Jump to: Board index » General » Fusion

The First Editing Suite to Implement StableDiffusion AI...

Learn about 3D compositing, animation, broadcast design and VFX workflows.
  • Author
  • Message
Offline

GuySwann

  • Posts: 7
  • Joined: Sun Sep 04, 2022 4:36 pm
  • Real Name: Guy Swann

The First Editing Suite to Implement StableDiffusion AI...

PostTue Sep 20, 2022 8:47 pm

...Will literally get a million+ downloads/purchases in a matter of weeks. And because StableDiffusion is open source it will literally be a matter of weeks before this happens.

There is already a web based software attempting to do exactly that and honestly it will blow all of the professional editing tools out of the water on about 70% of the "prosumer" sort of editing. It could take a huge gash out of high end editing by expanding what you can do from a more simplified app that has like just 9 basic tools but integrates all of them with AI.

Partly because I just invested a lot of time learning Davinci Studio, partly because I have really enjoyed Fusion, and partly because I like this product, I REALLY want to make sure this is on your radar because I think ignoring it, when its freely available, is extremely reckless from a software business perspective.

This thing is moving at an insane pace.

https://runwayml.com/
https://twitter.com/daniel_eckler/statu ... -Hf1z2wu1A
Offline
User avatar

Bryan Ray

  • Posts: 2478
  • Joined: Mon Nov 28, 2016 5:32 am
  • Location: Los Angeles, CA, USA

Re: The First Editing Suite to Implement StableDiffusion AI.

PostThu Sep 22, 2022 12:12 am

Bryan Ray
http://www.bryanray.name
http://www.sidefx.com
Offline

GuySwann

  • Posts: 7
  • Joined: Sun Sep 04, 2022 4:36 pm
  • Real Name: Guy Swann

Re: The First Editing Suite to Implement StableDiffusion AI.

PostThu Oct 13, 2022 3:26 pm

Nice! Thank you
Offline

GalinMcMahon

  • Posts: 715
  • Joined: Sun Sep 29, 2019 10:14 pm
  • Real Name: Galin McMahon

Re: The First Editing Suite to Implement StableDiffusion AI.

PostThu Oct 13, 2022 11:31 pm

Other than making "Normal human hands with 5 fingers" and getting some sort of alien hand thing with like 18 fingers, what would this do for a video editor? Even once it's up to speed on that and can create quality video, what would be the benefit of creating those within Resolve? I suppose if you could create a scene and edit it before it's even visualized, that would be pretty cool but that isn't really video editing IMO. Also, Blackmagic does in fact use a ton of ai in Resolve already and it keeps getting better.
Water cooled Windows 11 laptop
i9 12th gen - 64GB RAM - 16GB 3080ti
2TB 4th gen nvme main - 4TB 4th gen nvme scratch
Micro panel (thank you BM :) ) - Stream Deck
Resolve Studio 18.5
BenQ ultrawide - DeckLink 4k mini via Sonnet - 48” LG C2 OLED
Offline

Kel Philm

  • Posts: 604
  • Joined: Sat Nov 19, 2016 6:21 am

Re: The First Editing Suite to Implement StableDiffusion AI.

PostFri Oct 14, 2022 7:43 pm

I am guessing it may eliminate the need to purchase stock images?
Offline

Hendrik Proosa

  • Posts: 3015
  • Joined: Wed Aug 22, 2012 6:53 am
  • Location: Estonia

Re: The First Editing Suite to Implement StableDiffusion AI.

PostSat Oct 15, 2022 7:10 pm

There is one very practical use case for these diffusion models: inpainting. Creating a relatively good looking cleanplate for paint work is always helpful, even if one has to do some additional touchups.
I do stuff.
Offline

birdseye

  • Posts: 349
  • Joined: Fri Jun 12, 2020 2:36 pm
  • Real Name: Iain Fisher

Re: The First Editing Suite to Implement StableDiffusion AI.

PostSun Oct 16, 2022 7:11 pm

How would copyright laws interact with that. If it produced an image of a Disney character or a location in Star Trek, how close visually could it be before the legal fur starts to fly.
Offline

Hendrik Proosa

  • Posts: 3015
  • Joined: Wed Aug 22, 2012 6:53 am
  • Location: Estonia

Re: The First Editing Suite to Implement StableDiffusion AI.

PostSun Oct 16, 2022 7:36 pm

birdseye wrote:How would copyright laws interact with that. If it produced an image of a Disney character or a location in Star Trek, how close visually could it be before the legal fur starts to fly.

Applies the same way how it does when you make an illustration yourself, this is all well established. Copying someone’s style is a gray area though, I don’t think this can be solidly protected. And arguments like oh, everyone whose stuff model was trained on is now entitled to fees or whatnot won’t fly far, just as no-one is paying for ”stealing inspiration” right now.
I do stuff.
Online

VMFXBV

  • Posts: 579
  • Joined: Wed Aug 24, 2022 8:41 pm
  • Real Name: Andrew I. Veli

Re: The First Editing Suite to Implement StableDiffusion AI.

PostMon Oct 17, 2022 4:54 pm

Hendrik Proosa wrote:
birdseye wrote:How would copyright laws interact with that. If it produced an image of a Disney character or a location in Star Trek, how close visually could it be before the legal fur starts to fly.

Applies the same way how it does when you make an illustration yourself, this is all well established. Copying someone’s style is a gray area though, I don’t think this can be solidly protected. And arguments like oh, everyone whose stuff model was trained on is now entitled to fees or whatnot won’t fly far, just as no-one is paying for ”stealing inspiration” right now.


But its not just inspiration. Its literally scanning stuff off the web that they have no right to (including very private stuff like medical records and such in some cases). And if you look closely behind the scenes on how the stuff actually works its even more disgusting. I've seen the bots even trying to copy signatures and stuff and parts of the original artwork. Its kinda disgusting.

Hope rules are put in place for this kind of stuff.

Can hardly wait for stuff I shot for stock sites to appear in little Timmy's videos where his contribution was "blablablabla in the style of blablabla" (see Greg Rutkowski) in some prompt and then calling the result his own.
AMD Ryzen 5800X3D
AMD Radeon 7900XTX
Ursa Mini 4.6K
Pocket 4K
Offline

Hendrik Proosa

  • Posts: 3015
  • Joined: Wed Aug 22, 2012 6:53 am
  • Location: Estonia

Re: The First Editing Suite to Implement StableDiffusion AI.

PostMon Oct 17, 2022 5:19 pm

VMFXBV wrote:But its not just inspiration. Its literally scanning stuff off the web that they have no right to (including very private stuff like medical records and such in some cases). And if you look closely behind the scenes on how the stuff actually works its even more disgusting. I've seen the bots even trying to copy signatures and stuff and parts of the original artwork. Its kinda disgusting.

This is not how stable diffusion works, not sure where you see ”bots copying signatures” and so on. I have looked pretty deeply into how latent diffusion models work and I don’t see anything disgusting there.

What’s the difference with little Timmy copying someones art style in illustrator to fill his ”Timmy Art” gallery? Try getting little Timmy pay for that, good luck, this field is filled to brim with easy money.
I do stuff.
Online

VMFXBV

  • Posts: 579
  • Joined: Wed Aug 24, 2022 8:41 pm
  • Real Name: Andrew I. Veli

Re: The First Editing Suite to Implement StableDiffusion AI.

PostMon Oct 17, 2022 7:10 pm

What?

A good chunk of stuff the model was trained on was from Pinterest, Deviant art, Fine Art America...What do you mean this is not how it works? Its exactly how it works...

The bots I'm referring to are the Midjourney discord bots that do exactly this...

The other chunk was trained on LAION-5B. And LAION-5B contains stuff copyrighted stuff , like medical records, that explicitly say "don't use this stuff for anything other than medical stuff".

Hendrik Proosa wrote: What’s the difference with little Timmy copying someones art style in illustrator to fill his ”Timmy Art” gallery? Try getting little Timmy pay for that, good luck, this field is filled to brim with easy money.


The difference is its not just the style that's being copied but a good chunk of the original art, SIGNATURES INCLUDED. Even if the "AI" tries to garble it.

"In many of the results there have been traces of watermarks and signatures, these programs are explicitly designed with the function of removing such marks that can circumvent intellectual property”, Jon Juárez, an artist who has worked with Square Enix and Microsoft,adds. He’s referencing examples of AI-generated artworks appearing to have signatures in their corners, suggesting that while drawing from pieces they have been fed they’ve simply copied the signature — albeit imperfectly — as well."

And I'm not even talking about stills here, but what comes next for video (See the google imagen stuff). Which clearly uses stock footage to do the combining. And when little Timmy gets to go around to pass chunks of my videos as his own and charges money for it then it becomes a huge problem.
AMD Ryzen 5800X3D
AMD Radeon 7900XTX
Ursa Mini 4.6K
Pocket 4K
Offline

Hendrik Proosa

  • Posts: 3015
  • Joined: Wed Aug 22, 2012 6:53 am
  • Location: Estonia

Re: The First Editing Suite to Implement StableDiffusion AI.

PostTue Oct 18, 2022 9:28 am

VMFXBV wrote:A good chunk of stuff the model was trained on was from Pinterest, Deviant art, Fine Art America...What do you mean this is not how it works? Its exactly how it works...

Trained == Little Timmy went to gallery
VMFXBV wrote:The bots I'm referring to are the Midjourney discord bots that do exactly this...

The other chunk was trained on LAION-5B. And LAION-5B contains stuff copyrighted stuff , like medical records, that explicitly say "don't use this stuff for anything other than medical stuff".

Crawlers that gather images are not part of diffusion models themselves, this argument crumbles to dust with training dataset change.
VMFXBV wrote:The difference is its not just the style that's being copied but a good chunk of the original art, SIGNATURES INCLUDED. Even if the "AI" tries to garble it.

"In many of the results there have been traces of watermarks and signatures, these programs are explicitly designed with the function of removing such marks that can circumvent intellectual property”, Jon Juárez, an artist who has worked with Square Enix and Microsoft,adds. He’s referencing examples of AI-generated artworks appearing to have signatures in their corners, suggesting that while drawing from pieces they have been fed they’ve simply copied the signature — albeit imperfectly — as well."

And I'm not even talking about stills here, but what comes next for video (See the google imagen stuff). Which clearly uses stock footage to do the combining. And when little Timmy gets to go around to pass chunks of my videos as his own and charges money for it then it becomes a huge problem.

Let me give you an analogy for this. Little Timmy goes to gallery. Timmy is a bit simple in his head but can read. He reads the titles and descriptions of art pieces and memorizes them to some extent. Then Timmy goes home and thinks, why don't I paint an image with melting clocks and giraffes and fruits. He does that by roughly referencing all the different paintings he saw with clocks and fruits and whatnot, and then signs it in bottom right corner as Piscaccolo because simple Timmy saw that most paintings had this kind of stuff there. Which way is Little Timmy now going to jail or paying someone? This whole quote above screams of misunderstanding of how any of this works.
VMFXBV wrote:"In many of the results there have been traces of watermarks and signatures, these programs are explicitly designed with the function of removing such marks that can circumvent intellectual property”

Where exactly is there "explicitly designed function" for removing watermarks and signatures in Stable Diffusion model? If it is explicit it will be very easy to find I assume, since source code is public... It's worth noting I think that whole ML model and framework built around it are different things. Frameworks can do "stuff" but it is not essential part of the diffusion model itself. Killing off one will spawn three new ones.
I do stuff.
Online

VMFXBV

  • Posts: 579
  • Joined: Wed Aug 24, 2022 8:41 pm
  • Real Name: Andrew I. Veli

Re: The First Editing Suite to Implement StableDiffusion AI.

PostTue Oct 18, 2022 2:29 pm

I don't mind the tech itself.

I do mind the stolen content that's being used to train it. And its pretty obvious where the images come from...Otherwise "in the style of etc" wouldn't work as a prompt, because there is nothing "Ai" in how these work.

Machine learning is not "intelligence"...Typing in Spiderman and getting an image of Spiderman as we know it means it def used copyrighted stuff.

There is no misunderstanding how this works. The "Ai" is too dumb to remove the signature from the original artwork and it uses it in the mangled garbage it spits out. Lets not pretend its anything more than this.

I will support this for serious work when they make their sources public (not just the code but every material they used for the training). And when that stuff contains materials that the mega multi billion company has paid for or its in the public domain I will gladly support them.

Until that happens they can f off with magnificence.
AMD Ryzen 5800X3D
AMD Radeon 7900XTX
Ursa Mini 4.6K
Pocket 4K
Offline

Hendrik Proosa

  • Posts: 3015
  • Joined: Wed Aug 22, 2012 6:53 am
  • Location: Estonia

Re: The First Editing Suite to Implement StableDiffusion AI.

PostWed Oct 19, 2022 6:35 am

If something is in public domain, referncing it is not stealing. There are obviously cases where owner does not allow some uses as you pointed out earlier, but this simply means that training dataset needs an overhaul. And while gathering training data and training itself is processing heavy, finding financing for it these days or crowdfunding etc is a question of days, if not hours.

There is no intelligence in any of this, I don’t think anyone argues there is. But typing spiderman and getting spiderman is not a priori producing copyrighted image either. If it is not the same image someone else drew and it is not direct piecewise copy either (as can be easily demonstrated by dissecting the algorithm), only thing left is the trademark ”Spiderman” and its appearance, which are protected exactly the same way as in cases someone manually drew spiderman. Lets say I want an image of pig in pink spiderman costume and I call it SpiderPig. Whether there is any copyright infringement is pretty hard to tell, and fact that actual spiderman images were part of training data does not make it any clearer.

AI is not too ”dumb” to remove the signature, because removing it is not an objective. Signature simply has no meaning in this context, it is just a piece of image as any other and thus gets encoded to latent space representation.

There is a curious case that someone might want to get a handle on, and it’s the fact that if a copyrighted piece is encoded to latent space, decoding it will produce the same image. But to actually get anything here, one must demostrate that these encodings exist in the model weights data. And this is practically impossible, because they don’t.
I do stuff.
Offline

Sander de Regt

  • Posts: 3500
  • Joined: Thu Nov 13, 2014 10:09 pm

Re: The First Editing Suite to Implement StableDiffusion AI.

PostWed Oct 19, 2022 8:14 am

Try googling 'SpiderHam' :lol:
Sander de Regt

ShadowMaker SdR
The Netherlands
Online

VMFXBV

  • Posts: 579
  • Joined: Wed Aug 24, 2022 8:41 pm
  • Real Name: Andrew I. Veli

Re: The First Editing Suite to Implement StableDiffusion AI.

PostWed Oct 19, 2022 8:34 am

If something is in public domain, referncing it is not stealing.


But that's my point. The materials they used are NOT in the public domain. Just because they're on the internet doesn't make them public domain. And I'm pretty sure no one gave their permission.

training dataset needs an overhaul


Bingo, that's what I want. As it is right now, its full of copyrighted material. If the dumb "Ai" understands what Spiderman is then its ben fed a ton of Spiderman images and it was "told" its Spiderman. Unless the guys who fed it drew those pieces themselves then its thievery.

Imagine you sell books in some market. Books you wrote and then printed yourself. You set up your little shopping cart in the public market. Then some a-hole comes, takes all your books from your shopping cart and starts scanning them with a really bad scanner in front of you then prints and sells them at a cheaper price right next to you. You ask him
"what are you doing?" and he goes "well, these books were laying here on the public domain so I just took them, don't worry they're not 100% the same copy, what are you gonna do anyway?".

And I'm happy places like Shutterstock and other stock selling agencies have already banned Ai "art".
AMD Ryzen 5800X3D
AMD Radeon 7900XTX
Ursa Mini 4.6K
Pocket 4K
Offline

Hendrik Proosa

  • Posts: 3015
  • Joined: Wed Aug 22, 2012 6:53 am
  • Location: Estonia

Re: The First Editing Suite to Implement StableDiffusion AI.

PostWed Oct 19, 2022 9:11 am

Tons of spiderman images are in public domain, not free as in free beer, but public in the sense that everyone has access to them (admittedly I use "public domain" loosely here, not in its more specific meaning). What you argue is that Little Timmy must not draw any inspiration from them. This is impossible argument to win. Best one can win in practice is misuse of Spiderman trademark and whatever else is protected using well established methods, AI methods don't change one bit here. Drawing Spiderman and calling it Webmister won't save you in either way if Marvel wants to come after you. But Marvel isn't interested in coming after Little Timmy who likes spiderman and puts his illustrations to Timmy Art gallery either, unless little Timmy tries to sell them or step on Marvel's feet in other ways.

Problem with your books example is that it assumes these new books sold at cheaper price are the same books. They aren't. No writer will win in court saying books that are similar in their style, use same words in similar order and similar names are actually their books and thus infringed. All the crime book writers could sue each other to death based on this. Whether you bought the original books or read them in library or at your friends place or in bookstore or maybe grabbed it from shelf and made a run makes no difference. Same with music, there is a reason why similarity of tunes is specified and not every song that "sounds like mine" is something to cash in on.

Would you be also happy if Shutterstock banned repetitive "art" content where people rip off each others style in drawing, illustration and videography? "Your stock footage submission was removed because we already have a sunny day with palmtrees beach wide angle cinematic shot". Most of the "art" people produce is neither original nor artistic and they are angry because it turns out their craft is something a dumb box under desk can do in seconds. Clients really don't care if someone sweats over their dog grooming business logo for weeks or they get it by button push if results are indifferentiable.
I do stuff.
Online

VMFXBV

  • Posts: 579
  • Joined: Wed Aug 24, 2022 8:41 pm
  • Real Name: Andrew I. Veli

Re: The First Editing Suite to Implement StableDiffusion AI.

PostWed Oct 19, 2022 10:25 am

Hendrik Proosa wrote: But Marvel isn't interested in coming after Little Timmy who likes spiderman and puts his illustrations to Timmy Art gallery either, unless little Timmy tries to sell them or step on Marvel's feet in other ways.


This isn't about little Timmy and his experiments. We're talking about implementing this tech in a professional software where people use it to make money.

Hendrik Proosa wrote:Problem with your books example is that it assumes these new books sold at cheaper price are the same books. They aren't.


But in my example they really are. If you take a book , scan it and change the font used inside, is it not the same book just because its "written" in a different way visually? Or if you change some words with their synonyms its ok? Because this is what this "Ai" is doing.

There's a difference between a human imitating something and a piece of software that literally scans other people's work and reshuffles it. If it spat things out of thin air then ok, but its not.

Hendrik Proosa wrote:Would you be also happy if Shutterstock banned repetitive "art" content where people rip off each others style in drawing, illustration and videography? "Your stock footage submission was removed because we already have a sunny day with palmtrees beach wide angle cinematic shot".



They already do that and I'm happy for it.

Hendrik Proosa wrote:
Most of the "art" people produce is neither original nor artistic and they are angry because it turns out their craft is something a dumb box under desk can do in seconds. Clients really don't care if someone sweats over their dog grooming business logo for weeks or they get it by button push if results are indifferentiable.


None of the artists that are "angry" about it gave their permission for this piece of software to forcefully scan their stuff. It doesn't matter if you think their stuff was original or artistic or if that dumb box spits it out in a second. The OpenAI people surely think it was artistic since they used it.

Its like going to a movie set as a visitor, using your dumb box of a camera that spits out things in seconds (that you hid from the security guard), record everything that's going on, reshuffle it when you get home and then sell it because "its not 100% like they did it, its just similar and they haven't caught me so its ok".

Where do you draw the line? Its ok to steal stuff because its digital and no harm done and "no one cares" (the clients) or how does this work?

I get that Marvel would be safe, but what about the guy where his livelihood depends on selling his awful non artistic art that this bot blatantly stole and reshuffled?

Why does the bot /model / etc need to steal it / scan it / read it in the first place if its soo amazing?

Or is it ok if it only affects those loser artists? What happens when these corporations come for the rest? Musicians, Web Coders, Developers and so on?
AMD Ryzen 5800X3D
AMD Radeon 7900XTX
Ursa Mini 4.6K
Pocket 4K
Offline
User avatar

Chad Capeland

  • Posts: 3017
  • Joined: Mon Nov 10, 2014 9:40 pm

Re: The First Editing Suite to Implement StableDiffusion AI.

PostWed Oct 19, 2022 5:47 pm

VMFXBV wrote:
Why does the bot /model / etc need to steal it / scan it / read it in the first place if its soo amazing?



Interesting anchor for the argument. If the output is "original", why not use that as the input?
Chad Capeland
Indicated, LLC
www.floweffects.com
Offline

Hendrik Proosa

  • Posts: 3015
  • Joined: Wed Aug 22, 2012 6:53 am
  • Location: Estonia

Re: The First Editing Suite to Implement StableDiffusion AI.

PostThu Oct 20, 2022 7:14 am

VMFXBV wrote:Where do you draw the line? Its ok to steal stuff because its digital and no harm done and "no one cares" (the clients) or how does this work?

I get that Marvel would be safe, but what about the guy where his livelihood depends on selling his awful non artistic art that this bot blatantly stole and reshuffled?

Why does the bot /model / etc need to steal it / scan it / read it in the first place if its soo amazing?

As I have said numerous times, the procedure for ”stealing” protection is well established already, nothing has changed that. But deriving ”inspiration” is not stealing. Your argument about simple reshuffling and cloning of pieces of original artwork does not align with the reality of how diffusion works.

Why do humans study art, constantly consume other ideas and fuse the stuff they like, not live in a vacuum and produce fully original art? Can’t they do without stealing if they are so amazing artists? You are trying to attach some ethereal qualities to people while at the same time trying to point out these qualities don’t exist in AL models. They don’t, and the problem is that the more things evolve, the more these qualities evaporate from human art, because it turns out most of it was just wishful thinking. Most human activity is based on copying something that already existed and then adjusting to feedback, which produces evolvement.

As Chad pointed out, these systems could also be fully self-consuming, if feedback loop is established in a way that in the end it produces results meaningful outside this self-encapsulated system, for human users. Using already meaningful input is just a shortcut, just as studying art or engineering or cooking or whatnot is a shortcut to not having to reinvent everything every time.

If it were fully self-consumed system with no pre-existing training data and it in the end still produced images very similar to copyrighted artwork, would it somehow change the situation? Would it be original in this case?
I do stuff.
Online

VMFXBV

  • Posts: 579
  • Joined: Wed Aug 24, 2022 8:41 pm
  • Real Name: Andrew I. Veli

Re: The First Editing Suite to Implement StableDiffusion AI.

PostThu Oct 20, 2022 9:18 am

But that's what the model does. Its simple machine learning. Please don't tell me you're attaching any sort of intelligence to this. It copies from what's being fed and outputs based on some rules ''this is a pole, this is spiderman, this combines this way based on what we stole from Greg Rutkowksi''.

You're ignoring the big picture. Just because some humans steal doesn't mean its now ok to steal at a faster pace by using a machine.

Unless you say all humans steal which is mumbo jumbo.

Using already meaningful input is just a shortcut, just as studying art or engineering or cooking or whatnot is a shortcut to not having to reinvent everything every time.


How is this even remotely the same thing? Sounds like an excuse a chinese knockoff manufacturer would use.

I would be fine if the the model used data that they paid for or the ''engineers'' made themselves. This way the artists would give their permission for this machine to obsolete their work.

But as it stands, the current model is worth jack without the high quality STOLEN input its being fed.

Only people that are happy with this are those that can't draw/color/etc.(mostly coders i bet) Until the mega corporation steals their work too and automates it. Then you'll see the tantrums...
AMD Ryzen 5800X3D
AMD Radeon 7900XTX
Ursa Mini 4.6K
Pocket 4K
Offline

Hendrik Proosa

  • Posts: 3015
  • Joined: Wed Aug 22, 2012 6:53 am
  • Location: Estonia

Re: The First Editing Suite to Implement StableDiffusion AI.

PostFri Oct 21, 2022 7:47 am

I can't make it any clearer that I don't imply any intelligence in any of this, as I have said before.

This is an argumentation of semantics. What does "copy" mean here exactly? Everyone copys one way or another, but not all of this is stealing. There is a huge gray area between drawing inspiration and copying with intention to produce an emulation of existing piece. Me drawing in style of Marvel the best I can and me copying a specific illustration from a comic and telling this is my own creation are pretty far from each other.

It is pretty hard to come against the "consuming" end of this, if one can't prove that copyrighted artwork is somehow stored as direct replicas inside for example Stable Diffusion model. So it falls to fighting against the "producing" end, the output. And as I have said before and will say again, protection of intellectual and other property in this end is well established already. Same mechanisms as fighting chinese knockoffs apply: you can't make anyone pay for dissecting a Nike sneaker, only fight the produced knock-offs themselves.

We are in the phase of "cars steal the jobs of horse coaches" and "photoshop kills artists", I don't see much horses around these days, but I don't see many jobless horse coachmen either. But as with everything, these are opinions and everyone has their own, so I'll leave this at it. Time will tell where it all goes.
I do stuff.
Online

VMFXBV

  • Posts: 579
  • Joined: Wed Aug 24, 2022 8:41 pm
  • Real Name: Andrew I. Veli

Re: The First Editing Suite to Implement StableDiffusion AI.

PostFri Oct 21, 2022 9:17 am

I can't take you seriously on this. Its not even a matter of semantics.

The machine learning algorithm cannot be inspired. Its not alive. Its not intelligent. All it can do is copy what's being fed and literally photobash things into an output. Its a glorified all-in-one scanner printer.

And no, not everyone copies one way or another. That's again mumbo jumbo. You could go all the way and say we have eyes and store images in your brain therefore we copy everything we see. This kind of reasoning is abstract, borderline absurd.

It also doesn't matter if you can find stored art in the actual model itself. Deleting the material used after you've built the model doesn't excuse you from thievery. None of the artists gave their permission for their work to be used in "training of a machine learning model that spits out art and other stuff".

Crawling the web for pictures that you don't own IS THEFT. No matter how you try and spin in. They used this specific dataset and web crawled to build their stuff that its now used by others to make money.

And it not even a question of "did they use copyrighted material". There are already cases where people have found their private stuff in the dataset that they used (medical records). So this isn't even an argument.

Not to mention the prompts "in the style of" which couldn't happen without the actual art being used. And I don't need more examples, Greg Rutkowski is enough. He wasn't on board with this therefore its theft.

Once a thief, always a thief.

[edit] https://haveibeentrained.com/?search_te ... 0Rutkowski
Well there goes that...


But you have your opinion. Its ok. I'm not here to change your mind.

Hope this garbage never gets implemented in Resolve. Not the way it is right now.
Getty Images and Shutterstock banned Ai Art. Many communities banned it as well. Hope all of them do.
AMD Ryzen 5800X3D
AMD Radeon 7900XTX
Ursa Mini 4.6K
Pocket 4K

Return to Fusion

Who is online

Users browsing this forum: No registered users and 39 guests