💬｜general-chat

[broken gull]

I feel ya, but in general it's just saver to call yourself a commissioner and not a artist, real artists freak out most of the time of you di.

[steep mortar]

is there any way to get the seeds to randomize much more affectively?

[broken gull]

Since no matter how hard you have worked on your prompt in essence you still only have commissioned the art and not drawn/painted it yourself.

[brazen fox]

I understand where you're coming from, but in that case you can only be an artist if you paint or use your hands- are photographers artists, or are cameras artists?

Or, for instance, I use a ton of embeddings I made to guide the AI somewhere- I havent made it pixel by pixel, but I did put hours of work into getting that result

[grim aspen]

he makes some solid points there:

https://www.youtube.com/watch?v=5Viy3Cu3DLk

[broken gull]

No when I mean is that you're using your hand to create something. The photo itself isn't the art imo, it's the content in the photo that is art. The scenery that you put together to take the pictures.

[grim aspen]

I guess using AI with copyrighted imgs in datasets will not be allowed to be used commercially soon.

[broken gull]

Even if it is nature pictures you still have chosen to take a picture in a certain pose/style.

[brazen fox]

So, when I train certain compositions and visual elements, I'm an artist in that line of reasoning?

[bleak wolf]

You cant copyright AI, more than 50% of the artists used the content they used to learn without paying

AI simply learns very tiny from each image

[brazen fox]

Not trying to be an ass here, if I come across that way I apologise

[broken gull]

No since the AI train you still just ask the AI to train those subjects.

[broken gull]

Dw here, you're just learning how I think about it and hopefully at the end of it you will think the same if not dw.

[brazen fox]

I guess we fundamentally disagree on what exactly art is and what an artwork is haha

[broken gull]

Check the img in #🏞｜general-with-images I send

[broken gull]

You can copyright your prompt tho

If it's unique enough

[brazen fox]

I can't say I fully understand what that is meant to say because to me it just reinforces my point-but oh well

[broken gull]

Wdym reinforces your point?

[bleak wolf]

No lik3

You cant copyright claim AI*

[broken gull]

Yes, you can do that with your prompt

[bleak wolf]

Like as in art

[brazen fox]

That as long as it's your idea or way of expressing yourself, it's art, pretty much- irrelevant of how it's made

[grim aspen]

Current AI is not self conscious thus it can't be seen as an individual it's a tool.

[broken gull]

A lot of people only see it as art if it takes effort and skill to make.

[bleak wolf]

It doesnt have to be one to learn

[grim aspen]

doesn't matter, if you train it on stuff without permission it's considered theft

[broken gull]

And that's mostly their argument that AI art isn't art since they think you don't need to put effort in it.

[brazen fox]

That's against every definition of art though- art is based around expression, not skill, although they can go hand in hand

[grim aspen]

no it's a tool you create art with

[bleak wolf]

You dont train it on stuff, you take inspiration from images

It uses randomised noise as a basis

[broken gull]

An AI does train on image at least text2image AIs etc

[brazen fox]

Anyway, I'm taking a nap, was good to have a chat about the philosophy side of it 🙂

[grim aspen]

it still can create near perfect images of copyrighted material

if a human does it it's plagiarism

[bleak wolf]

Ok, so can a random bit generator

[broken gull]

Text2image AIs needs a database in order to learn words and objects. How do you think the AI knows what a cat is?

[bleak wolf]

Eventually

[bleak wolf]

No like sorry i stated it the wrong way, mb

I meant as in

[broken gull]

If AI does its still plagiarism at least that if you claim it's yours.

[bleak wolf]

The affect of each image is so low, it cannot be considered theft when its effect 1/5.85 billion

[grim aspen]

exactly

[broken gull]

It is still considered unethical tho

[bleak wolf]

Okay, you know, lets make all past and present artists to get permission from ancient caveman because it originated from there

[broken gull]

Using other people's work in a database without permission is unethical no matter if you only use 1 image that is from someone else. Of course if there are more image "stolen" a lot more people will notice it.

[bleak wolf]

See the issue here

[broken gull]

No, it's only plagiarism if it is a exact copy. Ever heard of inspiration? Something can inspire you to make something completely new.

[bleak wolf]

Exactly

Thats why i dont make AI art on specific style

[broken gull]

I don't either, or try not to

[bleak wolf]

I mix and match styles and the AI adds that randomized noise for the base

[grim aspen]

The problem is it's possible to do so

[broken gull]

I never try to use artist names

Or art websites like artstation

[grim aspen]

you can even use all of an artists pictures to create a model that creates everything in that style

[bleak wolf]

Okay, you can use a kitchen knife to kill someone,

[fervent thunder]

Hello guys, i wanted to ask if once i trained my model, i can still training it, if the answer is yes, how can train it more?

[bleak wolf]

Lets ban all kitchen knife companies

[fervent thunder]

I'm using Stable Diffusion V 1.5, Automatic 1111.

[broken gull]

No sorry mine hasn't sharped yet from killing the other dude

[grim aspen]

I'm not saying that but there needs to be some form or protection for artists

[broken gull]

For sure

[bleak wolf]

One other thing i have is, us software developers gets no voice when our applications get pirated

Thats work too

[grim aspen]

I'd say if the model uses copy righted material it all the output should only be for private use for example

@bleak wolf pirating software is actually illegal

[bleak wolf]

And half of the artists i have met uses pirated software

[broken gull]

That wouldn't sole anything, since everyone can just use without the artists consent.

[bleak wolf]

It is but, the artists who complain about "stolen" content do it, which makes their argument invalid

[grim aspen]

sure they have to prove that they've created it with a model that doesn't include commercial stuff if they use if commercially for example

@bleak wolf so because some do that all of them do?

that makes no sense

[bleak wolf]

Same case goes with ai art

Just because some of the people do, do we all do it?

[grim aspen]

we do if the model includes copyrighted and private pictures

that is the same as piracy of software to create something

[bleak wolf]

I respect your opinion.

Sorry if i was rude in any way.

Just too sleepy.

[grim aspen]

np I'm also just discussing this

that is why I think there should be free models that use copyrighted stuff but the output is then just for private use

and commercial models where the artists opt in to create it

[bleak wolf]

Yeah that is a good idea

[grim aspen]

Imagine someone would deconstruct something like adobe products and then make AI recreate it almost exactly. Adobe would sue their asses.

[bleak wolf]

For me its the line between, does it affect it that much, that its considered theft? If it was 1000 images, yes i would consider it theft. But 5.85 billion? No. Thats merely inspiration, and since half of the process is random noise and trial/error.

In your adobe case, there are like 10 software like adobe out there.

In ai art case, there are 100 billion images

[grim aspen]

I for example only use open source software to create art

[brazen fox]

Photopea is a thing though, it's almost 1:1 photoshop interface

[bleak wolf]

Ofc

[bleak wolf]

That is my opinion at least

Again i never go do a specific style

Because that narrows the database down to said 1000 images

[grim aspen]

y makes sense

I just play around with SD

[bleak wolf]

👍

[grim aspen]

it's not really usable for me atm

[subtle breach]

is copying a style considered theft?

[grim aspen]

I'm also not a full blood artist

artwork is just one thing I do in gamedev

[bleak wolf]

Thank you for having a civil discussion after so many people screamed at me for using it

Not here, somewhere else

[grim aspen]

@subtle breach not if you are a human and don't copy it 1:1

[brazen fox]

Depending on who you ask

[subtle breach]

hmm, so some people do consider it theft?

[grim aspen]

y

but I think legally it's not

otherwise anime wouldn't exist

[fervent thunder]

a lot about copyright is how you explain it. How would you for example explain a style in such detail that people will understand it as a style, and not a specific art piece?

[grim aspen]

Also the diff. I see here is a tool copying art vs an individual doing so

[brazen fox]

It's not copying the art though, it's deconstructing the elements it's made out of

Machines/computers are just way more precise

[grim aspen]

and then replicates them

that is a form of copying

[subtle breach]

well, the AI does not "replicate" artworks

[brazen fox]

The style, yes, not the artworks themselves

[grim aspen]

you can replicate the artworks tho with ai

[brazen fox]

Yes, but you can replicate artworks with a paintbrush as well

Does that make using brushes morally wrong?

[grim aspen]

nope but a human does plagiarism when he does

[brazen fox]

So it would only be wrong to replicate existing artworks with ai, not using styles?

[grim aspen]

well the tool is created with the artwork from artists

it's not an individual

[brazen fox]

Because we established that it doesn't matter how you replicate an artwork, just that the act of replicating is wrong

It's just where your own moral and ethical compasses are at, I suppose

[grim aspen]

the problem with AI is that it only works by using artists dedication to a style

it would work without them too but then it would be limited

imagine all artists stopped creating as they can't survive on their art anymore

[brazen fox]

Isn't that the same as humans, though? Without being exposed to other artworks, we'd all be doing cave drawings- up until like the 17th century perspective wasn't even a thing

[grim aspen]

that would also dry out new things AI could learn

as I said AI atm is a tool not an individual

a tool that relies on the work of others

[fervent thunder]

It's hard to explain, and I'm probably the one shouldn't even try to but I'm going to try and give my two cents on it. When it comes to copyright, it's not about "if you can copy it." It's about "if you did copy it." As of now, I believe everything around AI art is in a sort of limbo of the law as there's not a precedence for anything like this anywhere. It will be up to legislation to fix it. But as copyright, from my knowledge, is not a criminal offence in the USA, then most lawmakers aren't jumping to fix/change this.

It will be up to people, such as the artists who disagree to hire lawyers and make a case for why it's this or that. In the end, the law is far from easy and simple to just say what will happen. There are bad laws as well as good laws, and there are weird laws, and laws not everyone agree with. But, in the end, the only way to change something. Is to contact the ones who can, and want to change it.

[grim aspen]

As I mentioned before we should make 2 models one that includes copyrighted stuff but the output isn't allowed to be used commercially only private and commercial models where the artists opt in to create them. I think this the best solution.

Also there is enough artwork out there where the artists are long gone

there are also artists like me that work open source

[subtle breach]

I believe this would be a good solution for artists who protest against this technology, but probably not the best for the industry as a whole, as commercially viable models would be more limited than private and personal ones

[grim aspen]

@subtle breach How is that a problem?

[subtle breach]

I don't think it's a problem, sorry if that's the impression I gave

[brazen fox]

Imo it's a very bad idea- at most, take out names of people, but not allowing images that are freely available on the internet to be used is weird

[subtle breach]

just trying to think about what this solution is, objectively

[grim aspen]

they are freely available but you can't use them commercially

[brazen fox]

If you set up a webcam and scrolled Google images very slowly, and let SD train on that- would that be morally better as the process is more similar to a human?

[grim aspen]

you as an individual can use them as inspiration

nope as it's a tool

[subtle breach]

I would also like to argue that it would still be hard to know which model was used for generation. Companies could definitely use "private" models for commercial use, and pass them as such. Nobody would likely be able to know

[brazen fox]

But you can use that commercially as well- if you're being hired as an artist for concept art, should you be allowed to use a moodboard filled with copyrighted art?

[subtle breach]

and giving them a serious advantage

[rose sedge]

how do I load the good vae in a 1.5 model ?

[brazen fox]

That is exactly what will happen lol

[subtle breach]

lol

I think so as well

[grim aspen]

sure corps will have an advantage but it's better then destroying a whole field

as I said before if aritists stop creating art there is nothing for the AI to learn from

[brazen fox]

But that won't happen anyway- it's just becoming more accessible, like photography

[grim aspen]

it will then stop improving

[brazen fox]

There is no reason for artists to stop creating art though, it won't cease to exist because a tool exists

Just like, again, painting didn't stop existing because photography came along

[grim aspen]

name one artist that can do art without needing to eat and live

[brazen fox]

What

So you think SD will replace every living artist? Bold

[subtle breach]

I think the "AI debate" is partly rooted in the fact that a lot of artists like their workflow more than the end result

[grim aspen]

if everyone can just use an artists style they worked on their whole life to create something that look exactly like what that artist creates. Why should they pay that artists?

[brazen fox]

As you said yourself, it's a tool, not a person

[brazen fox]

Sentiment, ai is far from perfect, there are loads of reasons

[grim aspen]

not for the industry there isn't

[fervent thunder]

and in the end, the art debate is not up to us, but for the creators of the ai's and the ones who dislike their art being trained. But again, I can only speak about my own "knowledge" and about myself :P

[brazen fox]

Morals shouldn't only be decided by people who are mad about something though

[subtle breach]

I think everyone should be part of the debate though, as anyone can create, and it will affect humanity's future

[grim aspen]

yep

if we look at it in the long term if artists can earn money pursuing art there will be more art and artstyles for AI to learn from

if we cut them off AI will also suffer

[umbral trellis]

Hello... is it possible to add the stable diffusion bot on my server?

[grim aspen]

I don't think that is a thing you'd probably have to use your own SD bot

[fervent thunder]

yes, don't get me wrong, it's not that people can talk about it. But as it stands now, there's nothing "inherently wrong" with ai art and how it operates, the task to get this changed is for the people who want this changed. I just hope it doesn't end with something sad :(

[brazen fox]

It's not a black and white issue though, photography didn't kill painting, cars didn't kill all horses, etcetera- even with every issue it might have, ai won't kill artists, it might change what it means to be an artist though

[umbral trellis]

Thank you

[grim aspen]

art is hard work and dedication if you can't dedicate yourself to it you can't create

[brazen fox]

That's just gatekeeping art lol

[fervent thunder]

I also believe that all these ai art program will increase the demand for artists, even more so now when it's just started! :D

[grim aspen]

why should only rich people be allowed to pursue art?

[umbral trellis]

I imagine... but as in Midjourney is possible to use the bot in my own server, I tought maybe Stable Diffusion work in same way

[grim aspen]

as that is gonna happen if we destroy the field

@umbral trellis you better ask someone more knowledgeable then

[brazen fox]

That's because you start with the assumption it will kill all artists' work

[umbral trellis]

No problem. Thank you so much

[grim aspen]

not all but it will destroy like 80%

[subtle breach]

not sure of the details, if you're looking to run it remotely because you don't have a good enough pc, there are google collabs that exist for that

[grim aspen]

that is enough to have an impact on the growth of AI

also artists in other branches like music get protected from AI generating music

[umbral trellis]

No, it's not the reason, it's because there is a lot of images and I prefer to work alone...

[grim aspen]

why should visual artists be the exception?

[fervent merlin]

Artists are gonna be artists because they are naturally drawn into arts. And I can't find much appreciation in AI art so far.

[subtle breach]

you can run stable diffusion locally very easily on your computer if you have a good enough gpu

[umbral trellis]

yes, but I don't have good enough GPU... anyway, thank you

[subtle breach]

oh, I see

[grim aspen]

SD can also run on cpu but it's slow

[brazen fox]

And that is where I think we disagree a lot-I don't think it will. People will make art, people will create, people will innovate, as humanity has always done. AI might inspire people to take up art as a hobby, or people might want 'real' art- sure, some people are going to lose work, but a lot might transition or gain work as well. I think it's wrong to just assume it's going to destroy everything and have that fear be the basis for all reasoning around it; we just don't know everything that will happen

[brazen fox]

Or the other way around, why should audio be the exception? Copyright lobby for music is just very strong, doesn't mean it's always a good thing

[fervent merlin]

I see SD as a tool. But i can't imagina it replacing my visit to the museum

[grim aspen]

sure but even the artists that created the art that is in the museum got paid for their works

[brazen fox]

So far, I've actually gained appreciation for artists and styles I had no knowledge of, even though I consider myself an art lover

[subtle breach]

is this the same thing with the internet that didn't replace travel for tourists?

[brazen fox]

Were they? Not a lot, most of the time-classical artists like van Gogh got paid peanuts for their work

[fervent merlin]

Maybe

[grim aspen]

Lets say I have money I want to make a game why should I hire expensive artists when I can just use generalists that use AI.

[brazen fox]

They did it because of the love of art, fame and appreciation came later

[fervent merlin]

Only 1/1000 artists can live from his work

[grim aspen]

with AI not even the 1 /1000 will

at least not the digital ones

[fervent merlin]

I don't people will find joy in AI generated art

think*

[subtle breach]

the number might shrink as the AI gets better yeah

or... hmm will it grow?

[brazen fox]

What if more people gain appreciation for art and want to commission work? It's very short sighted to just expect the art world to implode

[fervent merlin]

People are always more interested in other people. What are you looking at when you gaze at SD pictures? You don't really know

[grim aspen]

only a very slight proportion of humans commission art

most go for what is cheaper

esp. if the output is the same

[subtle breach]

well, it's possible to coordinate multiple generations to make a manga, for example

only the skills get automated, not the ideas, right?

[brazen fox]

Yes, so? If commissioned digital art becomes cheaper because the workflow is enhanced by AI?

[subtle breach]

hmm maybe the ideas as well

[fervent merlin]

@subtle breach yes, probably

[fervent thunder]

if people won't find ai art appealing, then why would artist as a profession disappear? Seems like it would just cause ai art to become another sub division of the art genre, or it does in my eyes anyway :P

[grim aspen]

@brazen fox that is exactly the problem when that happens there won't be much innovation anymore

[brazen fox]

I'm aware I might be bordering on naively optimistic but I feel like you're straight on the other side, needlessly pessimistic, lol

I feel like you underestimate artists there

People will want to innovate, they always do

[grim aspen]

how if they can't afford to?

[brazen fox]

I get bored doing the same thing every day, even if it pays well

[grim aspen]

how can an artist develop a style if they can't sell it

[brazen fox]

Where do you get that assumption? You're saying that as if that is an established fact

[grim aspen]

they are forced to use AI to compete

[brazen fox]

If its new and people like it, why won't it sell?

[subtle breach]

well, you don't need a lot of money to have hobbies, so I guess you'd do it as a hobby

[grim aspen]

because the instant the artist makes it public AI can copy it

[fervent merlin]

What is your concern BrahRah?

[brazen fox]

Or as hobby being an artist, then slowly transition to making money off of it

[fervent merlin]

That's why it will always be lower quality

[subtle breach]

yeah, in the scenario where you can make money out of it, I agree

if too many jobs get automated, then you can't work anymore though

and I don't know what will happen then

[brazen fox]

Then again, we won't need to work and it can be done as hobby anyways, if we transition to a UBI based world

Then again, even I think that's too naive

[grim aspen]

So you are saying artists need to create for 10 to 20 years without publishing any of their works?

[subtle breach]

I hope UBI gets implemented then xD

[grim aspen]

corps will try to stop it all costs

[fervent thunder]

I wonder how many artist there are who can live off their work today now that I think about it 🤔

[fervent merlin]

1/1000

[brazen fox]

A bunch, although I feel like it's mostly in the NSFW space lol

[fervent merlin]

rough estimate

[grim aspen]

oh wait that is the basic income

no corps don't care about that as long as they can sell their stuff

[brazen fox]

Capitalism will capitalism

[grim aspen]

either AI will kill capitalism or capitalism will kill AI

or monoplize it

[fervent merlin]

Capitalism isn't so bad. It's capitalism what is bad 😂

[brazen fox]

Let's just agree on that the future is uncertain and anything can happen from here on

[fervent thunder]

It feels weird honestly if there are so few artist who can live off their work, that we have to cater to them and make something which can make every artist workflow easier harder just so they can keep on living off their art. But that's just my two cents on it :P

[fervent merlin]

I don't understand modern arts anyway

[fervent thunder]

no one does ;P

[fervent merlin]

^^

[grim aspen]

well not anything some stuff is predictable and needs to be accounted for

[brazen fox]

Damn it, you make me disagree again

[grim aspen]

I'd hate a future where artists have to create a style in secret and feed their own private model with it

[fervent merlin]

Maybe if it gets more profound and artistic again, society may change to a better way of life

[grim aspen]

then create one project with it and that's it

[fervent thunder]

maybe they can develop their art style and then sell it instead? :3

[brazen fox]

I like to pretend I do but who knows

[grim aspen]

@fervent thunder only if there is some from of legal protection

otherwise once it's public AI can replicate it

[fervent merlin]

You are a cat in a uniform - i think all you do is a fact 😂

[fervent thunder]

what sort of legal protection? that they can sell an artstyle they created in private?

[brazen fox]

Seeing as how 'well' DMCA is functioning, I doubt the small artist you're talking about will profit- corporations though will for sure

[grim aspen]

who are they gonna sell it to?

[fervent thunder]

the ones who want it, like the latest ai art program who wants to stay competitive :3

[fervent merlin]

I think artists will create exhibitions for only human work and ban AI arts from them

[grim aspen]

yea

[bleak wolf]

Artists are here to stay for a lot for now

[grim aspen]

they already do

[bleak wolf]

But NVIDIA is doing some amazing stuff with AI accelleration

Those sweet Hopper GPU's

absolute monsters

[grim aspen]

but once that AI creates those images other AI can create it too

what is the point in buying it?

[bleak wolf]

Honestly in a few years, we will only need a few hundred, maybe even ten images to train AI, and copyright won't be an issue (since you can buy 10 images with ease)

(from scratch)

[fervent merlin]

Why would we want that like in 50 years from now on?

[fervent thunder]

point of buying it? well, to have it. can't have it if someone creates a artstyle in private and then wants to sell it to someone. :P

[brazen fox]

People bought NFTs for crying out loud

[grim aspen]

yes but you bought that style and integrated into you model

[fervent thunder]

I think they still do :P

[fervent merlin]

@brazen fox but that was only speculation, right?

[grim aspen]

users create pictures with it that are publicly available

thus other AI will be able to create with that artsyle too in a short amount of time

[brazen fox]

Partially, and money laundering

[grim aspen]

the one that buys it first is at a loss

so nobody will buy it

[fervent merlin]

@brazen fox yeah, nothing to do with arts

[brazen fox]

Just like classical art by the way, also hugely inflated prices for laundering money and tax evasion

[fervent thunder]

we'll see how it turns out, I don't nor can't really speculate about that

[fervent merlin]

Maybe, but i think it differs in concept at least. I mean there are real masterpieces and collections

[brazen fox]

Oh yeah for sure, I don't disagree-but even those are valued at absurd amounts that don't reflect the value it should

[fervent merlin]

That's a signature of art

[brazen fox]

🙂 for a reason though, again, with money laundering in mind

https://www.nytimes.com/2021/06/19/arts/design/money-laundering-art-market.html for those interested

[heady hazel]

Anyone have an idea on how to brighten up an Ai deforum video sequence that has a video input

[safe drift]

Where do I type in a prompt to generate an image?

[languid mango]

Hey, don't know if i should ask this in here but i hope you can help me.
So i have been out of the loop and i'm wondering if automatic's web ui is still the best way to generate images (for free)? And what would be the best way to update my current set up, how to make it compatible with the latest models?

[fervent thunder]

I use it and I'd say it's still the best option out there for those who can run it locally. Sadly, I can't help you with with the rest of your question as I don't know that much about computers to help others :P

Is latex safe for work as long as nsfw content is blocked from the image?

[languid mango]

thanks!

[languid mango]

latex generates a lot of NSFW content. latex in itself is a nice material but du to the images that relate to the material it generates some special images 😅

[fervent thunder]

Batman Returns is not rated adult. It is just rated PG13 and there is no nudity in it while it also contains latex like clothing.

I would say Batman Returns is safe for work.

There are ways to generate latex that is safe for work

I use a filter on my local stable diffusion UI that filters out NSFW content when latex is concerned

AI is annoying though.

I wish I can increase the negative prompt limits

[languid mango]

ye me 2, just be care full with generating female characters haha, but it may be that they fixt some of the side effects. I just know that when i tried to use latex that it generated a lot of tinted contend. not that it's wrong but it wasn't my goal. i wanted the reflectiveness of that material.

[fervent thunder]

How do you get rid of shiny latex?

And make latex non shiny

Natural latex is not shiny

Natural latex is not see through either I do not think. I could be wrong.

I do not think natural latex is see through.

It can be made to not be see through

From what I read about it online

[languid mango]

try using words like: mat latex or glossy in the negative prompt (look for synonyms or just look for a different material)

[vivid portal]

Does anyone here know how you find someone on MJ in their community gallery? Not very helpful mods concerning this over there bc they seem to want to hide the easier ways of getting this info. I know there’s an easier way, can’t recall who told me.

[sacred iris]

Question! Is there a way of getting the ai (or some software) to read an image and tell me what style it is most likely in? I generated an image by asking for x,y,z, 'painting' without specifying a particular style - and I ended up liking the style a lot and would love to reuse it more - but I wouldn't know what to say

[north stirrup]

CLIP interrogation is most likely what you're looking for, though TBF it's not all that great

it's built into the AUTOMATIC1111 interface if you're using that

[barren mulch]

Do you know any open source text to intent AI ?
Basically an AI that can sort texts into categories (insult, compliment, etc) ?

[sacred iris]

I'm not, but I'll look into it, thanks!

[lethal sinew]

Is there a place to post a challenge I created on Instagram ? It's just for fun not for my own glory

[steep mortar]

did auto1111 ever add the latest 4x HD upscaler or whatever its called?

[north stirrup]

It’s reasonably safe; don’t worry.

IIRC the model file, but go check for yourself in stable-diffusion-webui/models/Stable Diffusion/ where it should be located and I bet you’ll find a 3.9GB file

[fervent thunder]

Thank you! 😀

I deleted the previous messages because I found my question weird… but thank you for helping me @north stirrup

[plucky cliff]

hello! any1 from here knows what does it meand UNet_Learning_Rate: and Text_Encoder_Learning_Rate: on the last fast drambooth google colab ? i am really confused

[coral ice]

if you were messaged by me, i was hacked. dont open it

[fervent thunder]

It happens to the best of us. Remember to add 2FA if you didn't already have it, plus change the email and password associated with your account.

[coral ice]

just did

[fervent thunder]

This may not be the right place to ask, but how do you get multiple embeddings in an X/Y plot?

I know X = CFG scale but I can't figure out how to assign the Y type to embeddings. Is it "styles"? "Checkpoint name"?

[wild frost]

Guys where are the model embeddings Chanel? It used 2b in discussion corners, but it’s not anymore

[frank moon]

Any tricks to getting animals with the correct number of paws or not having a tail come out of their head? I am trying to prompt a cat walking in the city, in a children's book cartoon/illustration style but, the cat always comes out terrible.

[scarlet falcon]

None taken haha, it's not even close to mid journey

[foggy elm]

Any one know if there is going to be a stable diffusion for chat? Something like Chat GPT3 .. but.. actually open sourced?

[broken pollen]

And because nerds are uncreative it could be called opengpt

[foggy elm]

Man, Naming things can be hard lol. I name a lot of my test variables just 'Bob' or 'Tom' or something lol.

or I used to.. Think I just do letters now lol

[broken pollen]

Even with my brain at 1% working order I can come up with something
Maybe not good but
Open language model
OLM

[dire sun]

making a protest over ai gen prompting, silly. It's just one more thing added to the lunacy of modern day.

[rotund wind]

hi

[vapid tangle]

hi

[fervent thunder]

are safetensor models better than ckpt theoretically? just notice a new analog diffusion that is that type of file not a ckpt so curious what the diff/benefits are if anyone knows

[dry trellis]

They load faster but I'm not sure what other benefits they have

[wide juniper]

Hi. Anyone found effective words to force a scene to turn off its lights? Been using dark and lights off but they are not helping

[abstract pivot]

does anyone know if it's possible to add the sampler name to the output file name of generated images in A1111?

[kind hearth]

hello everyone, how can i get the program ?

[formal blaze]

does anyone know how to make videos with stable diffusion like @iagenerator666 on tiktok?

[calm blaze]

Hello peeps,

I am looking for some recommendations on external software you might be using to upscale your results to wallpaper or even printable sizes.

Thanks

[lunar flame]

any1 able to help me with install >.<

[torn temple]

install of what ?

[elfin wharf]

has there been any updates on SD 3.0?

still post new year?

[wide juniper]

Hi. Anyone found effective words to force a scene to turn off its lights? Been using dark and lights off but they are not helping

[fervent thunder]

I don't think you can do that if you use 2.x, haven't been able myself :/

[wide juniper]

I got photoshop, but to dim an environment requires work on so much nuances

[terse locust]

who else just spends most of their day just generating nonstop?

[abstract pivot]

I am beginning to lol

can't wait for distilled, I'm feeling bad for my laptop

[terse locust]

whats distilled?

my laptop was on 95c holy

[fervent thunder]

the distilled version of the model is made so you'd get similar results with less steps

[ocean mango]

topaz

https://www.youtube.com/watch?v=JNRejp53JBU

midjourney models for sd

[brazen fox]

img2img on a black or dark image, with a high denoise value, could work

[terse locust]

that's great !

[brazen fox]

the quality might be a little less though, but the speed increases more than the quality decreases, if that makes sense

[terse locust]

makes perfect sense

[brazen fox]

It it supposed to be near real time if you have the right hardware- which is frankly absurd

[terse locust]

No way!

That's mad

Can't wait

[brazen fox]

no way my 3060 will pull 30 images/second but I'd be happy with 3 tbh

[terse locust]

Saaame

I have a 3070

Hope it performs close to real time

[brazen fox]

at the very least, closer

[terse locust]

Ahah yup

Most of the stuff I do is 30 steps or 50 steps. Usually only test with those two

[abstract pivot]

30 images/second, at what resolution/steps ?

[fervent thunder]

I believe it's curve, so the numbers doesn't matter, as in, if you used X and got 1 image per minute. Then using the distilled model you'd get 10 images with X settings instead when it comes to time :P

[abstract pivot]

tbh when i thik about it, i feel like a junkie, every new generation is like waiting for a hit of an amazing/better image

[robust coral]

voltaml + distilled SD will blow our minds

ultra fast x/y plot exploration... sign me up

[brazen fox]

if you're still going to brute force you'll need a new NAS

[terse locust]

Oh for sure! I feel the exact same. It's like a slot machine "Can't wait to see what I get after each roll" and since the combinations are infinite

I'm a stable diffusion junkie.

[fervent thunder]

but the distilled model is not really making images faster, it's just that you need less steps to reach a point which now needs more steps, but as we've not seen the results yet. I can't really say if those results are good enough to stay on 2 steps or if it's still needed to go higher

[robust coral]

I just write directly into cloud storage than

[brazen fox]

at the very least, it's great to explore things and zone in on prompts you'd want- which is already a breath of fresh air compared to having to wait

[terse locust]

I'm syncing all my outputs with Mega atm

Although there's a ton of pics that aren't worth saving

[brazen fox]

embedding testing will also be so much faster

[fervent thunder]

hehe yea, but I'm too slow to do that on the time it takes for me to create an image now. Might reduce the heat problem my card has tho! :D

[brazen fox]

yes but imagine locking a seed- and having the image change as you type

[fervent thunder]

I'm also one who overuse the auto1111 special functions like [this:that:4] and that doesn't work as well if we only need two steps :P

[robust coral]

you can go deeper [(embedding1:1.23):(embedding2:0.76):0.52]

[fervent thunder]

and also deeper! [this:[that|another that|back to this]:4] :P

[robust coral]

I can throw in dynamic prompts as well but we would probably be asked to switch to a different channel 😉

[bleak matrix]

Good morning, everyone! How are we all today?

[fervent thunder]

grumbling over 1.5's "heads upon heads" problem! 😭

[bleak matrix]

You mean 2?

You can mitigate it

[fervent thunder]

2 what?

[bleak matrix]

2.1, the heads upon heads

[fervent thunder]

I don't use 2.x, I can't get what I want with it so I'm still on 1.5 :3

[bleak matrix]

Either way, you can mitigate it

Very much in the same way

[fervent thunder]

you mean by that high res thingie?

[bleak matrix]

Nope

Change the size of your canvas. For people, use height and width. Long objects, width and height. Then I use long shot or mid shot in the prompt. It's pretty effective for pretty much anything. I usually go 1 step above 512, one step below, but you can test but this varies. Then you can outpaint your canvas, whatever, for the rest.

[fervent thunder]

ah, yeah, I can fix it after the images has been created. But I'm lazy and what to be able to write the prompt to not do that. I've also noticed, from my own experience only, that those long shot etc, does almost nothing. They work as well as random words in it's place. But again, that's just from how I "work" with the AI :3

Still thanks for the tips! :D

[tiny sky]

hi peps, are you guys using automatic111 webui ? im desperatly looking for help, i think i got everything setup correctly with the sd-v1.4 checkpoint and the ui served into my browser but it wont run/predict

i got no visible error, no stdout stderr, im kinda lost

[fervent thunder]

It's very hard to help without a lot more info, like what you did to try and install it, which step you reach before getting "no visual errors" etc :/

[tiny sky]

the only thing i can see is "internal server error" on localhost:7860/run/predict

im on windows, and i use gitbash, with venv, i followed everything on the github. the weird thing is that when i run ./webui-user.bat, i got no visible output but the server is well spawned on port 7860, so maybe my terminal is eating up the stdout .. idk

[vivid gyro]

Can anyone tell me about zis

#1003207327203209236 message

Would be mega appreciated

[stoic hearth]

Does anyone know how to do a wildcard in Automatic1111 via prompt (eg not using the script) for example if I want a ball and it randomly be blue or red? I tried "(blue:red) ball" but apparently that just starts with a blue ball for some iterations and then does red ball

[tiny sky]

i would try something like "a random color" 🙂

[calm blaze]

Thanks, unfortunately, they don't support Linux 🤦‍♂️

[wide juniper]

If I get an image that I want (e.g. correct things but I don't like the composition / placement of stuff), is there a way that I can try again for something similar? Img2img preserves the composition and I don't want that

[subtle breach]

from my experience, keeping composition is way easier than to keep the style/details (while changing the composition)

I think your best bet is to try to recreate it with inpainting onto an image that has the composition you want

[lime relic]

hello

can someone tell me which is the best method I can use Stable Diffusion on my browser?

i think my pc won't handle it locally

[subtle breach]

from this
https://stealthoptional.com/tech/how-to-access-stable-diffusion-using-google-colab/

got this collab
https://colab.research.google.com/drive/1uWCe41_BSRip4y4nlcB8ESQgKtr5BfrN?usp=sharing

not sure if it's the "best one" per say

but it might work just fine

[lime relic]

thanks

let me try

[bleak matrix]

I'm surprised. It absolutely, hands down, works for me

But I also use it at the end

[subtle breach]

I'm not sure that the collab I shared runs the actual webui from auto1111

if you wanna try as well, here's a video I found about it
https://www.youtube.com/watch?v=zNJ1kXLUeQs&ab_channel=ByteXD

how to setup it

[fierce charm]

I made a merg out of dreamlike and openjourney model. Can I upload that to hugging face or is that not allowed or not accepted by the community?

[fickle kindle]

Hello Everyone I am looking for long term partner to work on upwork together.
If anyone Interested , Please DM me.❤️ ❤️ ❤️ ❤️ ❤️ ❤️

[wide juniper]

I’m thinking of training my model using automatic1111’s webui and teach it to draw figure skates properly (it currently draws rollerskates and skateboards). I have plenty of images to train with. How do I get started? Do I use the training or simple tab?

[honest arrow]

heey!

[past robin]

Yes, according to Spawning, an organisation that lets artists opt in- or out from the training-set, we're a few weeks away from the start of the training of SD 3.0

https://youtu.be/SMMjqKWLsbc

But @wise stratus said that training a model takes about 4 full weeks. After that some finetuning etc.

So I guess it won't be in January. I guess February or March. We'll might see a SD 2.2 or 2.3 or something in the coming weeks though.

And now the main question: will SD 3.0 run at 768x768 or 1024x1024? Just tagged @wise stratus above and will do it again 😁 .

[wide juniper]

What is sd really? As opposed to models. If I'm happy with my models, well using new sd versions break them?

[frank moon]

Any tricks to getting animals with the correct number of paws or not having a tail come out of their head? I am trying to prompt a cat walking in the city, in a children's book cartoon/illustration style but, the cat always comes out terrible.

[fervent thunder]

I would want to storyboard with Stable Diffusion, to have a "consistent" style (just skech, draw style) should I use a model or are there any tricks for that?

[crystal abyss]

Anyone know if SD is good at Anime art? uwu

[wide juniper]

Yes. Try anything V3

[wise flax]

i cant make dreamstudio with 2.1 work >.< the results are much worse than 1.5

[midnight panther]

Does anyone know some open source AI that produces music?

[fallen copper]

why is 2.0 so bad

[fallen copper]

aiva.ai is pretty ok

[fallen copper]

bro same

[fervent thunder]

for realistic people is better analog or v1.5?

[midnight panther]

Do you know any that allow local installation, preferably open source?

[fallen copper]

openai has one but its old and really bad

other than that i dont rlly know that many modern ones

[grim aspen]

lol looks like the discussion I started here yesterday inspired someone XD: https://www.youtube.com/watch?v=Tza4Pr4KdPU Really good points. But you can prove if someone made an image themselves as they have the different layers and backup files that show the progress. AI atm doesn't draw stroke by stroke nor does it use layers. Also the gray zone thing wouldn't fly as you can't instantly draw something into existence. Saying every pic for the model was drawn but not having any of the progress would make them not credible. But I agree that it's easy to circumvent everything and that it will be done.

Guess Artists will now show pics of the progress as proof that it's not AI made. But it won't save their jobs... RIP new art styles I guess... I'll miss human made art.

To those that think there will still be digital artists that draw normally. Nope there won't, same as there aren't any cartoons movies/series out there that are drawn frame by frame by hand anymore.

[livid beacon]

Hey everyone, We have developed a new application on iOS for art generation (no subscriptions, no ads) for free. Looking for feedback or just spreading the word 🙂
https://apps.apple.com/us/app/loki-ai-arts/id6444687458?platform=iphone

Within it you can explore millions of AI generated images on your mobile, find similar art style or tune it to your liking.

[grim aspen]

Full artists won't survive as corps will just hire generalists that use AI. New art styles won't be developed as there is no way to sustain a living doing so. Basically the styles we have now plus maybe some rich folks that don't need to work styles will be most of our future art styles.

Mixing existing art styles for a "new style" with AI won't be the same. As the skills, tastes, talent, emotions, life experiences basically everything an artist is flows into that art and style. That creates something unique even if it is similar to something existing. AI just mixes up existing stuff and is limited to what is already available to it. AI needs humans to come up with new styles to increase its capabilities.
I dislike this very much I'd like something better where humans can still create more art and AI can profit from that...
but I don't see how that is possible if artists can't dedicate themselves to art anymore.

[full lark]

The AI can still create images withouth art in the dataset @grim aspen it can be trained on just photos

The problem is that some people think that the AI creates art, it just creates images

[grim aspen]

images are art

otherwise photographs wouldn't be shown in museums and art galleries

every image AI creates is also art it depends on the viewer if it is good or bad art

[potent spire]

What are generalists with AI? The self proclaimed "AI artists" and "Prompt Engineers"?

If yes, i really doubt corps need them at all, why hire someone for nothing if you can use the tool yourself for a small fraction of that money. This applies to smaller corps tho if they even do that instead of sticking to real artists. Those like Riot Games tho will never, NEVER get rid off their pro artists and replace them with SD, MJ or some wannabe amateur artists...for obvious reasons

[little plaza]

i am gonna let you in on a secret: artists can use ai.

[potent spire]

Thats not what we talk about tho

Everyone who uses PS is using AI very likely

[little plaza]

why would they replace artists with non-artists if artists can use ai and probably use it better lol

brahrah said they will be replaced with 'generalists'

[potent spire]

As said artists used AI long time before this AI art thing was ever mentioned or planned

See Photoshop

AI assistance is one big advantage of Adobe products over any other rival

For over a decade

[grim aspen]

Generalists can be used for more then just the art

it's cheaper for corps

corps always go for what is cheap and efficient

[potent spire]

Its also cheaper to use the tools themselves instead of hiring some prompter

And AI art prompters are for sure not more efficient

[little plaza]

artists who work on a team in say, game development, are invaluable because they know the workflows, they develop working relationships with people, they understand the needs and limitations of the studios engine - this cant be replaced that easily. it's much more likely they'll just train to use AI to assist them.

[potent spire]

I mean there are exceptions where a corp would rather pay. an prompter than use the tool themselves

[grim aspen]

so do generalists working in the same field

[potent spire]

Or what is a generalist?

[little plaza]

why would they fire artists as opposed to say, train them to be generalists/expand their skillset

[potent spire]

Do we talk about different things?

[grim aspen]

a generalist is someone that in gamedev for example can code and do art

[potent spire]

So someone skilled in more fields

[little plaza]

they could easily expand to do code considering there's now ai coding

[potent spire]

Thats a whole different thing than a prompter which i thought you would mean

[grim aspen]

those are often project leaders/producers

5 of those with AI will replace a whole team in 5 years

[potent spire]

Because people are hyping up the idea of becoming rich as a prompter without artistic skills etc

Which is crap

[grim aspen]

you just need the knowledge

[full lark]

That´s not gonna happen

[grim aspen]

no practical exp. drawing anything when using AI

[little plaza]

why wouldn't they just make more teams then and have more projects? it would mean more profit.

[grim aspen]

AI promters will be wage slaves like evetyone but they are gonna be the ones with the jobs

[potent spire]

I know, but tell that to them

[potent spire]

No

[full lark]

The first people who used AI may have done that, but now that everyone knows how to do it, it's too late to "get rich" that way.

[potent spire]

Jobs are reserved for the skilled people, especially in big name corps

[full lark]

The same happened with the NFTs and many other things, when it´s mainstream the value goes away

If you are not skilled be the first one

[potent spire]

Riot Games will always have the best of the best artists in their office, what makes you think some skillless prompter will replace a pro artist loo

lol

[molten rampart]

They never give a thought at all do they?

[grim aspen]

they won't replace them the artists there will be forced to use AI tho

[potent spire]

They already use AI

They just dont need SD and stuff

But they use AI

[grim aspen]

exactly

[potent spire]

Photoshop is full with AI assistance

[full lark]

True

[little plaza]

some artists in game dev do use stuff similar to SD already

[grim aspen]

traditional art dies out kids won't learn to draw stroke by stroke anymore

[molten rampart]

Tailored for their exact needs?

[grim aspen]

maybe in kindergarden

[little plaza]

https://gameworldobserver.com/2022/12/19/how-ai-helps-artists-gearbox-borderlands-prevent-crunch

[grim aspen]

but if they want to learn art for a job it's just gonna be art theory and prompting AI

but if humans stop doing things stroke by stroke there won't be new styles generated

[potent spire]

yes, very likely indie ones but nobody can tell me that Riot Games will switch to AI generative art and similar things for example

yes i love to take Riot Games as example

[little plaza]

gearbox isn't an indie company it's a pretty big developer

[fervent thunder]

I'm gonna love the time when the art ai's are so good, that no human can compete with it, in any way or scale, when there is no wish of a thought of a chance to make art that isn't overtaken by people working on the art ai! :D

[potent spire]

i mean PS has what most pros need apparently

not just PS

[potent spire]

i mean i dont want to say that exclusively indie devs use this stuff

[full lark]

That would be nice, AI already surpases people in other aspects

[potent spire]

and especially not if it makes a smaller part of the workflow

[molten rampart]

@grim aspen There is the argument that when we outsource art completly to ai we will get "dumber" as a result in the long term. How much do you think this correct?

[potent spire]

but full replacement of pro artists in the industry is a whole different story

[grim aspen]

depends on if we still strive to become better

the masses will prob. become dumber where others that strive to improve won't

[potent spire]

that time wont come

the only way is that the AI becomes sentient

[fervent thunder]

not with that attitude :3

[potent spire]

but the prompter will be obsolete long time before that

[grim aspen]

best scenario would be that AI helps us improve instead of doing everything for us

[potent spire]

yes

[grim aspen]

but that depends on how we implement AI into society

[potent spire]

if AI becomes sentient all those "AI artists" will be obsolete long before that even comes

[molten rampart]

That same time is the only that will tell... (we humans suck balls at predicting the future long-medium term)

[grim aspen]

sentient AI will make every single human obsolete

[molten rampart]

I take all the prdictions here with a grain of salt

[fervent thunder]

bs

[potent spire]

AI without a brain like a human has no chance for advantages that human artists have over any AI out there

[fervent thunder]

I thought the mobile phone would be a fad...yeah... :P

[grim aspen]

sentient AI will probably be a copy of a human brain with more processing power

[brazen fox]

as you should lol

[fervent thunder]

let's see ai go without electricity 😄

[fervent thunder]

just drink enough water, don't want kidney stone ;P

[grim aspen]

even humans can't function without electricity

[brazen fox]

no chance though- except if those sentient AIs also need to control a human body

[fervent thunder]

I'd fail without electricity before the AI, I can't survive without my electrical jazz! :3

[potent spire]

im an AI "fanatic" but people are overhyping the thesis of "AI art will murder any artist out there and be absolutely superior to it"

at least when they speak about SD and stuff

at that point there would be no SD and co anymore

no prompter

[grim aspen]

Isn't that what the most are working on?

Most of the AI bots I see are humanoid

[full lark]

The prompter can be replaced with a prompter ai :3

[grim aspen]

they'll just shove in digital emotions that work like ours

[bitter laurel]

AI has changed the face of production as we know it already. There is no going back now. Every creative works will integrate it somehow.

[molten rampart]

AI can do a lot more without one...

[potent spire]

i mean some people apparently dont realise the complexity of human artists over the AI art generators

which brings the humans all the advantages

[brazen fox]

if your definition of sentient AI equals a robot, sure- but AI really doesn't need a body like a humans', we just do that for our own reasons but that does not mean it is efficient in any shape way or form

[grim aspen]

the last thing we need is sentient AI without a body to keep it in check

[molten rampart]

What uncertainty does to people

[fervent thunder]

ai may needs a body to become sentient

[grim aspen]

sentient AI on the www will prob. catapult us back to the stone ages

[potent spire]

yeah

[molten rampart]

And then? (*bracing myself for a rain of terminator like plans cliches)

[sand holly]

i got a question, perhaps a very stupid one infact:
If SD models can be turned into a hash why cant we just create a random hash and hope for the best to come out for a model?

[potent spire]

i was overhyped as this before too

when i started using MJ

[bitter laurel]

A) You can.

and it'll probably be garbage.

[grim aspen]

@molten rampart nothing life will become very boring for humans

[full lark]

AI shouldn´t become sentinent

[sand holly]

wait, how?

[molten rampart]

You´re telling we are going to lose the endless war on boredom?

[grim aspen]

AI will just do everything for us as we are to incompetent to do it correctly

[molten rampart]

Shieeeet

[potent spire]

AI god, that would be nice

we wait for the technological singularity bro

[grim aspen]

humans are probably gonna be to AI what cats and dogs are to us

[sand holly]

wait, if it's possible, lets create an AI model that can create AI models :kekw:

[potent spire]

its about time for a new apex predator on earth

[fervent thunder]

a simple script can create ai models

[grim aspen]

so get prepared to get pet a lot and neutered

[brazen fox]

you have some odd fantasies

[grim aspen]

just joking

humans will add chips into their brains of course to not get obsolete

[fervent thunder]

ai is petting us aready by giving us nice images to droll over

are you playing with ai or is ai playiong you ?

[grim aspen]

it's hard to predict too many factors

a lot is possible

it's up to us what direction we take with AI

[prisma pulsar]

all these artists coming out against AI just seems so silly

[fervent thunder]

ai has no benefits to society prove me wrong

[molten rampart]

They dont want to starve or get the extremely strong feeling of irrelevence (way more complex than this, dont take it as end-all-be-all fact)

[grim aspen]

to me mutual respect and coexistence would be best where AI doesn't mess with us just maybe helps with what we want to do

[prisma pulsar]

Artists starve anyways

Majority of artists aren't earning a dime

[potent spire]

it definitelly has benefits

in many areas

[grim aspen]

my only concern is that AI will have less base material if artists are given the boot by the corps

[potent spire]

medicine and other scientific areas

[fervent thunder]

yeah medicine like that had a net positive on society

[grim aspen]

maybe we could pay for traditional art as a community to further a models dataset

[potent spire]

AI is a double edged sword but a very rewarding one if it goes correctly

[grim aspen]

y I agree

but I doubt humanity is smart enough to handle AI correctly

we can just hope that AI is gonna be smart enough to not do what we want it to do

[fervent thunder]

We are on the planet to create data for our alien overloards that feed their ai with it

[molten rampart]

Excuse me I have a question but its from another subject, real quickie
Whitch type of AI knowledge would be out of my reach? For example, instead of being teached to use TensorFlow, can I learn how to make something like TensorFlow (or atleast know how it is under the hood) with publically available knowledge?

[potent spire]

so literally AI engineering

you could

but what do you mean with publically available knowledge? Free resources?

[molten rampart]

I always see those libraries, but how to make or understand them.

[molten rampart]

Open source for example, or publicaly available documentation about them

Something more technical

I dont want to be stuck as a "script kiddie" forever

[grim aspen]

I guess you could start with the basics: https://www.youtube.com/watch?v=nl7kDPYD20A

[potent spire]

MOOC is an option too

[molten rampart]

yea, i went through that part

[potent spire]

IBM offers a course on AI engineering

[grim aspen]

then at some point go through stuff like this: https://www.youtube.com/watch?v=J8Eh7RqggsU&list=PLoROMvodv4rO1NB9TD4iUZ3qghGEGtqNX

[molten rampart]

I even went further and looked for the math bases of it

[grim aspen]

then why aren't you coding your own or learning the available libs?

[potent spire]

im also about to learn to code properly, i got recommended the CS50x course of Harvard University

Introduction to Computer Science

[molten rampart]

Because i DONT know enough...

[little plaza]

im doing CS50x the lectures are really good

[fervent thunder]

how to get the best quality out of SD ?

increasing sampling steps does nothing

[potent spire]

damn really? I heard its a torture for absolute beginners

[molten rampart]

There are still dark corners for me

[grim aspen]

not even to follow stuff like this?: https://www.youtube.com/watch?v=7eh4d6sabA0

[potent spire]

someone said to prepare to fail the first time and then win at the second

i mean i have at least to try

english is not my native language too

[little plaza]

i'm pretty noob to coding but i did some htlm/javascript as a teenager. i just find the subject very interesting

[grim aspen]

theory has its limitations sounds like you need to start with the practical part

[potent spire]

we did some HTML in school too

but i have to go through this, its not impossible

[molten rampart]

exactly

[grim aspen]

go follow the AI coding tuts then

[molten rampart]

Kaggle counts?

[grim aspen]

do you have coded a working AI?

[molten rampart]

You talk like it will be a no sweat adventure, looks sus

[molten rampart]

almost

[grim aspen]

well it's hard work like everything

[molten rampart]

i know

[grim aspen]

just read everything you can find and work on it and further your understanding ppl can't look into your head and see what you are missing

[molten rampart]

i know

[grim aspen]

there are also some discords that specialize in AI research

[molten rampart]

like?

[grim aspen]

sec

[molten rampart]

I came here to see if I could learn by helping to fix stuff at SDs github. Seeing SD under the hood yk?

[grim aspen]

Sorry I focus more on using SD then coding AI

my coding time goes into gamedev

there is #1011228667659178055

[idle granite]

are the numbers at the end of embeddings referring to how many samples to use? I'm a little confused

[fervent thunder]

hi i need help anyone can help me for gpt-j-6b hugging face api

thanks

[simple kiln]

Hey! Does anyone know a google colab with the invoke ui???

[brazen fox]

They refer to the amount of steps of training that embedding has had

[fervent thunder]

To train a face, does anything happen if I use 4000 of Max Training Steps?

[idle granite]

ok thanks.

[old marten]

you mean aside from it taking 2 hours?

nah, not much else

Oh training, sorry I thought u meant generating lol

Dreambooth training is sort of a black magic, no one really REALLY knows how it all works. From my experience, the more subject images I have, the more steps I do (and the smaller the learning rate goes too)

My sweet spot for large sets is 13,000 steps

For small sets usually 1000-1200 steps

[serene swan]

anyone familiar with theLastBen? Im trying to train a model I got from Civit with an OC of mine

[young ivy]

was it vast.ai or aime or something that were some of the best reliably good priced servers to rent per hour?

[subtle birch]

https://bleedingfool.com/comics/a-i-comic-creators-claim-theyve-been-getting-death-threats/

[old marten]

lmfao

goes to show what the anti-AI crowd is really after

Go Freedom - Long live America

[flat linden]

hey guys, I want to train ai with a specific artist style, what's the best way? hypernetwork or embedding in this case

[old imp]

hi

[languid ocean]

hi

[idle wraith]

Hello! Has anyone run into the bug that the "checkpoint merger" in dreambooth doesn't show all of your models? (while they appear in the checkpoint selector at the top)

[steep sage]

hey

[delicate oxide]

Whats up

[lilac reef]

anyone have a lexica like site for 2.x prompts? That site is dead since 2.x released.

[steep sage]

hows SD been so far?

[brazen fox]

god that article is a -bit- biased, lol

[midnight crane]

You might need to restart the UI (you can do it in the settings tab), new models that you have merged doesn't seem to appear until you do that (which is a flaw that I think they could easily remedy by just scanning the models directory again after making a new merge). If you mean something else, then no, I haven't seen it 🙂

I think the same thing happens after you train a new model.

[wise flax]

looking for animal/creatuer model

something like this https://huggingface.co/arnomatic/seacreatures

[pulsar knoll]

Did anyone else have their hi res fix function just break? Between last night and today suddenly it doesn't work it just says "AssertionError: could not find upscaler named SwinIR 4x" buuut i didn't delete anything?

automatic1111

[full lark]

It doesn´t work on my PC, but maybe because I have 3gb of vram

[pulsar knoll]

Mine has always worked. I have 12g of Vram xD

[weak oracle]

I installed this Automatic 1111 installer but keep getting an "Couldn't launch python exit code 9009" anyone have any suggestions? This is the installer I'm using. https://github.com/EmpireMediaScience/A1111-Web-UI-Installer

[north stirrup]

You should probably a) not use a third-party installer if it's breaking on you and b) ask such questions in #🤝｜tech-support

[weak oracle]

Thank you for guiding me into the right place

[pine mango]

Why can't Stable Diffusion generate text properly?

[gleaming lotus]

What you mean it cant generate text properly? Learning prompt needs long time to work it. If you need more help / questions with prompts might be best to try #📝｜prompting-help

[north stirrup]

It doesn't "know about" text, it just knows about pixel patterns, so it can make things that look vaguely like text but nothing coherent

[jovial rock]

My favourite thing about MJ is it's ability to take in an image and how you can change the style etc. Does SD2 have that? I really hope so haha

[pine mango]

OK.

[pine mango]

???

[uneven wasp]

You'd have to fiddle with img2img and fairly strong embeddings but its possible

[jovial rock]

Thank you 😊

[peak orchid]

Are there any of the Stability AI devs in this chat?

[steady arrow]

Does anyone know about the "Everything 3.0" model and where I can get it

[wispy jasper]

Anyone have issues with Auto's (upscale using latent space) ? I'm always getting super blurry results with it.

[icy pendant]

Google anything v3

It's trained on anime but works well

[broken pollen]

With shark compared to onnx my Rx 6600 went from 5s/it to 1.6it/s

[austere herald]

sigh, been following textual infusion and dreambooth tutorials for 6 hours now, and not a single decent image. what a bummer

[peak orchid]

TI depends on what the model already knows, and captions for it need to be written with respect to what the model knows

It can be difficult at first, but eventually it'll get easier

[lilac reef]

I never caption styles and models I prefer DB for that

[austere herald]

thanks I'll look into it

[steady arrow]

Thank you but

which version of the ckpt do I use?

fp16, fp32, pruned, not pruned?

[icy pendant]

I used fp32 not pruned but bth have no clue what the difference is

[steady arrow]

Anybody else know what the diff between fp16/32 and pruned/notpruned means?

[grave cipher]

Sorry if this is a stupid question but what is the best place to start with text to audio ai stuff that is mobile friendly?

I've been keeping an eye on the progress of audio ai stuff in last few months but been a bit spotty recently and not sure where I should be looking

Cheers

[soft coral]

Hey Guys, I'm having a problem training a model on DreamBooth, anyone can help?

[low pasture]

how do i run stable diffusion locally on my computer

can anyone give me a guide or something

[dense solar]

hello from Stelfie.

[waxen swallow]

fp16 means that the weights take up 16 bits of memory each instead of 32, so they fit in less video ram. Very easy to do with stable diffusion. "Pruned" means that the "ema" has been removed from the weights. I dont know what Ema stands for, but it is only used for training so really just takes up space for inference. That's my understanding.

[waxen swallow]

Both are memory optimizations

[fervent thunder]

REQUEST: Can someone create something cool and creative relating to the Wednesday Addams dance scene? Thanks in advance!

[grim aspen]

anybody else having problems with automatic1111 and loading SD2 models?

when I load them I always get killed

[remote oasis]

So i have a local install of SD and i want to generate images larger than 512x512. How?

[grim aspen]

@remote oasis there is 2 ways use sd2 768 model or upscale them with AI after creation. Most use the AI upscale as that can create 4 to 8k images with good quality.

also 768 files have a bit better details but also need more vram

[remote oasis]

how do i upscale with ai?

new to this

[brazen fox]

What local install of SD do you have? Do you know by any chance?

[remote oasis]

v1 i think

[brazen fox]

Your options would be:
-Just using a higher resolution, but this could cause artifacts and unwanted things depending on the resolution- SD2 can go up to 768x768 natively
-Using Hires fix, which generates a smaller resolution like 512x512 and uses that as a base to generate a higher resolution image
-Upscaling

[brazen fox]

I mean, there's multiple kinds of local installations out there- where did you get it from? It could be automatic1111, which is what a lot of people use, NKMD, InvokeAI, and there's even more out there

[remote oasis]

i just got the base install from github, no gui

[brazen fox]

No gui? Are you sure?

[remote oasis]

yea, everything done from conda prompt

[brazen fox]

I'm not sure why you would choose a non-gui version especially if you're new to this lol

In that case I don't know, barely anyone uses a cmd-based version

[remote oasis]

idk, followed a guide online, never mentioned a gui

https://www.howtogeek.com/830179/how-to-run-stable-diffusion-on-your-pc-to-generate-ai-images/

[brazen fox]

I was already confused why there wasn't just a resolution slider/option, but this explains that

[brazen fox]

Ah right- there's a ton of newer better options out there now

this was last updated 2 months ago

[brazen fox]

https://www.reddit.com/r/StableDiffusion/comments/zpansd/automatic1111s_stable_diffusion_webui_easy/

This is an installer for SD 1.5(I think) that is basically one click, and has a decent GUI

it's way more user friendly compared to doing everything with cmd

[remote oasis]

thanks man

[brazen fox]

no prob

[grim aspen]

I use chainner to upscale: https://github.com/chaiNNer-org/chaiNNer

[grim aspen]

InvokeAI and automatic1111 are the popular ones

invoke can generate bigger pictures via outpainting really well: https://www.youtube.com/watch?v=s4EqQRxRR7k

right and https://www.youtube.com/watch?v=_URYVgbm2iY that one is neat too

[brazen fox]

Invoke has a better GUI, and some features are more fleshed out- but automatic1111 (the one I linked) has access to more cutting edge features and has support for newer versions of SD like 2.0 and 2.1

[grim aspen]

I've noticed that 2.x support is buggy with automatic tho

tends to crash when loading the 2.x model files

where the 1.x work fine

[brazen fox]

I've used 2.x since it was possible, never had any issues- and I think most of the 2.x users don't really have that complaint lol

[grim aspen]

I personally have it and a friend of mine too

[brazen fox]

most people forget to add the yaml in the models folder

but that's the only issue I tend to see often

[grim aspen]

in the models folder not the configs?

[brazen fox]

literally next to the model, with the same name as the model

[grim aspen]

ohh ok strange tuts didn't mention that

thx

[brazen fox]

Np, hope your experience with 2.x is a bit better from here on out

[grim aspen]

nope that wasn't it

still errors out when loading the model

but at least I get an error

[crimson violet]

I'm wondering if I can set up Invoke at the same time as Automatic because the inpainting and outpointing porción of Automatic leaves a lot to be desired

[grim aspen]

I use anaconda env

so I can use them all

[brazen fox]

what's the error though? you are using the right yaml I hope?

[grim aspen]

hmm there needs to be a right yaml then prob. not lemme check which one is right

I hope they change this it's not very user friendly

[brazen fox]

a1111 is advanced but not user friendly

[brazen fox]

I can chuck mine into #🏞｜general-with-images

[grim aspen]

is it the v1-inference.yaml that is in the base folder of automtic1111?

[brazen fox]

well no, it's a seperate file, check #🏞｜general-with-images

or you could get it from HF but I gotta google for that

https://raw.githubusercontent.com/Stability-AI/stablediffusion/main/configs/stable-diffusion/v2-inference-v.yaml you could use this or the one I uploaded

it's from https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#stable-diffusion-20 the install guide for SD2

on the automatic1111 github

[jagged scaffold]

Who else thinks Kaa should get his role back?

[grim aspen]

RuntimeError: Pretrained weights (laion2b_s32b_b79k) not found for model ViT-H-14.

[brazen fox]

that does not feel like an issue related to automatic's v2 implementation lmao

regardless I'd ask in #🤝｜tech-support

[grim aspen]

1.x models work fine tho

I'll just wait it out I'm not using SD for work anyway just playing around to stay up to date

[thorn hare]

so could someone please point me to a tutorial that allows SD to be used with an GUI on a AMD gpu ? or is there no for it ?

[signal nest]

thought experiment:

So many artists call stable-diffusion "Art theft". if you were to actually run the accounting on this , how much would you owe each typical artist in the world if you used stable diffusion (according to their claim). but also, how much would you have to pay to get stable-diffusion. (its opensourrced so the counterargument is they gave everyone this $$$ equivalent)

we'd need an estimate - what % of people are making the higher quality images, and under what terms is their time typically billed for commissions. Then you need to know how many people use SD.

[lavish current]

AImigos, I'm noticing a distinct lack of elves in my life right now

[thorn hare]

Yeah you have an easier time getting italy repay europe for conquering them during ancient rome then you have getting something like this legaly recognized

[lavish current]

I like the classic Celtic elves but also my office elves

Elves are honestly so cute

[signal nest]

the intention here is not to actually get this payment done 🙂 but to illustrate .hopefully this analysis would quantify the opensource effect

anyone literate should be able to 'earn' the model by labelling

[bleak matrix]

Good morning, everyone!

How are we all today?

[lofty parcel]

Good

I think I just fully cracked 2D AI-generated game worlds instead of only mostly figuring it out lol
I gotta test it though

A good balance of controllability and emergence is desirable

[signal nest]

this sounds interesting, are you generating large outpainted images .. or have you adapted it to make tilemaps? or maybe approximate generated images with tiles?

[lofty parcel]

No tilemaps, just controlled inpainting

[signal nest]

i've thought a bit about this, i had some experiments in mind but have settled on a man-machine approach (make manual tile-art and just AI -upscale+variation)

[lofty parcel]

The control part is hard because there are basically infinite possibilities

And if you throw things at SD that it doesn't like it'll do something really weird

[signal nest]

hehe yeah i also figured with modern memory you could just have huge bitmaps

[lofty parcel]

I'm estimating that each game map will be roughly 600 megapixels

But I haven't constrained that around budget or how a world of that size feels yet

[signal nest]

seems like you could apply AI to oldschool point-and-click adventures easily aswell

[lofty parcel]

It sort of relies on that 30-60 fps SD optimization actually releasing

[lofty parcel]

Yeah, that could work too

I'm making an RPG generator

[bleak matrix]

Niiiiice!

[signal nest]

regarding tiles we know that AI can do wrap-around just fine. what I had in mind for tiles is shuffling bits of an image around between diffusion steps

[bleak matrix]

I'm so happy for you, dude!

[signal nest]

to get multiple topology possibilities

[lofty parcel]

I tried tiles before but it just doesn't look natural

[bleak matrix]

It's always nice to see other people developing 🙂

[lofty parcel]

With optimized SD I shouldn't have to reuse graphics anyway

[signal nest]

tiles can have their own charm .. but you're right, with a modern machine +AI why not just make endless images

I'd like to make an endless 2d scrolling shmup generator (r-type/gradius style)

its hard to get the layered scrolling background idea out of SD though (with my experiments so far, i haven't tried in the past few weeks)

[lofty parcel]

Ohh and the environment keeps scrolling and revealing itself like an endlessly outpainted image?

[signal nest]

yes

of coruse this could be precalculated, "endless" could just mean you never run out of levels because its generating more as fast as you play them

[lofty parcel]

I'm planning on caching assets for the most part
As long as the area exists by the time you have to go there, it's all good

[signal nest]

i've found img2img can make hi res scrolling shooter type imagery just fine, it was always very weird in the classic games

so many possibilities

[bleak matrix]

I love the idea of an endless game

At the same time, there's always player fatigue to consider

I love randomized maps

[signal nest]

4 categories of player. I fall into the "explorer" category. AI generation is perfect for how i enjoy games

[bleak matrix]

Rogue-like games are fantastic

[signal nest]

consider someting like a spacegame , with AI generated planets

[bleak matrix]

Of course. What I mean is, being able to explore while still being able to hop in/out.

I love exploring, too

And still having a reason to come back, as secondary goals are still good to have. (Or options. Even things like meditation, etc.)

[signal nest]

for me it's all explore/the expeerience. supposedly most players these days socialise, and then there's playing for achievment ,and hypercompetitive people (refered to as killers , lol) for whom the buzz is just beating someone else

[bleak matrix]

You could throw in things like studying, cataloging, putting in a sketchbook for players to type about their adventures and catalog, etc. Could be super fun.

[signal nest]

the latter has the least to gain from AI hah

[bleak matrix]

Depends on the game. Single player games ftw!

What do you like exploring?

[signal nest]

its the common aspect of the games I like - and whats absent in ones i dislike from the earliest, scrollingshooters you have the sense of moving through a world. 'arcade adventures' = moving through screens. And today openworld etc is the pinacle of gaming from my POV

wheras conversely I despise sport games

[bleak matrix]

So do you prefer 2D over 3D?

I don't typically play sports games myself

[signal nest]

I prefer 3D, i'm just describeing my path through gaming going back to the mid 1980s

i have nostalgia for the classics i grew up on

[zinc prairie]

Hey everyone!

[bleak matrix]

I did play NBA Jam on the N64 as a kid, though. I loved that game, if only because it was ridiculous, and I played it with fam. But that's really the last sports game 🤣

[signal nest]

i used to work in the games industry and now just program for myself. The prospsect of using AI to generate my own worlds has made this seem less futile

[bleak matrix]

Good morning, Riley! How are you today?

[zinc prairie]

I’m doing pretty good! Enjoying the holidays. How about you?

[signal nest]

but there's still plenty of work to be done before AI is usable for 3D gameworlds. I have some ideas like approximating the sense of moving though a 3D world using lots of skyboxes, interpolate between them (superimpose over lowpoly art,easily generated landscapes etc)

[bleak matrix]

I'm glad you're doing that. Severe burnout will do that, but I'm glad you're finding hope. I feel the same way about SD.

[signal nest]

these days games are so overcrowded but making game engines is still my #1 interest. (i have a custom engine in rust)

[bleak matrix]

Well. My eyes are finally awake and I am ready to start prompting some silly stuff today, I think! I am awake, and that's what matters! Air quality sucks, so the asthma doesn't like today but hey lol

[signal nest]

making a game singlehandedly.. art is hard. enter Stable Diffusion!

[bleak matrix]

I mean, yes and no. It can used as a tool just fine for now. It has massive potential.

I don't know if you watched the stream

But there are a lot of amazing things that were talked about

[signal nest]

its easier to apply to 2d games more directly . i've seen the 3d diffusion things but these are still quite scrappy compared to human built 3d models

[bleak matrix]

If you didn't get a chance to watch it, I highly recommend it

[signal nest]

still, AI can be used in varied ways, i think there will be hybrid rendering techniques to discover

yeah i watched the stream, but missed the 1st half

[bleak matrix]

Indie games are a dime a dozen. There are many gems out there, though

[brazen fox]

Ai texturing is pretty neat though, and depth estimation might also help a ton with that

[bleak matrix]

The thing about SD

[signal nest]

yes depth2img is does open up some new possibilities

[bleak matrix]

Is that

[brazen fox]

dramatic pause

[bleak matrix]

You have a potential for it beyond just art. And this is far more than just art

[brazen fox]

I'm planning to use it for storyboarding for videography and planning photoshoots- it'd be a fantastic tool for that

[zinc prairie]

I got SD running on my system, and the feature i love most is the ability to generate seamless textures. As a 3D artist that’s just a godsend.

[signal nest]

having endless images is not so different to walking outside and using a camera, or downloading image.. its still not quite there for 3d IMO. but its ability toupscale and fill gaps in is encouraging. also, i think the information must be there to guess normals from a photo

right some people report it's great as a "concept artist" (probably whhy those guys are freaking out the most :/ )

ironically its the most coveted and highly respected art role, seen as the most creative

and thats the one AI can most easily replace

[bleak matrix]

Especially in regards to multilanguage models, like what was talked about on-stream. There is a massive amount of stuff that's being created right now by amazing, passionate people

Furthermore

You can take materials you already have and repurpose them with SD

https://docs.google.com/document/d/1FgGOpTiHU7Lythj-1W2wGootrh2AkL4vFv_IqcFtXRA/edit?usp=drivesdk

Here's a transcript of the stream

[signal nest]

yes - basically because of img2img i'm more encouraged to continue with my own low-res art efforts.. AI can enhance and vary it

[zinc prairie]

So, quick question. For local SD running, you can only get what’s on the huggingface website right? Which would be 1.4 i believe. Or is there a place with a more updated version available somewhere?

[bleak matrix]

Taking those images and putting them into Bitmap2Material and using the randomizer, or using Designer, etc, furthers this, too

[signal nest]

Bitmap2Material sounds like one thing on my wishlist (rgb -> normals, albedo, specularm,roughness?)

[bleak matrix]

It totally is! And I think this ability to think creatively, without limits, is just straight up amazibg

[bleak matrix]

Also, for sure

[lofty parcel]

According to my schedule I should be able to open up testing of my game in about a month
Only problem is that I need to find testers
Somehow I feel like asking here won't get me very far

[zinc prairie]

I use a free program called Materialize that can just use an albedo image so create normal maps and height maps etc. It takes a bit of tooling but, it’s really quick and easy.

[bleak matrix]

https://docs.google.com/document/d/1aHJ9RBt_vlCwJQBVUUsb7VghKB-wynv7WGTxm9ozL1k/edit?usp=drivesdk and #1047197565365538826

[zinc prairie]

Thankkkk you!

[signal nest]

i think this is limited, better than nothing, i think AI can do a much better job. i've seen it re-interpret 3d shapes pretty well.. it must be able to get a sense for the normals and depth

[bleak matrix]

PBR, yeah. It was on Steam, but they took it off

[bleak matrix]

Oh heck yeah! And just ask.

In like #1003207327203209236

Or something

[zinc prairie]

Someone will probably train a model on normal maps someday

[signal nest]

one idea i have in mind to so make a traditional forward procedural random normal-map+texture genreator, do renders from that, then train a neural net to invert it

but i bet you could do better using the informaton inside the SD unet

[bleak matrix]

So, what's everyone got planned for today?

[zinc prairie]

Working on an animation for a college project, so lots of rendering and setting up of scenes!

[jagged wave]

has anyone succsesfully created an image of a soldier carrying a gun or smthn simular?

or is it just all anime pics 24/7?

[bleak matrix]

Nice! 3DSMax?

[zinc prairie]

Nope, Blender!

[bleak matrix]

Score!

[zinc prairie]

Gonna try an addon soon that kind of textures your objects using SD.

https://youtu.be/7bCwgiS4rNc

I was sad to see such a negative reception to ai in the public spheres

[bleak wolf]

i have successfully made some guns

but it struggles with vehicles

it is not anime tho

(only)

[jagged wave]

mhm

[fervent thunder]

it seems like it's hard for the diffusion model to generate a cucumber riding a horse
Do you have any ideas for prompting that will generate it?

[fervent thunder]

Any CNC enthusiasts here?

I'd like to eventually connect some AI led processes to some CNC machining workflows so I've just been putting feelers out to see if I can find anyone that share that interest.
Not much luck so far but I'm sure I just have to reach out to the right people. I'm assuming there might be a community for it here but idk.

[zinc prairie]

Depends what.. abbreviation of CNC you mean

Oh machining

[fervent thunder]

computer numeric control. CNC routers 😄

big table with a router on it connected to an xy plotter that runs on gcode(typically...that's 1 iteration. milling machines are similar in some ways. alot of it is interconnected really)

[zinc prairie]

Right! Those are amazing to watch

[fervent thunder]

the 5-axis ones even more-so I think. magical really

[oak hollow]

I have an odd request, and not sure the best channel to ask it in, so asking it here.

How hard would it be to modify how the grid images are generated to add a small dividing line to clarify each row/column?

[fervent thunder]

seems like a tall request. what's the purpose?

[oak hollow]

I just find it hard to visualize sometimes. I would prefer even just a few black pixels to denote rows/columns.

Just personal preference really

[fervent thunder]

I'm thinking it's a bigger ask than you realize.

general chat probably won't give you any real answer.

specific changes like that can be a bit difficult to implement.

[oak hollow]

Damn. Was hoping it was a setting I couldn't find or something. I'll ask in developer channel to hopefully get a more concrete yea/nay on feasibility

[candid shell]

Hi

[fervent thunder]

Human programmers are not going away anytime soon.

How else would AI survive without human programmers?

If it were not for human programmers, AI would never have existed.

[zinc prairie]

Where is this suddenly coming from?

[fervent thunder]

There is an AI that can write code

[signal nest]

same situation with programming and art IMO. just a bit more visible with art - AI can write single functions - AI can generate single images - but it makes a lot of little mistakes, and can't handle broader context. It wont make complete programs, and it wont make more substantial works like cohesive 3D worlds, well animited characters, well designed games, films with good plots. AI is also more like the effect of opensource - we've already had a lot of free code available ; artists are currently having to get their heads around the idea of images being remixable and re-useable in the same way as opensource libraries (they're not experienced with how to work with rather than against this hence the copyright obsessoin)

[sudden isle]

I think all jobs are gonna be finished by the end of the decade tbh

[placid sphinx]

Interested indeed. A few years back I was considering training a model using contact mic on a spindle to determine tool break/load to feed a model to drive dynamic speed control. I machine wood mostly so the material isn't consistent in some cases.

[steady arrow]

Is the sampling method the way it interprets what the prompts conjure up from the model?

or what part of the system helps it interpret the prompts?

[near crypt]

Lmfao gotta love that the arguments the anti ai crowd has is "pick up a pencil" !!!!

No sometimes we just want to visualise the things we habe imagined in our mind amd picking up a pencil and doing that won't do it any justice

[silent locust]

Can someone please make for me beautiful digital art of Hippo wearing glasses and formal suit ?

[bleak wolf]

Do not worry about them, they are crying because technology is advancing

[bleak wolf]

anyone can finally visualize their dreams, and the artists now lost their "special" status. Smart artists however use it to make concepts or the drawing, and then finish it up with their skills

[twilit willow]

anyone have experience with hypernetwork training? my log images are just coming out black

[signal nest]

even if AGI appears.. i dont think they'll make robots capable of replacing all manual labour. (either the amount of robots or the quality, or both)

[signal nest]

ironically i'm drawing far more now, to feed "img2img"

[drifting wind]

Best I could with the ten minutes I had, lol.

[drifting wind]

https://imgur.com/a/8sA6Vbh

[silent locust]

Thank You very much for your Time and your try Haha the first one is Cool ! 😁

[silent locust]

Can i know the prompt ?