🏞｜general-with-images

[west tree]

Mage.Space

[wispy nest]

That an extension or completely different?

[west tree]

It's a web host. It's well worth the price of admission. I think they're doing $35 a month for unlimited gens. I use it to save money on colabs, and it's extrememly convenient.

[wispy nest]

Is there no local version?

[west tree]

Mobile-based web-app is great, too.

No, but how else are you going to run Stable well on a phone?

Ooh, and their animated GIF models are absolutely gorgeous, sometimes better than runway's footage.

I have fallen in love with Karras Sigmas.

[oak osprey]

it's a myth, you're wrong, and if you insist that you're right, provide evidence

SD 2.x text encoder had issues with anatomy, it's not at all related to the fact that it can't do a damn areola

because, guess what. people fixed its anatomy issues without teaching it tits.

[stoic pivot]

Folk often mix up anatomy with proportion or body shape.

[oak osprey]

like, SD 2.1 has everything about booba except the areolas

it knows anatomy, it's missing a texture

maybe Bad_Anatomy wanted weiners in the mix or something, in which case not even 1.5 will do what you want without fine-tuning, they purposely keep penises out of the training data because of what happens with them

[west tree]

I'm dealing with it this morning. Mage started filtering most of their models. The NSFW ones just have better anatomy.

Maybe it's exclusive to 1.5

[oak osprey]

maybe it's just BS lmao

sucks that they put more care into fine-tuning models that will be filtered

[weak sage]

this is what happens when you set denoising too high when you use upscale resampling

[solid gate]

the new trailer of Messi's project uses AI

[oak osprey]

wow that got old way faster than i expected lmao

[steady skiff]

They need better music behind that, It felt like gears ⚙️ grinding

[zealous socket]

Roop is soooo good. All the time I spent trying to train models is irrelevant now

[steel apex]

I think Stable is improving every day

[weak sage]

a picture is worth a thousand words

[west tree]

Did I encroach on your freed'ms or something?

[west tree]

Here's the Mother of God. You should stop generating Waifus and pray to her, instead.

Stop generating, start venerating.

[oak osprey]

oh lawd

she even has fingies

[west tree]

The fingies came out nice.

[west tree]

Working on Tsar Nicholas II cutting down Bolsheviks on the alien planet they came from.

[weak sage]

how is that mary magdalene she doesnt even look jewish

[west tree]

How is that Ben-Hur he doesn't even look Jewish

[oak osprey]

the original image is base -> refiner,
and then 4 variants i made using the image -> base model (img2img) -> refiner (img2img)

https://giphy.com/gifs/yo-dawg-xzibit-ToMjGpM4h8DAsgL0fEA

[west tree]

That is some very uncanny valley. Prompt?

[oak osprey]

i was already made fun of for that prompt earlier

not saying it again

[west tree]

Nice. I show you mine, but you won't show me yours?

[west tree]

OH MY GOD I'M A GENIUS!!!!!!

@oak osprey Chad Kaczynski

Hey, baby, you want it with the hood on, or off? Why don't you let me obliterate the security guard that works for you, while you watch?

[oak osprey]

it's been long enough now. john cusack as a twisted jester

[west tree]

I don't get it.

https://tenor.com/view/icp-insane-clown-posse-insane-clown-posse-gif-20063408

It's lame and short, kinda like "ted kaczynski as a bodybuilder", but what else is there to make fun of?

[stoic pivot]

Was looking at the images scrolling past in the Bot channel, suddenly "Best quality" finally works 🙂

[weak sage]

this is where controlnet is nice

[west tree]

"Come here and let me teach you about the power process"
-Chad Kaczynski

[wispy nest]

glorifying a murderer is kinda fucked up

[oak osprey]

people tell me "it looks nothing like The Joker", maybe it's a meme, but the prompt is for a jester

[west tree]

Ahhh

[west tree]

Your name is Chaz.

[oak osprey]

let's respect each other

🤝

[wispy nest]

yep first name is Charles

[oak osprey]

first name King

[wispy nest]

but what does that have to do with posting pictures of the unibomber

he's not ironically cool, he was a murderer

[tropic shell]

🔥

[wispy nest]

https://tenor.com/view/banned-thor-mjolnir-gif-14742819

[oak osprey]

Sunny your reactions always make it worth it

omg nooooo discord restarted and everything is tiny now

it's probably coming for you all next

look

wait maybe that looks normal to you all

😩

[west tree]

Wait, me?

Did I post a video?

[signal lark]

yeah, mine says 4 dedicated

[oak osprey]

no, sorry lol there was a botty bot

thar

deleted

[west tree]

Oh, jeez, I didn't think I'd done anything terrible here.

[oak osprey]

you didn't ❤️ that i know of

[west tree]

Friejacked Nietzsche

Slavoj Zijacked

[oak osprey]

the pope's slime left his hat by accident

[signal lark]

@elder locust

[oak osprey]

them's rookie numbers

you gotta pump those numbers

[signal lark]

im a rookie. this is my 3rd day

[west tree]

Hilaire Ballcock

Am I the only philosophy nerd here?

Martin Heidegger, cause I couldn't come up with a pun.

Artur Schopenhauer

[wispy spindle]

Who ping me from my slumber

Raaaaaaaaaa

[west tree]

I did. Step on me.

I didn't

[wispy spindle]

Step on urself

My guess is @oak osprey

So take that

Ping tag

[west tree]

He bumped one corpo, not that I encourage or endorse doing things that are illegal.

Illegal things are bad. We should not do illegal things.

Helluva writer, though. Certainly persuaded me to a manner of thinking.

[wispy nest]

thanks for the warning

[west tree]

Aw. I was hoping it'd give you the sploojing thumb.

[oak osprey]

i missed your eucalyptus scent

[west tree]

Just smells like leather, rubber, and... I'm not sure from down here.

Can't get a good Hegel. I dare any of you except chaz to make a good bodybuilder Georg Hegel.

[obsidian island]

Potter, did you put your name in the goblet of fire!? Dumbledore asked calmly

[west tree]

Or Heidegger. Won't do heidegger's mustache right, and (hitler mustache) does nothing.

[split rover]

[oak osprey]

@grand walrus your inbox today?

[graceful nacelle]

I love transparency!

[steel apex]

Want

Also want

[steel apex]

Inside your brain are two signs

[mental mist]

you guys heard of warpfusion?

[west tree]

Ooooh.

[stoic pivot]

Guess the web site without cheating

[sterile temple]

facehug

[past pelican]

FacingHug

[stoic pivot]

So close I think you know it 🙂

[hoary summit]

alfan

[oak osprey]

GPT4 is so powerful

they turned out good

digital collage is a weird keyword

wonder which keyword made it like this

[reef vortex]

looks good

[cyan snow]

Also, @oak osprey, does SDXL satisfy expectations? Just curious

I'm not in my country so I can't try it out yet

[shut sinew]

People keep forgetting that 0.9 isnt the full release and shouldnt be rated as such imo

[oak osprey]

for me, yes

[cyan snow]

Do you think that people that prefer 1.5 than 2.1 will like SDXL better?

[oak osprey]

no those people are lost souls

they hate change

stubborn as fuck

not allowed to talk about it here

[sterile kiln]

oopsie

[oak osprey]

np

[cyan snow]

That's most of the community, if it's actually better than 1.5 on every term, then there is no reason to be stubborn

[oak osprey]

just keep it on the d-l

go try base 1.5 and compare That

because this is a base model

then think about how much 1.5 finetunes changed it

now extrapolate.

[cyan snow]

Um, I'm not in my country for the next two weeks xD

I can only assume

[oak osprey]

well, i forgot lol

sdxl makes incredible stuff out of the box, but people feel threatened by it because its so good that it obsoletes their old trained models

there, that ought to explain

personally i might not train sdxl. it doesnt really need it. its better than most fine tunes

[cyan snow]

Then, why not finetune SDXL? It should be way easier because it's supposed to be released with specialized fine-tuning tools

[oak osprey]

it doesnt need it

sdxl is so good that its more likely you damage it than improve it

[cyan snow]

I will be the judge of that =]

[oak osprey]

but you asked me lmao

[cyan snow]

Yeah, I just wanted to know what your opinion is, because I trust that you already experimented with SDXL

[oak osprey]

i have, a lot

[cyan snow]

Elder Magic?

And mini Shrek syndrome?

[oak osprey]

didnt try

shrek is superior to earlier models

look you can feel wall-e's despair

[cyan snow]

Out of the box 1.5 was somewhat more coherent than 2.1, and SDXL without the refiner on clipdrop beats all fine-tunes, MJ and Kandinsky. So I predict that SDXL will go very viral, this means there will be as much clout as 1.5 if not even more

[oak osprey]

no i think youre misremembering

[sterile kiln]

Wow, 1st shot with SDXL, without refiner 😮

[oak osprey]

1.5 is worse out of the box eslecially for photos

[oak osprey]

use it without refiner for a while to get the hang of it

try cfg 2

[sterile kiln]

2?? Oo

[cyan snow]

WHAT?

That's insane

[sterile kiln]

2 cfg scale?

[cyan snow]

Does that actually work?

[oak osprey]

yup

makes the output look more weird and realistic

try some artistic prompts at cfg 2 and photo prompts

add soviet 1977 tv show to the prompt too

[cyan snow]

Holy sh!t, that's probably the lowest cfg I have seen a diffusion model get to

[oak osprey]

sd2-1 does it too,,,?

[cyan snow]

SD 2.1 almost always makes wacky results, so this is not a fair comparison

SDXL seems to be MILES ahead of 1.5 out of the box, and it's easier to finetune(just speculation) I think that 1.5 models will barely be made anymore after SDXL gets more viral

Also dead chat

I think it's because of all the chaos that exists currently

[oak osprey]

@cyan snow what are you talkin about? base models, or fine-tunes?

i spent 2 months fine-tuning SD 2.1 because it was superior as a base model

fine-tuned 1.5 tends to have more "aesthetic look" but at the cost of no longer following the prompt. and please don't tell me i'm simply wrong, because i did so much validation testing and analysis

2.x training had the same issue for me at first where i could train it and make it look amazing but then, prompts i hadn't trained on or concepts i hadn't had in my data, got lost

[west tree]

You're wrong. Provide Evidence. Blargleblarg.

[oak osprey]

well for one bit of evidence, CLIP-L has a 768 long embedding vs 1024 for OpenCLIP-H

that additional precision goes a long way

[west tree]

2.1 is wickedly imprecise. The simple NL prompting kills me. I can't get poses correct, or good aesthetics. Artist references go over its head. blah.

[oak osprey]

ah

theres a thread here for artists that exist in 2.1

a lot of them actually work better than in 1.5

=p

bob eggleton and darryn eggleton are both known but you have to put one of them in your neg prompt

or it combines

1.5 also cant do poses and needs hires fix to do basic high res correctly, thats why OpenPose exists

i think you are confusing precision with ease of prompting. you cant have a widely capable model that is also super easy to prompt. that is due to the very nature of the training data distribution in the weights

it has billions of parameters that you are tweaking via a 77 token long embedding

just imagine how many prompts you would need to generate to receive every possible result from the model

you couldnt do it in your lifetime

[west tree]

Yeah, but how well will it do Martin Heidegger in a limnal pool?

[weak sage]

a bunch of fake barbeque and sandwich accounts are passing as real on Twitter right now, nobody can tell theyre AI, no one even skeptical i could see

i mean, nothing against ai but theres something weird about salivating over a food that no one can ever taste

[split valve]

i read about a similar thing today about AI fake humans..

And i've got obvious bot accounts following me on twitter, usually pretty women, with no actual tweets of their own or actual interactions other than programed one line repsonses they say to every man they follow.

[west tree]

Swollen Kierkegaard will teach you all about fear and trembling.

[weak sage]

as long as a platform doesn’t implement a two tier caste system of ai content, pretending to not be using ai tools is kind of unethical maybe? idk

like when youre punished by the platform itself for admitting you used ai i understand lying

[livid ermine]

So here's the before:

And after:

[west tree]

Swollen Kierkegaard is coming for your heart or your body. Y'know. Either/Or.

[west tree]

Fyodr Dostoy-shred-sky

[weak sage]

you should have kept the cute hting on the bed

[livid ermine]

i didnt notice lol

got this glitch or whatever u call it..

[ocean steeple]

A batman scene I made with azovya rpg model

[weak sage]

looks like u got some aliasing from using 20 textual inversions or you used way too much add_detail in your prompt

[livid ermine]

aah, well either way i thought it looked cool 🤣

What about zeroscape-v2-xl? All I've managed to do is add some barely noticable car movement. I need the camera to move around and stuff, how is that done?

https://i.imgur.com/xtqwr5K.mp4

Oh and some eagles. But it's choppy, too fast, and only lasts like 2sec.

[weak sage]

ive never even attempted #🎥｜animation tbh

spends 30 minutes upscaling
gets NaN vae error
T,T

[west tree]

Saint Thomas Aquinas preparing for a just war.

[stoic pivot]

AI hater "Today all can do great images with help of AI and it will take the job from designers".
The AI: ...

[green socket]

https://i.redd.it/jtddel5i6j0a1.jpg

[oak osprey]

low CFG art in SDXL++

@round yarrow old soviet WALL-E

[weak sage]

bahaha

[sacred fulcrum]

Prompt: A house, breath of fire

[steel apex]

An icy apocalypse

Warning to Pagophobes, you might hate me

[livid ermine]

Just made something similar, but incase of apocalypse, hold up

[livid ermine]

The idea is permanent human settlements on the Antarctic, mostly underground. Each settlement should achieve city status within their countries.

[wispy ether]

Goofy ahh

[wispy ether]

Photoshop has a ai

Everything has a ai

[livid ermine]

What's goofy? 🤔

[stoic pivot]

Yes, but that is a tool graphical designers can use, just as a standalone AI-image generator it will still suffer from AI current flaws.

[boreal falcon]

Well we will have so much more from Adobe Firefly when its out of beta and gets all the tools that are still in exploration and ofc Sensei AI assisted tools than from Stable Diffusion or Midjourney like. But we are also niche/target group and not really all the prompters you will mostly find in the SD and MJ community

The later ones are far more likely to stick to Stable Diffusion and/or Midjourney because they are different from all the artists, designers, etc. and have a different workflow

[steel apex]

This would made a great Fallout style game tbh

[eager jasper]

Some images from my oil painting app: https://define.pics/app/impasticator

[karmic peak]

@dry crow

[wispy ether]

get a dedicated gpu

[karmic peak]

I need to sell my kidney

[wispy ether]

If u can buy a cpu u can also buy a gpu

[livid ermine]

True that. Those sketches never fulfilled my quality demands so I gotta redo them at some point.

Trying to redesign LA in a way that's realistic and achievable using new algorithmic styles of architecture.

Unfortunately it would involve tearing down what's already there, but I think it could be done.

Those are just four of some 20-30 renderings that I'll share in the final version.

They're Zeroscape v2 animations although that can barely be noticed. If there only was a way to move the camera around 🤔

Before you burst out in laughter calling me names, I do work with the Norwegian company Snøhetta that designed Times Square, Ground Zero and are now trying to revitalize Detroit and Paris.

[west tree]

Can't we just shuffle LA off into the pacific ocean?

I think that's the fix.

[surreal thistle]

Can you recognize everyone? Who do you like the most?

This took me almost 3 hours.

[wispy comet]

difficult to choose. i go red and yellow bird

[surreal thistle]

I like them both as well, though something's up with Chuck-chan's arms.

And Bomb-chan's leg, it's way thinner at the end than it should be.

And Corporal Pig-chan's helmet should cover her eyes, but AI also can't do that.

Other than that they're close to perfect.

[wispy comet]

what model do you use?

[surreal thistle]

The Kizuki model on Catbird.

[wispy comet]

you should do clash royale characters too

[surreal thistle]

Wasn't big into Clash Royale.

Maybe I'll do SpongeBob characters.

[wispy comet]

or plants vs zombies

[surreal thistle]

Yeah, that also.

[fervent wing]

dream hacker in the rain

[steel apex]

Question: How would I recreate The Poolrooms? I love that liminal concept it my favorite

[west tree]

Like these?

[steel apex]

Something like that yeah

[west tree]

There's a decent LORA. Tried doing it without. Didn't work.

[oak osprey]

today's bike ride lol

i climbed about 1200ft 😁

[surreal thistle]

The SpongeBob anime.

[karmic peak]

is this thermal cycling?

will it kill my gpus

[oak osprey]

you been here long enough to know better, kiddo

[sacred fulcrum]

sdxl

[oak osprey]

dirty rooms

[sacred fulcrum]

sdxl is good at making technical drawings

[valid pewter]

this was the other creators piece

not using img 2 img i, but just txt2img, i was able to make this

@green socket

[green socket]

It turned out awesome!

[valid pewter]

like this picture here

there's a 3rd hand coming out and resting on her waist

if i send this picture to inpainting

and i draw over that 3rd hand on the waist, what do i need to make the settings to actually like remove the hand?

ohhh its inpaint + sketch

what about fixing the eyes?

[daring urchin]

i have issues upscaling i cant use hires fix, ill be met with a error.

[high hearth]

use Topaz Photo AI

[green socket]

Or use an online one

https://replicate.com/jingyunliang/swinir

[valid pewter]

@green socket so negative prompts are things you dont want right? in the pictures above with the sorceress girl with blue eyes, im seeing in their negative prompt the first thing is "glowing blue eyes"

but they clearly wanted it and used it

[green socket]

Yeah negatives are stuff you don't want

There must be something in positives making the blue eyes

[valid pewter]

hmm i see

is there a way to remove all the white of this image and replace it with like described terrain?

im trying to img2img this and make the art more epic, while putting terrain behind em

[shy fern]

soviet sportscar

[surreal thistle]

What are you talking about and why do you keep deleting the pictures?

[wispy ether]

Emberassing

[surreal thistle]

Let me guess: they look too young?

That can't be fixed, I'm afraid, because no matter if I put "woman" or "girl", they still look the same.

[sterile temple]

you also need to avoid scantily clad in suggestive poses

[surreal thistle]

Ah, that's the problem?

I assume there's a Discord that allows more lewd pictures made with AI.

[sterile temple]

yeah, also clothing that is basically just a second layer of skin leaving nothing to the imagination, semi transparent, etc.

[sterile temple]

yea, I dunno personally what they are though

[violet finch]

uzi made on warmed.ai

[wispy spindle]

I'm deleting them because they are inappropriate
#✍🏼｜rules-and-tos

[surreal thistle]

Yeah, alright.

Not all of them are though.

[wispy spindle]

Since your reply is "not all of my imgs are inappropriate!", it seems you realize that some are.
Let's avoid the huge boobed child anime charas with little to no clothing, erect nipples etc in the future

[surreal thistle]

Yeah yeah, no worries, mate.

But it's still quite hard to generate a good adult woman with an anime model.

[split valve]

how can a drawing have an age? like its just subjective interpretation and dependent on the quality of the drawing. At the end of the day its still a drawing not a human.

What they are actually is representations, and the representation will have subjective interpretation.

[split valve]

Those are good enough to make a 3D model based on them and sell the model on CGtrader.

[unique patrol]

https://tenor.com/view/rick-morty-show-me-what-you-got-gif-5257429

Reminds me of

[wispy ether]

ifunny.co

[oak osprey]

does this drawing not look like an old woman?

[oak osprey]

@wispy spindle he's still trying to argue in favour of depictions of minors

[split valve]

well as a matter of fact im arguing about how can a drawing have an age and that a representation of a human in the form of a drawing is not an actual human and its open to interpretation.. you know full well that it cannot generate nude images anyway so i dont see what your point is... Im not one of the folks that use trained versions of SD to generate those nude images that go up on deviantart. i only create art wit it! not porn.

[split valve]

no not really, it looks more like a butthole.

[wispy spindle]

it's not about a drawing having an age-- it is about depiction of media users choose to share online. A naked adult woman, and a naked child-- both have differing context. If you want to have a keyboard battle about the whole "it's not human, so it doesn't count!", I'd recommend having it somewhere else, as that attitude and writing it off as a "person to person interpretation" is the whole entire problem.
Bottom line we do not host a server for NSFW imagery or sexual-esque content.
I've had to warn you multiple times for being inflammatory about the same exact topics. If you'd like to be productive to other topics that don't revolve around "censorship", please feel free to do so!

[split valve]

I din't mean to be inflammatory, im just outspoken and like to discuss or debate things. But i shall not do it again if you insist.

[wispy nest]

theres a skeevy image just a few up

[wispy spindle]

Emote the message you want to report with "⚠️".
Right click on the message-> Apps-> Report to Staff

[wispy nest]

i mean two primary school kids having what looks a passionate embrace didnt set off bells for you

i didnt see sonos images but at least it sounds they were adult in form

[wispy spindle]

??? I was simply asking you to report it, as I do not see everything always

[wispy spindle]

they were not and this is an odd comparison

[wispy nest]

they both should be yeeted

personally i think all that anime stuff is sketchy as hell

[wispy spindle]

Instead of being critical of what I moderate , I encourage you to try and use the methods I listed in the future, as I am not always glued to this specific channel

[wispy nest]

my bad. i assumed you were scrolling through and knocked out sonos images so i didnt get why you passed over what looked like some other questionable content

[oak osprey]

fruit has anime pfp 😁

i would have reported it but i report damn everything on here and wasn't sure if at this point i just hate anime or if that's really a problematic image

[wispy ether]

funny

that looks so real

[stoic pivot]

Prompt was "Spooon" and I have no idea.

[sterile temple]

i found you can get more consistently good images putting much more weight in the base model and then just finishing up with a couple of steps from the refiner. but it's using different samplers and different numbers of total step counts

there's a large no mans land in between. you either put lots of steps in the refiner to finish a very noisy image from base, or do about 92% with base and then a teeny bit with the refiner to sharpen up the image

[oak osprey]

but i do 50%

[sterile temple]

the image you get from base is really noisy

the refiner has to do lots of work to finish the image

which is why it adds lots of detail

[oak osprey]

here's 65% refiner, 50% refiner, 30% refiner

i'm using some code i've written for Diffusers today that returns noised latents

so for example i have 20 steps i'm doing and the base does 10 and the refiner does 10

the refiner adds zero noise, to ensure it doesn't overwrite any details

it behaves pretty differently with the traditional img2img behaviour

[sterile temple]

yeah I played around heaps with doing it that way, but then tried to get a ratio that made the base to do most of the work in denoising, and found it has to go to about 92% with base, then hand over to refiner to finish

[oak osprey]

that produces total noise for me but i might have an error in my code

guess i did, cuz it works now

[sterile temple]

this is where i'm at, getting pretty good images with good details

and this also does pretty well, with fewer steps

[oak osprey]

lmao at 2 steps of refiner on 18 of base

[sterile temple]

see the total number of steps is different, and different samplers too

[oak osprey]

30 steps here with 28 on base

DDIM / normal

[sterile temple]

what;s the prompt?

[oak osprey]

a stunning portrait of john cusack as a jester at the twisted carnival

.5 is so good

so that's 10 base, 10 refiner

[sterile temple]

*awkward silence while we all wait for my 3060 to run*.... ahem..

[oak osprey]

tickle that pickle

[sterile temple]

i have a 4090 sitting in the kitchen, just got to put it all together

[oak osprey]

yeah and i have a girlfriend but she's canadian so you wouldn't know her

she goes to a different discord server

i'm getting a lot of these good ones

[oak osprey]

A mesmerizing 3D render that showcases a groundbreaking universal quantum experiment. The vibrant colors and intricate details bring the experiment to life in a truly captivating way

[oak osprey]

wow, the AI even got all the text right

well

it says GALAX

[sterile temple]

this new model is amazing

[oak osprey]

LMAO

the hell is that

fake

[sterile temple]

the cheapest one they had of course!

[oak osprey]

https://giphy.com/gifs/editingandlayout-galaxy-quest-i-knew-it-ToMjGpIqvqxK6e3hMsg

^ Galaxy Quest

[sterile temple]

lol check out his hat

[oak osprey]

keep the wizard hand strong

[sterile temple]

good movie

looks like something out of robin hood men in tights

[oak osprey]

lmfao

[sacred fulcrum]

Prompt: "Nick Veasey, [gun name]"

[frozen socket]

Dangerous question to ask I know but...

A little like going to an opticians which do you prefer please?

Image on the left or image on the right

[sterile temple]

image on left, but looks like it has a little bit of noise left in the image

[novel saffron]

My vacation from work officially starts today and I should be sleeping. Its 1:00AM and I'm playing with AI

[novel saffron]

Without makeup and with makeup. Both look okay.

My infinity cat prompt

a portrait of [a cat|infinity] in a scenic environment by [bob ross| andy warhol | sandra chevrier], hyperdetailed, trending on artstation

Now I need to see how well SD XL will make black and white clipart. My 3D printer draws svg images

[severe swallow]

Doomguy andvetures in wondeland

[twin dust]

Refiner really does a good job (diffusers):

[oak osprey]

@twin dust try my pull request for diffusers issue 4003

https://github.com/huggingface/diffusers/issues/4003

[twin dust]

Any parameters that need to be changed? At first glance, it is not a noticeable difference.

Dare I say, it kind of loses some interesting details, I'll do more tests with fixed seeds, but the water sparkles are gone.

[oak osprey]

yep, there's so much that can be changed. the guidance_scale for either pipeline. the num_inference_steps can be increased, but needs to be the same ending/starting timestep for both

guidance 12.0
guidance_rescale 0.0
steps 30

SDXL Refiner: On
num_inference_steps=30, first_inference_step=15 (old method: strength=.5)
steps 30
guidance 7.5
aesthetic_score 10.0
negative_aesthetic_score 2.8

[west tree]

Finally getting around to trying openpose. Cannot for the life of me figure it out. Why is this:
`

Outputting this:

[oak osprey]

@west tree that thing seems like an elaborate prank

[steel yarrow]

this new sdxl thingy is real good

[twin dust]

Ooh, yes, indeed, with finetuning, you can achieve the result you are looking for:

I have used the refiner pipeline extra settings only so far, it seems quite nice for finetuning.

[wispy nest]

ford transit custom 8+1

[oak osprey]

i love these!

[west tree]

Christ before the court of Antipas

[west tree]

I've got three versions of Christ before Herod. Which is best?

[mystic frigate]

o got this image from bugged inpaint, what can i do with it? My prompt only ruining it, controlnet reference doesnt rly help, I wont to change hair collor and more stuff

[crimson goblet]

does this hand look believable? I think I have an extra knuckle

[sacred fulcrum]

OpenTTD

[next surge]

left looks more realistic , how did u make it ?

[frozen socket]

image on the right is default SDXL then I ran that image through an img2img step using relaticvison3 as the input model

[next surge]

im kinda new so i understand right that u have those models on your own pc , so far ive jsut seen 3rd party solutions like dream studio...i will google that img2img part but can u tell me what resolution that is and how you can achieve pictures witih resolutions of 5k times 5k pixels or more ^^ ?

[twin dust]

sd-xl-base (DEIS) and refined with partial-diffusion (comparison)

[minor basalt]

Here's some dumb bullshit

[grim patio]

Might want to make sure that ControlNet is actually enabled and the preprocessor is set to None

[west tree]

Saint Veronika

[fleet geode]

@wispy nest I take it back. sai didn't just make the most of it, they did damn well, considering they are part of the corporate world

[wispy nest]

i mean i guess the company did in his best intrest. But it wouldnt have happened without all the backlash the company got.

[fleet geode]

currently working at a big corporation - and previously did IT supporting work for many medium sized companies
I'd give them an A+, if I consider how most other companies would have acted in this situation

[wispy nest]

IT is really hard to get into

I have a 2 year degree in IT and still cant get into anything because i live in the wrong area

its like i wasted 2 years of my life

[fleet geode]

A. - be a social chameleon - then you'll be valued for your ability to deflect most bad preconceptions that people have about IT workers
B. - be significantly better at IT than 90% of your peers - you're just useful
C. - dont target big companies, try to get a job at a small company consisting of 1~3 people, that do support work for other companies (this is area dependent though -so you might be out of luck here)
D. - expand into web area, then do support work for websites

although my experience is limited to european countries

[wispy nest]

Im american

[fleet geode]

well. in that case you still have A. available 🥲

have you considered living in europe? 🤣

[wispy nest]

I already gave up on an IT position i still use my IT knowledge to fix friends computers and stuff but other than that im a failure when it comes to IT

aka why my profile says My life is a syntax error (Mistake)

But ill stop talking about this now lol nobody needs to know my issues

[regal juniper]

Warning: You might get hungry, and your mouth may water. You been warned. 😂

[west tree]

I did it better.

[smoky oak]

Byrgyr

[smoky oak]

@dense tapir there is a guy semi close to me selling a 3090 for $700

I could likely get $700 for my 3080 and 3060ti FE

Then I could use some additional money I recently made to buy a new PSU

Just a though

[obsidian island]

L: single SDXL prompt R: submission for the dailies (dogs together in human clothes) after loads of img2img passes & inpainting :p

[pastel mango]

Photoshop generative fill (Firefly) while still low res and in beta because it produces errors and artifacts, it still doesn't look bad

for outpainting

the low res can be fixed in SD anyway

[wild sorrel]

why not use sd to begin with

[mystic sparrow]

been using SD since it was released

[pastel mango]

uh cause I was testing out firefly for outpainting? the original image was made in SD

the outpainting in photoshop is smoother, while the resolution is smaller and the server that renders it censors some words

so you can't generate nudity or gore

[boreal falcon]

If you generate larger than 1024x you will get bit more blurry results or stretched

When you outpaint for example do it piece by piece ideally

not more than 1024x1024

[daring urchin]

guys i keep getting these marks below but for some reason even with negative prompt it doesnt work

[stoic pivot]

Papa Smurf Longstocking.
Regrets, I've had a few.

[pastel mango]

yeah I know, I doubt it willbe like that in the final release though

[boreal falcon]

True, also all the other planned features

[pastel mango]

these guideline restrictions aren't gonna hold up long term though, since people work with all kinda of stuff in photoshop and some of it is gonna contain nudity or blood

[boreal falcon]

Photoshop itself doesnt restrict you, their generative AI does

At least their official models now

[pastel mango]

yeah which is a problem

[boreal falcon]

Later you can train it on your images

[pastel mango]

but I know there are alternatives

Imperators plugin for example, that his dev team at aistudio made

for photoshop

it's way more advanced than generative, it has no guidelines

[boreal falcon]

I will avoid MJ and SD by any means necessary for serious stuff

I mean i wont pay for MJ anyway

[pastel mango]

why are you avoiding SD?

what do you mean?

[boreal falcon]

Copyright risk and want to be clean

[pastel mango]

there's no copyright risk unless you use loras and models directly associated with another artist

or try to pretend to be them

[boreal falcon]

Plus no need for me to make it more complicated and time consuming as it already is for me with all the art softwares

[pastel mango]

the biggest risk is that you dont own the AI gen but that's only if all you did was write a prompt

if you posed 3d models, made a sketch over them ran that through control net etc

then you pretty much own it

[boreal falcon]

Its still not clean and companies and professionals avoid those

[pastel mango]

I intend to learn to draw properly though

and use AI as reference

[boreal falcon]

Plus the reputation

[pastel mango]

I dunno about that, a lot of companies right now in Japan use AI

often to generate the basis for backgrounds etc

then they extrapolate them into 3d and work on them

or take out the things they like about them

and redraw i

a bit more fidgeting with firefly

[boreal falcon]

Will get back to it later, gotta work 😄

[pastel mango]

okay, but still my point stands. As long as you don't mainly deliver just prompted AI art. Nobody is gonna care if it's part of your workflow as reference

artists use reference all the time

[strange jungle]

That's on fire. I prompted breathing fire.

[warped basin]

Yo I wanna make a hot instagram model with stable diffusion, anyone know how I can do this?

[oak osprey]

no one wants to help you romance scam lol

"STOP! In the name of love."

[iron field]

Any of the realistic models + ControlNet plugin

[west tree]

Saint Olga of Kiev

[regal juniper]

Futuristic 4Runner. Even after all those years, it retains the same style.

[sterile kiln]

SDXL is surprising on many points, it "understood" Toriyama's style of drawing, no model based on 1.5 does

[oak osprey]

@cold storm my SD 2.1 1024px 16:9 / 9:16 finetune

no lora, text inversion, refiner, or hires. fix

[boreal falcon]

@pastel mango i may and do use AI art as reference and idea generator and for fun i also use it for photobashing for example

I also swore tho anyway not to use Stable Diffusion due to the conflict with artists, at least not for actual and serious work

For inbetween me and friends i could, for what goes to the public i wont

[pastel mango]

AI art is very useful, it's never specific or easily changed enough to a customers (or your own) needs. However, using it for example to get a set of trees rendered from different angles of a fantasy sort or whatnot

that could be used, as an image you view only

you don't use it in your art

you just look at it

[boreal falcon]

The thing for me tho is that i as well can go to Artstation and Pinterest for ideas

Plus i dont want to depend on generative AI at all

[pastel mango]

but you do want to depend on other people having drawn what you need? O.o

I'd rather have as many options available to me as possible

[boreal falcon]

I dont

[pastel mango]

there's little point to shutting a door to an alternative

[boreal falcon]

I have a full time job already and may end up in IT sector sooner or later hopefully, i dont exclude the idea to become a full time worker in creative sector tho

[pastel mango]

as an artist you should never let them dictate your work anyway, but they can still be tools

[boreal falcon]

Nobody really dictates what i should do and what tools i should use, well people make propaganda ofc both artists and prompters

[pastel mango]

I mean I've seen artists that work for big companies, or some mangaka who I have no idea who they are. They might draw something and then instead of drawing the character in the same pose and draw 100+ outfits

you can generate concepts of those 100+ outfits

on your character

to get an idea of what looks good and what doesn't, in terms of like...some style area

like "ok this spectrum of this looks good, that looks bad"

then you throw em away and draw

[boreal falcon]

for example some artists are criticizing me for using generative AI at all, while prompters criticize me for being a "corporate shill" and ruthless towards the "small people"

[pastel mango]

that's something many concept artists seem to do with characters, like if someone wears a jacket. They might take the jacket they drew and run it with noise of 0.40 with dynamic prompts

just tell those artists they sound like the purists who were telling digital artists the very same thing

who were calling them out the very same way

[boreal falcon]

Im siding with artists here in general

[pastel mango]

"oh no a new tool that changes things"

it's not gonna replace artists, unless you're a shitty artist

[boreal falcon]

Many arguments are true tho and im siding with them

[pastel mango]

you gotta be a shitty artist to be replaced by an AI that can't really generate highly specific needs

you're just broad stroke siding?

no nuance?

[boreal falcon]

Im siding with artists in general

Not in every single case tho

[pastel mango]

I don't know what that means

the general case artists seem to have is "I'm going to be replaced, make it illegal"

[boreal falcon]

if you want it from other side of coin, im hostile towards a part of the AI art community

[pastel mango]

if you wanna go for the most generic statement

I'm really sick of AI art showing up in every search result when I try to find something and showing up in the place of the original

[boreal falcon]

Not really

Many are advocating for at least a "ethical AI"

Which means something like Adobe Firefly which also isnt enough for some artists tho

[pastel mango]

don't generate ai art of a famous anime/manga/comic and try to make it out to be yours or to be real

well some artists just need to stfu tbh, they're thinking with their feelings instead of logic. There are logical arguments, then there are emotional ones

I agree with plenty of arguments

[boreal falcon]

I dont side with every artists as said, i had some real conflicts with them

[pastel mango]

and I think plenty are also dumb

[boreal falcon]

Everyone looks for his interests, so do i

[pastel mango]

being afraid of every side of ai image generation and the possibilities of future digital art tools, is hindering

I mean look back at late 90s, early 2000s

when digital artists were being attacked, told they were ruining the art scene

cheating, using tools that gave them an unfair advantage and would put the other artists out of work

and so on

we're in a similar situation now

it's a bit more involved, yes

[boreal falcon]

In case of AI art, i dont care about other consumers of this tbh. Well i might for some but in general i dont care about their dreams, about what would happen if this tech gets "stomped" etc

Nor do those arguments about people with disabilities using this touch me that much if at all

Because those are often used as a shield

[pastel mango]

One of my arguments is that creating proof of concepts, demos and so on for a develoer or in my case a writer. To showcase, in order to attract attention and potential investors or donationsso I can hire real artists to work on it

ai art generation gives me that ability

I would never use it in a final product

[boreal falcon]

Oh i may use it as a final product, but unlikely for the public and if then retouched/edited manually and also labeled as such

For the serious stuff where i might and hopefully will end up having a second and passive income generative AI is actually irrelevant, pure optional tool compared to the stuff i pay a ton of money for

[pastel mango]

there's too much legal copyright convolution going on right now, but it's easy enough to make sure you own even AI art. Control net, sketching, photoshop layering and adjustment before more control net and so on

then what you don't own is pretty much the ink

and you can remove that too, then add shading with CSP

but yeah I still think that if you're an artist and you're afraid that AI image generation is going to replace you. Then you must be a pretty bad artist with a poor range and inability to adapt, adjust and so on to whoever pays you and instructs you

cause while ai image generation is quicker, that's just at creating a generic image. Often a pretty random generic image that under scrutiny makes little sense

gunblade!

[boreal falcon]

Oh well at the current state AI art doesnt come even close to some of the advantages real artists have, especially advanced ones

[pastel mango]

of course not, I mean some has the potential to create tools that will advance the workflow of 3d art

but not replace

advancing workflow, speeding it up or adding an additional layer/tool isn't the same as replacing

so yeh, it will become a tool

and it's up to artists to either show the way or be shown the way, or fall behind

[boreal falcon]

I just calculated my costs per year (in the future since i already paid some and some on discount)
2.600€/year for softwares alone if no discounts and deals.
Plugins and addons are extra and would cost me another 1.000 and more per year with drops afterwards because i get discounts on upgrades then

You see there maybe also why i dont want to pay for Midjourney anymore

And why i dont want to make it complicated with SD if id use it at all

I wouldnt want to play around with all the "more detailed" stuff of SD like installing, looking at noising stuff, A111 settings and what not

[pastel mango]

I get not paying for MJ

but SD is free

everything for SD is free (more or less everything, if it isn't then it has a free alternative)

good god I was trying to get a little violence into a prompt and I think I went way overboard

[stoic pivot]

AI is just a satire of America, enter "a president" and there is only one president type in the world.

[boreal falcon]

Yes, although i basically dont pay for Firefly either

[pastel mango]

I think that may just be your model

[pastel mango]

is that cause you get it from your company or cause you're jack sparrow?

it's also not as versatile at this point

[boreal falcon]

I pay for Adobe products myself

[pastel mango]

then you do pay for firefly

[boreal falcon]

Yes but i dont pay for Adobe because of Firefoy

Firefly

I paid before it was even released

[pastel mango]

okay

[boreal falcon]

Its one of the cost factors in the above mentioned cost

It came in handy and i waited for them to release their own generative AI stuff besides of those they already had

[stoic pivot]

My model, I used Bot-1 to make it just for fun, that was first result, but it differ a lot, some time they can look like this when enter "a president".

[pastel mango]

try to get "President Trump in a mecha suit fighting the Soviets in a grand space battle Gundam Style"

[stoic pivot]

Why? I just wanted a president and saw it tended to do what look like a parody of America. When I enter "a king" it do not do any parody of Sweden or England.

[boreal falcon]

He is so red

Did he yell or did he eat chili

[pastel mango]

I would totally make our old King fight in the berserker armor

[stoic pivot]

That is a good question. But to be honest I think it is for there are so many caricature of Trump with orange or red skin, and this show that most of AI is based on American internet search and sites within a time period of 2019 to 2023

[pastel mango]

well when he's outside he does go overboard

[stoic pivot]

My goal was not to talk about Trump but how AI work and is very limited doe on Internet culture, and that it do not depict America normally but like a satire due to politics there over a period been jus memes and downgrading caricatures.

[pastel mango]

well you cant really put Biden in a Mecha, he'd forget he's supposed to fight and keep pressing buttons while muttering "Damn computers...SUSIE! THE COFFE MAKER STOPPED WORKING!" then he's attacked and is just thrown and tussled and alarms go off "I'm sorry I swear I paid for this! I'm not stealing! Damn kids skating on the side-walks"

or he might start telling the onboard virtual intelligence a story about his legs

You see the road to not caring about politics is to just mock everything

[zealous socket]

Anyone else rocking the box fan setup for running SD in the summer months? 🥲

[past pelican]

On the plus side, you have this perfect little space heater for the winter months.

[weak sage]

AI doesn't exist, ANN exists, and its application to photo editing is simply one more tool in the digital artist's arsenal. Real artists are thrilled.

[crude sonnet]

Experimenting with stereoscopic images

[weak sage]

that means look at cross eyed and it turns 3d

but this image is too wide to do that on my phone

how u do it, layered canny preprocessing?

[novel cedar]

lmk what you think, any suggestions, I'm trying to get better. Prompt included

Neoclassicism, Dishonored Game Art Style, Goku, Vivid, 4k, outstanding quality, one person, full body shot

[dry cosmos]

Check the full image

[wet cypress]

how is this angle called?

[steel apex]

Log Cabin living room with big windows and autumn scenery

[turbid carbon]

when I add textual inversion like "easynegative" this happens

can anyone help me

[radiant matrix]

guys how the name of this ear plugs in english? im trying to generate a image but in my language its a name that doesnt fit for english translation

[keen sentinel]

Try earplug, earspool, or ear guages

[radiant matrix]

Ear stretcher means the same too?

[keen sentinel]

yes. But i wanted to give you the technical terms for them. Maybe they trained the model your working with on the proper names

[radiant matrix]

im gonna try, until now it only show little piercings or strange tool into the ears lol

[keen sentinel]

lol sounds about right bro. half of prompting is just figuring out what the hell the damn model understands. Use that deepbooru interogator. it hellps

[radiant matrix]

its my first time using this bot, what do you mean about the last phrase about interrogator?

#1101178553900478464 message that the closest I could get

[errant heath]

why the pics look perfect when not hires

and when hires they turn into mush

[woeful cedar]

can we run sdxl on automatic1111 later this month?

[oak osprey]

@cyan snow

[cyan snow]

Is this SDXL's competition?

[oak osprey]

ha, probably not, it's just its own thing

[cyan snow]

When I return home il make a comparison

[oak osprey]

nice

[cyan snow]

But heck, I bet Kandinsky has at least like a few backdoors

It's made by the russian government's banking team or smthn

[stoic pivot]

I use SDLX to test words, I still don't understand "naïf".

[weak sage]

the glowing fire effects might turn out well in sdxl, im not running that either tho

[weak sage]

ukrainian artist laptops gonna explode lol, unlikely but idk

[cyan snow]

It's ckpts..

AND made by the russian government's banking group or whatever they are called

[weak sage]

theoretically cant u just merge a ckpt with nothing and make a safetensors

[cyan snow]

Idk man, it's still super sketch

[weak sage]

thats a good name for an ai art program

super sketch 2.0

[cyan snow]

Plus, SDXL is way better , and no backdoors

[weak sage]

SuperSketch_2-0.ckpt

[cyan snow]

Totally_not_a_trojan2.2.CKPT

[weak sage]

🤣

[cyan snow]

Why would anyone use Kandinsky when SDXL exists?

That's kinda stupid

SDXL's architecture also allows it to fuse 2 images

[weak sage]

it's using layers, isn't it... ITS USING LAYERS AAAA

[cyan snow]

SDXL seems to be a masterpiece. I am certain that when 1.0 releases it will replace 1.5

Also safetensors are safer than ckpts and a little faster

[weak sage]

only if it runs on 4gb cards

[cyan snow]

Who tf uses a 4gb card in 2023

[weak sage]

T_T

[cyan snow]

You can't do anything with that vram. Even with all optimizations (as far as I know..)

[cyan snow]

Also, you can test it out. It is on HF

[weak sage]

1.5forever then, until i spend 1 grand on a computer upgrade

[cyan snow]

If you can use 1.5, you should be able to use SDXL

I have no idea how will you go about doing that though

[weak sage]

1.5 runs easy on 4gb

[cyan snow]

Then your good =]

[weak sage]

we will see

[cyan snow]

But you can try it right now, SAI released the same model that was leaked

It's an early version of SDXL

[weak sage]

thats one way to beta test lol

[cyan snow]

I heard from other people here it's great, way better than MJ and most 1.5 finetuned models and Kandinsky

But SDXL is 2 different models, the refiner and the base. It's a little tricky to set up

No one is allowed to guide how to do that on this server, so I won't start instructing how to do so.

[sterile temple]

You're allowed to help people with SDXL. They didn't want people to help people use the leaked version. But now the official repo is on auto approve, so anyone can apply and get access immediately

[cyan snow]

Oh, I already got the researcher access a while ago, I'm not at home for the last few weeks so I haven't tested it out yet.

[frosty wave]

stable doodle is very cool :) cute rabbit with carrots style: Fantasy art

it works well despite my doodle beeing terrible XD

[ripe bluff]

https://cdn.discordapp.com/attachments/325201038216790017/1129048924372606986/00011-54019016-anyloraCheckpoint_bakedvaeFtmseFp16NOT-824x1176.png

It work pretty well i swear

[scarlet finch]

Stable Doodle Prompt: a beautiful human

[pale crypt]

Images autogenerated to match the audio. Bit self-referential for sure 😂 Made in hypernatural.ai

[steel apex]

Pink Sunset Apocalypse with vibrant blue waterfall, Lisa Frank style

[oak osprey]

oh that's not true anymore (or never was) or

[stoic pivot]

Interesting. I entered some tags I often used when I and a friend uploaded photos to Flickr and I can almost identify where parts come, all images I posted there was under a CC0 license. The image SDXL make do match perfectly to how this area look like, that is super cool and I hope I contributed 😄

[wispy nest]

e make me this art -
**rainbow or ugel (or similiar or or candy (or really ang or very cl

[severe vapor]

It's too bad I didn't get this 2.1 model done sooner, it turned out pretty good.

[steep sleet]

I am trying to get Vladmandic AUTOMATIC1111 to work using SDXL 0.9 - anybody else? 🙂

[cyan snow]

Who ever the fuck wrote this article is a fucking idiot

[boreal falcon]

Lmao

[dim spruce]

written by midjourney no doubt lmao

[cyan snow]

They just can't accept defeat LOL

[vestal river]

怪物史瑞克风格 狗狗

[cold heron]

wherecan i use that

[urban ibex]

That table does seem... fairly... incorrect.

[cyan snow]

Are midjorney allowed to make their own finetuned SDXL model? Because damn, this seems to be the only way they will still stand a chance against stability

They reached a point where they start making incorrect tables for god's sake

[scenic umbra]

where is this table published?!

[cyan snow]

Somewhere at their twitter

[scenic umbra]

doesn't look official and it's plain wrong on some stuff, this must be a ramdom guy I can't find it in the official twitter account

[cyan snow]

Damn

Idk then

I have screenshot this on twitter

[dense haven]

Hi. How do I go about upscaling this image? Controlnet tile doesn't quite do the job for me. The blurred lines stays blur and I would like to sharpen them

[oak osprey]

yes

hence me not understanding anyone saying 'mj is so worried lololol'

they arent, they have no reason to be

[cyan snow]

Why not?

[oak osprey]

they can use sdxl?

[cyan snow]

I think there is no commercial license, correct me if I'm wrong

[oak osprey]

it has a license for 1.0 that allows it

0.9 is research release

[cyan snow]

Emad would never

There shouldn't be a commercial license for SDXL, shit's a masterpiece

Just it being open source is very generous

Let alone being commercially used without any license agreement? Heck no

[oak osprey]

part of their stated commitment to reducing co2 burden of the world

midjourney doesnt have to burn their gpu cluster to copy sdxl

i really wish you all would just move on from 'mj vs sai' because it isnt and never was like that

[cyan snow]

They wouldn't dare. They can't just take people's hard work and make tons of money over it

[oak osprey]

they worked together at first and in a sense, they still do because of the license etc

yes they can do that and why not

people arent forced to use midjourney

they do it because their service adds value over running it locally

[cyan snow]

Idk then, whatever MJ is using sure isn't SDXL, heck, not even close. If they have access to it why wouldn't they just deploy it?

[oak osprey]

people are free to download sdxl and use that. but they would rather it be easy, and pay midjourney

because sdxl 0.9 is the current release

man, discussing this is frustrating because you are so attached to your idea that midjourney is evil or undeserving, when they and stability jus endlessly recycle others work and claim innovation

its all theft of art that made these weights possible. sai isn't inherently better or more deserving because they more carefully curated their stolen data

[cyan snow]

You sound like anti AI tiktok girls right now. When a person learns how to make art, he is learning from other artists. Is this theft? I think not. So how is AI any different?

[oak osprey]

because it is different, lol. explain how they are the same

i train models, please tell me how anti AI i am

[cyan snow]

Let's say, you want to learn how to draw, right? Just the process of learning how to draw involves looking at other people's work.

[oak osprey]

stability used a dataset of 12 billion images

if you learnt to draw by looking at 12 billion images, i have questions

its not the same at all, stop pretending it is lmao

[cyan snow]

Humans learn art, AI learns art. It's just good at it.

[oak osprey]

I do not see how this has anything to do with midjourney being allowed to use sdxl

they are allowed to use sdxl, and emad licenses the weights that way

both companies steal endlessly and parrot innovation

i have no clue why it matters who wins

we already have the weights, stability can shut down today and liquidate and i wouldnt be phased

[cyan snow]

Well, good luck with that, I guess =\

[oak osprey]

good luck with what? i wish for nothing, i go with the flow, i am unaffected by midjourney or stability

you are the one with a stake in this it seems, i would be happy if sdxl became midjourney

their pictures would look so much better

i see mj images all around the web so the fact they could improve would make life better

[wispy ether]

:/

:/

[frigid torrent]

A lot of the MJ portrait/bust images I've seen look like default TF2 profile pictures

By which I mean, "smooth rendered person in a void with abstract effects around them, usually looking determined"

Not a complaint or a dig just an observation I've been dying to share lols

[wispy ether]

😭

https://tenor.com/view/thewolverine-logan-yashida-gif-18620261

[frigid torrent]

https://cdn.discordapp.com/attachments/531959975514341386/1129465872239562802/00419-1990056235.png
https://cdn.discordapp.com/attachments/531959975514341386/1129466052946960444/IMG_20230714_123323847.jpg

I printed out this small segment I generated using a hypernet trained off scans of my own drawings, and now I'm drawing around it with a pen

Meta?

[boreal falcon]

Its theft if you sell copyrighted images that dont belong to you and claim to be the owner of those

[cyan snow]

Which is exactly what Stability doesn't do. When generating images it's created for the first time on this planet, it can't be copyrighted.

Just like when an artist is inspired by an art style -_-

That is why it's called ARTIFICIAL intelligence

Humans also learn, so does AI.

If AI art is theft, then almost ALL art is theft

[boreal falcon]

Stability AI sells those products

Midjourney always sells those, there is no open source

[cyan snow]

Stable diffusion is free. I never spent a single penny for AI =\

[boreal falcon]

Also lets see how the laws might be affected

[boreal falcon]

But they have products that they offer for money (used for credits)

[cyan snow]

Eh, you can also commission artists

[boreal falcon]

Ofc you as the consumer can use other peoples artworks for some stuff

As long as you dont claim to be the owner of it and sell it

Or publish in some cases

Depends on copyright license of the work

[boreal falcon]

Which is not the same

When i pay an artist to make a work for me its a whole different case

[cyan snow]

Nope. When it comes to a trained AI model, it doesn't work that way. NONE of the dataset is present in the model, just the AI learning from it, that's what humans also do..

AI art is just a better painting brush. The prompt is the creativity

[cyan snow]

Also, the dataset SDXL is trained on has over ~20 billion images. If they were all in the model it wouldn't be 10GB size.

No theft or anything to copyright here.

SAI don't own the dataset, the dataset is what it's learning from

Also, training data in the future is likely to be aesthetically picked AI art.

There will be a model that knows what images to add to the massive dataset.

[boreal falcon]

Thats not what people found out

Especially some images that were identical to the original

Plus the watermark

Which is also how Getty got to sue Stability AI

Ofc it also depends on if you do img2img or for example prompt something specific which makes it more likely to get eventually in trouble compared to some generil prompt

[cyan snow]

That's bs, when you generate an image using AI it's the FIRST time it ever existed. If they claim that the AI just made an image that IS copyrightable, they are fucking idiots

[boreal falcon]

AI doesnt make copyrightable images, but it can generate copyrighted material

Thats a difference

[cyan snow]

Artists can do that as well -_-

[boreal falcon]

You cant copyright an AI image unless enough human input, but you can generate copyrighted material which is a trouble

[boreal falcon]

And? How does it make better for this

If they do it they can get in trouble as well

Noone says artists have to be excluded from copyright laws

be immune to those

Nvidia, Getty, Adobe have gone a "bullet dodge" way which comes in handy for me

[cyan snow]

No one asked them what they were generating. They have the right to use it for anything for themselves, but if they made an image of let's say - SpongeBob, of course paramount can sue you, but they CAN'T if you don't use it publicly.

Same goes with artists that don't use AI tools.

[boreal falcon]

Yes ofc

I agree there

[cyan snow]

The AI is just a better painting brush. It's up to you if you use it for dumb bullshit

[boreal falcon]

Its not a better painting brush

Not today

[cyan snow]

It is, the best digital art images are made using AI.

[boreal falcon]

Nah

[cyan snow]

Heard of SDXL? Shit's crazy

[boreal falcon]

Generative AI doesnt beat manual done work, especially not those by professionals

Speedwise, yes but speed alone doesnt matter

[cyan snow]

That's an opinion. Ask most people on this server and they'll probably agree with me.

[boreal falcon]

Most of them arent even artists

And have zero touch with it

I mean thats okay

[cyan snow]

But they ARE AI artists. Skill is useless when it comes to art. It's creativity. Even the best artist in the world can't make good art without any ideas.

[boreal falcon]

I was hyping this too much at some point too

Until the reality strike

[boreal falcon]

And this is an example

Skill matters very much

Unless you draw the strawman alone to put it that way

[cyan snow]

If you consider creativity as a skill.

If that's what you mean I agree.

[cyan snow]

That was before SDXL started to become a thing. If you compare aesthetic score of past AI models like MJ or SD 1.5, it doesn't even get CLOSE to SDXL.

[boreal falcon]

No, skills like coordination of hands and very important for advancement in art: Fundamentals

Lighting, perspective for example

Color

Anatomy

Those things require skill

[cyan snow]

Still. People that are creative can make equally great images using AI. Prompting is also a skill.

[boreal falcon]

I talk about manual done workflow vs generative AI workflow

[boreal falcon]

They are far more limited

Ideally at least you d like to have editing skills

To minimalise the limits a bit

[cyan snow]

Both are manual. Prompting is an insanely difficult skill to master.

[boreal falcon]

Or at best you have both artistic and editing skills

[boreal falcon]

Prompting is not even close to fundamentals of art for example, plus you still get randomised result even if you get closer to much closer when you prompt for specific stuff

Luckily you can later even the more difficult stuff

Well some people give up

[cyan snow]

Again, that's an option. Most people here don't agree.

[pastel mango]

Prompting is easy if the model is well made

[cyan snow]

Not for specific stuff

[boreal falcon]

Most people here arent even artists tho

[pastel mango]

To get even more specific results you need control net, actual sketches, adetailer, layering, knowing how to properly apply BREAK, setting up models and scenes with 3d objects

[boreal falcon]

Have no touch with the creative industry

[pastel mango]

sketching stuff you want to add to the renders

[cyan snow]

Yeah, most people here are AI artists.

[cyan snow]

Also that.

[boreal falcon]

"Artists" yes

Prompters i call them

[pastel mango]

I wouldn't call use artists. Synthographers yes

[boreal falcon]

We agree to disagree on the labeling which is ok

[pastel mango]

a play on photographer

Synthographer

I think that better captures what someone dealing with AI imaging (alone, without any artistry involved) does

[boreal falcon]

Synthography is an new term for that

Yes Cryonicus

[cyan snow]

Call it whatever you like. AI art is still art

Even being good at that

[pastel mango]

then so is my stick figure...Or the book I told an AI to write

[cyan snow]

Technically, yes, it is =\

[pastel mango]

I do a lot of AI imaging, but I wouldn't call myself an artist until the day I can use it to enhance my own art

[cyan snow]

You can with cnet

[pastel mango]

but then using that technicality, you've invalidated AI as "Art" to mean the barrier of entry was a stick figure

something I kinda disagree with 😛

though these days someone can smear shit on a canvas and sell it for tens of thousands of dollars

or put a banana on a white canvas

[cyan snow]

Still, amazing pieces of art are made using AI.

[pastel mango]

an eaten rotting banana

and call it art

art is made, but we're not artists cause we do very little to insert all those degrees of detail

[cyan snow]

Yeah, it's art, but not good art, at least by most people's opinion -_-

[boreal falcon]

@pastel mango the bothering part is that people here sometimes thing they are superior to artists and are experienced artists but they didnt even set their foot in this area

[pastel mango]

now if I drew something and it was pretty much 99% finished, but I was really struggling with that 1% area and I put it in the AI and got a few references. That could open my mind to understand what to do

[boreal falcon]

Like 0

[pastel mango]

I hate when people do that using AI language models that can help you "write a book" by going just by a few variables you inserted

As a writer it pisses me off

I'm not threatened, just irritated

[boreal falcon]

Like okay you need certain skills for SD when you go the more manual route of setting models up etc. But dont compare yourself to professional or advanced artists and their skills

@pastel mango agree

Btw what kind of novels do you write if i may ask?

I used to do it as well

[cyan snow]

You can make equally good images using AI.

[boreal falcon]

Mine was a psycho thriller

[boreal falcon]

Depends on what you mean by that

If judged strictly, then not exactly

[pastel mango]

My actual first big opportunity or product was delayed by my cancer that just won't stop coming back. I mean I started writing again now and just found out I have two metastatic tumors growing -.- So I might be in for hardcore chemo and radiation

[boreal falcon]

Putting an AI image aside of some pro artists for a strict judgement will make AI image lose in a heartbeat

[pastel mango]

So since I got delayed it moved on without me

so now I'm gonna do it myself

[boreal falcon]

Oh man, sorry to read that. I hope and wish you a good recovery

[cyan snow]

That is biased judgement.

[pastel mango]

Oh I don't think I will recover but thank you

[boreal falcon]

By veterans that are used to the best of the best artworks

By whom else

[pastel mango]

My proof of concept VN will be based on a short story taking place in the same universe as the larger one. Since I can't afford, nor find any good artists to work for free (Legally I can't guarantee them any deferred payment etc, so it's for free.) My proof of concept will use AI art

then if I can get some cash flowing in, I can finish my bigger project as a VN. When I can afford artists

I'll also use AI Voices for the proof of concept. The story though is 100% by me

every word spoken, every action taken, every word written

[boreal falcon]

I used to publish on Wattpad

But thats long time ago

The storyline i wrote was quite a metaphorical one for my life etc lol

Basically it was symbolical

[pastel mango]

I'm gonna put it on several platforms, either as a link or uploaded

and include the prologue in written format on the sites to draw people in

The main character isn't going to exactly be a hero though, but around the line between anti-hero and anti-villain