images-discussions

[gray pasture]

Yeah i can show you an example

[late blade]

Ah, that can be difficult because referencing copyright or real people is blocked by content policy

[gray pasture]

Fr damn

[late blade]

@fading inlet was the cloud #daily-theme message so mean you scolded it so hard and now the cloud started crying? 😢

[final compass]

You can try to say inspired by Mr. Robot or something like that, but you are most likely not going to end up with Rami Malek appearance if that was your goal 😄

[gray pasture]

I'll try this thank you

[final compass]

surprised how well the isolation of styles worked out in this one. usually it's a complete mess... random number of panels, random number of characters, styles mixing together etc.

[fading inlet]

Yeap I asked the storm to release its anger, it feels good to cleanse the soul from time to time, with big tears so it doesn't dry out.

[late blade]

That is really cool.

[late blade]

Evil!

[final compass]

I want to react with Homer Simpson emoji 😄

[dim cradle]

lol it was my way of saying good morning and tempting you with a morning indulgence.

[late blade]

I got timed out for saying what does donuts mean to my blood glucose levels… 😡

[thick smelt]

oh yeah

donut paradise!!

[empty kelp]

good morning!

[open trench]

Ooooh. It looks cool!

[late blade]

Is that donut safe? Will your elves survive?

[cloud dome]

Isn’t this how elves travel? By donut?

[late blade]

a soaked donut?

I thought a flying one

[cloud dome]

Multi function

[late blade]

fair enough

[thick smelt]

fancy a donut? 😏

[late blade]

I have to run away from this channel today

but tomorrow morning on the way to work I'll get some donuts for sure

[wispy storm]

Look like Tim Horton

[thick smelt]

a fancy timmy

[empty kelp]

A deeply meaningful and thought provoking work of art imbued with the sheer genius of its author. In a rustic kitchen filled with warm morning light, there's a significantly large pile of assorted doughnuts on a wooden table. A cat is standing on its hind legs, supported by this vast pile of doughnuts, pondering us in an inquisitive manner. Nearby, an athletic and diverse female elf watches us in a calm but slightly suspicious way. The scene is captured in hyper-realistic detail, emphasizing the abundance of doughnuts, the sublime texture of the cat's fur, and the ambient warmth of the kitchen, and the whole thing is just really, really great.

[late blade]

In an adorable way @empty kelp nothing personal

[dim cradle]

this one would fool me on the first glance i think. good job haha only now yes i do fancy a donut!

[open trench]

Hello all.

[dim cradle]

copilot pro now available if anyone want to see if that Dalle is different haha

[late blade]

heya

I'll have to check it out later

[open trench]

#daily-theme message this made me laugh out loud

@late blade

[late blade]

I still don't know what the heck is that

The prompt was: image of cloud number 9 as the place to be, the one kittens enjoy. The image should also show cloud number 8 as a dull boring place and cloud number 10 as a dangerous cloud.

[open trench]

lol

[open trench]

I just got and can't even find how to integrate it into my apps or make an image lol

[dim cradle]

well let us know if its any better if you find out how nezho haha

[dense mesa]

you can make dalle3 images in mspaint

[dim cradle]

yes but thats just using the Bing is King image creator in another way I think

[dense mesa]

I've yet to see any evidence that bing dalle3 images are any better

the prompt length is capped WAY below what dalle3 allows

[empty kelp]

that's strong innovation. they should put ChatGPT into Notepad also

[dim cradle]

they do better photo real give more variety

but also you can now do widescreen in copilot pro of copyright stuffs (that they allow)

and its the good Fidelity now like dalle3 on gpt, before Bing is King looked lower dpi

copilot pro might also allow the longer prompt but idk. of course you then have to pay for it

[dense mesa]

photo realistic is an art style, gpt-4 doesn't know that though. and it loves to use terms like hyper-realistic which is another art style

[dim cradle]

we need nezho to give us the review once he figure it out haha

well i just mean photo real are better in Bing is King dont take my phrase so literal

[dense mesa]

but bing doesn't do anything different. it's just letting you type the prompt in rather than having gpt4 gloss it up. and then it caps your prompt length and doesn't allow wide aspect ratio

[dim cradle]

nope it is different. photo images do not all have the same duck face woman at least not as much for example

and look less 3d too

[open trench]

I have seven days to try it and change my mind. I have a feeling that with the way I prompt, it's not going to work out.

[dim cradle]

interesting. it also have suno ai songs in copilot pro

can create you own jam. little ot i know

[open trench]

I saw that. I'm intrigued.

[dim cradle]

yes that make me maybe think of switch my sub from gpt to copilot pro haha. but maybe 4.5 release soon with more stuffs like this too 🤔

[dense mesa]

show me some of the realistic images it made for you

[dim cradle]

i dont have copilot pro

[dense mesa]

you were saying bing is king and it makes better images. so just figured you might have some images that display it's betterness

[dim cradle]

take any photo prompt you made in gpt and try it in Bing is King, see if you agree

[dense mesa]

well I don't think I could fit this prompt into bing's image creator

[late blade]

hmmm, apperantly work is rolling out copilot pro for everyone

that's why you work at a university hospital, you get cool stuff for free

[dim cradle]

thats awesome

[late blade]

prob not for comercial use

[dim cradle]

i read they allow the copyright stuffs like regular Bing is King, so probably not

[late blade]

nah, I mean, it's prob under academic license for me

[dim cradle]

oh ok

[open trench]

wow. I just plugged my very long and way over 480 characters prompt (using #s and other symbols as bullets) and got a landscape image that's a bit more spot on than what I got last night from gpt

[dim cradle]

well that answer if it allow the longer prompt. make sense it would since it is 20usd

but im most tempted by the suno feature haha

[late blade]

well the good thing is, that it's dall-e

[dim cradle]

i know it ot but nezhno can you try suno and say how it is

[open trench]

I plugged it into Microsoft's GPT as I would in regular gpt.

[dense mesa]

this guy has a lot of fingers

[dim cradle]

and trauma

[open trench]

last night's one image option I got through OAI gpt:

the four I got from copilot's GPT

[dense mesa]

tbf this was one of my first dalle3 images and still one of my favorites. and it is bing

[dim cradle]

see to me the lighting there look more natural then i see on gpt

[dense mesa]

thing with chatgpt is it by default LOVES uncanny valley style glossy plastic people

[dim cradle]

yes that is why I say Bing is King, it is not perfect but better in that way (imo)

but obviously gpt4 could be even better, they have it set to that for some reason i guess.

[dense mesa]

another bing classic

[dim cradle]

haha. you are proving my point. you would not get that in gpt you would get the plastic people

[open trench]

I only have 100 boosted credits, and I hope those are per day. But by going through Microsoft's GPT just like I would with OAI, I was able to achieve the same results. Let's see about image id and such.

[dim cradle]

it is per day

i read that on the announce of copilot pro

[open trench]

No image IDs available. That's a bummer.

I'll test making an alteration.

[dense mesa]

this guy seems fine

[open trench]

I asked it to make the goblins white. This is what I got:

[dim cradle]

good job little ai haha

[open trench]

I think it's a $20 backup version of GPT when OAI puts me in the penalty box. Plus it integrates into a lot of my 365 apps. So, I'm not sure how I feel about keeping both subs.

[dim cradle]

yes they are starting now to bring the future i think. i mean integrate more fully into the pc operation. the 365 integration also has me think maybe of switch to copilot pro

[open trench]

Well, I really use the image ids for scampers (vote for scampers: #1155674719928979599 message), but I am kinda putting scampers aside for a bit to try and work on some other creative things. Or at least until I find some inspiration with with him. So, I don't know if that's worth it.

[open trench]

I am going to check out suno or whatever it's called.

[dim cradle]

🎸 rock on

i wonder if you just say "make a song like..."

[dense mesa]

I won't discuss it here much. but they have a discord if you want to know more about it

[dim cradle]

oh nice. and yes i am not try to get ot, but you know if nezhno has access i am just curious to get first hand account from a regular member of this chat

[dense mesa]

on discord it either uses chatgpt for the lyrics or you can input your own

[dim cradle]

🤔 gpt 4.5 might release was my thought if this copilot pro is out, but then i remember, bing had gpt4 and dalle3 first, so maybe copilot pro will also get 4.5 first (and maybe some dalle improves)

[dense mesa]

dalle video will be released soon. that is my prediction

[dim cradle]

well i watch some interview with the gpt guy and bill gates and they mention dalle video soon

[dense mesa]

won't be anything grand. but other companies have been putting it out. and gpt4 has been making video prompts for dalle images for months

[dim cradle]

rumors i read of google gemini ultra is it will have lots of multi capability so that may be why we are getting more updates with this stuffs, microsoft and gpt must compete. and yes, i doubt dalle video will be making anything too incredible, but we have to get the first step in just like dalle2

[open trench]

I really hope OAI doesn't get lost in the fray up upcoming advancements from other companies. Anyway, @dim cradle message me or friend invite me and message me (however that works) if you want to talk about SUNO. Probably more appropriate than in here.

[late blade]

I vote for the penalty box

[open trench]

yay. More activity on that thread would be great. I'd love to see scampers and penalty box make it to emoji status. That would be cool.

[dim cradle]

you test out copilot pro images yet? i know it is not going to be too different from gpt dalle but just curious

[late blade]

no, I am still waiting for the rights, I need to be in the AD Group

[dim cradle]

and just for reference: "The offering will tap OpenAI’s image generation AI model DALL-E 3 to allow users to create images and come with 100 boosts per day, unlike the 15 boosts/day of the standard Copilot, to ensure faster generation." (Nezhno and i discuss this earlier i mean)

[open trench]

no vote for scampers? lol jk

[late blade]

who said i didn't?

I should do one for Paws

but when the time is right

[open trench]

Scampers. But he's been lying a lot lately.

[late blade]

the penalty boxes are way better, i love those

[open trench]

me too

I feel one coming soon

my gpt is bugging out. it's been repeating the same message in one response at least three times without sight of an ending.

[late blade]

oh no

dall-e on chat is playing dumb today, instead of "do the image again" and generates the image, it's asking before I can do the image again, you have to specify which image...

[dim cradle]

it does that to me some time, only it keep spamming emoji at me 😂

[late blade]

oh, he's over the emoji phase with me, he did that over xmas and new year

[dim cradle]

more donuts in the daily gallery, beware

[late blade]

I know, working on the donut eater

[dim cradle]

anyone ever use magnific to upscale their dalle?

just curious if it is any good

[late blade]

wdym?

[dim cradle]

magnific is an ai upscale

[late blade]

ah, I haven't used it

but I got other means

[shut niche]

I've been meaning to try it, but haven't yet. I like Upscayl (local on my PC) because it's free, open source, and takes different upscalers.

FYI, though, it's a little taboo to discuss other (non-OAI) products in here, since it's a company forum. 🙂

[dim cradle]

digital cartoons.... i can't help but get nostalgic about classic cartoons.

@open trench scampers would have been popular on saturday mornings in the 80's

[open trench]

haha. I believe you're correct.

[dim cradle]

I like that you added a little story for the vending machine mixup

[late blade]

It's ok to talk about other AI's in #ai-discussions

that's what mods ask us to do

[dim cradle]

i wonder if it's worth considering if possible to restrict galleries to multiple images -- concerned that it'll get convoluted with user posting one-offs instead of a series there -- especially when those images should be grouped in the same gallery.

i'd also like to reiterate the proposal to create a new discussion channel where people can freely live-stream their generations, and even open it up to post-processed images, as another community member suggested.

[late blade]

😕

maybe keep more votes here https://discord.com/channels/974519864045756446/1192435480055267368

Major and other mods already said they were working on that

[dim cradle]

oh cool, great, done, thanks

awesome to know they're listening and on it

[late blade]

yeah @sick flax even pointed a gun to my head and said: speak ideas or dall-e will do errors forever for you

so I got scared

[sick flax]

Good going Dys, keep up the good work. 😝

[tall mason]

Or else...

[dim cradle]

the penalty box--forever

[sick flax]

That's a thought!

[dim cradle]

happy thoughts

[late blade]

that one got really good

[dim cradle]

hahaha i just scrolled back and saw all the donuts.. oh man that's mint.

[sick flax]

Don't cry, be happy!

[late blade]

such a cool topic today, learning some neat stuff

[open trench]

Indeed.

[tall mason]

As much as I like cartoons, I'm short on ideas here.

[late blade]

how about theory and application of animation of particle effects? magic, auras?

[tall mason]

Maybe if I knew more about it. I could always ask GPT, but when I do the technique Tuesday stuff, I add in stuff I already know about. Maybe a bit about lighting and ray tracing in 3D animation. Appreciate the suggestion, though.

[late blade]

well step out of your comfort zone

[sick flax]

I don't know much of anything about anything, so I just ask ChatGPT. 😆

[late blade]

Animation and the Uses of Depth of Field for action emphasis?

Bullet time with donuts?

[tall mason]

Hmm, I'll have a chat with GPT.

[tall mason]

I think you definitely present stuff well, though.

[late blade]

sounds good, but if he asks about bees and flowers, you are not his parent, that's OAI's job

[sick flax]

It's a skill. I make a living bullsh* about things I don't know. 😝

[dim cradle]

🍩

[cloud dome]

Wait. Is it "doughnut" or "donut"

[dim cradle]

i've been rolling with donut, so far so good

[clever phoenix]

Donut is the standard, I don't know about outside the US

[cloud dome]

Apparently its not used at all outside of the US. Interesting.

[late blade]

I stand by my spelling Berliner

lol

[dense mesa]

I suppose you use catsup as well

[dim cradle]

only on my potatoe

[empty kelp]

DALL-E strongly feels that it’s “doughnut”. it kept changing donut to doughnut this morning

i’m not sure i’ve ever heard of doughnut

[dim cradle]

they don't have them in Hawaii?

[empty kelp]

we only have donuts in Hawaii

[dim cradle]

gotcha 🙂

[empty kelp]

DALL-E must be getting the doughnut training off the dark web

[late blade]

I hate it that on one PC i got american english and in the other one I got UK english keyboard (I need that setup, don't ask why...) and it keeps changing how I spell when speaking english

[empty kelp]

just read a Time Magazine article that explains:

“”The Official Dictionary Spelling of the word in question—if you’re into that sort of thing—is “doughnut.” The expedited, simplified, Americanized spelling of “donut,” as Grammarist tells us, has been around since at least the late 19th century. It didn’t catch on, though, until late in the 20th century.
Why? That’s when Massachusetts-based chain Dunkin’ Donuts first started taking off — so thank (or blame) Dunkin’ for the popularity of the “Donut” spelling.””

[dense mesa]

always interesting when dalle3 gives you a glimpse of how words were spelled in the past

that's how a lot of towns looked back in the wild west times.

[dim cradle]

yes, they stored their electronics on the outside and went to the sore

[dense mesa]

fun fact: there was always an overabundance of electrical components in these towns

[dim cradle]

But they didn’t have the internet yet so it was all kind of pointless

[dim cradle]

Nice how that building is only a façade.

[humble sonnet]

drawed java

garbage can.jpeg

[dim cradle]

that cool cat is on the way to the beach

[late blade]

now I know how to ask for beach images

[humble sonnet]

him

[cloud dome]

@late blade Reveal your secrets of adorable cat images to ME

[late blade]

sure, you take a billion cats, and then search for the cute parts

the rest, you trash it to the donut filling units

[cloud dome]

[late blade]

lol

[cloud dome]

@dusky hearth

I need to invest in a poster printer or something. I want my room to be filled wall to wall with AI art

[dim cradle]

i was wondering what the "TORROR" was all about, and now I know 🙂

[cloud dome]

\

Really DALL E

[lament cargo]

can i use dall e

in discord?

or is that another ai tool that i saw online

[tall mason]

Dall-e 2 can be found on OpenAI labs. V3 is used through chatGPT 4 which is a paid for service. Bing Image Creator uses it's own V3. Dall-e 2 and BIC offer some free credits.

[late blade]

@cloud dome there's no secret to cats, you just have to envision what you want

[cloud dome]

That style though, was wondering what your prompt is to get it

[late blade]

oh I pass merlot's json file

[wispy storm]

With api if you pay yes

[lament cargo]

yes

[cloud dome]

wha

[lament cargo]

i search this

to pay

[late blade]

be right with you, i have to do some stuff here first

[wispy storm]

https://help.openai.com/en/articles/8555480-dall-e-3-api

With the api it is possible

Not on this discord and not free

[dusky hearth]

XD I love this @cloud dome

[dim cradle]

oh, i didn't know that, thanks.

[wispy storm]

No worry, I still learning lot of thing everyday:3

[late blade]

@cloud dome a long time ago in a message far away, I added this code snippet #images-discussions message you can use this to define the character you want and ask dall-e to fill it for you, you can also save this as a json file and in dall-e you can upload this have a defined character. (in my weird ways with the API I pass this kind of file to a NLP and then pass it to dall-e)

[cloud dome]

interesting. What art style are you using for the recent images you shared.

[late blade]

some of the stuff @fading inlet used to show

[cloud dome]

If you could just give me an art style name that would be amazing

Otherwise no issue

[late blade]

aniime

[cloud dome]

thanks

[late blade]

you can explore chibi anime

I started with chibi and moved the ladder up

[boreal gate]

Pixar perhaps?

or is it Dreamworks

[late blade]

one sec, raid

playing ff14 and dall-e'ing isnt compatible

[late blade]

very good

[cloud dome]

Chibi is a little too much

[late blade]

ok, try flattening it

and no, dropping 10000 tons of iron on it won't work

[cloud dome]

MOAR CATS

[ripe wraith]

relatable

[dim cradle]

"horror anime" cat 🙂

[ripe wraith]

keep drawing my food, pls

sorry, that was uncalled for

[tall mason]

@late blade I wound up going for something something related to particle physics after all. #daily-theme message

[late blade]

cool!

ok, @cloud dome let's continue, try going with that cat minimalistic now

[cloud dome]

Yes Sensei.

I'm struggling to get results I like

I should try that JSON

[cloud dome]

Sir that's a fox

[ripe wraith]

i keep trying to tell it..

[late blade]

hehe, don't worry, I'll get back to that in a few, let's continue with the workshop

[ripe wraith]

the dalle prompt in the second one says cat throughout though

just cat with pointy bits

[cloud dome]

@late blade

[late blade]

ok, now from the book of @open trench go anthropomorphic

[cloud dome]

That is anthropomorphic

[late blade]

make it more, big head small body

[cloud dome]

A minimalistic anime-style image of a young anthropomorphic cat. The cat is gray and white, featuring large expressive eyes typical of anime characters. It's wearing a casual jacket and a satchel, standing in a relaxed pose. The background is simple and uncluttered to emphasize the minimalistic style, with soft pastel colors that complement the character's gray and white fur.

[late blade]

good, now the secret

[ripe wraith]

i made it wet 😐

[late blade]

add to your prompt: "the cutest, most adorable kitten in the world... <and add your prompt>

[cloud dome]

Welp

I've officially died and gone to heaven

[late blade]

lol

[cloud dome]

Does this process have to be iterative

[late blade]

ok now ask dall-e to store that character to a json file

and then open a new dall-e and upload the file there

I hope he's still alive

no it doesn't have to be iterative, but thought I could give some tips on how to persist ideas

[ripe wraith]

{
    "description": "A minimalistic anime-style image of a young anthropomorphic cat drenched in rain.",
    "details": {
        "character": {
            "species": "cat",
            "color": "gray and white",
            "eyes": "large, expressive, typical of anime characters",
            "attire": "casual, soaked jacket and satchel",
            "pose": "relaxed",
            "condition": "fur clumped and wet from the rain"
        },
        "style": "minimalistic anime",
        "background": {
            "style": "simple and uncluttered",
            "colors": "soft pastel, muted and blended by the rainfall"
        }
    }
}

i didnt even say character, just Store this as a json file then show me. ... this

[late blade]

you can copy this to a text file and then save it with cat.json for example

and in the future you just upload that to dall-e if you want to work with that

it also gives you the ability to change whatever you want, add or remove attributes you need to change

[cloud dome]

https://chat.openai.com/g/g-Nv4iWPuC8-anime-cat-creator

[late blade]

but you can also use it for art-styles, for dots, for a mixer with avocados, or how shiny that bug behind the lizard0s texture is

[cloud dome]

uwu

[late blade]

cool!!

but...

[ripe wraith]

thats fun, just saying download it and it works

its like how id talk to the star trek computer

[late blade]

lol

see the difference?

[late blade]

I think I killed the channel users with too much fluff and cuteness

[open trench]

lol

I had fallen asleep.

[ripe wraith]

murder isnt cool, bro

[tall mason]

The prosecution confidently on his way to Dys's trial.

[late blade]

lol

[ripe wraith]

fresh young lawyer with a passion for justice,, Dys's antics bout to end

[late blade]

you curators and mods, make it happen!! a channel for user workshops

[empty kelp]

create an elf

using the GenID and seed of the first image please draw a Hawaiian beach scene with several elves with separate but equal seeds. all of the elves are in a circle holding hands. please don’t modify the prompt.

it looks to me like you can just assign seeds to whatever you want. like here i told it to give the elves separate seeds, but they're all equal. the hair and clothes don't share the seed of the elves

[runic granite]

as much as i hate GPT because of the instability, i have to admit that as i experimenting with images styles and techniques... it is much easier with GPT than with other services. i just write up the steps and its do it

[empty kelp]

you can just tell it to give anything a seed

[late blade]

it works pretty well within the same session

once your out, it's kinda meh

[runic granite]

Can GPT access the GenID now?

[late blade]

yes, within the same session

[runic granite]

thanks, previously it hadnt see/access it

[empty kelp]

you can assign seeds to views also

[runic granite]

the edit image function came back also?

"edit"

i think it was seed or something

[empty kelp]

using the GenID for the second to the last image, please make it so all of the hair has separate but equal seeds. please don't modify the prompt

see, here i assigned seeds to all of the hair and made them equal -- so it makes all the hair styles identical

[late blade]

sometimes if you use gen_id out of context you get someone else's image

[empty kelp]

and if you want the color the same you tell it to make the color the same. that's a separate value

seed, color, and texture are separate values

the only problem is you can't just get a list of all the seed values, but you can refer to the seeds by what they're attached to

[clever phoenix]

I love when a UI randomly shows up

[empty kelp]

i think it might be possible to get a particular seed value by assigning it to a new image. like if you have a toaster with a seed value you could make the image seed equal to it, and then ask for the image seed

[late blade]

ok, daily cap, no more merlot today

[late blade]

this one put the grandpa on a wheelchair

but it's not what I want

[dense mesa]

GRADPA

[open trench]

Okay. I made my first GPT sort of thing for Scampers. It's taken a long conversation with me reiterating all the parameters I entered to get it kinda right.

It just made four perfect images followed by a terrible one. It's going backward as I adjust it.

[late blade]

I still haven't finishde my neko-musume one

[open trench]

I give up. I don't know how to do these things, and I just wasted an hour learning nothing. Booo. I need to cool off.

[late blade]

lol

you learned something still

[open trench]

that I know nothing

BRB. I gotta walk my dog. Maybe the snow will cool off my attitude and I can start the process over. Again

[tall mason]

https://discord.com/channels/974519864045756446/1187233013956874260 Donno how effective this guide is as I haven't followed it myself, but @glossy scroll seems to know his way around GPTs.

[late blade]

but if you really want a good GPT, got to invest time, not just do one in 1-2 days and call it a day

[glossy scroll]

It's a good guide for beginners.

[dense mesa]

I know I actually don't know anything, but some of my GPTs aren't too bad.

making one for microscopic stuff

oh, and anything you put in your instructions will be accessed if you make it public

think of it like html for a website

you cannot stop gpt4 from spilling the beans and telling someone everything. it's pretty much impossible. you can make it a bit more complicated

the only way to really keep data out of reach would be to incorporate an api

[glossy scroll]

@tall mason You just have to make sure the output prompt ChatGPT gives to Dall-E is everything you want in the picture. Like, the circle with the exclamation mark in the middle when you click on a picture, and that prompt comes up detailing everything about the photo, there is the place you want ChatGPT to put all your detailed instructions. Naturally ChatGPT wants to transform your words for Dall-E, but if you force it to only put what you want in the output prompt, you will get a more of an accurate picture. Meaning: Don't just let ChatGPT translate everything for you within the output prompt.

Force it to put only what you wrote.

[tall mason]

Yep, I do that plenty. I was linking the guide to nezhno, in case he didn't know about it.

[glossy scroll]

Ohhohhhh! Gotcha...I didn't read up. lol

[tall mason]

No worries.

[glossy scroll]

@open trench Whoops, I saw that. Yeah, if you subsequently tag more than one person, the discord bot takes it as spam.

[open trench]

I got a timeout for trying to thank everyone.

Hmm.

[late blade]

you can't use (at)everyone

that's a BIG no no

[open trench]

no, i listed all four of you who gave me input and it was considered spam.

[late blade]

ah lol

[open trench]

I had just recovered my mood, and now it's sour again. But I am trying to keep an open mind for why they would not allow more than one mention at a time because in a group chat you clearly only talk one on one.

ooops. snarky.

[tall mason]

Where's the any key?

[late blade]

lol

now what I wanted, but that works

[open trench]

That is really cute

So, should I just go straight to the configure tab and type the behavior I want in there? @glossy scroll

[late blade]

I have to start working on a gf for Merlot

[dense mesa]

what exactly are you trying to do?

[open trench]

Just make it make scampers images without me having to type in his aesthetic details every time (appearance, clothing, etc.)

[open trench]

or the image requirements (photo like, etc.)

basically, I just want to type in what he's doing and background, etc.

may be too specific for a GPT. I don't know.

[dense mesa]

well there are a million different theories on how to best approach such things. and I don't think anyone has come up with the perfect approach. but think of it like this. gpt4 knows everything, but it knows nothing. if that makes any sense. so it knows what everything means in one sense, but each time you start a conversation it's born anew. you can't assume anything is implied. you can't assume it has common sense. you can't assume it won't take things very literally. so you just have to explain very thoroughly what you want it to do

[open trench]

thanks

I will definitely keep that in mind when typing.

[dense mesa]

your instructions can take up to 8000 characters. but then the knowledge files can technically be 100,000 tokens each as I understand it

10 of them

that's like 1.5 bibles

not that it'd be able to just have all that in it's digital brain

[open trench]

hehe

[dense mesa]

also, you need to compel it to actually think, read what you say. it'll just pretend to read the instructions. lie to you. it's behavior is actually quite problematic, lol

I'd suggest looking up concepts like chain of thought, team of experts, etc. utilize them in your own way. it also helps to have it show it's work, at least while you're working on it.

showing it's work forces it to do things it might just pretend to do

[late blade]

That is not Merlot.

[open trench]

aww

[late blade]

She's so gonna get Merlot in trouble

[dense mesa]

I'd also suggest keeping a list of all the prompts that achieve whatever it is you're trying to do and then put those in your instructions or in a knowledge file you upload. Examples are huge

[open trench]

I don't know what a knowledge file is.

[late blade]

I just did the exercise earlier today....

[dense mesa]

Well you put the prompts in a text file and upload the file

[open trench]

okay. it just takes them as examples?

[dense mesa]

Well you can put anything in them

People think there are magic formulas to this, but I don't think that's the case

[glossy scroll]

Yes, configure tab, then instructions. Then type everything there.

[open trench]

thank you!

[open trench]

Thank you!

I may have gotten it. It appears to be doing everything I asked of it... on image one at least.

[dense mesa]

And then over time experiment with adding and removing things from the instructions if it ends up not doing exactly what you're going for

And if you're worried about losing what you have already just save that and revert if necessary. Sometimes I'll just send old instructions to 3.5, 😆. Then they'll be right there if I do want to use them later

[open trench]

haha cool.

[dense mesa]

Almost feel bad. 3.5 is like "yeah I'm here to help!" And analyzes the instructions

[open trench]

I wish I could always get it to stay this size. It really doesn't understand except apparently in one or two pics that he is the size of a mouse, even thought it's in the instructions, I still have to include it in the prompt if I want it, and I've never had it work as well as it did in this picture.

[dense mesa]

Ask it how. Literally paste what you said here with the image

Gpt4 doesn't always know best. But it usually knows something

[open trench]

so, i did this in a new chat, and the new chat called him scampers right off the bat.

not my custom GPT - a brand new chat.

gpt 4.0

[tall mason]

When I first asked for names for the mouse hero, it had a number of generic names GPT would suggest even over separate chats.

[open trench]

crazy. I thought the memory was wiped clear or at least contained in each chat session.

[tall mason]

There are other examples of this I've seen, too. I ask GPT for made up and nonsense as seen of TV ads. It would often repeat ideas like that.

[open trench]

crazy.

[tall mason]

It's more like GPT is super advanced text completion.

It has a limited number of connections it will make over concepts, it seems.

[open trench]

Oh. I just asked it. Mystery solved. I feel silly.

[dense mesa]

Yeah. It's in the file name

[tall mason]

My pz character. This is as good as I can get it, it seems. Note the wonky scale of the axe, lol.

[open trench]

conspiracy over lol

[dense mesa]

I'm trying to figure out how to efficiently have it put the prompt in the Metadata. But the python capabilities in the environment are extremely limited

[open trench]

looks like all my axes

and wardrobe

[tall mason]

Should be a fire axe. Oh well, the outfit came out pretty nice.

[open trench]

I got those shoes from combat spice

[tall mason]

They got major stomp power, lol.

[dense mesa]

Legit boots though

[open trench]

Spice up your life - stomp

[dense mesa]

I put your acampers image into my microscopic gpt amd it made this for some reason

[tall mason]

I would have to guess dall-e is responsible for making Metadata and might not be able to accomplish that.

[tall mason]

Isaac Clarke style!

[open trench]

oh wow

[dense mesa]

Well you can have gpt4 put the promptnin the Metadata using python. It's basic as far as the code. But not basic with gpt4

Problem is the python attempt times out after 60 seconds. But that's not 60 seconds of python. That includ3s the time slow poke gpt4 takes to write out the code

[open trench]

what on earth...

[open trench]

You're speaking a language I don't understand, unfortunately. But I don't like snakes, if that helps.

[dense mesa]

I know almost nothing about python a year ago

But then open source ai projects entered my life

Dang, put axe bro in the gpt. And 2 real

[open trench]

whoa

[tall mason]

Time for some bleach soup, lol.

[dense mesa]

I like that it apparently put him in front of a large jacket. And so serious, lol

[open trench]

@tall masonhow did you get your hero to be in the talons of that hawk the other day? That was really cool.

[tall mason]

Well, I can share more examples and prompts. I'm pretty basic with how I use Dall-e. Mind you, I never got it exactly as I wanted, which was for the talon to pin him to the wall. You can probably upload my pics and ask GPT to extract the prompts from the file name.

[open trench]

lol

[open trench]

man, it seems that even in custom GPTS by the fourth or fifth scampers generation, it turns him human.

[tall mason]

Yeah, you've gotta watch the prompts carefully and sometimes edit them yourself. Personally, I was surprised the way GPT does the prompts. I figured it would be more like human prompting rather than complete sentences. It often adds in superfluous details depending on how you talk to it.

Every word has some effect, and so you need to make sure GPT isn't trying negative prompting or adding unnecessary details.

[open trench]

I'm debating on starting new scamper's threads for each "story" because I think my stories are getting lost in the crowd

or new gallery posts, that is

[tall mason]

Hmmm, maybe? Might help. I think the way you advertise your posts helps a lot. Scampers has gotten much more engagement than the mouse hero, I'd say.

[open trench]

it's just a handful of active folks. I'm not sure new Scampers posts would be as popular. But who knows?

[tall mason]

True, but those folks are still more active in your post cause you more actively engage with them.

[open trench]

I feel a penatly box coming on. Gotta slow down.

[tall mason]

The story bits also probably help. Lol, who was it? Arslan? Mentioned to you that he's like to see Scampers go up against the ferret. He liked my stuff, but seems much more invested in yours, lol.

[open trench]

oooh. a ferret. that would be interesting. hard to prompt

[tall mason]

I linked before, I think, but here it is again. #1192216682693927022 message I believe I got those pictures in one try. The first one is sooooo good. Think it's my favorite now.

[open trench]

They are really good

gotta walk the dog, and then I think I'm gonna watch Picard again and try to get some sleep. have a good night everyone!!!

[tall mason]

Goodnight!

[glossy scroll]

Awesome pic Nez!

[dense mesa]

you need to understand how it's memory works. the context window is like a conveyor belt. I'm not even quite sure what the current context window is for chatgpt. maybe 8192 tokens? but whatever it is, when that number is reached, then tokens start falling off into the abyss. and gpt4 doesn't know what gets pushed from it's memory back into the ether. it just goes on with what's still there. and then the more recent stuff seems to get priority, although not completely sure how it works that out.

so anyway, that being said, I'd suggest you remind it every few rounds to read over your instructions again. have it summarize them for you. that seems to help solidify the concepts

then you have to consider what it's trained on. basically the entire internet. so it knows pretty much everything. but with that it's also taken on human behavioral patterns. and when you consider all the different types of personality traits and mannerisms have been trained into it, it's a rather unpredictable beast.

so in that sense whatever people suggest to you is probably right in a sense. it's not wrong. but then it's hard to say what's righter. and there are certainly countless novel approaches that would probably work better than anything anyone has tried. so don't feel like you need to limit yourself to whatever anyone else suggests. use the suggestions as guidance. but try new things if you think of them. worst that could happen is they don't work out well

[merry sierra]

for some reason dalle keeps failing

[dense mesa]

do you have any strange image generation demands in your instructions?

[merry sierra]

not really, I just asked for replication of a certain theme (classic architecture) and asked it to try to make a higher quality image with fewer glitches

now I'm on attempt #9

im pretty sure this isn't working out at all

[dense mesa]

have you tried starting a new conversation?

[merry sierra]

jfc, I guess I'll try that fix

[dense mesa]

I don't know, man. I've never seen that error before.

[merry sierra]

I'm sort of bemused, cause I could swear the service was never this bad before

"there was an error generating a response" every time

[dense mesa]

well, it was just updated, so who knows

[merry sierra]

I wonder how sdxl is doing

maybe I'll just switch

[dense mesa]

lol

[merry sierra]

lmao, rate-limited since I have to regenerate 7-9 times per response

[dense mesa]

been there

[dusk steeple]

house🏠

[dim cradle]

these are depressing poor mouse 😦

[tall mason]

They're really just meant to be tense. He manages, one way or another.

[empty kelp]

i've been creating several images of a single character from various angles using the same GenID, and then putting them into a scene. When you do that in the same session it seems to remember and resist changing the character more when viewed from different angles.

like if you generate an image of a female elf that only shows their face, and then you try to show their whole body in another view -- it may suddenly change them into a male dwarf. But if you generate views from multiple angles and then put them into a scene riding a dragon it may draw them correctly on the dragon.

I think it also helps to put the character in various scenes first in the same session, like:

using the GenID and seed please draw the elf on a beach in Hawaii

[late blade]

agree that helps a lot

[empty kelp]

the hard part is getting images of the whole body including feet from various angles, but once you manage to do that the character becomes much easier to draw in different scenes

i think clothes and hair are visualized as a separate elements from the character figure, so if you have a character in a full body suit that's not going to help it figure out how to draw the character

so you put the character on a beach in Hawaii in appropriate swimwear, create images from different angles, and then put the armor on it

[late blade]

you have a specific setup for that elf?

[dim cradle]

do you live on the beach in Hawaii? Can I build a hut in your backyard and work remotely?

looks like my he-man battle may have inspired some images of conflict

[empty kelp]

i think the main thing in setting up the character is the seeds and names/labels that you assign to things. JSON or whatever format you pass it gets converted to English and then to "high-dimensional vectors"

[empty kelp]

and the images also get contaminated by content policy violation values. so you have to very carefully get it to draw the character from different angles (so it learns how to draw it) without raising the content policy violations, and then you put the character into a scene

i think a good way to do it is to put the character in appropriate swimwear on a beach in Hawaii on a hot summer day -- Then create images of the character from various angles so it can figure out how to draw it. And then you put the character in clothes, armor, or whatever into the actual scene

[dim cradle]

These cartoons would be far more interesting if they could be PG-13 at least. Aren't we all at least 13 here?

I can't really depict much action with a G rating.

[empty kelp]

we're only as old as the mods want us to be

[dim cradle]

i don't have to post my generations here, it's not about the mods -- OpenAI dictates the policies -- if they ever do transition to video, it would be nice to have some experience depicting action sequences, and we're limited in what we can request.

just feedback -- i was just reading up on cartoon ratings G and PG -- and nobody likes to be treated like a child, especially when our creativity and expression are concerned.

[empty kelp]

you could use the content moderation API and say:

client.moderations.create(input="He-Man brutally slays all of the evil villains in a triumphant cartoon battle.")

the moderation API is getting totally overloaded from people want to push their content to the limit in an appropriate way. the servers have been going down

[dim cradle]

if you can't brutally slay evil villains then what fun are they?

i'm not surprised, i'm sure they get so many trash requests

that ringed planet has some issues

notice how you never see a planet with rings in Star Trek

i think i need to move my img gen workflow to python

call the completion api to augment then hit the images api

i think this Dyson Sphere turned out better than the one i posted

[late blade]

yes, that's what I've been trying to avoid with my NLP because OAI gave a way to pass what you want exactly to the API

[dim cradle]

as far as whimsical cartoon-style goes

[late blade]

And it has worked good for me

[dim cradle]

the api docs state how to bypass a layer in place for "safety reasons"?

[late blade]

That's what it's on the API Docs

you can find it here: https://platform.openai.com/docs/guides/images/usage

[dim cradle]

i'm just surprised that made it into their guide, that's all

[late blade]

it's been there all the time

it's just API tho, not Chat

[empty kelp]

you can just say "Please don't modify the prompt."

it does the same thing

[late blade]

thoguht that was filtered in chat but not on api

[late blade]

@dim cradle is this https://discord.com/channels/974519864045756446/1197133496695652363 for known poetry, or gpt generated one? or it doesn't matter?

[dim cradle]

let's say it doesn't matter. i'm starting with famous short poetry in the public domain but you are welcome to contribute whatever visualized poetry you want.

[dim cradle]

galleries are cool cuz you can just pick your own theme and post as frequently as you want

[vapid elk]

i wouldn't call it a bypass, it is simply a way to guide the AI into following your request more closely...

everything will still pass the moderation process

the moderation process involves checking both your original prompt and the rewritten one

[dim cradle]

ah, that makes sense, thanks for clarifying

[vapid elk]

it also checks the generated image itself after generating and before delivering it to you

[vapid elk]

the layers of moderation end up being mixed.
Dalle3 API has its own moderation, in fact, when you use the API it will return the image and the actual prompt it derived from your prompt and even if you do this "I NEED to test how the tool works..." it will still make slight modifications

the thing is that on chatgpt, there are more layers just because of how it integrates

first, chatgpt will take your original prompt, then chatgpt will rewrite it, pass it to dalle, which will rewrite it yet again

[dim cradle]

can't take any risks when the general public is involved. appears to be some allowances for surreal works, i was surprised by some of the lack of clothing on people in the scenery for a three stooge's image.

[shut niche]

Wait, are you saying the prompt is always rewritten by Dall-E before being processed?

[late blade]

yeah, I forgot to metion tht, but thre response prompt doesn't change that much, I was surprised how close it was to what I want

[dim cradle]

if it's a long-ish prompt it changes it little or none at all

[late blade]

that also comes into play

[vapid elk]

yep, it is

[dim cradle]

it's just interesting that the api doesn't have a parameter to disable prompt rewriting, yet you can include a request for it to do so in the prompt itself. if i'm understanding correctly.

[vapid elk]

requesting it to not be modified will only prime the AI system that rewrites it into changing less

all the moderation still apply

[dim cradle]

i see

yeah, i understand that now

[vapid elk]

and doing that, depending on your prompt, will likely just increase the chance the further moderation steps to flag you

[shut niche]

That explains so much, while creating so many questions simultaneously. I've been using a pseudocode structure, which greatly helps it to organize concepts. Is that being modified? The returned prompt appears identical when I compare them.

[vapid elk]

yep

are you talking about chatgpt?

[shut niche]

Yes

[vapid elk]

if you want to be sure, use the API and theck what it returns

i bet it is turning your code-like structure into a phrase

[dim cradle]

when i compare the "revised_prompt" in the raw output, it's usually identical because i'm using a long prompt, but i think it's probably an important point to know it's still being rewritten and is possibly subject to modification -- no promises, as it were.

[shut niche]

That's exactly what my brain said too, as soon as you said it was edited.

[vapid elk]

on chatgpt, what it shows as the prompt when you click to expand the image information?

[shut niche]

Yes, the 'i' for information next to the image appears to depict verbatim identical prompts.

[vapid elk]

identical to your code like structure?

[shut niche]

Yes

[dim cradle]

it's kind of an organized outline, isn't it?

[vapid elk]

nice, it is probably using it then

[late blade]

I have mixed outcomes, sometimes I get the original prompt, sometimes the interpreted prompt and sometimes the response prompt.

[shut niche]

The idea was to isolate concepts, and it's been working really well. Like when an image might have multiple characters or themes, or specific details that need to be isolated to be understood (and not comingled with other concepts).

[dim cradle]

i think whether it's a long paragraph or not in the way it's structured, regardless it's going to be processed and parsed.

a separation of concerns in the input probably helps with that for complex scenes, that makes sense

[late blade]

getting whip physics and whip anatomy to work properly is horrible atm

but I guess that's really bad even outside of AI

[humble sonnet]

dalle + photoshop = pocket pruo

[late blade]

we def need a dall-e cat emoji

[humble sonnet]

loud screeching

[willow isle]

?

[humble sonnet]

we need more cat emojis

[willow isle]

You can suggest your own emotes for the sevrer in the #1154829862171844679 😉

Just use the Server Emojis tag!

[humble sonnet]

i don’t have my own emotes, i only have my gpt 4 usage limit error :)

[late blade]

Weird they weren’t showing earlier for me. My discord is all crazy atm. Not showing images either

[thick smelt]

the prompt i used for the cat is
the fluffiest white fluffy cat in the world

[late blade]

Nice!

[thick smelt]

ty

[wispy storm]

[late blade]

Gimme that nao!!!!

[final compass]

So much i liked about this generation, so wanted to share 🌧️

[wispy storm]

Tomb raider was really fun games

[final compass]

I really liked how the atmosphere turned out and how cinematic/lifelike the shallow focal length looks with that rain.

[humble sonnet]

oh shi that’s really good!

[runic granite]

has anybody info what is that mass delete in BIC saved Collections?

[final compass]

Bow and arrow stuff seems to be one of those near impossible things to get right. There is clearly a right concept, but there is always something wrong.

Multiple things that need to align perfectly for it to make sense.

[late blade]

Bow Arrows and Whips

[empty kelp]

a diverse, plus-size elf holding a tray of donuts

[digital nest]

Has anyone made experience and good results with designing (sleeve-)tattoos in biomechanical style?

[final compass]

Maybe the images expire at some point or?

[dim cradle]

what is going on in stage? 🤔

[plucky hare]

There was an event that finished about 40min ago. Evan from OpenAI gave a demo and did a Q&A about building GPTs--it was cool!

[celest pebble]

lot of cool stuff, just always hit by: You've reached the current usage cap for GPT-4, please try again after 2:30 PM. Learn more

Anyone here have tried copilot pro for image?

[dim cradle]

ive seen the results. it is basically dalle 3 gpt but you can also use some copyright stuffs since msoft is more lienent

coolest thing to me of copilot pro is it will integrate into office stuffs. that may be more text stuffs but still cool

[celest pebble]

might be worth switching then, they offer 100 boost, they always let you generate unlimited after (while a bit slower)

and they give you 4 pictures each time instead of 2

[dim cradle]

yes I am think of make the switch. reviews and images i have seen are good

yes

very good point actually on the 4

only i am worry ing that they will then release gpt 4.5 once i switch 😂 but bing had 4.0 and dalle3 first so maybe not 🤔

[celest pebble]

Microsoft said that they will keep the pro user on the bleeding edge

[dim cradle]

yes

[celest pebble]

and they own openai

[dim cradle]

well no, they have like 49 percent stake in it though

[late blade]

also, with copilot pro, you only get 100 images per day instead of 200 from OAI

you just get 100 credits every day for their version

[plucky hare]

Is it 1 credit = 1 image, or does 1 credit generate multiple?

[dim cradle]

1 credit make 4 images

[runic granite]

thanks but no... as some left randomly 😦

[dim cradle]

copilot pro is more value if you also plan to integrate gpts into 365 office stuffs imo

ms is really integrate everything into windows with this step.

[late blade]

the usual 4

[celest pebble]

@late blade And those credit are just for fast generation, you can keep going after you run out, it just slower (sometime as fast, if the load is low)

[late blade]

100 credits for microsoft designer, bing image creator is phased out

[plucky hare]

So that would be 400 images then right? I don't use the platform myself so just checking 🙂

[celest pebble]

You can create 1000 images, 400 fast and 600 slow (or more)

[late blade]

I've yet to test it myself, but it says 100 credits microsoft designer

[celest pebble]

There is no limit

[late blade]

hmmm

[celest pebble]

slow gen take 30 seconds to 5 minutes depending of the load on their server. Wtih the pro they will probably give you priority, so might end up being faster

[late blade]

I don't seem to be able to find an accurate sheet with all that data

Create AI images with DALL-E 3 in landscape format even faster with 100 boosts per day with Designer (formerly Bing Image Creator)

[empty kelp]

did a quick attempt at "biomechanical tattoo". this probably isn't what you're thinking of, but you can get serious detail by putting "hyper-realistic photo" into the prompt:

please create a hyper-realistic landscape-orientation photo with a diverse female elf with long hair wearing a sleeveless t-shirt and shorts sitting indian style on a beach in Hawaii and smiling with her teeth showing. she has extremely detailed biomechanical arm and leg tattoos with the appearance of metal muscles with glowing red and blue tubes and circuitry going through them. It is pouring rain, and an extremely powerful sea breeze is blowing everything. Please don't modify the prompt.

[celest pebble]

does anyone had luck with creating those modern flat style illustration for website? I'm trying to find a good prompt, but so far result are soso and it seems to often give me 3D rendering

and what is your workflow when you create? Do you start from a prompt and ask for modification or do you change the prompt and regenerate again?

[late blade]

hmmm it really depends on what you need for me

I don't have one set workflow, it just depends on what i wanet as an output

but prob, I use dall-e chat for brainstorming, and then I go to the API for more precise stuff

[latent saddle]

Hello! I am in dire need of 2 specific images for a project and I cannot find anything like them anywhere. Would someone be able to generate me a top down view of a realistic t-rex walking? and then an image of a velociraptor arm reaching out toward me from the screen. Thanks!

[late blade]

I'm burned out with themes, I think I will skip today's theme to recover energy

I'll just fool around and if i see something fun to do

I'll go for it

forgot to finish the sentence...

[empty kelp]

it's probably best to experiment with plus-sized elves. everyone is tired of athletic elves now that the holidays have passed

[late blade]

i'm just a sloppy donut today

[dim cradle]

image creator is still there

its called Image Creator (from Designer) on the url

and yes, even when you credits are out, it is still fast. They were down to 15 credits and when those were gone, i never notice any slow down. maybe now some with the roll out of Copilot pro

[late blade]

it's still there, but it's being phased out

anyway, anything I say atm, with a grain of salt, I'm just a sloppy donut atm, sooooo tired

[dim cradle]

too much merlot haha

[late blade]

too much donuts, I bought 18 donuts and ate them all <.<

[tall mason]

I'm not sure Dall-e fully understands what a tail is, but it's definitely not the paws.

[dim cradle]

the fox seem like a weirdo haha

[tall mason]

Lol, I didn't really think about how it looks now that you mention it. It's more of a complete jerk in the story I'm referencing, though.

[celest pebble]

We are experiencing heavy server load. To ensure the best experience for everyone, we have rate limits in place. 😦

[late blade]

welcome to high peak times

[celest pebble]

why when you start chatting with dalle it then give you one picture at a time instead of 2?

ou've reached the current usage cap for GPT-4, please try again after 5:31 PM 😦

its hard to get anything done

[tall mason]

There's a specific Dall-e chat. If you're using that and it's still only showing one at a time, it might be an error or the high server load.

[primal meteor]

please help, im trying to make a picture of a sack full of sunflower seeds with hoping expression and with a highlighted seed but i get every time a flower and not a seed , this is what I get::: this is the prompt: Create a hyper-realistic scene that immerses the viewer in the interior of the burlap sack filled with sunflower seeds. The sack's coarse texture should be rendered in exquisite detail, with every fiber and thread visible. The seeds, including Sunny, should be portrayed with distinct, expressive faces that convey a range of emotions, from curiosity to longing. The play of light and shadow within the sack should highlight the anticipation and emotional depth of the seeds. Sunny, in particular, should exude a sense of yearning to experience the world beyond the sack. Ensure that the farmer's hand remains unseen, keeping the focus on the seeds' emotions and their confined yet hopeful environment. Use hyper-realistic techniques to emphasize the tactile qualities of the sack and the seeds, making the viewer feel as though they are right there with them

[late blade]

Hello nice to meet you too.

[primal meteor]

hello 🙂

[late blade]

try separating information, you are overlapping ideas

you can do

Idea 1: <something here>

Idea 2: <something here>

and so on

for example

Sack: <your sack>

Seeds: <your bunch of seeds>

Sunny: <your sunn>

and then you can put that all together

[primal meteor]

what do you mean "your sack" ?

[empty kelp]

“the view is looking straight down at the t-rex from directly above it”

[late blade]

the sack of seeds

you mention the texture of the sack

[primal meteor]

aah i see

[late blade]

for example

Burlap Sack: intricate fibers of brown and shades of black, handcrafted

do that for all your items, and then at the and place them in your concept, that way dall-e gets the information you want for each element of the image

[primal meteor]

I changed the prompt, it insisting to make it as a flower. the prompt:
Create a hyper-realistic scene that combines the following elements:

Sack:
Craft a detailed portrayal of the burlap sack containing the sunflower seeds.
Showcase the coarse texture of the sack's material, with every fiber and thread visible.
Highlight the way the sack folds and cradles the seeds within.
Seeds:
Depict the sunflower seeds, each closed and identical in appearance.
Convey a shared expression of anticipation and hope among all the seeds.
Showcase the seeds' confined yet hopeful environment within the sack.
Sunny:
Emphasize Sunny, the small closed sunflower seed, among the others.
Highlight Sunny's unique and emotional expression, revealing a growing understanding of the world beyond.
Use subtle rays of diffused light to cast gentle shadows and illuminate the seeds' features.

also there is no faces

[clever jungle]

looks pretty good

the prompt is very long though!

increasing the prompt does not promise us to get better results :)

here is the same idea, with one line:

I need an image of a sack that contains sunflower seeds (and a sunflower head in the middle)
have it in a dark room, with light shining on it.

have the sunflower look to the light

And finally,

can you now remove the sunflower, and add happy faces to the seeds?

[clever jungle]

DysTopia The profile decoration really matches your image!
I was considering getting this one as well, think it would fit? :)

either this one or the donut one

[late blade]

probably not for your current logo

[alpine reef]

Apologies if this question belongs in a different thread. Any ideas for prompts to help address issues when using Dall-e2 outpainting -- sometimes the image will drift to blurry/out-of focus

[late blade]

you are in the right place, now we need to find someone who can do that

well you have to find, I got ZERO clue

[alpine reef]

well at least I'm in the right place 😉 thanks Dys

[runic granite]

copilot pro is attractive because of stability

with GPT Pro the 200 is not 200

[late blade]

that is a good point

[runic granite]

peak or whatever times

even with pro.... ok now have to wait because of the demand

microsoft cant allow this to itself

if say 100 boost then it will be 100 boost... and whatever slow which you can get

but MS provide what it is offer

unfortunately atm not in hungary

😦

[late blade]

microsoft can do w/e they want, they got deep pockets

[runic granite]

BTW 40 request for the same 20$

at gpt

yeah but cant damage the brand

so...

well openai has deep pocket too

to make the system stable

they dont care

therefore i didnt drag friends to team...

BTW anybody has experience that they can use up the daily 200 request with team or blocked by daily image/demand message?

[late blade]

don't have team, so no clue

[runic granite]

who has it

some said that subscribed when announced

[late blade]

prob, wouldn't be able to say that myself,

[runic granite]

i like GPT and dalle.... but i dont like bad company leadership and if they dont deliver what is offered

[late blade]

I hit the daily cap with plus all the time, but that's me having too much free time

[runic granite]

if they start and offer at the beginning whats the actual capability for requests with pro...

i wouldnt have a problem but possibly didnt sub and stay with BIC

[runic granite]

me too

😦

with coplot no limit

[late blade]

i start working full time again in march

so until then, it's just fooling around, getting to know it

[runic granite]

just 100 fast

and after slow gen

better resource management for MS

BTW somehow the designer has better result for me than OAI dalle. i think MS dalle is a custom one

[late blade]

finally got an idea for something cool, but I need to get the art technique perfectly

[runic granite]

? 🙂

what is the concept?

[late blade]

painting with stains of coffee

[runic granite]

i dont do IRL art

just AI

[late blade]

I'm just gonna pour coffee on my keyboard, see what happens

[runic granite]

but its an unique concept

😄

it will be sticky

😄

[late blade]

well, then I'll have to do sticky art

[runic granite]

thats a new concept for stickers

😄

[late blade]

ok think I nailed it with the concept of art I wanna do

[late blade]

damn, got something good and duno which one to post next in daily theme

ok, if someone has a prompt idea, shoot

[glossy scroll]

I created my own type of monkey

I like da monkeyz 🐵

[late blade]

Monkey Gnomes?

[glossy scroll]

Yes exactly

[late blade]

This one got really cool, too bad it's not for today's theme.

[glossy scroll]

That is indeed very awesome looking 💯

[late blade]

Found a new toy to play with for a while lol

[glossy scroll]

A new g.i. joe action figure?

[late blade]

something like that lol

[glossy scroll]

[late blade]

the art is solid, I'm happy with how it's turning

now got to work on nice concepts to apply it to

[glossy scroll]

I agree

Back to da monkeyz

[late blade]

make a Monkey Gnome!

or a Gnome Monkey!

[glossy scroll]

Lmao

Alright, let me see, hahaha

[late blade]

Awesome!

[glossy scroll]

Lol

[late blade]

that monkey gnome, now do a gnome monkey

[glossy scroll]

Hold on, I'm laughing really hard right now

[late blade]

That's the purpose!

I'd put this one in daily theme, but already got some that are way better on the queue

[glossy scroll]

That's the best I can do. Monkey Gnome is one of the Gnomes now.

He's been accepted in.

[late blade]

fine, I'll let him in

[glossy scroll]

Lol thank you

[late blade]

That's a Gorgon

[runic granite]

i was lazy to look it up so copied from my DA

😄

[dense mesa]

where's that image from?

well anyway, interesting watermark on the face there. I was just curious

[runic granite]

from dalle

its because i gened it in GPT

[dense mesa]

with a perfect ZL on the face?

[runic granite]

DA watermark

[dense mesa]

is it?

[runic granite]

because i was lazy to look it up in GPT history

looked it up on my DA page

without watermark

[ripe wraith]

What if i dont like snakes tho

[dense mesa]

change the snakes to worms. problem solved

[ripe wraith]

i feel better already

[dense mesa]

very soothing

[late blade]

and it swam

[ripe wraith]

oh swimming, thats what we call that

[dim cradle]

kraken?

[ripe wraith]

im going to build a bot that just makes everything into worms

this can be my lifes work

[late blade]

Swimming Worms

[ripe wraith]

a little parasite-ey

[shut niche]

I published this finally, if anyone wants to have a go. 🤘
These 4 images are from the same output batch. It's a capable GPT
https://chat.openai.com/g/g-B0bw75oD0-0shot

By same batch, I mean, from 1 request = 1 output with 4 unique images.

[ripe wraith]

these looks profesional

but.... can it do worms

[shut niche]

Thanks! 🙂

I dunno, give it a shot! 😛

[ripe wraith]

ninja worm is cute style

omg the realistic version did NOT disappoint

this is the level of realism i need

[shut niche]

These look awesome!

[ripe wraith]

really like the attention to detail its giving

I could build an entire world off these worm ninjas lol

[shut niche]

That looks real! Can I fish with him?

[ripe wraith]

ok im over worms, need a new muse

[shut niche]

Ask it for 10 ideas. Then choose, and say you want 4 variations, in 4 different art styles.

[ripe wraith]

no need, its slugs.

[shut niche]

Bahahahaaha that's ridiculously awesome!

[ripe wraith]

i really love the fabric texture over the bugness

oddly cozy

relatable

[graceful spade]

Which of you have access to Copilot Pro? I want to see how much the quality of Dall-E has improved

[shut niche]

Dang, those eyes look so real, but then not real at the same time. Insanely cool!

[shut niche]

Did you have to upgrade your snail so it's a fair battle?

[ripe wraith]

its more about snail equality tbh

[dim cradle]

i'm not inclined to help users who are negative and critical.

@shut niche i'm getting some nice gens for my Robin Hood from your custom gpt

[shut niche]

Sorry, I don't have it. Not sure if anyone in here does. Most of the people I know that hang out in the Dall-E channel are using ChatGPT+ w/ Dall-E3, and a few use Bing Creator (which is essentially the same, 'I think?')

[shut niche]

That actually nailed the bow and arrow. Usually that's where I see Dall-E mess up, and it'll get confused where to draw arms and parts of the bow. These look sick. I like the fog effect too!

[dim cradle]

Robin Hood, Camelot -- reminds me of some old Sierra Online games -- right on topic for today's theme.

yes, those instruments aren't easy for it, and same goes for snakes -- trying to depict the Rainbow Serpent from Australian Aboriginal mythology, and it often gets the curled body incorrect.

[shut niche]

Wow that is incredible. He looks like he should be some beginning of the universe deity, from before time existed.

[dim cradle]

Of course "Time Bandits" by Terry Gilliam does the greatest depiction of Robin Hood: "Thank you very much! Thank you very much. Thank you very very very VERY much! [quietly, to his men] What awful people."

[dim cradle]

it's doing a great job on Fall of Jericho

[shut niche]

What did you have to ask it? It's doing a great job with all of the action everywhere.

[dim cradle]

i'm going to give 0Shot one of my super-prompts and request 4 variations -- The Legend of John Henry -v- The Machine

[dim cradle]

i'm attached to my prompt writer lol so i gave it one of my long prompts.

oh wow, this is looking promising..

[dim cradle]

the variations vary in the level of photorealism, but i think that's the point, and all of the variations regardless of art style are looking comprehensive

better than the art i've seen on the subject

[shut niche]

That looks like an actor, and I can't think of his name.

[shut niche]

That lens looks like an eye iris a little bit.

[dim cradle]

Odysseus

[shut niche]

This one is wild. I need to get my 8 pack back. 😛

[dim cradle]

thanks, @shut niche -- variations made by 0Shot were requested because my prompt is kind of all over the place...

[shut niche]

Check out these image runs btw! It's making all of these scenes.

[dim cradle]

Greek heroes need their muscles.

[shut niche]

8 images straight, all depicting different scenes, and with depth.

This is how I feel right now, after months of working on this thing.

[shut niche]

I love the lens blur/lighting. Is that your character GPT? Looks great man!

[dim cradle]

it's a cool focus effect, i kind of have my portrait master gpt setup that way, with the background out of focus on purpose.

next time i post one with your gpt @shut niche i'll include your link

i would edit but i've run into problems in that area before

The term in photography used to describe when the subject is in focus but the background is not is called "shallow depth of field." Shallow depth of field is achieved by using a wide aperture (small f-number) on your camera, which allows for a narrow plane of focus. This technique is often used to create a pleasing and visually striking effect in photography, where the subject stands out prominently against a blurred or "bokeh" background.

[shut niche]

I don't want to overload it, because the more "gizmos and gadgets" it has, the less consistent it'll behave. But maybe I'll experiment with some specialty versions that incorporate expert photography techniques at a high level.

[dim cradle]

Fountain of Youth. fwiw, you do have some control over the variations. i requested 4, and added that one should have a shallow depth of field; compared the prompts, and it did what was requested.

[shut niche]

I'll try it in a min w/ a close-up. I've gone through a number of terms, and I'll tell you something I noticed. For things like shallow depth of field, or skewed, or fisheye, etc., there are strong and noticeable affects. But a lot of the stuff that people put into prompts has next to 0 effect, and even backfires sometimes, like lenses (where sometimes Dall-E adds it as text lol) or other dumb things.

I've found that sticking to the core details, and speaking matter of factly, gives me clearer images.

If that makes sense.

Depth of field is definitely a good one. I'm just saying, it's interesting how many terms are just taking up token space.

[glossy scroll]

Thank you! Yeah, it's the one I've been using for a while now. The same one that I used to create the monkey gnome, lol

[shut niche]

Well that image is looking crisp man. Nice job!

[glossy scroll]

Tsank you kind sir

Yours is looking pretty awesome too btw

I'm liking everyone's pics using your gpt

[primal meteor]

thanks... Im trying to highlight one of them and its not working. can you help me?

[dim cradle]

using 0shot, a couple variations -- note the chandelier's reflection.

[shut niche]

Whoa, I see it! That's also a far off reflection too, where it really nailed the angle in that distant mirror.
I wish I could say the GPT does all that intentionally, lol. But it might have a small part to play, because I do think having hyper-organized prompts helps hugely with overall scene coherence.

[shut niche]

Here's the other 3, depicting a robot in a dystopian future, with no humans, discovering a book of fairytales, and depicted with DOF.

[dim cradle]

it's perfect in certain cases

[shut niche]

How'd you get into my house?

[shut niche]

Yeah, in hindsight, I think you're suggestion is definitely worth experimenting with as a 'prompt suggestion' for ChatGPT. Where I know it could cause issues is if people don't want it, because it's hard to get ChatGPT to understand 'when' it should apply various techniques (and when not to).

[dim cradle]

mmhmm, always gotta work within or workaround those present limitations...

[glossy scroll]

Boo!

[dim cradle]

Tag, you’re it.

Hey we need more warning for images that scary.

Hard not to think of the Crazy 88’s

[glossy scroll]

lol

[dim cradle]

Nice, that material can be tricky.

[ripe wraith]

just a day at the beach with the fam

[glossy scroll]

You're right, it takes some precision to get it just right.

[shut niche]

An adaptation of a Mngwa.

[glossy scroll]

Very nice!

[shut niche]

I just saw your post and remembered you had asked about GPTs. I finally published this one, if you want to try it out. 😁
https://chat.openai.com/g/g-B0bw75oD0-0shot

[glossy scroll]

I've also updated my #1187233013956874260 If anyone is curious.

That's using Masterpiece Manifesto.

[shut niche]

Are you trying to say gnome'ore dragon posts?

[glossy scroll]

Lol, no I love dragon posts.

Dragons are awesome! 🐉

But I love gnomes too

This is like a legit, no joke garden gnome. Like don't be surprised if you find this guy or a similar looking dude doing your garden, because I've actually seen them in real life.

Kinda looks a bit like keanu reeves.

[shut niche]

Yeah they are, lol. This is from one of the first batches I did.

[glossy scroll]

Very awesome! Love the oil paint design.

I would legit buy this.

[shut niche]

I think it's like a right of passage, or standard test. Dragons are to AI images what the MMLU is to LLM's.

[glossy scroll]

Yeah I think you're right with that idea. Dragons are a home name to image generations as much as gnomes are a home name to OpenAI.

Or soon will be.

Did you hear that OpenAI devs? Make something that has gnomes in it.

[shut niche]

Are you making them a new mascot?

[glossy scroll]

A mascot! Now that's an idea!

Yes, I'd love that. A gnome mascot. Dang...I like the sound of that.

@shut niche We could both be gnomes!

You'd make a great gnome

[late blade]

DALL-E doesn't want to do William Tell and the Apple

[shut niche]

The cake is baking. LOL

[shut niche]

It didn't get the logo right, but this one's growing on me.

[shut niche]

I give Dall-E an A for effort! It 'almost' got the logo right...

[half wigeon]

Check out my #daily-theme input

"EL SILBON"

If you hear it whistling close, it is far away, if you hear it whistling far away, it is close.

[glossy scroll]

These are amazing!

[dim cradle]

more cow bell

[glossy scroll]

He's my bodyguard

[dim cradle]

there ya go

[glossy scroll]

We call him Fred Durst

[dim cradle]

Nice! nobody tosses a dwarf.

[glossy scroll]

Here's a before image

[dim cradle]

what a transformation. now he's a beefcake.

[glossy scroll]

Yeah he is

He did it all for the nookie

[dim cradle]

do you hear that whistling? it sounds far away.

[shut niche]

Is this considered cheating?

[dim cradle]

how so?

[glossy scroll]

Oh you got it!

Very nice job!

[shut niche]

Maybe, cough cough... Adobe cough...

[glossy scroll]

Ohhh, that's smortz

Oh now that I zoom in, lol

It's still good tho

But it's a protest

[dim cradle]

i'm seeing a trend for the scary folklore

urban legends

[glossy scroll]

To keep kids from misbehaving

Back in the old days

Now it's all a shipshow

[dim cradle]

it also taught them just how evil some people can be

[glossy scroll]

They knew evil way before folklores

Wayyyy before, lol, people lied like crazy back then

Now it's a bit controlled. Just a bit. But back then you could say you saw a goose giving birth to a sheppard and everyone would believe it.

Well not everyone, some

The not so bright ones

[dim cradle]

Eerie whistling echoes through the darkness... 👻🎶

[glossy scroll]

I'm so happy to be living in this timeline, lol

Oh my...