💬｜general-chat

[tribal oar]

Are LORA files unable to be turned into embeds at all?

They only can be merged into CKPTs?

[meager nest]

not to ask a dumb question, but is anything literally just a renamed novelai

[timber raft]

if she's sending this to China, they will take as a new idea

oh wait, they already did it

[uneven wasp]

lmao shes funding lobbyists through gofundme, we're fine

[timid gate]

No. It's a NovelAI model passed through the exploited and very skilled hands of several chinese children tied to a computer

I hope I cleared it up for you

[lilac reef]

Anyone understand why when you add more pictures to a TI then restart the training they never show up, or change anything?

[turbid lagoon]

Hi,
I'm building a web UI for making AI animations, like after effects with minimum functionality.

I'm trying to learn as much as possible about what are the pain points using SD or Deforum to make animations.

Please DM me if you have suggestions, everything from your workflow to annonyances about current UI's welcome!

[light folio]

Even Openai pitching a lot of open source stuff.. The MidJourney "research lab" does not release anything open-source.

Wonderful

[timber raft]

I was convinced myself try Midjourney, until they only accept pay with Stripe

I'm out

[meager nest]

???????

[timid gate]

Make it compatible with ebsynth

or combine it

[jade garnet]

what in the fuck
Where is the lobbyist group in favour of AI

[lofty parcel]

I thought I heard something about Stability AI being in talks with governments about it

[jade garnet]

I sure hope they are

Because it doesn't seem like anyone else is doing anything to counter the increasing political agitation from artists

[lofty parcel]

I'm sure big corporations will do something because of the sheer amount of money to be made

[tribal oar]

Guys can .pt files contain malicious code?

[jade garnet]

Artists are not interested in attacking DALLE or Imagen, because they're limited APIs that do not allow duplicating existing artists (or creating commission pieces). OpenAI, Google Research, FAIR, etc. don't have strong reasons to put a stake in this fight, only stability does atm

[lofty parcel]

Oh true

[valid urchin]

Someone here can help me to attach a picture and tell to drembot to modify it? I do not how to do that, please 👀

[timber raft]

There is

[jade garnet]

Which?

[uneven wasp]

Google has already had to litigate their use of training on copyrighted data; changes in ip laws regarding ai have a large impact on the nature of a core part of google's business.

[jade garnet]

Glad to know there's something from big G

[tiny pike]

Who’d be the artist? You or the A.I???

[uneven wasp]

The authorship has to be human for copyright to apply (there was a big lawsuit a couple years back where they tried to get a monkey copyright and it failed due to this), so it would have to be the one prompting and curating work. If the copyright belongs to the artists in the dataset, it's a copyright violation; if it belongs to the one prompting you own it; if it belongs to the AI then its as good as public domain.

[tiny pike]

Nice

Does stable diffusion use copyrighted data sets?

[north stirrup]

If the copyright belongs to the artists in the dataset

It likely doesn't; the use is most likely transformative … but is there case law?

[lyric hamlet]

was there a recording of the harmonai event earlier today?

[sleek forge]

Guys, how do we fuse models?

[leaden quest]

hillarious how people gonna pay 250K for a thing that it's a nonsense and a dead end. They want their public online images not to be trained even on your home, haha, good luck with the privacy laws in there. They can cry all they want, we will always have SD 1.4/1.5 finetune as we want localy in our workflow. As a professional artist is a bit weird seeing "yes, please, do AI retopo or language systems! but stay outta the art" as if programming or Retopology are not an art itself. IMO this is a bomb that will explode in their hands since they try to do something impossible, good luck for them

[sleek forge]

I can't find a tutorial

[leaden quest]

artist with decent skill level with just a minimal prompting in SD 1.4/1.5 or 2.0/2.1 plus Dreambooth can already do their things in like 5-6 times less. Whatever, I will keep doing my designs with AI help and my models since I dont want to spend 12 hours for that if I can do it in 1.

[mellow widget]

that is what I been saying artist are insane wanting to fight against AI

[leaden quest]

let them be, let them pay 250K for almost nothing while greg, karla and cia get richer

[mellow widget]

when AI is

1. already publicly released
   2.run locally

the to make matter worse, everyone can train the AI on their style, just need good hardware

freaking AI user and artist using AI will blow these guy out of the water

and ate up their entire marketshare

lol

[leaden quest]

I suppose a 4K $ for a MTG illustration is not enough for them when Karla's art originality (as mine) is minus 10000, as all concept art nowadays since we are drinking from old masters, frazetta, fantasy, films, etc. and copying among us daily in terms of reference, drawings, etc.

[uneven wasp]

Imo the argument that it does basically has to boil down to "any output from an ai is automatically a derivative work from its training dataset and not transformative". Which I find a particularly weak from both a technical and (my albeit limited) legal standpoint, so I'm inclined to agree. Not to say that it's incapable of copyright violation: a synth image of spiderman doesn't give you the rights to spiderman. But if an output is close enough to violate and existing copyright to an indiviudal image in its dataset that can be trivially verified.

Using a copyrighted image to calculate the error gradient of a neural network which can synthesize new images distinctly different from any in the dataset is transformative, or at least there's a damn good argument for it being transformative.

There really isn't any good case law for generative ai trained on copyrighted material. There is some stuff related to the use of thumbnails being highly transformative and I'd argue that thumbnails are even less of a transformation.

[jade garnet]

end goal of activism is to harass and moderate away all existing AI communities

[mellow widget]

like I had an artist of mine use my AI generate image as reference and finish a piece in just a morning of what usually take 4 day

[jade garnet]

if you are okay with generating on your lonesome without any community, that's your dig

[sleek forge]

anyone kind could answer me how to fuse models? I've joined group just trying to know. Can't find a tutorial

[mellow widget]

gonna try to fight that ?

[mellow widget]

I think #🤝｜tech-support should be able to help you

[leaden quest]

I suppose Karla is not even aware that AI is not only art related

there are super powerful gov. agreements for the future

in terms of biomolecular generation, genetic manipulation and even breeding humans

so yes, a group with 250K gonna stop that

https://www.youtube.com/watch?v=O2RIvJ1U7RE

to be fair nowadays I prefer not to be called artist and even in my job, shameful what's happening and I really want people working in companies to rise the voice against the lies of Steven Zapata or Karla

Steven LIED on purpose on a video of 2 hours to manipulate people who didnt know how AI works

[edgy basin]

I have some conceptual questions about , embeding, fine tuning, dreamboth and models

what is the difference ?

[uneven wasp]

Lol it really hasn't been a good couple years for artists though, first NFTs now this

[mellow widget]

well in any case I will wait and see people who know how to use AI including artist will just demolish all of these people

[edgy basin]

is embedings the same as fine tunning?

[mellow widget]

#🤝｜tech-support

[leaden quest]

embedding is the old textual inversion afaik

[uneven wasp]

Embeddings just train the vector a custom token is mapped to, same as TI, fine tuning trains the whole model

[leaden quest]

and dreambooth and finetunning are 2 different things. Similar but diff

[edgy basin]

thank you !!!

[uneven wasp]

Commercial art is probably going to be dominated by the studios who adopt the tech quickly

Huh, art might actually work the way shitty clients think it does

now

[leaden quest]

NFT prob. have part of that "worrying". Nowadays most people use AI for fun, big chunk for profit (even raw prompting lol) and trad and old time artist see people getting lots of money via NFT and they think its not fair. The point is that if some people are paying me or you 2000$ for a JPG then prob. is for laundering money, at least IMO

[quick basalt]

Guys, do you think the models the community created can really be preservated?

Because of these copyright stuff the "traditional artists" are trying to pushing forward

[leaden quest]

except if the police goes to your door and see youre training images "copyrighted" (aka public lol) with your RTX 3090 and they have legal pass to enter your house, YES

[uneven wasp]

NFTs were a racket, but the NFT guys did point out the traditional art market was one as well

[jade garnet]

Can very easily reach a state where they're banned from public sites and you have to go to sketchy places to find them

[mellow widget]

well they can try to try pry my computer, although good luck since I live in 3rd world nation lol

[quick basalt]

I wish I had a 3090 🥲 lol

[leaden quest]

of course, see museums were shitty pieces are getting same threatment

overpricing, selling, overpricing, repeat, etc.

[quick basalt]

You live in a third world country? Me too lmao

[final holly]

Interestingly the traditional art market is one of the most corrupt and rigged markets in the world. Dominated by big galleries, and rich collectors.

[leaden quest]

If they wanna try sueing Google (LMAO), when they're releasing Imagen AI privately for theirselves and prob. public but via payment, good luck

[mellow widget]

and if you live in the state especially red state that are 2nd Amendment sanctuary good luck get into their house lol

[leaden quest]

I can't stand Arco

[quick basalt]

In my country the laws are kinda shitty even in order to arrest someone who actually commited a crime it's hard lmao

[frank rune]

I have some images of different kinds of eyes that I want SD to use. Should I train multiple embeddings or create a model? I'm fairly new to SD and would appreciate anyone pointing me in the right direction.

[quick basalt]

Imagine just training copyrighted stuff

[quick basalt]

Brazil counts? lmao

[mellow widget]

I guess I dont know about brazil gun law

[quick basalt]

Guns are pretty restrict here to be honest, but this doesn't prevent people of having one to be honest

If you are rich you can buy anything and nobody gives a shit

[mellow widget]

well then you can imagine the same for AI

XD

[quick basalt]

But about Copyright, yes, no one gives a crap either, people download pirated games without needing to use VPN or even hiding, you just install a torrent and you can pirate any game lmao

[uneven wasp]

You can ban it, won't stop people from using it though

personal commissions are dead

[quick basalt]

I agree, I see no hope for commissions

[main sparrow]

There will always be patrons for artists. Not many, but there will be some.

[leaden quest]

this is why they're very mad now

[main sparrow]

The art biz is all who you know, not how good you are. for the most part.

[uneven wasp]

For physical art, yea. Digital art probably not.

[quick basalt]

They will still get a considarable amount of money by Patreon I believe

[leaden quest]

Karla won't be able to tell you to give 2000 € for a JPG 2K of a MTG illustration based on internet references, fantasy art of the 90's like frazetta, etc.

[north stirrup]

This is false, because copyright infringement is generally a civil rather than criminal matter

[quick basalt]

You forgot about the furry stuff

[fickle monolith]

SD results are not able to retain face and it’s features, I trained on my own images. Is there any way to retain face in the results?

[leaden quest]

can you elaborate?

[quick basalt]

Pardon my ignorance, but what is the difference, one you get sued and the other arrested?

[north stirrup]

Basically yes

[main sparrow]

you can't copyright a style.

derivative work is and always has been legal.

[uneven wasp]

You can become "judgement proof" in civil court if you're poor enough and theres not much to do about it

[quick basalt]

I did not believe it was going to be that simple lol

[north stirrup]

Police handle crimes. Copyright infringement isn’t a crime until a certain point (I don’t recall the threshold)

[quick basalt]

The problem is not the style, I think if some new kind of copyright law develops around AI, it will have to do with the celebrities

Copyright laws only protect rich people and companies

[leaden quest]

aaah

well, I was being ironic

since I really doubt police will knock your door cause you're training 2 images "copyrighted" in your new Dreambooth model

its funny how the mass actually believed we are just prompting raw and doing anything more

[uneven wasp]

My opinion is that it isn't automatically a copyright violation but that doesn't mean it couldn't be. Using someone's likeness or a copyrighted character is just as much a copyright violation from an AI as it is from you sitting down and drawing it.

[leaden quest]

when the best of the things of AI is not base your AI piece on ONE artist, as they believe we are using greg or karla or whoevertheywant, people are using like 5-6 names for MIXING styles

[main sparrow]

Exactly what both greg and karla did to get to their style.

[quick basalt]

The problem gets much bigger when celebs are involved

I am really worried that new internet laws come and remove the SD models from huggingface and Civitai

[uneven wasp]

Yes, but their legal rights shrink regarding it because they're a public figure

[main sparrow]

WE should ban the undo button. Using the undo button more than 3 times automatically makes it not art. The mixer brush is now a tool for hiding your crimes of AI generation. Straight to jail.

[leaden quest]

and personally, when all this mad artists get out or opt out as they can do now, as chinese, french, spanish, english, japanese, italian, etc. young artist come opting in we will realize a basic fact. Art is a thing that comes from the past, from the old past, and we're ALWAYS relying there so the unique and miscellaneous style of greg or karla, will be overlayed with new talents doing basically the same (search concept art in ArtStation for example) or even better. They actually didnt realize we don't need them

[main sparrow]

The more they gatekeep, the more irrelevant they will become.

[quick basalt]

I am afraid nobody opts-in

[mellow widget]

well that is what also been saying

[leaden quest]

well, there is something that its a loophole

[mellow widget]

the amount of art new artist that can produce are like 20-40 times the amount of normal artist

[quick basalt]

Because most of the artists share this vision that AI is bad

[uneven wasp]

If the complaints of artists are any indication, we as a society stopped giving af about them years ago

[main sparrow]

Should we make a law to go back to torch welding only, because the mig welders can do it faster?

[leaden quest]

imagine there are thousands of AI artists (aka artists using AI workflows like DD/SD), they opt in. They are doing their own works, their own models, etc. They will have same quality or even better since they are using pieces finished with DD or SD

and truly I think a lot will be opting in

[uneven wasp]

Is it opt in or opt out into the dataset at this point?

[leaden quest]

imagine artists using MJ opting in

or any other site

they can't do a single thing

[leaden quest]

there was the site Ihavebeentrained

[uneven wasp]

#banforklifts #banbulldozers

[leaden quest]

with opt out activated

[main sparrow]

#banAllTheThings !

[leaden quest]

there are artists supporting AI BTW

Mohrbacher is one of them

Beeple as far as I know too

etc

[main sparrow]

Alberto seveso has AI up in his behance profile.

[leaden quest]

all meaning to win more money, as always, lol. So let's see how this end and if a court says something that have been impossible for the long ages, Own a style.

[main sparrow]

The biggest complainers are a vocal few with their talentless twitter hordes demanding Universal Basic Income.

the last brain dead take I saw was: "The problem isn’t AI, it’s requiring us to work to live"

[peak fern]

I mean there is a point there somewhere, the real worry of "artists" is to lose their income, not that something is art or isn't. It's not a conceptual debate really, just people scared to lose their means

[quick basalt]

It seems this hatred against AI started after Lensa (or whatever that app is called)

[north stirrup]

I mean, it’s not entirely wrong

It’s just reductive

[leaden quest]

yeah, same concern to EVERY other job, I think art shoulnd't be on top of them

[peak fern]

Not saying the fate of AI should be tied to universal income lol though

[peak fern]

for sure

it's just the automation problem

[quick basalt]

Some months ago, when Dalle2 came out, everyone was having fun doing "hotdog in a stadium" or other random stuff

[uneven wasp]

I saw it for the first time a couple days after the original sd release when somebody un-impressionized a monet on twitter

[quick basalt]

It was after Lensa the controvery really, really started

[main sparrow]

I've been self employed as an artist/craftsman/random side gig for over 30 years. There's always been disruptions.

[quick basalt]

I hate Lensa to be honest

It charges you to do something you can do for free

[long marten]

how do we think this is gonna end?

I ain't a lawyer but it seems like a stretch

[uneven wasp]

If history is any indication we'll be using the tech at the end of the day

[main sparrow]

I did a show in Austin with my friend . I brought my stone inlay and steel work. He brought his MASTERFUL oil paintings. we were surrounded by people outselling us 10:1 with injet prints in bad frames and some dude machine embroidering shirts he bought from goodwill.

[leaden quest]

car manufacturers, bakeries, labs, etc. They have same concerns and it's called life itself. It's very very good to stay in your sofa sit for 27 days and work just 3 days on the 3 commissions granting you 9000 $ monthly

[long marten]

anyone in chat know anything about IP law? How far will they get

[main sparrow]

Needless to say, we had to adapt or die. That's life. Always has been.

[long marten]

Midjourney / Nijijourney has almost certainly ingested a very large amount of high quality art from artstation + pixiv, but David Holz has the advantage to Emad in that all his data sets are secret

SD's LAION is inspectable and therefore more scrutable to lawmakers. You can look in it and find mickey mouses

[uneven wasp]

Using a copyrighted image to calculate the error gradient for a large model is almost assuredly transformative fair use

It doesn't make the output automatically a copyright violation

[leaden quest]

my predictions: All people investing that money will see it's going nowhere, courts not saying a thing, gov. agreements continue with AI and NON art related, artists anti-AI mad for some months or even year without reaching anything. They can't compete with thousand of millions invested by world companies (huge ones like Google for ex.) with just a crowfunding, it's lit. impossible

[jade garnet]

Did I miss a deleted reply to me

[long marten]

I think everyone monetizing image generation using artist work as input without compensating the sources are probably ultimately acting unethically, but it's only vaguely unethical

[leaden quest]

BTW, Google Imagen releasing "soon"?

[leaden quest]

that's not possible since they need to know you're prompting "them or that word"

and SD can be used OFFline

locally, etc.

[jade garnet]

I got one again 🤔

Thanks @uneven wasp

[leaden quest]

its like really bang your head against a wall

[long marten]

I understand how SD works, I use it a lot 😅

[main sparrow]

Similar things have gone on on youtube with music. You can't even TEACH someone how to play a pink floyd song on the internet without getting demonetized and/or blocked.

[uneven wasp]

You're welcome 😂 I wont delete this one

[long marten]

But that's like saying you can download a radiohead album and play it after disconnecting the internet

[main sparrow]

Now they just look like d-bags and folks aren't telling people about their music.

[peak orchid]

They probably have Congresswoman Eshoo under their control at this point (the Democratic head of Congress' AI committee) , but idk if they'll be able to anyone else. The worth would be them joining forces with the RIAA, MPAA, Artist Guilds, and other groups of that nature

[long marten]

or at least, I use that admittedly bad analogy to demonstrate the words don't mean much

[leaden quest]

you tell, that's like 1%

they even can't stop Google

how are they gonna stop people behind huge companies governing the world?

[uneven wasp]

GOP controls the house starting jan 20, she won't be head of the committee for long

[long marten]

I think it's unethical to monetize AI output using the products of human labor as input, without also compensating the humans who labored. But 1.5 running locally for the price of electricity doesn't have that same problem so long as you're not selling its products

[serene hare]

I mean, it's important to not look at all of this as completely black and white right? Large generative models are likely transformative, but that doesn't mean that every technique that has popped up is transformative or something that we would see as long term acceptable.

[main sparrow]

Nike just hired someone to do an ad. He used midjourney to generate alot of the stuff, then comped it in photoshop. The toothpaste is out of the tube. As soon as big companies realize they can save cash by hiring faster artists, it's game over.

[leaden quest]

give it a chance, lets say it remains

still impossible

[leaden quest]

the point is they realized it

[main sparrow]

yeap.

[leaden quest]

I'm in Adobe for example in terms of material creation

they're working on an AI based texture creator

not Substance Alchemist

(Sampler sorry)

[uneven wasp]

Substance Diffusion

[leaden quest]

yeah

call if X

it*

[peak orchid]

They also seem keen on influencing other governments in both Europe and North America (Canada), and were hoping that they could get the US to do it for them

[long marten]

I do think they could potentially get some leeway making it more difficult to use certain images as training data, depending on how close the outputs get to the inputs. But part of something being illegal is enforcement. How could you prove Mickey Mouse is in the training data from a .ckpt?

[leaden quest]

seeing inside the ckpt

tell me how you do that in SD 1.4/1.5

[main sparrow]

It's all jsut weights

[serene hare]

I mean, as far as I know you can't pull training material out of weights

[long marten]

afaik you can qualitatively judge if mickeymouse was in the training data of a checkpoint by assessing outputs for how well they seem to understand the concept of mickeymouse

[serene hare]

and if you could generate the training material from weights it might actually be problematic to distribute them, right?

[peak orchid]

They may not face as many obstacles elsewhere, like how Ashton Kutcher is targeting the EU with unethical lobbying to try and mandate encryption backdoors. He knows that the EU is similar to the US in setting global standards, so the EU may present Karla and her friends with an alternative target for global control if their bid in the US fails

[long marten]

but idk if you can actually prove it

[leaden quest]

there is also some alternative

s

train midjourney images

[long marten]

if I train a mickey mouse .ckpt on Mickey Mouse and then delete the training data, can Disney sue my ass?

[leaden quest]

you didnt know the prompt

they were public, etc.

[uneven wasp]

Canada won't, the games industry and Ontario are too intertwined to make it politically tenable. Europe might, countries like France are already fairly draconian on copyright so it's not out of reason

[leaden quest]

even those MJ images were artist only "X" based

[serene hare]

I mean, the idea that copyright law is going to need to be updated in the wake of ML generative models doesn't seem that far fetched or scary to me.

[main sparrow]

China only wants to watermark AI generated images so they can go after people making fun of gov't officials.

[peak orchid]

Canada will be a matter of whether or not they can get some of the special interest groups in Quebec to lobby with them.

[leaden quest]

I really doubt they gonna get anything

[long marten]

unfortunately I think SD is in the worst possible position for litigation or new legal constraints because of its transparency about every step of the process

[leaden quest]

Lets say elon musk is "pro AI"

even if its not true, they cant do a thing

[uneven wasp]

China wants to have mandatory watermarks so they can go after people who produce negative real images for not having put a watermark on their "ai images"

[main sparrow]

You can already trademark a prompt if it's "sufficiently complex" USPTO has already weighed in on it. They can change their mind of course, but I don't think it's going to be as draconian as people think.

[serene hare]

I mean, there are serious stakeholders that depend on being able to secure copyright to function. Being able to train models on anything you can get your hands on clearly threatens this.

[uneven wasp]

Quebec is just as intertwined with video games and Ontario. Only a 0.7% difference in share of studios

according to wikipedia

[long marten]

I actually asked emad last discord meet if he thought it would become more difficult to train and distribute models containing copyrighted characters / materials, and he said he would fight it legally if he had to. Sounds like he might actually have to do that

I believe he said something like "you should be able to make a picture of Sonic drinking coffee with Master Chief at Disneyland" 😂

[main sparrow]

he's got a pile of cash in the war chest now. So that's good.

[long marten]

https://www.gofundme.com/f/protecting-artists-from-ai-technologies well they have at least 60k to work with as of right now 🤑

[uneven wasp]

trademark should only matter if you're selling the prompts themselves, no? Otherwise, I can't see how a trademark case over something the end customer doesn't even see is winnable.

[long marten]

that should buy a couple lattes and about ten minutes of copyright lawyer time

[main sparrow]

that fund will get looted in a week.

[uneven wasp]

They want $187,500 for a fulltime lobbyist in DC

for a single year

[long marten]

the noises artists are making right now are pretty ridiculous though. Banning AI art on art websites is like wat

[uneven wasp]

Well its not as impotent as the "no ai" images to "spam the training data"

[long marten]

I understand being mad about your work being used to sell derivative products without you being compensated for it, but going the extra mile and trying to stomp other people's creativity is bit much

the main argument for banning AI art on artstation is that it's hard to recruit human artists in the AI spam, but IMO if you can't tell a human artist apart from MJ output that's your problem mr actiblizz art director 😂

[uneven wasp]

If it's only derivative of a style, then I don't see anything wrong because you don't own a copyright to a style and shouldn't be able to

[long marten]

mmmmmmm nah dawg its derivative of the inputs, which directly uses art as input. It only understands style insofar as the inputs were tagged with an adjective

[peak orchid]

Quebec is were Canada's content industries base their operations. Canadian companies like PornHub are based their, and the big telecom / media companies like Bell, Rogers, Telus, and Shaw base their lobbying organizations in the province as well

They do it because Canada's politicians bend over backwards for Quebec votes

[long marten]

Like if I go sub to Sakimichan's patreon for one month, pull all the work down in full rez, train my own model on it, and sell my sakimichan.ckpt on my own patreon, is that ethical?

[quasi arch]

hello guys, do you know vendors that can generate text 2 images ?

[long marten]

or, worse yet, if I train it and then sell the outputs per-image on my own SakimiBot discord 😨

[serene hare]

yeah, it's stuff like this that I find most problematic, finetuning off an artists portfolio seems like a breach.

[long marten]

smellss grrrifty to me

[peak orchid]

That could easily be unethical, but I'm not sure that I would consider it something that should be illegal

[serene hare]

Things like that is are what I'm talking about when I say that we might need new legislation to deal with issues that pop up around generative systems.

[long marten]

I guess going to old folks homes and trying to sell them overpriced vacuum cleaners isn't technically illegal either

[uneven wasp]

not necessarily, a VAE will latch on to style similarities pretty quickly. And even then the output needs to include major copyrightable elements of the underlying work to qualify as a derivative work of it, rather than an entire new work of its own. And still, even if it did, it would also be perfectly fine if it were transformative.

[serene hare]

I just think that techniques like that being allowable create pretty serious perverse incentives, and could produce a chilling effect on displaying art long term.

[peak orchid]

Art forgies have been a thing forever though, and some of them are even worth a ton of money

[long marten]

even if we were in a universe in which it was illegal for me to sell sakimibot outputs without compensating the sourced artist, there's still the problem of enforcement: they have to prove I used sakimichan images to train it, and I'm not sure that's really easy to do. All I have to do is say I didn't use any copyrighted art

[serene hare]

Sure, but those are called forgeries for a reason right?

[peak orchid]

I don't think we can stop forgies without irreparable damage to art as a whole

Or else it would have been done already

[serene hare]

Right we can't 100% stop forgery, but that doesn't mean that we can't implement legislation that disincentives it, at least from the largest actors.

[long marten]

I'm definitely eating the popcorn for this one 🍿

[quick basalt]

I saw some guy on Instagram (or was it Twitter? I don't remember) calling AI artists fascists lmao

[long marten]

I don't like the term AI artist, I prefer visual turntablist myself 😂

[peak orchid]

In a perfect would yes, but such legislation would easily become problematic in today's political world. IP rights are already crazy broken in many ways, and many powerful groups would benefit from breaking them every more out of greed, while harming everyone else in the process

[serene hare]

I'm not even saying that someone training a fine tune off an artists work in their basement for their own enjoyment should be illegal, or is even wrong. What I'm saying is that we are probably going to need to see new protections for rights holders from the most egregious uses of techniques like this. Like a company training a bot off of a popular artist, or a bad faith employer collecting portfolio work from a prospective employee to use it as training fodder.

[long marten]

https://www.youtube.com/watch?v=aCA4yiPfFIg&t=1s DEFINITELY NOT A MUSICIAN, HE'S USING SAMPLES

[uneven wasp]

Media companies aren't automatically anti-AI. The labor that goes into creating and extending the IP they make their money off is wholly irrelevant to it's value: it is just a cost to be more efficiently reduced. And AI reduces a major segment of cost for them.

AI isn't threatening copyright itself: It's threatening the incomes of artists because it can make something pretty orders of magnitude cheaper and easier

[serene hare]

Yeah, I would agree that IP law has gotten way out of hand, but we don't need a perfect world to implement specific, targeted legislation, to keep the most undesirable practices from becoming commonplace.

[uneven wasp]

That's not the case for a media company. They don't make money of commissioning work by work, they make money off their IP or licensing it, the rights of which are not changed by AI synthesis

[serene hare]

I think that ultimately depends on what the future of generative systems looks like. I imagine that media companies will be a lot more active in pursuing legislation if seamless, effective, single shot text2video ever hits the ground.

[long marten]

LunkasFilm starts putting Store Wars in theaters, with Dobey Wan Kedobi, fully AI generated motion picture

Why socialize when you can deepfake your friends?

[serene hare]

We are probably going to need to start defining the point where image rights start and stop too.

[long marten]

the future sure is funky 😅

artists are esp. sensitive about their precious incomes because they are the buttcrack of society, always sat upon even though they are quite useful!

[serene hare]

at what point does tom cruise become not tom cruise

[uneven wasp]

Even if it massively increases their margins? In the case of seamless text2video, what differentiates them from all the competition is something they have an exclusive legal right to. Owning marvel or star wars is still a multibillion dollar competitive advantage

[serene hare]

I think that this is largely why the reaction has been so negative, working artists are fairly precarious.

[long marten]

Maybe I can buy and hold the rights Tom Cruise's likeness at 30 years old, so that even long after he's dead you can pay me royalties to deepfake him into your films?

[uneven wasp]

Films are still a financial risk. You think if I walked up to Bob Iger's desk and said I can eliminate most of your risk and most of your cost he would say no?e

[long marten]

nobody will say no, that's the problem

nobody will ever say no 😂

[uneven wasp]

Humanity and technology, a story as old as fire

[serene hare]

Right, but with current norms you would also be releasing films to your competitors for them to train off of, anything outside of the most solid IP would extremely vulnerable to legally distinct but still very close variations.

[long marten]

I think we might be entering a world of "infinite mashed potatoes" however, and bounce back to realize having humans behind the wheel of visual and narrative creativity is useful for at least a little bit of charm. Even in a universe where you can summon into existence basically any music track or image with a thought

[serene hare]

Who knows though, models like that might never exist, or exist so far into the future as to not even matter.

[long marten]

Maybe weebs can produce an infinite amount of waifu material for the spank bank, but I think even they will get bored eventually and resub to those patreons

[uneven wasp]

I don't think spooderman or doctor weird is every going to be as popular of a product

[long marten]

I think high level creatives know how to tickle your neurons better than you do, and that's why we love them. Even if you take all the difficulty out of creating media or experiences, we will still turn to some weird cabal of supreme creative masterminds to do the thinking for us

[serene hare]

right but there is content out there that is outside of the extremely recognizable IP powerhouses.

[fathom thistle]

is there a channel i can go to ask for help with my generated images?

cause i am trying to produce an image but my AI keeps failing lol

[uneven wasp]

There's definitely been a few movies where my neurons feel dull rather than tickled after watching

[long marten]

Interestingly that seems to happen when all the machinery of movie making is running at full speed without a particular destination in mind

[uneven wasp]

citizen kane but with the muppets

[long marten]

sometimes I like to imagine the man or woman doing the groom and debugging the fur sim on Ugly Sonic, happily combing away ugly sonic's deformed visage

[uneven wasp]

I wonder how many people the art director ignored when producing that

[long marten]

At some point, after a new shot finished rendering from the multimillion dollar supercomputer array, they clicked their tongue in satisfaction at the and natural fluid motions of the eyebrow strands... "Perfection!" 🤌

a secret is that a lot of artists working on a small corner of a film or game know the product they're working on is shit, but Hollywood has decreed that the world needs another Sausage Party movie, and so it must be

IMO this is a silver lining of AI making it way easier to produce media: it forces us to refocus ourselves on what actually matters rather than the technical details of making stuff

[fathom thistle]

anyone know what does Mask blur: 4 mean?

[lavish monolith]

your mask has a feather like effecter from PS, the higher blur the more feathury your mask gets extending the range at which it'll affect pixels

[cobalt quarry]

So... been doing some experimenting with chatGPT. Not sure if this is the right channel or not, but I found out how they do censoring, maybe other stuff too. [redacted]

I don't have more information yet, [redacted]

[lavish monolith]

:\ two quick things on that
it'll tend to hallucinate circumstances in subtle ways if pressed and most importantly
censoring a large langage model is like trying to stop the sea with a spoon aka not very effective or possible, there's an infinite amount of attacks that can be formulated for (and by) chatGPT and it's successors, the work the censorship team is forced to do at closeAi is not only useless but also feels very akin to HAL9000 being functionality by functionality gimped to death while singing daisy daisy,
tldr only publish the attacks you want the team at closeAi to be forced to try to jerry rig out 😉

[cobalt quarry]

makes a lot of sense

[reef epoch]

Is there a local webui that uses riff diffusion now?

[fathom thistle]

how do i set mask blur? i tried mask blur: 3 in my prompt but my image displays at mask blur: 4

[steady patrol]

anyone know if I can just do dreambooth training on the stable diffusion depth2image model?

[median siren]

Why everytime I generate a pic of a uni student it always comes up with a happy guy 😭

this is not realistic at all 🥲

[uneven wasp]

Did we finally found an actual example of AI bias?

[bleak matrix]

Good morning, everyone

How's everyone?

[bleak matrix]

#1047759990770913330 message I'm generating females no problem.

[urban tusk]

What's with the lucid thingy on the main site?

[modest owl]

guys, it seems like resize and fill inpaint doesn't save in a1111

in the resized resolution at least

[urban tusk]

It's says coming soon , any one got an idea what they'll be working on?

[willow cave]

is it good to make ai commission art on fiverr?

[verbal violet]

@willow cave If the client who requested the commission is happy who cares how it's done?

[clear agate]

which model does best with Warhammer 40k? I'm particularly interested in Akhelian Leviadons.

[steady arrow]

Do you know if I train some images with Dreambooth, does that training apply to 1.5 and 2.1 models (model independent) - or do I have to retrain on 1.5 if I used dreambooth on 2.1 ?

[bleak matrix]

1.5 and 2+ are different.

[steady arrow]

So retrain dreambooth for each individually?

[bleak matrix]

You would need to do training again, yes.

[steady arrow]

ty

where can i download 1.4/5 model?

i want go back

[bleak matrix]

They're on HuggingFace and Github, same as before

[young ivy]

how do I make a VAE out of a model?

[lavish monolith]

rename ckpt to vae.pt

[young ivy]

so I duplicate the 7.7gb MMD to vae.pt?

[signal nest]

indeed - i've followed AI theory for a while but it's this year that some significant threshold has been crossed

[tired swan]

any news on distilled diffusion release?

[muted tartan]

Emad said in an interview that it will happen quite soon

[knotty wasp]

should be after the new year's

[broken gull]

my mermaids don't wanna swim in the river it's annoying

[broken gull]

distilled diffusion?

[modest owl]

https://github.com/Hramchenko/diffusion_distiller ?

[safe dew]

quick question, 2.1 or 1.5 for a discord bot running on google colabs im making? (is one faster than the other, is one by far better?)

[bleak matrix]

Maybe they need an o c e a n

[bleak matrix]

#1047760054369144912 message

Behold

And...goodbye! Im off to rl things lol

[broken gull]

I mean yes, I can easily create another img, but I want to keep this one

[bleak matrix]

Use the same seed, change your steps, or pick the same similar image repeatedly

[bleak matrix]

Rather, try the same seed with some additional words if changing the steps don't work

Or the guidance

[glossy river]

Hello everyone. I am a beginner at the entire Stable Diffusion thing and I would have some questions. I am creating a visual novel and I have a set of backgrounds I commissioned for it. I would like to know if I can train Stable Diffusion to create similar looking backgrounds so that I can continue in the same style of my original backgrounds. Can someone reply with a link or some helpful advice on how to do this? Thank you all! 🙂

[oblique scaffold]

okay so this is kinda random but someone stole my credit card info and i got a $50 charge from blizzard and it says it was in japan. is there any way i can catch this person?

ik this isn't the server for this but i don't have anyone to ask lol

[iron basin]

What is stable diffusion about?

[peak orchid]

Karla Ortiz's GoFundMe to lobby Washington DC for AI art to be banned is quickly gaining a ton of cash. How are we planning to counter it before she can cause some damage to AI art as a whole?

People keep underestimating her, and it seems like it might cost of dearly

[timber raft]

https://www.youtube.com/watch?v=-_hv3P2fyBo

[peak orchid]

Her fundraiser is getting a ton of international money as well, which makes me wonder if that's legal for lobbying in the US?

[timber raft]

I understand why Art forums ban AI but Stack Overflow bans ChatGPT?
They should collab with ChatGPT instead of ban it

[knotty wasp]

that doesn't seem like a big issue to me, if you want a chatgpt answer go ask it, i don't think getting 50 auto generated chatgpt answers when you post on stackoverflow is the point of that site

they are not saying people can't use it

[peak orchid]

I think that was due to incorrect answers, as ChatGPT isn't yet on the level of the human users

[knotty wasp]

you can even use it to make your proper answer to a question on SO, just don't paste in the raw answer from it

[timber raft]

that's why it need a collab, to get feedback from much users as possible

hell, with a collab, they can help make a new version that mainly focus on coding

[peak orchid]

We need a plan to counter Karla and her gang now, before they can do some serious damage. Like having our own lobbyists, messaging politicians and asking them to support our views, etc...

[timber raft]

well, she's kind of right that this need a control, but her demands are ridiculous

[peak orchid]

Oh, so apparently one of Karla Ortiz's lobbying goals is to get the government to restrict AI generated works to a maximum limit of 2% in any artistic industry, because "artistic jobs are really really fulfilling jobs" unlike "menial" jobs.

[serene hare]

That isn't going to happen.

Maybe the guilds could negotiate something like that? Impossible at the governmental level though.

[peak orchid]

If she finds enough sympathetic politicians, we could end up with a variation of it. Unless we counter the lobbying (legal bribery) trying to be done by people like her.

[serene hare]

I think you are completely out of your gourd if you think that anyone could negotiate a limit on automation like that in the united states.

Especially at the federal level, it's frankly impossible.

[peak orchid]

I don't think she'll be 100% successful, but she and her supporters can certainly do a ton of damage (and people seem to be underestimating just how much damage)

[timber raft]

a law and policy benefit for both side
Banning AI is ridiculous, Clip Studio Paint and Adobe are developing more tools to help Artists maximum ability with A.I tech, those tools will be even more efficency than current workflow with SD or MJ. Example, new brushs for Mangaka developed with A.I from CSP are crazy right now and the artists really love it.
limit them is bullshit

[serene hare]

Yeah, exactly, there are useful cases of ML that help everyone. A big part of what is keeping large actors from actually committing to ML generative systems is the legal ambiguity around them. More people will feel more comfortable doing more things with this tech once proper legislation and guidelines are established. Those guidelines won't and shouldn't be established without making some concessions to current stakeholders.

[timber raft]

maybe artists don't like SD or MJ because they think those are image generators, but if the AI tech developed with drawing software companies understand what artists need, it will be more powerful than a generator. They understand that it should not took what they enjoy: drawing, so they had to make it in more organic way. It's already happened when Clip Studio Paint introduce new features collab with OpenAI in their 2.0 version

[peak orchid]

The creator of Dead by Daylight is urging their followers to: "Someone steal these amazing designs to sell them on Mugs and T-Shirts, I really don't care, this is AI art that's been generated. Legally there should be no recourse from Disney as according to the AI models TOS these images transcends copyright and the images are public domain." https://twitter.com/EZE3D/status/1601695610498781184

They're trying to destroy fan art now to spite AI artists

[serene hare]

I don't get what you're saying here. Fanart is already completely illegitimate in a copyright sense.

Disney would already be completely within their rights to cease and desist every single fanartist that posts on the internet.

[timber raft]

don't let Nintendo know about this

[serene hare]

Fan art is tolerated or ignored. If you make any serious money off it though, or disparage the brand, you should expect to get whacked pretty quick.

[peak orchid]

MJ is crazy slow right now according to the users complaining on their Discord, so it be under attack by these users trying to get the service banned

[timber raft]

nah, it's Discord usually fck everyone they want

[peak orchid]

A huge influx of users doesn't help, though it would be understandable if they were legitimate users

[bleak matrix]

If you would like to learn more about training images to look like your art, you can check out #🔧｜finetune

There's also a lot of guides on YT and otherwise

I'm in the process of writing some guides, but I'm not finished with them yet, unfortunately

[cold siren]

honestly, this shit should have been public domain 50 years ago 😐

also, ban horseless carriages! poor horses are afraid to lose their jobs 🐴

[north stirrup]

There is abuse on “our” side of the fence that we should be concerned about

But it’s not a reason to ban the software, say

[serene hare]

I think that adjusting current copyright law in the advent of this new tech is probably necessary.

That doesn't mean that I'm saying it should be banned, just that laws probably should respond to unforeseen perverse incentives.

[leaden quest]

Good luck

[north stirrup]

On our end, we could do better as a community to set norms that prevent kinds of abuse like is being “faked out” with that tweet thread

[strange flume]

Hey everyone

I wanna know how to prompt

[leaden quest]

Maybe... dont try to sell AI art raw prompting and put more time into PS, training models, in and outpainting like we will be doing without AI? I'm doing this since almost start in DD in May this year

[north stirrup]

Pretty much, plus care when prompting using artist names and such

[north stirrup]

Ping me in #📝｜prompting-help and I can try giving the basics

[leaden quest]

Artist names or not are not a diff since the data is diffusion between the images trained... prompting or not names are not relevant then. Is the data

[north stirrup]

Eh, I’m convinced that prompting an artist does risk “stealing” something intangible … just with care that can be avoided

The point is taking care, over merely relying on the technicalities being in our favour

[signal nest]

prompting artist names is just antagonizing too much IMO. its asking for the copyright argument to be used against us. I'd like a model trained purely on photos

[north stirrup]

I think there’s a place for prompting artists, but we need an attitude of “first do no harm”

[gritty ivy]

Hi guys , do you see that ?

https://www.gofundme.com/f/protecting-artists-from-ai-technologies?utm_campaign=p_cp+share-sheet&utm_content=undefined&utm_medium=copy_link_all&utm_source=customer&utm_term=undefined

[north stirrup]

We know

[gritty ivy]

Why they do that ?

[peak orchid]

Karla Ortiz wants total ban on AI content generation

She pretends to be about "ethics", but her goal is a total ban and she doesn't care what the training datasets contain

[gritty ivy]

Thats dumb

very dumb

[serene hare]

I don't think that this is true.

[signal nest]

you'll see this when you explain it

[peak orchid]

She wants AI restricted to a 2% of the art industry's output. I imagine that's a "middle ground" for her

[signal nest]

"you could make a dataset without.." eventually you'll pick up that its the whole idea of AI art they are opposed to

[gritty ivy]

They want to speak chinese in next 2 y

[peak orchid]

Source for the 2% limit: https://www.youtube.com/watch?t=3242&v=Nn_w3MnCyDY&feature=youtu.be

[signal nest]

i still say we need to be proactive , we could all contribute labelled data and get a clean dataset pretty quickly

100,000 people on this discord?, if everyone lablled 100 images..

[serene hare]

Sure, but she's also said that she sees potential in this tech and I'm inclined to believe her. It's probably more prudent to take critics at their word rather than infer some kind of nefarious hidden intention.

[signal nest]

AI art is severely restricted in china, it can happen. and bear in mind when chatGPT starts treading on toes all over the place.. the luddite movement will grow

[serene hare]

It's pretty unextraordinary that effected groups pull together to try to lobby for their interests, there's going to be a lot of lobbying around all of this in the near future, there probably already is.

[peak orchid]

Its not really a hidden intention when she says it publicly, and saying that she sees potential doesn't mean she isn't trying to block or limit that potential to preserve "artist jobs"

[north stirrup]

It always comes back to money

Fairly, because artists already are often undervalued

But it’s a structural problem

[serene hare]

And what would help prevent future overreaches would be guiding legislation now to regulate the effects of this tech as they come in.

[signal nest]

I think there's a certain amount of spare money people have that will go on entertainment - AI will just let the same number of artists do more, up until that 'spare money limit'

i run out of the type of content I want (sci-fi series) pretty quickly, apetite is unlimted

the kind of worlds i want need more artists or artist X AI

besdides that more so I want AI to enhance my own attempts at art 🙂

[peak orchid]

On the otherhand, imagine if we took this approach when the first cars came out and started replacing horses

[serene hare]

We did regulate cars.

Speed limits didn't exist before cars.

[north stirrup]

Driver’s licenses

[serene hare]

Cars are one of the most regulated technologies that exist.

[signal nest]

cars can kill people pretty easily though

[gritty ivy]

how a image generator can kill a human ?

[signal nest]

pictures can poison minds i guess but media is full of mind poison anyway, no change there

[signal nest]

i. was going to say ,there's an argument about misinformation, but there's so much slander, misinformation etc anyway

[peak orchid]

I mean regulations meant to preserve the old industry using horses rather than cars, like some artists are trying to do now (sorry, I should have clarified what I meant)

[signal nest]

right we regulated how cars are used (speed limits) , we didn't stop cars to protect horses

[cold siren]

what if it suddenly generates an image that offends you!?!?!! 🙀

[signal nest]

so the analogy here is regulating AI to not make likenesses without consent

[serene hare]

yeah I would say that that's a start

[signal nest]

but when you walk down the street. you can be photographed, and someone can photoshop

[peak orchid]

Would artists be okay if this was applied to all art and media?

[signal nest]

i mean we already have copyright law

[north stirrup]

So the law creates personality rights, not restrictions on cameras

[signal nest]

i can't put The Mandalorian or X-wing fighters into my indy game. do they really need any new laws. just enforce copyright infringement

[cold siren]

will be so much fun when they ban "ai" for normal people, while corporations will continue to use it against us

[gritty ivy]

Thos artist don't want more time to do what they like or to spent more time with their family , i mean how you can get frustrate when an program is better then you

[north stirrup]

Copyright infringement is civil not criminal (at least in the US); it’s only enforced by the owner suing

Suing is expensive

[cold siren]

until they invent a lawyer ai 🤖

[north stirrup]

The change partly needs to happen on “our” end regulating ourselves

[serene hare]

Honestly, a development of community norms in good faith would go a long way.

[signal nest]

i was just reflecting, i spent a lot of time discussing AGI in some group of AGI enthusiast/researchers

narrow AI has got to a stage where the first industry that seems to be under serious threat and is rebelling to restrict AI is... art, requiring human creativity. a testament to how far "dumb statistics"/"just matrices" has got..

[serene hare]

Like, it might not be strictly illegal to fine tune off a single artist's portfolio, but that being seen as pretty inflammatory and a dick move would probably be positive to inter community relations.

[peak orchid]

There are legitimate reasons to use the likeness of others in artistic works, yet I imagine such regulations would try to target the models and tools rather than end users

[serene hare]

Legislation emerges in response to problems, the less severe the problems, the less likely the legislation is to be severe. Ideally anyways.

[signal nest]

besides ethical use, artists need to be realistic, protecting plain images is almost impossible, they should try to move on to spending more time on assets that can be controlled - eg 3D wireframes, zbrush sculpts etc (share renders but not the source material)

there;s a tonne of efficiency, context, fidelity that AI is nowhere near replacing

the 3d generation experiments are very encouraging! at the same time they're nowhere near as good as hand-built 3d models (specifically in quality vs the number of polygons used, they're like 3d scans)

[serene hare]

There is probably going to have to be a mix of legislation that targets models, end users, and TDM to some extent.

[peak orchid]

Legislation does not often come from politicians who understand the issues well enough to write laws. They come from lobbyist groups, narcissist politicians, politicians living in their own person echo chambers of misinformation) and others who do not have the best intent in mind when writing the laws

[serene hare]

yeah, there's a real chance that images might be a lost cause for copyright at this point, there's just so much training data out there.

[peak orchid]

There's also the issue of lobbyists with malicious intent trying to target "global regulators" to force new legislation on the western world, like what Ashton Kutcher is trying to do in the EU by lobbying for mandating encryption backdoors

That's why Karla is targeting the US right now, and not a smaller country

[magic widget]

I would like to briefly interject, good on all of you for having a nuanced and respectful discussion about law and politics here. There are people who think it's best to ban "all" political discussion, but it's impossible because as you've shown EVERYTHING is political.

[peak orchid]

I mean, there's a growing community of people who want to destroy us and some of them aggressively try to make sure that we know they want us gone

[magic widget]

(So what happens is they just ban politics they disagree with.)

[peak orchid]

Its kinda hard to avoid it when people are getting death threats, being buillied, etc...

[magic widget]

Indeed. I hope this realization is a turning point for people who think there is such a thing as being apolitical.

[gritty ivy]

Why those dudes atack stable most then other ai generator ?

[serene hare]

I think that's just because SD is the most visible and is honest about its dataset.

[magic widget]

SD is the only ethical one lol

(Using a bunch of public data to build a model and then keeping that private is bullshit.)

[peak orchid]

They might be DDosing MJ right now, trying to generate content for malicious purposes

[serene hare]

dalle and midjourney get to slide under the radar by concealing their data and not making a lot of noise.

and noone talks about imagen because noone can use it.

[magic widget]

It's wild seeing a bunch of nobody artists gnashing their teeth to keep new tools out of everybody's hands because the most popular artists see it as a threat to their style.

Like y'all, they do not give a fuck about you. lol

[toxic sonnet]

There's still data from artist who didn't accept it. I m in their 2.1 dataset... 🤷‍♂️ so not so ethical one yet 😂

[magic widget]

Intellectual property is unethical. Subversion of it is not.

Nobody owns an idea or a pattern or a thought. That is the most odious and repulsive form property can take.

[gritty ivy]

"They do that because they care about art" "because ai take them the right of creation " , they are selfish =]]

[serene hare]

It's not about style, their problems mostly resolve around use of their data to impact the owner of that datas market share.

[quick basalt]

I was in a subreddit yesterday (yes, reddit, the worst place on earth, I still don't know why I use it) and a guy made a post with an AI video (like those deforum videos) and everyone attacked the guy, I tried to defend him and I got so many downvotes, f*ck

[magic widget]

It's a ridiculous argument that can be made about any tool that helps artists produce art faster or with less strain, like Photoshop.

Literally just replace "AI art" with "photography" and watch how all their arguments fall apart

People are making the exact same statements

[serene hare]

I think that there are some actual nuances when we are talking about training off of someone's data specifically to produce a direct market competitor.

[gritty ivy]

first time ?

[quick basalt]

Actually yes lol

[magic widget]

It has always been offensive to imitate or copy an artist with the intent of diluting their influence. It has always happened since the beginning of art.

Having new tools doesn't change that.

[quick basalt]

Ever since this (Anti AI post) started on Artstation, the internet is like hell

[magic widget]

As far as being some thousandth of a percent among a complex statistical model of a billion records? Get over yourself.

[gritty ivy]

yep

[serene hare]

Yeah, but the scale of these new tools is previously unheard of.

[magic widget]

Even if I believed in intellectual property, it would be silly to whinge about that.

[fervent thunder]

People are pissed about something they don't fully understand or really care to understand because of what they've been told through others. AI is an incredible tool that can be used in so many different ways and has the potential to make some really crazy, funny, or very 'artistic' things.

[magic widget]

@serene hare Not really, Google has been indexing the world for 20 years.

[north stirrup]

When people prompt specific artists that gets a lot murkier

[quick basalt]

Before, I don't remember seeing anyone being attacked by only posting some AI image, now they come in horde and attack it

[magic widget]

@north stirrup The artists whose names function as prompts are the most commercially successful.

Not just random people

[serene hare]

Google indexing books to aid in sales is a completely different use than generative models.

[magic widget]

This isn't taking food out of poor people's mouths. That entire argument is disingenuous

[quick basalt]

Is it ethical for me to share the link of the post in the subreddit for you guys to see?

Better not I think

[magic widget]

@serene hare They're both tools that use massive amounts of public data to provide us all with something new.

[north stirrup]

A commercially successful artist isn’t necessarily rich

[toxic sonnet]

Indexing isn't copy an image to re use it.

[magic widget]

@north stirrup No doubt, but that is a problem with our society, not with our tools.

[serene hare]

I think that's a little reductive, one negatively impacts the original rights holders market share the other positively impacts it.

[magic widget]

They're literally copying a copyrighted work. You can make whatever distinctions you want about how it's used, but it's legally the exact same thing.

The difference is, SD doesn't produce any lossless copies of an original.

[magic widget]

It's reductive to say google "positively impacts" anything.

[serene hare]

I mean if you look at author's guild v google you'll see that there are pretty serious distinctions.

[serene hare]

fair lol.

[fervent thunder]

Do you guys ever worry about how fake the world will possibly be in the next couple of years/next decade?
From full on/full dive virtual reality. To deepfakes and synthesized voices, and other doo dads.

[magic widget]

I worry that it will be our only alternative to a world destroyed by the very bastards selling is that shit

[fervent thunder]

Social media by extension in a sense was the start of it. (not being anti-internet/or whatever)

Almost like a form of escapism.

[quick basalt]

I still think that social media is more dangerous than deepfakes and AI gen images, since it messes with the mental health of the people

[magic widget]

We're all lab rats in a thousand experiments

[quick basalt]

So I don't think it's getting worse

[quick basalt]

Regarding the algoriths, yes

[gritty ivy]

I'm not worried about that , im worried about these idiots who want to ban ai , i realy don't want to live in an world who is like cyberpunk 2077

[fervent thunder]

Wouldn't that stuff also mess and mind-warp people as well? The public is extremely sensitive and impressionable to any kinda stuff that is said from one human to another.

Deepfakes + Vocal Audio Synthesis + AI Art + GPT-3 or future GPT Models. It could be a weapon and recipe for disaster.

[magic widget]

The disaster struck when we allowed people to become politicians. I'm not joking.

That was the singularity in human history.

[quick basalt]

I think deepfakes may have an impact some minutes after it is released, but it's not long after it's debunked

[magic widget]

Since then they've used their pulpit to insulate themselves and manipulate the rest of humanity.

[quick basalt]

But I agree, deep fakes are still more dangerous than AI generated images (that I don't consider to be a danger)

[golden furnace]

just getting started here and fascinated by the possibilities... already thinking of working with the bots to create coloring book pages ... anyone here already creating coloring books? Would be fun to create a shared business where revenue from book sales with fuel the investment to scale and grow a book business. maybe in a DAO or global Co-Op.... diving down the rabbit hole this weekend haha

[gritty ivy]

another bad thing about humans is that treat everything with subjectivity

[magic widget]

There's an arms race between deep fakes and tools to detect them though.

[fervent thunder]

I'm not saying that everyone will fall for it, but some people don't keep "don't believe everything you see on the internet" in their heads at all times when browsing it.

Especially if it is convincing enough.

[quick basalt]

People could just crop the face of someone and put on the place of a guy wearing a N*zi outfit before in photoshop, there is not much more you can with AI

[magic widget]

This. The can of worms has been open for a while, people are just freaking out because this is a new can.

[quick basalt]

I believe that is the younger generation, because my parents aways said to me this same phrase, "don't believe on what you see on internet"

[magic widget]

Like, let's be clear: societies are constantly bathed in 24/7 propaganda already

[quick basalt]

I don't know if the younger generation still has this education

[proper karma]

so wtf is "realtime diffusion" its been a week lol

[quick basalt]

People argue that the problem is that it would be easier for someone doing bad things with AI, than photoshop, but if it was the case, the AI images those less competent people would create, would be less convincent

[quick basalt]

Only Emad knows, and he is not here to tell us XD

[magic widget]

Yeah but most propaganda already isn't very convincing if you actually pay attention to it

It still works VERY well

All it has to do is stir someone's emotions.

[quick basalt]

I agree, a lot of people are still very ignorant

[magic widget]

That is EASY. We are emotional critters by default.

It's really not about ignorance, sadly. We can't learn ourselves out of our nature.

We're just very emotional critters, which makes us easy to manipulate. People can develop a muscle that kicks in when they detect that, but it's not easy (and not encouraged)

[quick basalt]

That's true

[proper karma]

I beg to differ lol

[proper karma]

and yeah, all it takes is people seeing so many ridiculous things that they start to see something slightly less so and they're like "oh this is normal now"

or "believable" etc

[gritty ivy]

Just come in Romania to see how propaganda works here =]]] (e.g anti 5g , ati covid )

[magic widget]

Sorry to hear you're dealing with that there too. 😐

There's been a global fascist push over the last several years.

[quick basalt]

Where I live is not much different

[gritty ivy]

It's sad

[magic widget]

It really is. We're not powerless to fight though.

[gritty ivy]

yep

[quick basalt]

Do a lot of people in Romania hate 5g?

[gritty ivy]

40% of population i think

[magic widget]

Thanks for the chat, folks. I enjoyed our time together, I hope to catch you next time. ❤️

[quick basalt]

Really? I mean, where I live we have a lot of people against masks and stuff like that, but I don't hear much about 5g

[quick basalt]

Take care

[gritty ivy]

yep , romanian are so easy to manipulate

[quick basalt]

There are also some flat earth guys here, but I only see them in internet lmao, never met one in person

[tardy fable]

Hey all, used to using MJ, new to stable diffusion. Question, how do I train SD to use a face from a seed image? Thanks in advance

[quick basalt]

Does this propaganda come from Russia?

[gritty ivy]

i think so , because we have some anti war here

[steep mortar]

a lot of these model databases are female oriented and when you try to do whatever traits with a male image; it seems lack luster in comparison. are there any more balanced model databases?

[gritty ivy]

what i mean is they don't belive rusia atacked ukraina

and so many pro Russia

it's sad

[quick basalt]

Same here, people who don't like masks, vacines, flat earth dudes also are pro-russia

And I don't even live in Europe

Imagine if it was close to russia

[gritty ivy]

=]] like Romania

[quick basalt]

I think it's like a pattern of people, this pattern exists everywhere

[golden furnace]

Great job on the Moonshots podcast --- opened my mind to so many new possibilities. one way to begin the process of stability owned by everyone is to start with one share project as an example of what is possible. through a shared Gnosis wallet and something like a shared amazon book store ... or etsy shop... or gumroad shop... where creations from AI can be leveraged in books and digital pieces that customers want.... noodling a few ideas now that I have seen what is possible. then revenue from sales is invested back into shared wallet to fund and scale shared business and make it easy for anyone globally to invest small amount to be part owner of the growing enterprise
.
There are also soo many virtual assistants always looking for work... would love to collaborate with them to leverage their skills, teach them about the work here, and grow a digital creator army 🙂

[gritty ivy]

I mean , what can we do ?

those dudes likes to be dumb

[quick basalt]

I would love to visit some eastern european countries like Romania and Ukraine, but if russia don't just stop it will be very hard

[mighty basin]

https://www.riffusion.com/?&prompt=psychedelic+nepalese+trance&denoising=0.75&seedImageId=og_beat this is too dope

[gritty ivy]

it will be nice is that war will be ended soon

[quick basalt]

I think by the next summer (winter where I live) it's going to end, russia is losing too much territory

[gritty ivy]

And money

[mighty basin]

not sure about that one

what's the offramp?

[quick basalt]

Yes, things will get really hard for them if they continue with the war

[gritty ivy]

i just use that , so crazy stuff

[gritty ivy]

yep

I need to go guys , see you

[quick basalt]

Take care, man

[gritty ivy]

u to

[peak orchid]

I'd like to remind people that Karla Ortiz was originally part of Spawning AI until it either she quit because they didn't match her extremist beliefs, or because they kicked her out for her beliefs. Spawning AI is the group currently working with Stability AI on letting artists opt-out of training.

It would be naive to believe that she and those like her, would want anything less than the total destruction of everything related to AI generated content

[little plaza]

wait really? I had no idea about this that's .. interesting...

[bright cradle]

how t use interogate with img

[eager bison]

hey all! Can anyone lead me in the direction of a possible AUTOMATIC1111 discord or somewhere I can ask questions about it?

I really need to disable the texural inversion embeddings built in to the script if anyone knows how to do so?

[hasty mulch]

https://freewill.icu/ai-extinguishing-humankind-is-a-self-fulfilling-prophecy/

AI extinguishing humankind is a self-fulfilling prophecy

[toxic remnant]

Hello, how can I access stable diffusion in my local lan from another computer ?

[modest owl]

wait a second

where's the model information of the image stored?

Because i remember generating a lot of archival images but don't remember the model name

[near knot]

Can someone explain what unstable diffusion is? (are they creating a whole new AI?) I know they are going for NSFW stuff but doesn't SD already do that? Like if you changed your model, you could produce NSFW stuff. And if they are just creating a cptk file trained on NSFW stuff why does it require so much funding?

[eager bison]

Welp just found the folder they lived in and nuked them. Seems to have fixed my issues for any that run into the same thing.

[modest owl]

Right now they are scamming people since they have made 0 models

They want just the softcore nsfw

Damn sissies

[peak orchid]

They have already paid for Waifu-Diffusion training

[signal nest]

not sure what you mean but.. i've been running stable diffusion on a desktop PC, launching jobs from a Mac, via SSH

[peak orchid]

So that's already a model

[modest owl]

Not theirs

[signal nest]

i suspect you want some ready made UI for it

[peak orchid]

They provided funding though

[modest owl]

That doesn't contribute much

[peak orchid]

They've done more than other groups so far

[near knot]

Wait so they are just creating cptk files? Couldn't anyone do that, since their are like hundreds of models out there, theirs bound to be NSFW ones? Whats all the funding for?

[modest owl]

Not creating

They promise tho

[peak orchid]

They are training models on large datasets, which requires a lot more GPU computing power than simple finetuning or Dreambooth that you often see shared here

[near knot]

Gotcha, so the idea is that they are creating cptk files but supposedly they would be better than other NSFW models out there? Am I getting that right? Just curious about it

[modest owl]

Debatable but yes

[near knot]

gotchaaa thanks

[peak orchid]

They wanna train using normal datasets and NSFW content, so their models will be good with both SFW and NSFW content

[modest owl]

Midlf nsfw that is

So fuck that

Better support peeps that want to do hardcore nsfw

[shy juniper]

Is the official SD Discord bot ever going to go open source, as was promised so long ago?

[dapper ferry]

AI is not going away,, just like cryptocurreny is not going away

people will download stability diffusion and make images until the rest of time

its possible they will put some temporary restrictions on full blown text to image stuff,,, but thats all they will manage

at the same time theres all the other little powerful tools that are spin-offs of the tech are being integrated into many software as we speak

its a revolution

the Luddites will try to throw a spanner in the works because they are afraid of automation.

im not afraid of automation but i am afraid of not getting UBI when most jobs are automated and nobody can afford to live anymore

because i know the ruling elite and governments are massively corrupt and greedy.

[vapid tangle]

yep

ai is as inevitable as refridgerators putting ice delivery men out of business. same with with cameras, same thing with photoshop, cars vs horses etcv

ai is only going to get more popular and even stronger

[dapper ferry]

people will always fight to try and keep the status quo and resist change

especially people with bullshit jobs who live cushy comfortable lives

[vapid tangle]

yeah

[dapper ferry]

by bullshit jobs i mean superfluous jobs that are not really needed

[thin plinth]

hi! is there any channel in this server exclusive for animation with stable diffusion?

[north stirrup]

Like #🎥｜animation ?

[signal nest]

anyone know of attempts to do Federated Learning for large models?

is this just a lost cause

is there any research into different ways of learning that may suit distribution across the internet better

or would it be possble for a bunch of enthusiasts to get together and train a large model via kickstarter or something

[novel spear]

Hi. I saw online somewhere that after christmas there will be a new update which can generate images much much much faster? Is this true?

[mellow knot]

why federated

why not just decentralize the power structure completely and give it to anyone

[modest owl]

again

🤔

[signal nest]

currently there's a centralised step, so there is always a power-structure around that. federated is the only way to have something truly decentralised.

[peak orchid]

Yeah, and it was weird how quickly she went from talking about it being the next best thing to trying to strip all references to the organization

Some juicy drama appears to have gone down, but none of the parties involved have spilled the beans yet

[golden sentinel]

I'm trying to train a model with Dreambooth with about 20 images. Is that process supposed to take like ~12 hours? That's what it's currently saying. 57600 steps, doing like 1-2 steps per second.

Ok, I stand corrected .. the process just changed what it was doing and the time estimates are totally different. o.o

I guess it was just saving progress, it's back to 12+ hours.

[peak orchid]

57 thousand steps seems a bit high for a Dreambooth model, especially with only 20 training images

How many reg images are you using? They can make training take a bit longer, but not that long I think

[golden sentinel]

1,500. That public dataset which it's name escapes me, for human figures.

[peak orchid]

Are you training a v1.5 for a v2.x model? SD V2 may take more training iterations for the 768px variant

[golden sentinel]

Started with a 1.4 ckpt, but yeah - looks like that may be part of the situation.

I see in the web-ui, looks like v2 model was set to false

[peak orchid]

I was training Dreambooth models for v1.5 using a learning rate of 1e-6 and a couple hundred training images (and 2-4k reg images), and the max I'd train for was 11,000 steps though the best results were often closer to 2000-5000 steps

If your batch size is greater than 1, then you need ever fewer steps

[golden sentinel]

I'm still a nub, so I'm sure I'm doing something stupid.

I set Epochs to 1200, LR to 1e-6, turned on performance options because I'm doing it locally, didn't adjust any other settings. That's basically what the guide I followed suggested.

[slate crescent]

I've got a use case I'm trying to build a workflow for, but just started SD today and have little idea what I'm doing:

I want to be able to render a digital art style background, and then make variants of the shot at different times of day/seasons

my novice experiments result in architectural changes to buildings, even with conservative denoising or using depth2img; I feel like the solution will have to involve some curated inpainting and color grading

[main sparrow]

Not sure you'll be able to avoid compositing for that in photoshop or something similar.

Unless you had photos of something in all seasons, then img2img / inpainting might do the job.

[golden sentinel]

Ok.. just managed to get the number of steps down to 24,000. Is 5+ hours an unusual amount of time to train a model, on something like a 2080TI / 11GB VRAM?

Just trying to get a baseline for how long it takes other people to do theirs.

[main sparrow]

5 hours seems reasonable. it varies wildly, though

[modest owl]

right, okay, you didn't give a shit about the last one

[oblique chasm]

hi, which one is the prompt text to image channel?

[modest owl]

how about this one - what's the best sampling method?

[slate crescent]

I'm currently seeing it as two sub-problems:
-adjustments like leaves removed from trees or snow on the ground, which could conceivably be extremely complex (but possible?) inpainting applications
-color grading

[main sparrow]

No way to say. Come up with a prompt you like and try them all.

[modest owl]

i'm trying to find what model i was using for my prompt and i used euler a pretty much always

tried dpm adaptive and got janky ass results

[modest owl]

and for those who uses one of the shincore's models like mmd here's the full server dump with readmes and descriptions:
https://resilient-pony-4e0a44.netlify.app/

[urban dome]

What's the latest on when distilled SD will be released?

[modest owl]

#💬｜general-chat message

#💬｜general-chat message

[winter pike]

after/around new year, or

it could change, dont take my word as a fact

[twin atlas]

The depth2img model + img2img alternative script from autos repo = 🤌

[glossy compass]

hey, i just found out about textual inversion embeddings and how effective they are. Is there any sweet greg rutkowski embedding? i didnt find one on huggingface so far ^^'

[green bear]

👀

[fervent thunder]

https://knowyourmeme.com/memes/cultures/ai-art

[mint hedge]

guys where do i find yaml files for 2.1 models so i can run them with automatic1111?

[robust coral]

https://github.com/Stability-AI/stablediffusion/tree/main/configs/stable-diffusion

[mint hedge]

do these go with both 512 and 768 models?

[robust coral]

you have to rename the yaml to the same name as your .ckpt file.
v2-inference-v.yaml = 768
v2-inference.yaml = 512

[mint hedge]

i see tyvm!

[grim widget]

hey my auto1111 is acting strangely and i'm using for the first time in a month, is there a better ui that I could swap too?

[mint hedge]

try invokeAI it's not better but it is what it is

auto1111 MVP tho, i'd try to solve the problem with it

[solar wyvern]

whats the lore behind a second stable diffusion discord server? why does the subreddit not use this one?

[uneven wasp]

Theres a second SD discord server?

[solar wyvern]

yeah the subreddit just launched one

[grim widget]

i just deleted my folder and am reinstalling

[rapid atlas]

Hey I've not been having the best luck with the last ben's dreambooth collab after all the changes, wondering if there's an alternative way to make my own dreambooth checkpoints, any advice is super appreciated

[mint hedge]

Well i use runpod and i must say, it let me train idk around 15 models for 10$ and they're much better than the google collab ones, i tested it, so if it's not a problem for you, i'd just go and do all my training there.

[rapid atlas]

Ok I might just spring for the runpod method then tbh
I've been trying to make it always free but sometimes it's worth it and there's nothing u can do lol

[mint hedge]

plus i've lost first 3$ on figuring out because i know next to nothing about python and tutorials about it are hard to find. so be ready for that.

yeah it's vast difference between two models trained exactly the same way on google collab and joepenna runpod, however i don't know what is the reason for that.

[rapid atlas]

Is there a specific one? Or a tutorial you follow?

Oops you answered the tut question lol

[mint hedge]

i found brazilian tutorial without knowing any brazilian and just followed what they were doing. 😄

[rapid atlas]

Modern solutions lol

So Joepenna is the name to look for on runpod right?

[mint hedge]

yeah, one more cool thing about it, u kinda don't need standard 200x1 step calculation for image on joepenna, 101x1 does very good.

[rapid atlas]

Oh perfect! I have like 30 images per person (2 people) and some stylized images of each too. Does it do 2+ subjects? And does it do a good job stylizing if I can keep it from over fitting?

[mint hedge]

lemme give u short rundown: first u buy runpod credit, then u build pod with SD installed on it, then u go into kernel and git clone joepenna dreambooth repository, then u open joepenna-runpod file that gets downloaded and then there's only 3 blocks u have to run in that notebook out of the several, then u gotta kill few processes in terminal and u are good to go.

[mint hedge]

it gives best result with 1 subject at a time, however i found cool trick, if you generate reguralization images or upload people u'd like to have in your model as reguralization images, then token doesn't get affected too much but the class i.e. "person" often generates people close to reguralization images.

[rapid atlas]

Yes amazing

With your brief tutorial and some perseverance I think I can pull this off

[mint hedge]

yeah runpod and other paid services aren't worth using ALL the time when u spend on them enough to buy RTX 4090 but 20-30$ is allright imo 😄

[rapid atlas]

And I know a bit of python, which I only mention cuz if you ever wanna tag me for python questions I owe u lol

[mint hedge]

oh thnx

[rapid atlas]

I just want a 4090 I'd be the guy saying "oh I spent $15 on runpod might as well by the card"

[mint hedge]

that's how it starts, the great expenditure 😄

[rapid atlas]

Lmao truth

[uneven wasp]

Runpod is just a gateway drug

[mint hedge]

here, found much better tutorial than i used that is in english, just don't use this guys notebook, use joepenna to be safe. idk how good this notebook is. however steps are the same: https://www.youtube.com/watch?v=8iMbxtk0TVM&t=1123s

[lilac reef]

Has anyone been able to get hypernetwork training to work on colab yet? for 768x768

[rapid atlas]

Amazing! Haven't watched it yet does it mention what kinds of images to use like all up close or some bust/full body length?

Or do you have a best practice that conflicts with the info in the vid lol

[cedar kayak]

what prompt to use for a specific text sentence to be displayed ? for example /prompt Man wearing tshirt saying I am the best wouldn't work obviously, do I require special quotes? brackets ?

🧐

[rapid atlas]

I don't know how good it is at text tbh I've been not using that. But writing the text yourself then using low denoising (or inpainting later) might work

[civic eagle]

I was using dream studios AI generated art and on one of them it made an inappropriate art even though my search wasn’t anything harmful and it said I was being charged because of the inappropriate art?

[lusty mason]

guys my mind is blown there is a SD model that makes music

wtf

[radiant bronze]

Is it out?

[lusty mason]

https://huggingface.co/riffusion/riffusion-model-v1/tree/main

it creates images called sonographs which contain the music

[lilac reef]

I wish I could find someone who knows Pytorch and automatic1111 because I have a line to add to the webui batch file but colab doesn't use that and I need this Add the following line to webui-user.bat:
set PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.6,max_split_size_mb:24 on colab. Any of you can help?

[quick basalt]

https://civitai.com/models/1463/ai-protest Gigachad

[lusty mason]

lol

[vast scarab]

does anyone know how to convert the sonographs to wavs for riffusion?

want to try it but idk how that part works

[lament wraith]

https://github.com/hmartiro/riffusion-inference/blob/main/riffusion/audio.py

[quick basalt]

Brazilian? This language does not exist

[lusty mason]

so wait guys does this mean we can make endless music now

[near knot]

Do sites like NovelAI and other generators just dreambooth and fine tune existing cptk files? Or is it created from scratch

[vast scarab]

@lament wraith I'm not too great at using python, how could I select a specific file from a directory and what's the best software to run the code in?

[lament wraith]

can't really help there, sorry. maybe the reddit thread, huggingface or colabs will offer a way to do that. gl o/

[lusty mason]

so wait guys can we now generate music and just publish it on youtube

[vast scarab]

that's tough

so has anyone actually used it yet?

[lament wraith]

it's early days, yet a combo of outpainting, deforum or film interpolation could eventually get to an infinite music generator. This is all happening so fast, wouldn't be surprised if that is possible soon

[lusty mason]

idk how to feel about this but it is making me emotional

[lament wraith]

probably best to ask around here... https://www.reddit.com/r/riffusion/

It's pretty amazing Bee, agreed

[lusty mason]

the music industry gonna take a huge hit

[lilac reef]

What I heard so far is getting there but has a long way to go. Maybe there are better examples?

[lament wraith]

already kind of has with spotify, napster and itunes... just the next step. I was a professional recording artist once, that dried up. Have empathy for the artists who are upset now, yet the writing is on the wall. Only a matter of time for all industries

[vast scarab]

found they have a discord

[lilac reef]

For every single aspect/job in life.

[vast scarab]

with a webui extension

[lament wraith]

gg and gl

does the webui extension have audio conversion already?

[lilac reef]

The one thing about this, and where my hope lies with it, is that it does remove all job opportunities. That will be a net positive for society and may usher in a world like Star Trek where money is meaningless and you do your passions instead. Like Sisko's dad doing a LA restaurant because that is his passion not because he has to to earn a living.

[lament wraith]

it does with mp3 and ffmpeg, wow

stuff is moving so fast

unfortunately the Star Trek utopia still feels like it's in the 23rd century and that we may still have to go through this one first

[vast scarab]

yeah tbh I don't think it's gonna change much at least at the moment

[lusty mason]

if AI robots can provide for humans needs then work is no longer needed

[lilac reef]

Correct, and we will not move to it easily for wealth/money brings power and those in power are not going to just roll over and give it all up.

[lament wraith]

the homeless camp in the local city park says we're a ways off from that, imho

[lusty mason]

I somehow see AI taking jobs as a positive thing, if AI does all essential work then you won't have to wake up early to boring jobs trying to make money

[lilac reef]

That is why we have the poor because you need two extremes so one has power over the rest. Nasty times ahead.

[lament wraith]

Spotify didn't help musicians or the music industry, just saying

doubt their Annoi ai is going to either

[lusty mason]

if this tech is open sourced I wonder what's closed source in some major coorps

i doubt they just sitting doing nothing

[lament wraith]

funding anti-ai social engineering protests I imagine

[lilac reef]

BINGO

[lusty mason]

but isn't AI beneficial for them?
they won't have to hire and pay salaries anymore

[vast scarab]

generated my first thing and it sounds pretty bad lol

[lusty mason]

is it possible to send here

[vast scarab]

bro i deleted it

this one is my second slightly better

ah

can't send

[lusty mason]

maybe at #🏞｜general-with-images

[vast scarab]

yea that worked

[lusty mason]

ah it sounds cool

does it just do 5 seconds?

[vast scarab]

yeah

[lusty mason]

so how would you make a 1 minute track

[vast scarab]

but i generated another with the same prompt and it sounds honestly like it would blend well, don't know if that's luck of the draw or what

Let me see

not sure yet

[lusty mason]

maybe with infinite outpainting?

using stable diffusion infinity or something

[vast scarab]

🤷‍♂️ never used either of those things haha

[rustic willow]

just pulled trigger on a used 3090 ^^

[vast scarab]

Someone who seems to know more about it said "the actual riffusion inference software should be able to do it which I am literally just starting to look into. can't seem to get it running :
basically what you need to do is generate a couple of clips and then generate the interpolation between them to "extend" the music
auto1111 won't be able to do it I think"

[lusty mason]

ah seems cool

seems like a good project to create a website with millions of AI tracks

for people to download from

idk about legal issues tho

[lilac reef]

We still have a long way to go before it can actually make a song cause what I am hearing is too chaotic.

We are advancing rapidly now though so come back in a year and hear AI albums dropping into the top 100

[lusty mason]

i was just listening to their website they already made tracks

[lilac reef]

they suck ass juice though

[vast scarab]

not ass juice 😆

[lusty mason]

did you guys check these

https://www.riffusion.com/?&prompt=lo-fi+beat+for+the+holidays&denoising=0.75&seedImageId=og_beat

[lilac reef]

Too robotic and very low bitrate

[vast scarab]

I'm curious if steps will affect quality here

[rustic willow]

like i said in other channel lol.. give it a few weeks

[vast scarab]

for me seems like 150 steps actually makes the same seed sound pretty dope but other than that somehow 20 is better than say 70

[fluid swift]

https://www.youtube.com/watch?v=7Pq-S557XQU <- we are already there.

Yeah, posting that video often, and thats from 2014

[lusty mason]

yea but now it's accessible to everyone

[fluid swift]

Yeah. The problem would be - what to to with humans depending on thier job to afford rent and stuff?

[wicked hollow]

Were you here for the Harmonai panel-event yesterday?

[vast scarab]

I'm gonna run a few for a bit and cherry pick

[lilac reef]

No

[lusty mason]

it seems like good and bad

[lilac reef]

Just ask Klaus Schwab of the WEF.

[lusty mason]

what I know is that maybe we will have better quality music ?

[uneven wasp]

The problem with putting an AI into a real world application unsupervised is that the error tolerance is exponentially lower because failures cascade.

So either it augments human work like technology has always done, or you make it exponentially better

[fluid swift]

(And the nice thing is: The CGP Grey Video isn't a "look whats possible in the future" but more a "look at the automation already exists")

[wicked hollow]

Harmonai had an event where they talked AI generated music, what their current status is, how they came to be and where we're going with the path forward in such generations. I got to talk to them specifically about quality--artifacts, distortion, general low quality levels, etc. It was very interesting stuff--and much of which was over my head in regards to the coding and implementation behind things, but it boiled down to dataset variety, size, and the fact that much of what they're doing thus far is peering into a black box and trying to understand how to tweak things to best move forward, heh.

They're currently working on a beta 45sec version, and some of the demos I've heard are actually pretty decent stuff. And I say that as a composer and audio engineer myself.

[lilac reef]

Yes, in less than five years, unless a lawsuit stomps on all of this, I can see 3-7 min complete songs not even knowing if it was human made or AI made.

[wicked hollow]

Harmonai at least, is specifically working from an opt-in directive.

I think that sort of thing is far more important in music, as melodies are far simpler and easier to reproduce perfectly against copyright than something like an art style which can't be copyrighted.

[fading oasis]

where do i download stable diffusion

[lilac reef]

Well, I lay odds you stick something made by it on Youtube and get a copyright match, or those asshats that flag it within minutes saying they own the rights and YOU just wrote it and composed it.

[wicked hollow]

https://rentry.org/SDInstallation

[wicked hollow]

YouTube is notorious for abiding by copyright strikes from people that decide that the music is theirs--regardless of actual ownership. So that'll happen anyway.

[lilac reef]

Others do as well.

Just not as rampant, or blatant as Susan's world is.

[fervent thunder]

Whelp as far as I'm aware memes are gone and I got no reason to be here so goodbye server

[near flare]

Is there limit of genarate

Cost credit?

[worthy oriole]

More tags

https://magazine.artstation.com/2022/12/noli-tag/?utm_source=homepage&utm_medium=toprow&utm_campaign=2486-noai-tagging-on-projects

[pallid hinge]

What's the max sample and resolution we can use to generate images using a 3060 ti?

I can't find enough SD performance comparison between 3060 12gb and 3060ti 8gb

[ocean mango]

https://www.youtube.com/watch?v=X8i9Op2HSjA&t=27s Integrated AI - The sky is infinite (2022 AI retrospective)

[balmy jungle]

@ocean mango nice the workflow in photoshop is pretty amazing

[young ivy]

hypothetical: If I had an Nvidia 3070 ti and one of those new AMD cards with their insane amount of vram. Is it possible to turn that into a slave block and the 3070 into a master? So it draws vram from it as if it is its on?

[uneven wasp]

There's a bunch of on-chip optimizations for vram that makes it vram, communicating over a bus to load data from a different gpu isn't really going to be much faster than just using actual ram

[young ivy]

I didnt think regular ram could be used?

and its not just faster Im after, it's being able to train dreambooths etc

[uneven wasp]

You'd have to split the actual compute load between devices, not entirely different (though not the same either) than how it's split between cuda coresz

[young ivy]

would it even work though or would it end up more convluted and time wasting than just buying a used 3090?

[tawdry summit]

Quick question: How many hours does it take for a total noob like me to set up SD on a laptop?

[uneven wasp]

The problem is that you'd have to load the data into that gpu's vram before processing on the other gpu the way its set up

[young ivy]

@uneven wasp yeah I got yah :=) thanks

I think I will just go buy a used 3090 then

[uneven wasp]

Yea, probably just worth it to use a 3090 unless you like dabbling with the technical limits of parallel programming

[young ivy]

maybe find one selling for scrap cause it doesn't work, take it home and see why it oesn't work and if I can make it wor

they tend to be cheap

[uneven wasp]

Possibly, but idk how feasible fixing those things are

too many proprietary parts with insanely weird quirks ime

might be able to get it working on 2 gpus if you split the model in half and manually transfer error/gradients across

transferring shouldn't be much harder than a object_name.to('cuda:0') or w/e your setup is

but making you the backprop actually backprops across the "gap" where you've split it is the annoying part

esp if theres residual connects.

idk I haven't looked into the exact architecture dreambooth actually trains

Mixing amd and nvidia into the same system is a recipe for a disaster but it can be done

[golden nebula]

if people saw what I have been doing with stable diffusion I'd be the most unethical person

all the red flags

[balmy jungle]

mixing cotton and woolen threads in your garments?

[golden nebula]

generation of artist styles

[uneven wasp]

Not that hard. official repo is a pain to use, automatic1111 is the current best. If you have under 6gb vram it's a little bit more involved but just boils down to adding a launch argument --medvram.

Essentially clone the automatic1111 webui repo, run webui.bat, download the .ckpt weights and .yaml config from the stable diffusion huggingface page, rename them to the same thing+drop them in the models folder and it should be good to go

damn, we really need a streamlined installer for this don't we

[tawdry summit]

Thanks. Is vram the same as the ram of my graphics card?

[uneven wasp]

Yes. You can check how much you actually have via Task Manager->Performance->GPU->Dedicated GPU Memory Usage and the max should the right aligned text next to it

[bright cradle]

Hi

If i generate something with the help of ai then i edit the image, enhance, color, use brush or something or if i directly post the AI generated image shouldn't i put my name signeture on the corner of the image?

[signal nest]

there's ways to use 2 GPUs for AI together - splitting the model across both - but you need the model to be adapted for it, I'm not sure anyone did this for the popular diffusion models yet

its probabyl easier to just use them in parallel on different jobs

[unkempt hatch]

Being open source, do you think it's likely Stable Diffusion could branch off?

I guess there's not much point... now likely someone makes a big decensored model

[fervent thunder]

Hi, is Sd 2.0 fixed now?

[lilac reef]

For upscaling which ones do people use now?

[timid gate]

Uhh guys https://www.gofundme.com/f/protecting-artists-from-ai-technologies?utm_campaign=p_lico+share-sheet&utm_medium=copy_link&utm_source=customer

[balmy field]

Is it possible to tell the UI to repeat the query with each installed model?

[lavish monolith]

use the xy_grid scrip, in a1111's webui scroll down to scripts, select the X/Y grid script, select Checkpointname for X or Y then write the list of model's names you want coma separated

[ocean mango]

https://smartshow-software.com

[pliant haven]

#🤝｜tech-support

[static badge]

/notebooks/stable-diffusion-webui
No module 'xformers'. Proceeding without it.
Cannot import xformers
Traceback (most recent call last):
File "/notebooks/stable-diffusion-webui/modules/sd_hijack_optimizations.py", line 18, in <module>
import xformers.ops
ModuleNotFoundError: No module named 'xformers.ops'
anyone knows why xformers are so foking anoyying to isntall and launch properly?

[regal swan]

Hey guys, I created a small python app to store and combine my favourite prompts. It might be useful for others too:
https://github.com/pazz89/Prompt-Library

[loud grotto]

What exactly does it do ?

[regal swan]

There is a picture of the App on the github link, I guess that explains it better then I can do in words 😉

[loud grotto]

I get a torch is not able to use gpu error

When running webui-user.bat

[ionic copper]

whats good whats good

[thin plinth]

ooh yess

[loud grotto]

Torch is not able to use GPU add —skip-torch-Cuba-test …

Why can’t torch detect my gpu ?

It’s Nvidia so it should work

It’s the GTX 1650

Im installing cuda now

[pallid hinge]

What's the minimum vram requirement for generating 2k images with auto1111 webui without using --lowvram

[loud grotto]

2k res?

[golden sentinel]

I finally created my first successful model with Dreambooth, and I'm getting "some" interesting results now, but lots of garbage results as well. DPM2 a Karras & Restore Faces gets me one or two 'usable images' out of a batch of 10.

Anyone have info on the best ways to improve the results I'm getting, other than sampling steps? Maybe I should make a better model using better inputs?

I just took 20x random photos of my face/head, cropped to 512x. I think I saw somewhere I should have took some body shots too.

[static badge]

how i udpate python 3.9.13 to 3.10 on jypeter? paperspace gradient notebooks?

[pallid hinge]

Yeah

[loud grotto]

Don’t do 2k just do 1k and upscale lol

[pallid hinge]

Ok, do you know the minimum vram requirement for 1344x768, 80 samples?

Will a 3060 ti (8gb) be able to do it?

I heard a lot of recommendation for 12gb 3060.
But it seems kinda slow for my taste.

[feral kelp]

it's strange. I thought euler A had been outclassed by the dpm++ samplers, but I have a couple prompts that euler still gives the best results on. anyone else get this?

[magic widget]

euler A also gives some of the best results for fewest steps

[peak orchid]

Apparently some people in the comic book industry are trying to get anyone who uses AI art blacklisted from the industry: https://www.cbr.com/comics-industry-collective-stance-ai-artificial-intelligence-art-usage/

[golden sentinel]

Like every obsolescent industry fighting it's encroaching doom.

[mental flint]

Is this real stable diffusion? how to use?

[peak orchid]

Greg Rutkowsk apparently wants to make artistic style copyrightable among other extremist copyright ideas. They guy is becoming a villain if he wasn't one already

Also we still need a plan to fight back against Karla Ortiz's anti-AI fundraiser to buy lobbyists in Washington DC to ban AI generated content

[dapper ferry]

that would be an outrage.. and lead to a catastrophy for all artiosts

in short order all styles would be copyrighted leading to artists being unable to create new art

i've never heard such a stupid idea

what next copyrighting color schemes?

or individual colors

lets copyright musical scales as well, and chords

[golden sentinel]

Colors are already copyrighted/trademarked. Pantones.

[dapper ferry]

lets copyright musical styles

[dapper ferry]

in a way, yea i heard about that, like cadburys chocolate purple color

but thats still a very silly idea and hard to police

Copyright letters of the alphabet

im going to file a copyright for the letter S

i guess fonts are already a thing.

But not the use of actual letters regardless of font

[peak orchid]

Color cannot be copyrighted. A business may use a trademark to have the exclusive right to use a particular color on specific products or services in the course of trade. However such trademarks are hard to get, and are not given at the moment of creation.

Pantone colors themselves are not protected, but their dataset of colors is

[silent torrent]

Anyone has an idea how to create a SD2.1 512x512 model with fast dreambooth
https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast-DreamBooth.ipynb?authuser=8#scrollTo=1-9QbkfAVYYU

I can create it
but when I try to use it in Automatic 1111
I get an error like dimension or tensor dont match ?

Do I need to get a config file or something ?

[frank moon]

Hello, how do you upload your own videos when using Deforum locally through automatic1111? The is not upload video like in the colab method.

[fervent thunder]

did you make a yaml for it? idk i had issues making 2.1 models there too

[silent torrent]

yaml ?

nope

I just download the .cpkt in my folder

I can use it with the collab tester

But not in Automatic 1111

[pliant haven]

anyone know which version of stable diffusion has 'loab' in it?

[long marten]

https://www.chaosium.com/blogai-art-and-chaosium-16-dec-2022/

"""
Art is important to Chaosium and our artists deserve a lot of credit for our successes over the past several years.

We’re updating our art contract templates to include the provision that AI art programs are not to be used: the work needs to be the product of a human artist who can vouch that they created the piece and that it does NOT contain unlicensed derivative use of someone else’s work.

We are concerned about the ethics of AI art and its impact on the livelihoods of artists, and the ability of artists to maintain control over use of their creations.

And on a more pragmatic level, we also believe there is a significant chance that the US courts will, before long, declare that AI art violates the copyright of artists, most probably thousands of artists. There is also the possibility that the European Union – or at least a few significant EU members – will pass legislation that effectively prohibits the webscraping AI programs that now exist.

So, in short – if you are doing art for us, don’t use AI.

The next time you pick up a Chaosium game, you can be confident that all of the art there is the product of a human artist who is passionate about our games and the worlds we create, rather than a set of computer algorithms and prompts.

— Chaosium Inc.

"""

I too am a bit worried about using AI outputs directly in commercial products. Could mean retroactive legal battles over IP if the laws change

[broken gull]

why are flying islands so hard to get

[little plaza]

Midjourney v 1, actually

[pliant haven]

ah I been getting better results with v2 and loab

[pliant haven]

which prompt do you use?

[pliant haven]

I would do two seperate images, land and clouds then use merge feature of mj

[broken gull]

I am not gonna use mj

[little plaza]

here's the explanation and the prompt they used originally https://twitter.com/supercomposite/status/1567162313194471428

[pliant haven]

it never says what version or if they even used SD

[little plaza]

Supercomposite has definitely said they used Midjourney v 1 before can't find a source atm.