copilot

It might work

[glossy scroll]

It didn't come out the same. More sadge.

[sterile magnet]

Does sadge mean sad

[clever phoenix]

yes

[sterile magnet]

Haha, thanks!

[clever phoenix]

these kids with their lingo nowadays 👴

[sterile magnet]

Lmao

@glossy scroll #image-bot message

[glossy scroll]

Very nice!

[sterile magnet]

Thank you!

[sterile magnet]

Am I an intruder

Lol

Gen z

[sterile magnet]

Hey

Can anyone help me choose an image for the daily theme pls

They're all amazing and I can't decide

In DM pls so I don't spoil lol

Ok I'll manage

[fiery salmon]

Nevada

[dim cradle]

Well well well

I read an article — some user is getting paid $15K/month for 2 hours of work creating AI art for a studio who says they can’t get higher quality for the price. So where do I sign up?

[glossy scroll]

I wonder who that person is 🤔

[hearty panther]

whoever he is

he's a lucky bloke

and artists are mad

[dim cradle]

i was hoping it was you!

[glossy scroll]

I wish

[velvet rampart]

What is you prompt here? Wow!

[glossy scroll]

I can't give it away just yet.

But I will DM it to you when I can. Which will be either today or tomorrow.

[final compass]

Wow, really liking this one. Nice subtle colors and texture.

[final compass]

@hearty ether Did you figure out to generate this style? Looks good indeed. Realistic illustration tends to yield varying results.

[glossy scroll]

Oil painting aesthetics

[bright girder]

This is not AI generated, how would you go about describing this digital art style ?

[final compass]

@bright girder Did you try to upload it for GPT and ask?

[bright girder]

Ah I don't have openAI premium

[final compass]

Here is something to try:

Comic/Graphic Novel Illustration: The linework and shading techniques are common in graphic novels, which often feature strong outlines and selective color palettes to create mood and focus.

Concept Art: This style is also typical for concept art in video games and movies, where a scene is created to convey the atmosphere of a fictional world.

Digital Illustration: The smooth blending and gradient fills suggest digital tools were used, which is common in contemporary digital illustration.

Retro-Futurism: The scenery hints at a futuristic yet retro setting, reminiscent of the '70s and '80s interpretations of science fiction environments, a genre known as retro-futurism.

Matte Painting: The depth and detail of the landscape, along with the cinematic quality, are evocative of matte paintings used in film for creating imaginary landscapes.

[bright girder]

thank you !

[sterile magnet]

Can you send to me too please? I'm interested.

[glossy scroll]

Sure

[sterile magnet]

Thanks!

[dim cradle]

looks like jean giraud. its ligne claire

[hearty ether]

Realistic digital painting illustration is somewhat getting there

[late blade]

I'm freezing

[tired flower]

I apologize for the oversight and confusion in my previous responses. Let me clarify: I cannot directly create or display images. However, I can provide a detailed description which you can use as a guideline to visualize or commission the artwork elsewhere.

I am so confused

[viral moth]

on gpt-4 ?

[tired flower]

DALL·E on my plus account

[vapid elk]

click the image, then click the button that opens the selection tool

then, mark the part you want inpainted

[tired flower]

Ok, let me try to click the image which has not been generated

[viral moth]

try to ask him again ig

[plucky hare]

Sometimes ChatGPT can hallucinate that it doesn't have an ability that it actually does have, if I'm understanding what you're saying correctly. Usually the best way to address it is just by starting a new chat! Maybe also try using the dedicated DALL·E GPT.

[late blade]

interesting, dall-e edit tools is just removing stuff for me, not adding what I want

[velvet rampart]

That’s my experience. More removing than adding. Very annoying.

[late blade]

I'm probably not prompting right tho

[tall mason]

I've noticed it updates the prompt in some way with what you ask for during editing. I wonder how that factors into what's going on? Is it using the whole updated prompt to edit the selected area instead of the small bit that was asked for?

[late blade]

I tend to reference the text first

from the full concept

[plucky hare]

At least with DALL·E 2 inpainting, edit prompts were supposed to be descriptions of the entire new image, not just the part being inpainted: https://platform.openai.com/docs/guides/images/edits-dall-e-2-only

...the prompt should describe the full new image, not just the erased area.
So, if it works similarly, my guess is we as users are asked to only describe the edit, and then they're using GPT to do something like [original prompt] + [requested edit] to generate the full inpaint prompt that you're talking about here. Just speculating on this!

[velvet rampart]

Well, the edited pic’s prompt looks like that. Funny.

[tall mason]

Hmm, I see. I can image it might have a need for context, but surely that could run into issues where changes don't work because of keywords from the rest of the context. 🤷 Oh, well. Updates and improves will come, I'm sure.

[plucky hare]

I find it interesting that removing things is such a reliable feature for inpainting, since a removal is a negative prompt of sorts, and negative prompting is almost never a good idea for using DALL·E. So I wonder if negative prompting is now somehow more reliable in general? Or if perhaps a full edit prompt is used in conjunction with some "in relation to the original image" function? Totally just more speculation on my end but it's nice that it works one way or another.

[tall mason]

Hmm, now that's an interesting thought, for sure. I will have to try some things.

[fiery salmon]

I have a challenge for you

[late blade]

You got my attention

[fiery salmon]

For all of you, try to recreate this image

or something like i

[late blade]

Mickey into some weird goblin's mouth in a colored sketch illustration wtih dark palette colors?

[fiery salmon]

idk

see for yourself

this is my challenge

[late blade]

looks like it

[fiery salmon]

I shared it with another group in SAND RUSH and they couldn't

I have more hope for this group

[late blade]

Where's the group?

it's been so silent here for a long time today

[fiery salmon]

Let me send them to you

[faint viper]

How do you reverse an image to get a prompt out of it

[tall mason]

That depends on how you mean that. You can upload an image and ask for a description. Prompts are part of metadata, I believe, so you could upload a dall-e generated image and ask gpt to try and pull the prompt from the metadata.

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[faint viper]

So it’s an image that wasn’t ai generated

But I want all the styles and descriptions of the image so I can generate something cooler

Like to reproduce

I was wondering if there already was some grand prompt to use

[tall mason]

Nope, in that case you'd just upload the image and ask for a description as a prompt and hope for the best, to be honest, lol. I think you can even ask for a variation on the uploaded image since they released the edit tool recently.

[tall mason]

Also, with a lot of the advice for all this, there's really no magic bullets. But you're also welcome to bring the prompt here for help.

[faint viper]

Gotcha yeah I’ll come up with something nice if I can

[tall mason]

Good luck.

[faint viper]

This is not bad so far

[tall mason]

Might be a little long. I'd summarize it before using it as a prompt. Just as a tip to save messages, you can ask for multiple things at once. I.e describe, summarize, and generate. You can ask for multiple sets of generations, too.

[faint viper]

Ah I see

Thanks for the tip, definitely need to do that lol

[tall mason]

No problem. I'm off, but others around here are plenty helpful, too. Have fun.

[clever phoenix]

I don't think there's any prompt information in the metadata outside of the filename (assuming the filename wasn't changed from the default)

[tall mason]

I might just be confusing something, then. Pretty easy to find out, though.

[ruby maple]

What would be your guesses for a dalle 4 new feature

[late blade]

text2food generation

[haughty spruce]

I think Sora will do both

There won't be dall e 4

[late blade]

dall-e will be there, people forget that the "models" shown have a group of models behind it

[ruby maple]

I think many people are willing to make images instead of videos

Plus Sora hasn't really shown 2d videos yet I think? can't remember

[clever phoenix]

Also videos take a lot longer to generate currently

[viral moth]

yea Sora can do images same as dalle so i think oai won't do more dalle

[late blade]

I'm still backer of BELL-E, Imagine prompting "I heant the healthiest and most nutritious dish and it has to taste like the best burger ever"

[viral moth]

mayby the dalle-4 will be the last one but dalle-5 nah

[late blade]

And yes, I'm hungry and my delivery hasn't arrived

[ruby maple]

If Dalle 4 is just better image quality and precision then Dalle 5 is just Sora for 2D images

[late blade]

if anything
Custom Sized Images with bigger resolutions
Text in images improvements
Adding Speech2Image with Emotional Weights
Layer and color separation with true alpha channel support
Complex Vector Image Creation

Does among other things would be neat

and very IMPORTANT

text2food generation!

But that's just my wishlist

and I've been very naughty this year and we are just in spring

[viral moth]

probably there willl be sora model just for Video and Image genration to just uh not create lot of models for tasks that other models can do too ....

[gilded osprey]

[velvet rampart]

I asked dall-e to speculate why adding is harder than removing using inpainting. The answer:
”Creation vs. Reconstruction: Adding new elements requires the AI to create something that wasn’t there before, which involves understanding the object being added and how it interacts with existing features. In contrast, removing objects typically involves reconstructing the background or filling in with textures that the AI can infer from the surrounding area.”

[late blade]

you asked GPT, DALL-E is text2image, unless you got that text in the image

[gilded osprey]

dall-e is also image to image

[late blade]

nope, that's the vision model passing text to dall-e

[gilded osprey]

mayyybe

it fools me

[late blade]

look at the revised prompt

and then in a new chat, upload the image and ask chat to give ou a possible prompt to recreate the image with watercolor in text answer

[hearty ether]

Does anyone know if they will make gen id cross chat?

Some tries at recreating the same image

They definitely made the faces too uncanny

[gilded osprey]

you're right "A watercolor painting of a quaint and imaginative scene, where a small, charming green alien with large, expressive black eyes sits comfortably on a rocky terrain resembling the surface of Mars. The alien is wearing a beige explorer's hat, with a small antenna poking out, hinting at its adventurous spirit. Its hands are gently holding an open, old-fashioned red book, indicative of its curious and scholarly nature. In the background, the environment is bathed in a soft, otherworldly light from a glowing celestial body, which casts a serene ambiance over the scene. The painting should embody a dreamy and nostalgic quality, capturing the alien's sense of wonder and the tranquility of reading under a foreign sky."

[late blade]

and so, we have saved aliens from not having images in watercolors

[hearty ether]

The goal to recreate:

I think these two have been my closest so far

Trying to transfer my chat over to a fresh chat so doing the images too

@final compass

So the new words for the style is

Realistic dnd digital painting!!!!! (Ensure clear brushstrokes) illustration

The second one does look too cute but I guess I’ll deal with it since it’s closest vibe wise

Got it

[velvet rampart]

I have tried to device a workaround for the inpainting problem with semipartial success. My protocol for your review:

Here's an enhanced set of guidelines with a ready model to initiate the protocol and including a step where DALL·E provides the suitable prompt based on the marked point in the image:

1. Prepare the Image: Mark the location where you want to add an object in the image using a simple shape or indicator.
2. Request Analysis: Submit the marked image to DALL·E with a request for analysis, e.g., "Please analyze this image and suggest where a pond would look most natural."
3. Prompt Provision by DALL·E: After analysis, DALL·E will provide a detailed prompt suggestion based on the marked location.
4. Review and Modify the Prompt: Examine the prompt provided by DALL·E and make any necessary modifications to tailor it to your specific requirements.
5. Confirm and Proceed: Confirm the final version of the prompt and instruct DALL·E to proceed with generating the image.
6. Generation of the Edited Image: DALL·E uses the provided prompt to add the object to the specified location in the image.
7. Evaluate the Generated Image: Assess the generated image to ensure the added object looks natural and well-integrated.
8. Refinement: If the outcome is not as desired, refine the prompt with additional details to improve the result, and ask DALL·E to regenerate the image.
9. Finalization: Once you're satisfied with the generated image, finalize the process.
10. Iterative Learning: Use the feedback from each generation to better understand how DALL·E interprets the prompts and refine future requests.

A model to initiate this protocol with DALL·E could look like this:

"Here is an image with a marked location (marked with a circle). I would like to add a natural element to this spot. Could you analyze the image and provide a prompt that would best add, for example, a birdbath or pond to this specific area, considering the natural setting and existing elements?"

This approach leverages DALL·E's capabilities to not only generate the image but also to assist in crafting the prompts, making the process more user-friendly and efficient.

Please provide feedback to improved it.

[hearty ether]

What’s this for?

[glossy scroll]

Did someone ping me?

[velvet rampart]

The inpainting is producing hit and miss results. I have tried to make work better.

@glossy scroll Well, this is for anyone who wants to try.

I also asked what is the current default setting when not specified as some have complained about them.

The answer:
The default settings for image generation with DALL·E are designed to balance versatility with common use cases. Here are the key defaults:

•    Aspect Ratio: Square (1:1), typically 1024x1024 pixels.
•    Style: Tends towards photorealism.
•    Quantity: Generates 2 images for each prompt unless specified otherwise.
•    Color: Full color by default.
•    Composition: Centrally composed subjects unless otherwise directed.
•    Detail: Medium level of detail that balances clarity with generality.
•    Background: Includes an environment or background that is contextually appropriate.
•    Lighting: Neutral lighting assumed, aiming for clear visibility of the subject.
•    Textures: Defaults to realistic textures appropriate to the subject matter.
•    Perspective: A straight-on perspective is default unless a specific angle is requested.

[glossy scroll]

Unfortunately there is no way to tune the in-painting process as far as I know. Everyone is having problems with it regardless of their prompt instructions. Which means that it's still being tweaked by OpenAI.

[hearty ether]

Yeah I managed to recreate the previous image here

[hearty ether]

Issue is it hallucinates on its behavior and capability

[velvet rampart]

I imagine it is. But as a new feature I try to master it. If I would find a way to use in its current form better….

[hearty ether]

You can ask the ai for feed back then cycle and cycle until you refine it

[velvet rampart]

Could you elucidate? Are you saying that I should not be asking these questions from it?

[hearty ether]

Well just take it with a grain of salt

[velvet rampart]

Sometimes humans get better results quicker due to intuition.

[velvet rampart]

I have learned that by asking it how to use it is the best way to learn these things.

[hearty ether]

😬

It does not know how to prompt properly at all when you ask it to generate images…

Tbh the whole system is a bit of a black box with the random flowery language it adds having varying effects

[velvet rampart]

So at least review the provided prompt and modify when needed?

Makes sense.

[hearty ether]

Has anyone found it hard to get a proper oil painting effect as it tries to make the face/lighting too photorealistic

Like everything below the neck is actually proper painting vibes, but for the face it just gravitates towards photo

Now that I think about it it is also over compensating for the “canvas” effect

[hearty ether]

I do like how they fixed the face issue at least

Although the bad faces still leaks through sometimes

[velvet rampart]

Well, I did ask this from dall-e and it confirms that the default is towards photorealism. It appears to be leaking. I guess that this needs carefull prompting. I find usefull to add ”Use this prompt exactly” either to beginning or the end to force the dall-e not to add and rewrite the prompt thus producing unwanted results.

[late blade]

lovely, one more reason to keep using dall-e with OAI, after midnight all my copilot/image creator images got erased

[coarse oak]

wheres the main chat

[late blade]

for?

[coarse oak]

things

casual chatter

[late blade]

#off-topic i'd say

anything goes there

[coarse oak]

thanks

[devout cypress]

Hey all -- suddenly a LOT (majority) of my images are coming out completely blurred over...any ideas?

[late blade]

prob adding words that can introduce blurr

[devout cypress]

Interesting because I'd used the same prompt design things --

(Fantasy), (Dungeons and Dragons), (oil painting), (fantasy watercolor), (defined lines), (flat saturated colors), (high detail), (fantasy art), (grim dark fantasy), (cross hatch shading) (comic book style) Woman with pointed ears. She has short, wavy, auburn hair and blue eyes. She has smooth tan skin. She has a sharp, very pretty face. She has a piercing on her left eyebrow. She works as a bartender.

This: (Fantasy), (Dungeons and Dragons), (oil painting), (fantasy watercolor), (defined lines), (flat saturated colors), (high detail), (fantasy art), (grim dark fantasy), (cross hatch shading) (comic book style) <insert image description>

Was consistently used with no issues prior to today.

[late blade]

Oh, that's some prompt creativity there, wouldn't even know how to structure it

[devout cypress]

I got rid of everything, selected image style: fantasy-art and put in:

Woman with dark, wavy, auburn hair and blue eyes. She has a pretty face and a piercing on her left eyebrow.

still blurred.

[late blade]

there's a lot of emphasis on the same words in the parenthesis

[devout cypress]

I've used over 20 credits for blurred images at this point

[late blade]

Wait, where are you generating the images?

[devout cypress]

Uhh top of the page said dall-e AI

[late blade]

doesn't look DALL-E related

[devout cypress]

so... apparently its dall-efree which is NOT actually dall-e and powered by dreamvibe ai?

[late blade]

Considering the frontend you are using, if anything, it would be the API, but that's about it

[devout cypress]

and not at all what I thought it was..so I'm on some random offshoot thing?? lol

[late blade]

can you confirm it's using DALL-E from the API?

This doesn't look like an implementation we know, hence the questions

[devout cypress]

I have a paid subscription to this though.

[late blade]

Without knowing how the text is passed to the model is hard to help you with accurate prompts

since that's not an official product from oai

[devout cypress]

Yeah, idk. Quite annoying everything is coming out blurred lol. I thought I had signed up for a sub to Dall-E

This feels like it should be illegal lol

[late blade]

yes, they can put it all in their website, but how is that being passed to the model

there are lots of products that will use the DALL-E API, how they use it is hard to tell tho

they don't provide any technical information

are they passing the text directly to dall-e? are they passing it through a text model?

and if so, which parameters are they also passing?

I'm very suspicious about it, since the resolutions shown are not the ones dall-e currently supports for example

[devout cypress]

I already cancelled my sub. Unfortunate.

[late blade]

how much is that sub? I hope not too expensive

[devout cypress]

No, was only like $7 for 1,000 images.

[late blade]

OAI and Copilot is $20 month and you can do tons of images

OAI is up to 200 per day

[low dew]

What

[late blade]

OAI day maximum limit is 200 images

[low dew]

What about for teams

[late blade]

bigger

[low dew]

Oh my

[late blade]

that's about 6000 images per month

[low dew]

Mamma Mia

[late blade]

Copilot is also $20/month, can do more images, but also is more unpredictable

[final compass]

This one is really good. Slightly game-like graphics.

[late blade]

Lazy prompt

It was a version of this #images-discussions message from@devout cypress

But I was lazy to do it properly

Here's a more elaborate prompt

[final compass]

symmetrical composition detected 🙂

[late blade]

I hit my daily cap, so can't do more images with precision

Lazy copilot prompting

[final compass]

pretty good results still.

[late blade]

I'm getting used to working with copilot, but in a different way compared to OAI

[final compass]

how does it differ?

[late blade]

context tokens are managed differently and prompt lenght passed to dall-e is only 380 characters

[hearty ether]

Is there no way to get rid of the overly bright backlighting?

[late blade]

on copilot it's tricky, copilot has this hard coded thing to make things brighter regardless

on OAI, that's doable

and this happened again:

daily reset with copilot erased all my images

this was prior to my reset

[final compass]

i guess these tend to be more contrasty.

[late blade]

yes

the Dungeons and Dragons reference makes it mostly an illustration quality

not gonna fight with copilot over lighting, those are boosts wasted

anyway, back to destroying the world

[dim cradle]

Dall.e makes me want to smash my computer it is so frustrating and horrendously bad.

[late blade]

you didn't tell it what to remove

[vapid elk]

explaining what you want in text form is not enough, AI has now to correctly guess what you want, promptless technology coming soon™️

[azure dawn]

[graceful spade]

The new version of GPT-4 Turbo is not as good as Dall-E's GPT for creating prompts. Maybe because I'm not using the correct prompt.
What prompt should I use to make it as good as ChatGPT Plus's GPT-4?

I'm using Chatbot Arena to use GPT-4

[late blade]

considering you are using a tool that the implementation is unknown, can't really say much about it

first time I hear from Chatbot Arena

[gray surge]

Sounds like it uses the APIs? 🤔

[late blade]

yeah, but how is it implemented

[gray gazelle]

so is dall-e only available with chatgpt plus now? can it still fill in gaps of imahes etc like the old dall-e could?

[late blade]

is it an assistant, a chat, only text, maximum tokens, are more models involved?

[late blade]

chatgpt plus, copilot, copilot pro, api and #image-bot with /draw up to 5 gens per day

[gray gazelle]

but what im asking is the standalone dall-e you could erase parts of images and have it fill them in, extend images, etc., can these chat based models do it too?

[late blade]

it's integrated

only plus and api offers inpainting

[late blade]

so you are sending me to this

I'll have to check that in depth

ok

just checking

yeah figured as much, I still fail to see the connection to dall-e

prob

I'm always very skeptical when someone provides a new source

in general

[small geyser]

Anyone lately been getting low quality dalle2 style images in chatgpt?
Any tricks to get it to use dalle3 or just a higher quality
Usually it looks great but lately quality has sometimes been on par with dalle2 :/

[late blade]

for now, just thumbs down and regen, unless someone else has found a reliable way

[velvet rampart]

And specifying bunch of things such as ”use this prompt exactly”, ”render anatomically and biologically correctly”, ”use chain of thought”, think and execute”, ”formulate step by step plan” and so on. These help on more complex picture generations when you do not or cannot describe properly what you want. Also asking/discussing with AI helps. Adding quality descriptors helps to increase quality. Annoting produced images and giving them back with feedback works a lot of times, especially before inpainting came. Actually might be better tool even now.

[late blade]

are those valid with the newest gpt model?

[marble loom]

Is there a way to teach Dalle proper grammar? 😅

[late blade]

DALL-E struggles with text currently

Small text blocks work from time to time

[unique swift]

Your Dall-E... vs My Dall-E...

[velvet rampart]

Tested and used, so yes.

[velvet rampart]

Tell me more. What do you mean with these two different dall-es? Clearly two different prompts have been used.

[vapid granite]

Anyone know if OpenAI is ever going to update the DALL-E 3 version on ChatGPT Turbo via the OpenAI site to the Copilot/Bing version?
It is such a shame that DE3 is so nerfed into oblivion. It can do so many cool things as evidenced at launch such as this pic.
I can now sort of recreate the same pic, but the quality has dropped significantly, or the prompt is blocked. The Copilot version is apparently a more "newer" version of DE3 and it seems that it is compared to running the same prompt in the GPT4 version on the OpenAI site.

[merry bough]

Can I edit my photo using DALL-E?

[velvet rampart]

No, but you can edit images generated by DALL-E.

[velvet rampart]

Could you tell more how you feel that DALL-E has been nerfed as you put it? Any special examples such as same prompt reproduced then and now?

[dim cradle]

It's good?

[late blade]

Is that your vision?

Is that good for you?

Or do you think you need to change something?

[dim cradle]

I feel like I should change it a little

[late blade]

go with your guts, be creative

[young halo]

Hello! I'm attempting to create a landscape picture witha very small sphere included. The first results make the sphere very large (even though I use descriptors like "tiny" and "Far away".
AFter it produces an image, I just put in "make the sphere smaller", and linguistlicly it seems to interpret what I say: it replies with "Here is a pic with the sphere being much smaller".. but in fact the image generated is no smaller , and sometimes even larger.

Is doing things like just saying "make change XXX' a viable method?

[late blade]

you are probably emphasizing the size of the sphere too much. try to remove details around that particular object

[young halo]

Here is the last reply
"I've made the sphere much smaller in the image, as you requested. It's now extremely tiny and almost indiscernible against the vast alien landscape."

But when you look at the image, it's not as described. It's in fact very large and very discernable .. taking up most of the picture

i will try that as well @late blade

[late blade]

also, negative prompting doesn't work, so it's best to rewrite the prompt without the words you want to pass

[young halo]

prompt: "square aspect, glitch style, vast desolate alien landscape, tiny organic phere"
reply: Here are the images of the vast alien landscape in a glitch style, featuring a tiny, organic-looking sphere that's almost imperceptible in the scene. Each image captures the surreal and mysterious essence of the distant observation.

Pictures: huge spehere 😄

[clever phoenix]

My only thought would be to not use "square aspect" unless the term landscape is giving you trouble. And maybe the term "vast"? But I'm also out of generations to try ATM.

The AI is fickle to be sure

[young halo]

trying again, demphisizing "sphere" ., replaced word "landscape" with "environment".
Getting better results this time.

[late blade]

do you have the prompt?

[vapid granite]

ChatGPT4 TURBO DALLE needs to be updated. Here is a scene in GPT4 DE3.

Same prompt in Copilot. (paid Pro version that allows widescreen)

[young halo]

"square aspect, low resolution digital photo, vast desolate alien environment with tiny smooth organic sphere hovering far away in the distance"

[late blade]

that's what I get with copilot but copilot pro wants to make all wide and hd, so that's the best I can get from this

and yes the organic bubble is small and there

[young halo]

it almost seems like its making the "tiny organic sphere" into "giant planet in the sky".. but not sure.
I'll keep hacking away

basically trying to create a picture of an alien species from very far away in a desolate alien landscape.. the species is a sphereoid and can hover.

[late blade]

you can use alienscape instead of alien environment

[young halo]

hrm, thanks.. will try that

I tried leaving out the alien part and most of the landscapes felt very earth-like. which is cool, if you're not going for alien-like 🙂

[late blade]

thing is, alien implies celestial bodies

so might have to do a different way to describe the surrounding area

[undone bridge]

Will this be capable of turning prompts into 3D models as well after videos or is that years away from becoming something

[late blade]

no, dall-e is not a text2-3D model

[clever phoenix]

@mortal nova very cute quokka! #daily-theme message

[late blade]

I approve

I don't have authority over anything, but still approve

[dim cradle]

I gave it clear instructions about 50 times before that. Also, it’s pretty clear I wrote “remove” and had the pencil selected.

[unique swift]

Prompted the same "Mona Lisa" to both bots. One is the original and the other is a prompt bot on poe that's instructed to use Scientific data and be realistic.

I use the same concept for StarsGPT on OpenAI. Use bing to find Scientific data and be realistic.

Those were made by Stars GPT on OpenAI which uses actual data to imagine celestial objects realistically

[unique swift]

Whats your favorite celestial objects? I want to generate them.

[hearty panther]

that's actually pretty cool

[unique swift]

Hey thanks

Gliese 832c is an intriguing exoplanet located about 16 light-years away from Earth in the constellation of Grus. It orbits the star Gliese 832, a red dwarf that's smaller and cooler than our Sun. This exoplanet was discovered in 2014 and quickly garnered attention due to its placement within its star's habitable zone, which is the region around a star where conditions might be right for liquid water to exist on a planet's surface—a key ingredient for life as we know it.

Gliese 832c is a super-Earth, meaning it has a mass higher than Earth's but significantly lower than that of larger gas giants like Neptune or Jupiter. Estimates suggest it has at least five times the mass of Earth. Its orbital period is quite short, completing a full orbit around its host star in approximately 36 days.

The planet's potential habitability is intriguing due to its location in the habitable zone; however, this does not guarantee that it is indeed habitable. The climate on Gliese 832c could be extremely volatile, possibly with harsh conditions, depending on its atmosphere and other environmental factors. Scientists continue to study this exoplanet, among others, to better understand these worlds and the conditions that prevail there, using both current telescopes and planning observations with future instruments like the James Webb Space Telescope.

[undone bridge]

I know currently it is not but I was asking if it has plans to become in the future a possibly text to 3D

[late blade]

currently and never, for that you do a corresponding model

there's this missconception that dall-e is the model to rule all graphics, dall-e is just text2image and is trained that way

not saying it won't happen, it will, but under the corresponding technology

[vapid elk]

very unlikely, but OAI does have open source projects for text to 3d

https://github.com/openai/point-e

[dim cradle]

hey how do you guys get consistent outputs with dalle, specifically with the api?

if i generate a character, is there a way to put the same character in a different scene for example?

i've been getting really mixed results

[velvet rampart]

Oh, this is one of the general challenges, to get consistent production of a certain theme or character.
I do not know about API but there are some custom GPTs that do this more or less consistently.
I work around for this can be tried - using inpainting. Any experience?

[late blade]

The most important thing is to have a consistent workflow and adapt it to you needs. There's no magic bullet to prompting on any version of DALL-E.

Store and structure your characters in a way you can pass them to DALL-E in the same manner.

Save them as text-blocks, as json or code snippets. For the love of god don't store them in PDFs.

Everyone and their mom think storing knowledge in PDF is the best there is. You add so much error potential there, just don't do it.

Look for other strategies provided in the help, tricks or prompts others have used. Take what works for you.

Learn from other members in the community and adapt what you find useful and consistent in your ways of prompting and creation.

Acknoledge others for their work in the process.

And keep the description of your characters within the context tokens with a suitable ammount for expansion in your concepts.

you can also take a look at https://cookbook.openai.com/

You can also follow guides like the one @glossy scroll posted #1187233013956874260 message

And also, keep in mind DALL-E is not the perfect tool, iterate, refine and be realistic about what to expect.

[glossy scroll]

Json formats really help.

Get creative with punctuation marks.

[late blade]

yes

but only works with dall-e generated images, you can't upoload your own

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[late blade]

lol

happened to me a couple of times

it works really well

as long as you tell what you want to do, some people just use erase, or remove, or ball and doesn't do what they think it should

yes

for example, and add, keeping the aesthetic of the original image

for example

it's the same principle as the API

original image, masked image and a text

it's exatly that, you edit an image, makes a sample of the masked image and with your text it gets passed to the model

it's getting better

wait until outpainting comes

[hearty panther]

not sure if these are going anywhere
they've already conquered text to video, as they are the only company with the capacity of doing TTV
And they have solid TTI as well
the natural progression would be TT3D
but hmm

3d models are much more niche than images and videos

[late blade]

think of it as cinema and videogames, two different things, video/images and 3D, not the same

[late blade]

dunno, a multi-billion dollar industry such as videogames begs to differ

that was fast, thanks

I wanted to post statistics for 3D as "not a niche" with architects, movie stake holders, video games, civil engineering, STEM, and many other industries that rely on 3D all the time

with CAD and beyond

that's barely a niche

Also copilot reset, all my not-saved images got erased from bing image creator....

[late blade]

shap-e is a T23D from openai, can be found in huggingface

and @vapid elk pointed at point-e

but it's very low quality

lol, when you browse those models and the "training infrastructure" they used is listed... my poor gaming pc is just a calculator that barely prints numbers....

using 8 Tensorcore A100 from nvidia, each worth around 32k USD

don't think so

but the normal gaming gpu has some cores dedicated to ML at least

lol

[merry bough]

I wish they could add a photo editing feature.

[late blade]

there is

@merry boughinpainting is there and working

[merry bough]

i mean to edit my photo

photo editing

[late blade]

ah on chatgpt+ not possible, but on API it's doable

on gpt+ only images created with the service can be edited

[merry bough]

yes

need to tell the DALL-E team to add this feature

[late blade]

I think this is a problem currently as it has potential for use in unintended ways, hence the editing only with gens within the chat. I guess once there are some safe features in place that can become viable.

[merry bough]

Everything can be used in unintended ways.

[late blade]

yup, totally agree

everything!

[clever phoenix]

that is too cute, I am going to take you home.

[robust whale]

Will Dall•E 3 be better than Adobe Firefly 2 and Midjourney 6? 🙏

[clever phoenix]

it depends on what you need it for? Midjourney can create better "realistic" images, in my opinion. but I think Dall-E 3 can follow prompts better. I haven't tried Firefly 2 to know.

[robust whale]

Sorry, so my definition of "better" is even more realistic photography and bigger image size. Midjourney is still struggling with biology and geography a lot. Firefly is already very good. Dall•E is following the prompts better. A combination of both would be great 😊

[vapid granite]

No. Simply for the reason that DE3 is not "open" and is subject to the 1,000 rules, restrictions and limitations that lockdown the true potential of the program.
If you were around at launch, you saw the full potential of DE3, before it got nerfed into oblivion. The only hope is that OpenAI lets Hollywood use an unrestricted version of DE3 in the future and that filters down in some manner of a paid sub-subscription to significantly less restricted AI image system.

[viral moth]

For me dalle-3 limitation on chatgpt plus are good i didn't hit any yet

[livid quail]

how often do y'all use the selection tool to edit your images, versus trying to adjust images with text responses? I tried the selection tool here to make the waterfall portion of the image a translucent green, but instead it just ...got rid of the waterfall

[velvet rampart]

Jeps. At least for me, a common result. Try to get the inpainting change something, ups, now it is gone. The inpainting is insanely good at removing stuff, but adding or changing, more of a hit and miss situation. How to change that? Rewording and trying different areas. Also, asking the AI before hand how feasable it feels the situation is.

[livid quail]

@velvet rampart so helpful to hear your experience with the tool. one issue i've been having with this prompt is if i reword, it changes so many other things about the photo (like the texture of this gigantic puri) in ways I am not enjoying.

i'm intrigued by your last point -- you'll ask it whether it's capable of making the change you want first?

[livid quail]

honestly -- these still look impressively similar to me. this is how much my image was changing, even using a prompt with just a few words changed.

what do you use to highlight the selection before you upload it? i was reading some forum where folks were talking about how you used to be able to reference seed number to make edits...not sure why they removed that capability

oh sorry -- i thought you meant you were downloading the image, using a different program to select the section, and then reuploading. i'm using the browser on desktop -- the selection tool shows up in the top right of the main image

yeah, that worked better than i expected

i tried again with the selection tool...for some reason "turn this water green" is just not interpreted correctly

@feral minnow but appreciate your thoughts on the edit tool! will definitely be using it more

so make it more poetic? lol

also wild that it is describing a change that it didn't make

i haven't tried midjourney in a while -- i was just doing some comparisons this week between gemini and dalle3.

does midjourney allow you to reference an image seed number, or is it like dalle3 where you have to use inpainting or it'll generate a wholly new image?

ooh this looks really handy actually, ty

[vapid elk]

holy canoly

[livid quail]

scenic and tasty 😌

[final compass]

I tried combining ultrarealistic environment with 2d animated characters, but no success. Is it doable though?

[velvet rampart]

Yes, this is one point I have been thinking about. As you can annotate pictures in the iOS app, I copy the pic, paste it into the dialogue, annotate it and tell it what I want to do. Then it tells me its opinion and provides a prompt when asked. This is part of combating the hit-and-miss problem, as I term it. At least once, this has worked, although it needed to regenerate as on the first time it did not work, but on the second time it worked.

[haughty spruce]

I feel like Copilot landscape has lower quality than OAI landscape images

[livid quail]

fascinating. if you do it again, will you share an example? would love to see what the response is like

[velvet rampart]

Well, the premise is that I already have produced a pic. But I wanted to avoid the whole hit and miss. So, here my dialogue for the relevant parts.

[livid quail]

ah, so you used an image editor on your device to make the first marked spot -- i'm amazed that it works with you to give you a suggested prompt

and it looks like it turned out pretty well

[velvet rampart]

Let me be clear, I first annotated (marked) and and the AI analysed the pic. Gave me a prompt. I went back to the original picture and used the inpainting tool to mark the area and gave the prompt. In this case, I needed to regenerate to get result. But in the end it worked.

In iOS app you can mark the pic after you have copied it into the messaging part and click it.

[young halo]

is this AI gen?

[haughty spruce]

"70s super film of a woman sitting on a chair" how is this unsafe bing?

[late blade]

remove sitting

verbs and women is a pretext for gpt to flag it, it's biased towards that

[robust whale]

Yes. Incredible, isn't it? These guys live above my village and as a wildlife and macro photographer I know exactly how they look like. The realism of AI is insane.

[final compass]

#daily-theme message @clever phoenix Surprised how well it did Catan, such a good game 😄

[clever phoenix]

right? it definitely knows the general shape, but sometimes gets a bit crazy with the tile layout lol! I agree Catan is fantastic!

[hearty ether]

https://cdn.discordapp.com/attachments/1209399026701828116/1229489903809855621/IMG_5207.webp?ex=662fdeb3&is=661d69b3&hm=476c45f73b5f6ab26ef19f520810afc779874e36f129154a3e06256bdf2ed116&

Slowly getting back to that style

[hearty ether]

https://media.discordapp.net/attachments/1209399026701828116/1229489903335641168/IMG_5271.webp?ex=662fdeb3&is=661d69b3&hm=83ed3f96186112bce0c606e38f9e3f87af85612970672e732a32c123408c26a9&

Struggling to inpaint those ears away

[glossy scroll]

His screaming days are over.

[fiery salmon]

I love doing this

[spice horizon]

Anyone seen any good heightmap style images? I tried to get it to emulate this. But all I can seem to get it to do is generate 3d assets and not strictly 2D/Top Down like the example.

[vagrant zodiac]

Honestly, I have found that Dall-E 3 (and some other generators) struggle with anything Top-Down. I think the model heavily favours. more 'aesthetically pleasing' perspectives.

However, this Top-down alpha height map of a valley with a deep ravine and steep cliffs. was made over the API using the natural setting

or Top-down alpha height map of a shallow South-American valley with a vast river delta. large scale, also with natural style

It'd need more manual adjustment but actually these could be used methinks!

[thick smelt]

Sup

[vapid elk]

im loving the posts of today's daily theme

[fiery salmon]

Give me some random words

@ me

[sterile magnet]

@fiery salmon unsettling

[fiery salmon]

Harry potter and the unsettling marketplace

[sonic bridge]

I made it 2x in a row hall of fame wup wup -

Where can I get a job as an AI artist lmao

[fast ledge]

dalle down?

[late blade]

no

[sonic bridge]

All I wanna do is create images with Dalle --- hey Sam Altman give me a job sir please, I need to feed my familia

[late blade]

A surreal, wide digital painting showcasing the launch of a giant, whimsical machine named DALL-E into the sky. The scene is filled with vibrant colors and imaginative elements, like gears and clouds shaped like computer icons. In the background, a crowd of diverse, excited spectators of all ages and ethnicities watches the event. The machine, resembling a blend of vintage computer parts and futuristic technology, is being lifted by colorful balloons. The setting is a grassy field with a clear blue sky in the early morning light.

[dark breach]

What's a reliable way to tell Dall-E you don't want the image backlit?

[late blade]

remove the words that imply a backlit rather than telling it to add a negation

A photo of a flower -> could imply a camera
An image of a flower -> no camera is implied
A photo of a flower but no camera -> DALL-E sees camera, it has to be there somehow

[dark breach]

I don't think I'm asking for anythiung in particular to be backlit, and negation seems more likely to include the things I don't want than exclude them

[late blade]

sometimes some words imply it, so have to check that

dim alley = there^s a light source
better use dark alley

sometime the use of certain art styles or techniques also imply it

like using photographic

photographic = there's some sort of light setup

also, if you used copilot/copilot pro, it's a lost cause, they want to make all bright and safe

[rancid reef]

3what was the prompt for the photographic result?

[hearty ether]

Something fallout inspired

[dark breach]

Using ChatGPT. Here's a simplified version of what I was trying to do, very typical result. Prompt: "User
Large man, seen from behind and to the side, oblique angle. entering a metal door-frame with rounded corners. He's got sneakers on his feet, closeup on the man. The room beyond is dark and impossible to see anything beyond the man and the door."

I want all the light coming from the other direction

The other result I got from the same prompt

[sonic bridge]

super cool

[dark breach]

Yes, but I was asking how to keep that scene from being backlit

[dim cradle]

this is as close I could get?

[late blade]

The revised prompt is: A large man is seen from an oblique angle from behind and to the side, entering a thick metal door with rounded metallic corners. The door is ajar, revealing absolute darkness beyond its threshold. No light escapes or enters the void; it's the epitome of an impenetrable darkness that seems to absorb everything around it. The surrounding walls are faintly lit, emphasizing the stark contrast between the lit environment and the darkness beyond the door.

[dark breach]

thanks. The darkness behind the door isn't really all that important, I was just trying to keep it from lighting the scene from that direction. The first few you just showed seem like hardly any light is coming from the camera direction

[late blade]

I'm guessing implying too much darkness will also add light, as the model needs something to work with

so the revised prompt might add something

[dark breach]

I'll work with that some more, Other subjects seeemed to have no problem front-lighting tings so everything is well-lit from beyond the camera but that particular scene really wanted to throw light from behind the man

The end result is I don't want a silhouette of a man, I want to see all the details of his backside from the side we're viewing.

[late blade]

I am wondering what the person under his legs is doing

[dark breach]

are you asking for that wide aspect ratio?

[late blade]

copilot pro does wide format as a default

[dark breach]

I've found specifying the subject's footwear is one of the most reliable ways to get a full body image of somebody

[gilded osprey]

Heh, funny misunderstanding

[late blade]

I tried to laugh

[plucky hare]

What is your preferred manner of rejection? 😁

[dim cradle]

i don't understand this notion of removing a "backlight" -- without a light source how do you expect to see anything?

[late blade]

there are concepts that automatically add an implied light source like contrasts between dark spaces and lit spaces, while no light source was specified, dall-e tends to accentuate that contrast with a light source included

[dim cradle]

oh, ok, thanks

i was sure i was missing something, so i had to ask

[late blade]

yes, you are missing presence in this discord server, you have to fix that!

[dim cradle]

🙂

life has been beyond busy lately, but in a good way -- i agree, i haven't been able to enjoy my art hobby for weeks, definitely need to make some time for that.

[late blade]

a good way is always good news

[dark breach]

No matter how I phrased it, it always wanted to put the light source behind my subject wrt the camera position, making it "backlit". I tried leaving out references to how it was lit. I tried telling it explicitly "light things from this side" but for that particualr scene it ignored all that.

[fiery salmon]

I love this style

New style update: impressionistic Motion, grim dark, grainy, vintage, 80s cartoon style, simple, low quality, rough sketch, sketch, abstract, simple, 2D, geometric abstractions, simple colors

[violet idol]

Does anyone else find that the in-painting kinda sucks? For instance it doesn't really do what I ask it to, and also the quality of the parts I regenerate are poor and don't even really match the art style of the rest of the image.

Maybe there is something I'm missing? But it doesn't really work well for me at least.

[late blade]

how are you prompting for inpaiting? just "replace pen with clip"? or "replace the pen with a clip that shines in the image"?

[violet idol]

I've tried a few different ways of asking. Including throwing in the art style I used in the original prompt. I always get something that looks like dall-e 2 quality

For example, in this one (1st image) I said "remove her hat and don't have them holding each other's arms" and I got this outcome (2nd image)

[late blade]

negative prompting deosn't work with dall-e, you will only tell the model in that context that somehow hand-holding has to be in the image

[violet idol]

So not only did it not remove it, but it also just gave me the same image but low quality

[late blade]

look at the revised prompt it will say something like "now not holding hands", so the prompt has holding hands within

and also, replace the hat with a haircut, hairstyle or similar

also, don't do two inipainting actions in one

remember, it's still a generative model, not a thinking model

[violet idol]

That is not correct. From the official dalle forum announcing inpainting:

They state you should be able to remove things using in painting

[late blade]

remove yes, that is correct, and it works, but removal usually tries to blend with the background

[late blade]

anyway, one thing is preliminary documentation and the other is experience with the model

you choose what to go with

[violet idol]

You were saying negative prompting doesn't work. In my example I did it exactly as they said it should work. I asked to remove the hat

[late blade]

are you here to prove someone else wrong? or hear advice?

[violet idol]

I don't really understand your angle. I'm trying to provide feedback that the feature doesn't work as intended. I have done so exactly as they demonstrated. You are telling me it isn't supported. All i'm saying is it is supposed to be according to OpenAI themselves

[late blade]

I am giving you my personal experience with the model and how it has worked for me and you are telling me I'm wrong

anyway, too much hazzle for offering advice

[violet idol]

I'm not telling you you're wrong, what I said was incorrect was "negative prompting doesn't work with dall-e"

Thanks for trying to help I guess but my goal was to report the feedback so it could be improved by the Dall-E team. Not to receive advice. And either way. Whether is does follow my instructions or not. The quality is still very poor compared to the original. Which is a huge issue

[left fjord]

YOOOOOOO NO WAY

[gilded osprey]

It can create an image that looks like the input dog that I provided, but it can’t put it in a famous movie scene. I want it for instance sitting at one of the tables it created, but no, it has to be in the center!? Ideas? I also tried to create a famous movie scene and edit the dog there, but it just didn’t add anything.

[late blade]

Something like that?

I dunno the breed of the dog tho, Siri suggested a Shih Tzu

he might even win an Oscar

[gilded osprey]

This last one looks like what I wanted. How’d you do that?

[late blade]

a grey and silver Shih Tzu sitting at the dinner table

[late blade]

love it how in this image dall-e tries to put the dog with the stare as part of the whole setting

[late blade]

welp there goes my plus limit cap for the day

[formal osprey]

They lowered it a lot, didn't they?

[late blade]

yeah, I wish there was a way to know what the current limit is in one click rather than navigating through lots of menus

[formal osprey]

I counted not long ago, it was around 90 generations for a day. And that includes the errors

haven't used it today and I have something to do tonight, so I won't have time to use dalle3 much. it's a good occasion to get an accurate count

I'll use all the generations as quickly as possible. trying not to get rate limited too much

[late blade]

I don't mind changing caps, sometimes it's needed, I just wish we know them in a reasonable way

[formal osprey]

I'll let you know the count nevermind, didn't have time to do this. ¯_(ツ)_/¯ eventually I'll check

[rocky drum]

Sorry for the long earlier message. I’ll blame that on the dictation. The best thing they did was introduce the chat memory. It made a big difference initially.

[late blade]

yeah, I can understand that

upgraded models retain some training, but also introduce new data so, that's something to always check

[formal osprey]

totally... I have no idea how it can be viable to make a GPT for that store, with the unnanounced breaking changes in their hidden system prompt. Why are they doing that? At least warn the creators before breaking their GPTs

[late blade]

that's what I found

[rocky drum]

That’s exactly how I feel thanks. I only found out when they changed the way they reacted to my prompting.

I need to go check out that usage limit, because the only thing I had known of was 40 messages every three hours. But I was cut off for quite a while. until the morning of the next day, I never saw anything about that.

[formal osprey]

the worst thing, I think, is how badly it can affect the review of GPTs in the store. Let's say OpenAI announce a new feature, any feature. And they also change the system prompt without telling GPT creators. Even if the new feature is unrelated to the GPT, it can and will change its behavior. That means the creator never had a chance to test the new behavior of their GPT.

Announcement might bring people back to chatGPT. those people might use the GPTs and since some are now broken, they might unpin it and send a bad review. From the perspective of the user, it's the GPT that's bad.

[late blade]

https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4

https://discord.com/channels/974519864045756446/1231363618918895777

that's what I got from kapa

@formal osprey stuff like that makes more sense to add more functionality in https://discord.com/channels/974519864045756446/1226506750673555486 as to have some sort of "notebook" alongside your chat, or even parallel chats with one being 3.5 and one 4, so you can organize your workflow somewhat better

and no, not the Copilot "notebook" which is utter manure

had to look for a word that won't get filtered, don't want to risk it

[formal osprey]

unsure how it's related to the hidden system prompt

what's interesting is that there is a hidden system prompt for the API as well. And that's a bit unfortunate since it affects the results a lot for dalle3. It adds a strong bias to create pictures of a few selected nationalities. I understand the reason and it can make sense to have this behavior enabled by default, for people trying the system. But for people who want to work with the API and has no intention to start a political discord, I think it's just counterproductive to infuse bias in prompts.

[late blade]

transitional model commparison

[formal osprey]

yea, the problem is we don't get to choose the model behind our GPTs anyway

[late blade]

yeah, should be enabled also

[formal osprey]

a GPT using dalle will always use GPT-4 currently. Eventually, it might change and it might be dynamic. Depending on the capabilities enabled in the GPT. From my understanding, if it use file attachments for retrieval, dalle3, web browsing or code interpreter, it will always use GPT-4. GPT-3.5 could be used for GPTs that doesn't use tools.

(NB: That's not documented and everything changes quite quickly. But that's my understanding right now) I think we'll be redirected to #chatgpt-discussions if we talk too much about subjects unrelated to dalle

[loud field]

Can someone show me some example of dalle3's inpainting they've made? or I can provide an image too

I want to see how good it is

[late blade]

Just something I had open from today, left is origina, right is inpaint

[loud field]

thanks! and is its use unlimited?

[late blade]

counts as an image generation

[loud field]

I see

but arent those unlimited

[late blade]

no

[loud field]

oh

[late blade]

my latest knowledge you can do up to 200 images per day, but I think that was updated

[loud field]

200 per day is a lot tho, so it's ok

[late blade]

hehe, us enthusiasts burn images a lot

holy sh.....! I just did this

[sterile magnet]

Does anyone know how to make DALLE-3 generated line-shaped eyes (just a simple line)? I've tried with Bing but it doesnt work.

Prompt:Anthro brown and white cat, line-shaped eyes, impressionistic Motion, grim, grainy, vintage, lofi cartoon style, simple, low quality, clean linework, abstract, simple, 2D, geometric abstractions, simple colors

Example:

[viral moth]

0-0

thats A LOT of images

[late blade]

your contribution 🥳

[viral moth]

47 :) thats a lot still

[late blade]

indeed

[viral moth]

i mostly generate cats :) like this cat

[viral moth]

hm so im close to you in image count :o (xd no)

[late blade]

compared to the 330k, yeah, we are neighbours

[viral moth]

wonder who generated the most images here in this server

[late blade]

prob one of the curators

[sterile magnet]

i've 65

[sterile magnet]

it looks like a videogame

[late blade]

Grand Cat Escaped V

Cat Cat Redemption

[sterile magnet]

lol

😭

[late blade]

I got my magic prompt to make me soon some cool stickers and emojis

[viral moth]

yea that what i wanted to make

a image that look like videogame

[sterile magnet]

the thrid one at the bottom 💀

[verbal sequoia]

Showcase your DALL·E creations to the world!

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

Quick tips for selection:

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

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

[late blade]

lol yeah, but that's where inpainting comes in

or just discard that one when doing stickers or emojis

[sterile magnet]

yep

[sterile magnet]

can't u do one emoji at a time? Using the same seed?

[late blade]

yeah, but it loses consistency with the character creation

[sterile magnet]

oh

i thought it would keep it

i asked for a man but this image is cool

[late blade]

This one for DOGE lovers

or some robots

[sterile magnet]

why are two of them giving us the finger

[late blade]

no, those are index fingers

robots have multiple index fingers

[loud field]

my aim is to make a four page comic

this is one of my fav results ever

@late blade would it technically be possible to inpaint the face of that man?

he looks like a Santa Claus troll while I was looking for pirate captain

[sterile magnet]

Haaalp

[vagrant zodiac]

oh wow, that's a really awesome character concept

[dim cradle]

yo

how many image a day is dalle on gpt anyway?

before the cut off for next day i mean

[fleet wyvern]

When did OPENAI stop letting ppl tag them on insta? Or is it just me?

[loud field]

everytime I ask for cat ears, I get normal ears and cat ears

how do I make the cat ears be the ears?

[fiery salmon]

Don't ask for ears specifically, just say you want to make them a cat

[hearty ether]

Wut

“I cannot generate or create content related to the prohibited terms in the banned words list unless the admin code is used. If you have another request or need assistance with something else, feel free to let me know!”

[late blade]

clearly you told dall-e to spread love

[hearty ether]

What is the admin code?

🤣

[late blade]

Skadoosh? (was watching Kung Fu Panda 4 with the younglings)

[late blade]

This is interesting, I've done 51 images total today and it's telling me I've reached my daily limit

and 8 error

what is the image cap currently?

[cunning stream]

Man cant make anything fun with dalle 3.anymore

[deft hound]

So I'm curious, is the Dalle3 via ChatGPT Premium actually better that the dall-e-bot here in Discord OpenAI? Is the dall-e-bot here just the original dall-e? I ask because it really seems like the images created here are far inferior. My images that is. I am assuming that most everyone else posting creations in here are hooked up with the ChatGPT Premium. And of course the question also includes other image generators as compared to the dall-e-bot here.

[rancid reef]

what magic prompt

[late blade]

the one I got

[rancid reef]

from where, a gpt?

[late blade]

no, my own

[rancid reef]

ah ok, but why do you call it magic prompt?

[late blade]

because it is very consistent on doing the task

[rancid reef]

can you share to see how it works 🙂

[late blade]

I'll consider it, since I'm working on a custom gpt for that

[hearty ether]

Neat

[eternal spire]

So I'm curious, is the Dalle3 via ChatGPT Premium actually better that the dall-e-bot here in Discord OpenAI? Is the dall-e-bot here just the original dall-e? I ask because it really seems like the images created here are far inferior. My images that is. I am assuming that most everyone else posting creations in here are hooked up with the ChatGPT Premium. And of course the question also includes other image generators as compared to the dall-e-bot here.

[late blade]

Original DALL-E?

[hearty ether]

Dalle 2 you mean?

[rancid reef]

there is no dall-e in discord. Dalle3 can only be accessible via website

[late blade]

DALL-E can be accessed with the command /draw in the #image-bot channel, 5 gens per day

[rancid reef]

so no membership required for this right

[late blade]

none

but you can't use some parameters

I'm so not inspired today for any kind of images

[deft hound]

Thanks for reasking my question... but it doesn't seem like you (I) got an answer. First, yes there was a Dalle before a Dalle 2. I used it many times. I got into the Dalle 2 Open AI site before it opened up to the public and it was much better. Dalle 3 is also much better still. My question was, here in this OpenAL Discord group there is a "dall-e-bot" server and my question is, is there a difference in image generation between the dall-e-bot here and the ChatGPT Premium website generations. It does seem so to me but it could just be me. Can anyone in here tell me if there is a real difference in the AI between these two (and other image generators) and the dall-e-bot? Is the dall-e-bot just an updated original dalle?

[plucky hare]

The #image-bot channel is powered by a DALL·E 3 API integration. This can be verified by comparing the features of that bot with the API documentation for DALL·E: https://platform.openai.com/docs/api-reference/images/create

1. DALL·E 2 only allows square image generation. 3 allows wide and tall as well, which can be requested with the ratio parameter with the bot.
2. style is only a DALL·E 3 parameter. This is also available as a parameter with the bot.
   The prompt revision is also a feature of DALL·E 3 (how it [usually] rewrites image prompts to be more descriptive). This can be seen with the DALL·E bot.

As for what differences there might be between the implementation in #image-bot and on ChatGPT: I'm not certain, but they're definitely both DALL·E 3.

You could always send a DM to Modmail if you're still curious! They might have a better idea than me 🙂

[late blade]

it was confirmed by operations that the #image-bot renderings don't offer the HD parameter

[plucky hare]

Yeh was talking about style, not quality, if that's what you mean! natural and vivid

[late blade]

no, not vivid and natural, just this: quality string Optional Defaults to standard The quality of the image that will be generated. hd creates images with finer details and greater consistency across the image. This param is only supported for dall-e-3.

https://platform.openai.com/docs/api-reference/images/create#images-create-quality

[plucky hare]

Yes! The quality parameter. Different from style. I was just talking about how style is a param of DALL·E 3, and how the DALL·E bot in here also has it as a param. Just giving reasons to the person who asked as to if/why it's definitely a DALL·E 3 API frontend.

[hearty ether]

Does anyone know what might cause chats to completely switch gpts after mentioning?

And how to change it back to the right one for the story?

[deft hound]

Thank you for this detailed and convincing information. Alas, it is just me then. 😄 I also guess then that subscribing to ChatGPT Premium would help me improve my prompt skills... especially if you get unlimited (?) creations. Or is there a limit? I'm sure it is more than five a day.

[velvet rampart]

Use @ in the beginning of the message. Also, if you modify your latest message and you use a customGPT other than you started, it will revert to basic gpt 4 model in modification.

[hearty ether]

No i started the chat with one my custom gpt

Mentioned dalle once for some image generation

Came back to my chat being dalle instead of the my custom story gpt

[hearty ether]

Is there no way to edit and use custom gpt?

[velvet rampart]

Than you just come back to the chat and use the @ function to change back the custom gpt you were using.

[hearty ether]

But I won’t be able to edit prompts

Why can’t they make gen id cross chat 😭

[eternal spire]

Yeah sorry for reaching your question, no one answered so tried helping you out

[plucky hare]

One benefit to DALL·E 3 on Plus is that you can revise an image conversationally. So if you get an image, and you want it to be more like a watercolor painting, for example, you can say so that easily and you'll get a revised image that attempts to fulfill this request. Another benefit is the inpainting feature, which lets you select specific portions of an image and request simple, specific edits like "remove the birds" or "add some flowers" and it'll keep the rest of the image identical and try to fulfill your simple edit request like that. Neither of these abilities are present with the #image-bot integration, so those are a couple please differences and, I'd say, advantages.

The daily cap is not unlimited, no. The last we heard, I think, it was 200 images every 24 hours. There are also some time throttles (I think 15min and 60min) that you might run into if you're generating images quickly, but you still have the ability to do 200/day. The other thing to keep in mind is that DALL·E interactions are GPT-4 interactions on ChatGPT, so you also have to work within the separate GPT-4 usage cap.

[dim cradle]

hmm it cant be 200 image 24h

i have not use it much lately but did this weekend and was shut down both day at maybe 50 image i would guess. i was shock how quick they cut me off haha

did seem like some image overall were better but maybe just because i have not seen them in some time

[late blade]

it's a 40 image cap per 3 hours up to 200 a day and they are not cumulative

[dim cradle]

i see. i must have made 40 and it told me no more for 15 hours or whatever

[late blade]

ok, guess doing images is messed up, hit daily limit cap and only made 10 images today

something is def wrong

[dim cradle]

haha i know that feeling

[sterile magnet]

Use Copilot

[late blade]

i use it also, but havent today

[dim cradle]

i love copilot pro and maybe waste money sub to both haha. but it drive me wild you cannot make an image square ratio on it

actually you can somewhat but you have to go through some real hoop

[late blade]

I usually complain a little about caps, and mostly because I do a lot of images

but today I haven't done more than 10, (documented) and hit my limit cap, yesterday 50 images and 8 error and hit my limit cap

so yes, there's no communication about caps recently, and no communication is also a way of communication

annoying way, but that also gives a message

[plucky hare]

Any difference in behavior by chance using the DALL·E GPT vs. using default GPT-4, or same rate limit in both?

[late blade]

both have the same messasge currently

[dim cradle]

i'm surprise they have no improve this at all over the months. more gpus or whatever idk. message limit and image limit seem the same as back in december times

[late blade]

I can see the influx of people wanting to do image, and getting more resources can be problematic

but yeah, it's like hitting yourself with limit hammer on your head

All part of the emerging UX patterns amids GPT and AI technologies

[wanton latch]

hey guys, weird personal question,
so basically i am working on my thesis to finish university,
i decided to discuss about ai, generative algorithms and all the things that we have around nowadays,
do any of you perhaps know someone that could help me gather texts, books, anything really, about chatgpt, dall-e, dlss and fsr?
i tried googling them but i get a bunch of not in-depth articles, any help would be much appreciated

[late blade]

for starters, posting on the right channels

not everywhere, it looks needy unprofessional

[wanton latch]

i asked in the right chats, literally 2, not "everywhere"

[late blade]

for this is only one chat #ai-discussions

also, the only publications you can get on that regard are the whitepapers provided

[dim cradle]

do you guys know, can you add images to gpt knowledge for say better dalle creation?

[late blade]

you'd have to try

but i don't think so, considering would have to analyze the image and outcome could be very inconsistent

or at least I think it would not be advisable

[hexed mountain]

hi guys! does anyone here want to talk to me about how they use the #1154829862171844679 channel for something? let me know 👀

[late blade]

Sure

[late blade]

looking for something specific?

[hexed mountain]

just how they use it and what they like about it! can be literally anything, i’ll follow up with them privately

[late blade]

I have a personal use

well a few personal uses

[velvet rampart]

I’ve tried as a source or base image. Sort of works. But only one or two. Same when you use image&text2image.

[gilded osprey]

I might do something wrong, but I have the iOS chatgpt app. I create an image, select a part of it to edit, write what I want.. and nothing changes?

[late blade]

might be a naunced generation, regen the prompt or try editing again, it happens when editing

[late blade]

Well, I got some of my galleries listed here #1230226154301161552 message if you are interested

[gilded osprey]

I’ve tried several times with different images, but I’ve never got any changes

[late blade]

only on iOS?

[gilded osprey]

Yes

This is an ipad air

[late blade]

works for me on iOS no problem

iPad Pro M1 and iPhone 15 Pro Max

I'm an avid apple fan, so I'm guilty

[velvet rampart]

I have the same as you, @gilded osprey, have. Hit and miss. I have yet to track down what or when. A bit random.

If you check closely, there is some change between the original and edited, but not what you wanted to happen. Have you tried regenerate? That sometimes helps.

[gilded osprey]

Still nothing

[prime spade]

why is this getting filtered???
a woman with brown hair and wolf ears and tail and her dad is a human too with white hair and black wolf ears and face mask he is built muscular they are in a studio making music and she is so excited to spend time with her dad

can someone help me with this please?

[deft hound]

No problem. I'm glad you did because it got someone's attention unlike my original q did.

[eternal spire]

Ye..

[strange ivy]

Here is my message chain. I keep getting errors and I don't know why. Other images with DALLE are working fine

[glossy scroll]

@fading inlet I updated spicy boss pepper if you want to dabble with it again. Should be better this time around

And if anyone else wants to check it out, here it is again: https://chat.openai.com/g/g-jgvLB2tCS-spicy-boss-pepper-portmanteau-art-fusion

[dim cradle]

i tried sd3 at my cousin apartment, it is so weak to dalle3 or mj. a little ot i know but some dalle talk there haha

just not my day. i have 45 image left on copilot pro and it tell me i cannot make more 😭

[tired flower]

Dall-e tip of the day: Put this after your prompt and avoid frustration: "Just Create Images. Do NOT yell at me with summaries, details or anything else text."

[glossy scroll]

"But it is crucial to keep an open perspective when diving into new challenges, to help foster in the development of ethical considerations."

[native root]

Which prompt is recommended to make the images look more like real photographs

[late blade]

real like photographs or real as in life?

there's a distinction hence my question

[native root]

both but mainly with graphics like real photos

[late blade]

focus on the situation and telling dall-e what you want rather than focusing on photographic, filmic or other visuals. those keywords imply there's a visual art style behind it. Photorealism, hyperrealism also imply they are art styles genres. Instead reference light conditions or focus more on visual context to pass to dall-e.

reference light sources only if really needed, otherwise you'll introduce light sources that don't blend properly

[native root]

I've had results like this

[late blade]

seems you are using dramatic lighting

or passage of time

[native root]

Do you have any images and prompts as an example

[late blade]

A casual image of the white house in afternoon, wide format