AI is going to be our biggest game changer

WendyLuvsCatz

wolf359 said:

Gordig said:

Latest Corridor video:

And the behind the scenes:

 

@Gordig

There are already twitter threads where ”traditional animators” are furious about the Corridoor crew

restyling their live footage to look like anime.

the same cell tracers complained about 3D taking away their grossly underpaid work 

they even said Disney's layers of parallax painted glass panels was not proper animation

WendyLuvsCatz

from the Art Studio forum

Minam said:

Can we even use Stable Diffusion?

I think the EULA was updated? 
 

• License Restrictions. The Content is provided solely for User's use as permitted herein and in conformity for its intended purpose. User does not have the right to provide the Content to others in any form or on any media except as set forth in this Agreement. The Content may be copied in whole or in part solely for User's use as permitted herein. Specifically, User may copy the Content onto the storage device of an unlimited number of computers owned or controlled by User. The Content is for use solely by User and not by any other individual or entity. All individuals or entities must obtain their license to use the Content. Notwithstanding anything to the contrary contained herein, the Content License expressly excludes the use, incorporation, or input of any Content, in whole or in part, in connection with i) any AI engine, program, or system (including, without limitation, Image.ai, Nightcafe, Artbreeder, chatGPT, Shutterstock, DALL-E 2, Deep Dream Generator, Hotpot ai, DeepAI) with capabilities or instructions to auto-generate materials that are derivative, imitative, or otherwise plagiaristic of the Content; and ii) any activity contrary to the intended use of the Content License, or otherwise to circumvent the terms, restrictions, or safeguards applicable to this Content License. For the purposes of this EULA, "AI" means any method of artificial intelligence such as deep learning, neural networks, or any other similar technologies intended to consume and analyze content for the intent of auto-generating new content.

 I am presuming this EULA is only enforcable for Stability's subsequent releases not the various forks made from it when it was OpenSource

oh this is an addition to the DAZ EULA

generalgameplaying

WendyLuvsCatz said:

 I am presuming this EULA is only enforcable for Stability's subsequent releases not the various forks made from it when it was OpenSource

What license/EULA did the software come with back then?

Better safe than sorry... EULA for the software's source code excluding or independent of the models the same? Concerning lawsuits, people using the old software with the LAION~5B+- data set may face the same restriction (or dangers), if this already is part of a settlement at court. Open Source by itself could still be "all rights reserved" in theory, but they had an EULA/license back then, it should have been distributed with the source code (and maybe the download process).

 

Odd indeed that publishing "content" doesn't appear to make sense with that part of the EULA, for a non-lawyer. If the EULA doesn't specify anything better than personal use elsewhere, i don't see how distributing generated content would be ok with this one, except if this was legalese for "you distribute this EULA with your images". Given this EULA now, i would actually consult law-people, if i wanted to publish anything. Given that lots of non-law-people use the software, there should be some practical advice somewhere, for this to make sense outside of the court ruling scenario. LAION-5B ~ "for scientific purposes", seems plausible with the updated license.

 

For sure, someone with law background will create an article or a video about this very change. That's the least i would check.

(Question see below: where exactly is this EULA from?)

WendyLuvsCatz

generalgameplaying said:

WendyLuvsCatz said:

 I am presuming this EULA is only enforcable for Stability's subsequent releases not the various forks made from it when it was OpenSource

What license/EULA did the software come with back then?

honestly I have no idea, it was OpenSource which usually means free to modify redistribute 

WendyLuvsCatz

meanwhile 

http://'My wife is dead': How a software update 'lobotomised' these online lovers

Artini

It looks like it is a game over. Even my favorite channel returned to MidJourney:

 

generalgameplaying

WendyLuvsCatz said:

honestly I have no idea, it was OpenSource which usually means free to modify redistribute 

Go to the safe side :).

Open Source per se means only providing the source code, but it could still be "all rights reserved".

If it had been provided at GitHub for instance, there pretty much certainly will be a LICENSE file, e.g. from the university: https://github.com/CompVis/stable-diffusion/blob/main/LICENSE

To me this software License seems to explain cases pretty well in simple language, and i don't find the odd formula from the above post in there, but the license does contain some restrictions like "agree not to use in any way that violates ... law of any country... etc.", so in case of a court ruling, the use could already be limited for a certain data set.

 

Maybe the EULA from above in this thread is from the/an app or the/a cloud service?

(Or from the/a download from the site of one of the commercial participants, ...)

generalgameplaying

Artini said:

It looks like it is a game over. Even my favorite channel returned to MidJourney:

Commercially or ...?

I'd remain somewhat sceptical there. E.g. if the only woman from that province looks like Mickey Mouse (in the training data), literally everyone might end up with Mickey Mouse. That's just a small thing, but with random images from the internet such can always happen. Or the boy leaning against a brick wall ... how many boys are tagged (presumably many), but the selection only has one. Do i write "another boy" to get another face? The technique underneath presumably hasn't changed much, yet.

(Not judging how cool it is, to be able to do what you see in the video, actually. Training on photos would also yield abilities, and in artwork you'll probably have 'realistic' as a tag, often. So to me the surprise isn't huge per se. The examples are also somewhat limited so far, some portraits with blurred background, some landscape, and the only complex image with a person and a car seems already 'more difficult'. 'Good enough' is another question, which also needs answering by how the masses of images generated by different people will look like (or alike).)

 

Artini

Just a quick example made with the variation of the prompt from the last video

Artini

... and one more

Minam

generalgameplaying said:

For now, the technique is machine learning-based, as opposed to having a "strong AI", the latter of which would be like human intelligence. Using the term "AI" muddles the water somewhat and might lead to people expecting too much, and ads promising too much. Don't expect miracles, it'll rather be cold-blooded killers, like alpha-go or alpha-zero.

 

The current state of things isn't that great, concerning "AI"-assisted content creation, it's just "interesting", maybe on the verge of becoming something annoying. However, specifically as more specialized or cheaper and at the same time better, as well as more parallelized hardware evolves, actual tools that support you, will become better or even just feasible to build at all, meaning tools that employ machine learning (substantially, not just for the buzz-words). It's still a lot of work and research to build useful software, that actually does something with machine learning. So on the one hand, many more machine-learning-based applications pop out of the ground as we speak, specifically in audio processing, but of course also in the headlines with the top-notch stuff at the movies and somewhere in between with image/photo  applications. For Video and rendereing i'd assert, that the complexity is just one or another level higher than with typical audio and photo applications, so the big advances for video and rendering currently are rather with data centers or cloud-applications, at this point, or maybe just up to now. It's just not so easy to build something lasting, that stays useful, like "helping with animation", so not so many players will have succeeded by now, at the edge of the doable. Denoising probably is one of the simpler tasks, in comparison, but it's complex enough to not have been available at randomly low prices on each and every corner, a moment ago. In fact "AI-denoising" actually might be distributed more and more as a part of photo-software, that manufacturers of digital cameras themselves offer their buyers free of charge. So maybe we are there already, at that specific junction for photo, though there still is a lot of higher ground to cover, beyond denoising.

 

I see a problem with ai-assisted art-generating Software taking the jobs too quickly: copyright.

It's premade content but very limited, even if it's been fed "so huge an amount of content". That just for the content, and then it's further going to be limited by policies, like "no religion, no politics, no nudes". Blood and gore and violence likely will be ok on the menu, but probably will not really remain manageable. (Paint him without legs and with tomato juice and upside down, ok not, paint him walking upside down, with his leg taking a swim in the pond, we'll get there....). Let's not divert, the current thing is not enough, it needs more abilities and more content.

So where to train such a system from? There is thinkable sources, of course, but it doesn't look like it could easily work.

1. Copyright against the results. The result might be a copyright infringement, no matter how sophisiticated the system is, no matter if they own all the rights to the assets in use, the result can still infringe on other people's rights. That's probably not even special, BUT it's half a dealbreaker for the "free of charge" cloud-model, that feeds of user-generated content.

2. Deals with other platforms. Maybe platforms like deviantart ;), hopefully not though. So who posts their stuff there, will in future remove their own job by training the ai, or independently: many people train the ai (with what though...). Variations of this. Problem with user-generated content remains the copyright, which may at first seem indetectable within the innerts of the "ai" because you don't see it in the machine learning system, however it's not gone, ... infringing material from training will likely have a chance to generate infringing material as a result, in general.

3. This brings us to the old lie: "the user is safe". Likely they'll have to exclude all liability for generated images, or do it already. With current EU-copyright law there is no way to heal this, AFAIK, not having fair use, and risking high fines for repeated infringement. 

4. Contracts with the biggest organizations that hold rights on videos and images, like "the content industry" :), or in germany the VG-Bild-Kunst (compare to what GEMA is with music) and which again was it then. I don't see them making enough money to pay similar amounts of x*1000000000$ to parts of content industry and the publishing industry in europe, on short to medium term.

5. There could be peculiar alliances (content industry + big tech), which may have some idea of how to pull something off, with using somthing like "the whole internet" for it.

6 . All this will lead to upload-filters and "other kind of restrictions" of all sorts, including showing skin at all, and all the false positives we know from such systems, while the users will have to live with the uncertainty of infringing other people's rights all the time. Can anyone solve this?

 

In essence copyright looks like half a show-stopper for any fast development, but that might not hold true on the medium term. The other very big point is, that for the artist, it probably makes more sense, if the ai-system trains and learns with YOUR very asset library, because you can control the training data this way, and the tool will have to be more of a tool, than a guessing thing. Unfortunately that is a different type of beast "AI-wise". It would have to learn on DAZ-side for instance, on the one hand to train on a lot of assets, but also for retaining assignability for used products, in terms of license-management. Possibly DAZ would free their users from some of the edges, allowing lots of stuff or derived stuff in,  and/or from there it would have to learn detached, just interacting with the user and their differing assets, which may also include assets in different formats or from different vendors, which is where it gets complicated again :), but that's on the user then. The guessing thing may be fun and may be terribly good some day, but it'll likely be "like with smartphone cameras: good enough for a lot, but not everything...".

Then again, if they added a section for adults, nobody knows what'll happen. In that instance, in theory, it could become instant consumption, destroying other markets :). So i am curious if for instance producers of adult movies would license content to such a platform, knowing that it might knock them off the feet the next day. Complexity-wise generating videos (really new ones) is another level, yet to even scratch with less than a super computer.

 

So in the end... what do i think DAZ Studio could profit from, concerning machine learning?

- I think it can :). It usually means a lot of extra work, if it is something substantial, so i don't know how they'll monetize, or if they have already priced it in, or if at all...

- Of course some of the already mentioned algorithms like denoising may come, and maybe helping with animation from poses, helping with muscles from the time line and the posing, helping with dforce cloth, hair, what not, ... all the stuff that's still itchy to do with classic algorithms, but could in theory be mended at a glance. And you might still turn it off, if it can't handle something.

- Might create a learning-capable smart content tool :). Simple things like selecting or advising the best matching categories, e.g. during character setup, after deselecting, reducing clicks. Smarter sorting order with multiple panes, e.g. when i'm looking for shaders for a specific type of item/thing, there might be better choices than showing shaders or morphs from products, for which those are not the primary thing.  Selecting replacment shaders for some contexts. Machine learning might actually help in such instances. Better im- and export of formats, maybe ai-assisted (assigning textures, grouping...). Maybe internal tools to export better to other formats for gaming contexts, e.g. using ai to transform the non-compatible parts like some shaders or rigging into whatever the goal is. However, typically, if it has to run on "some laptop", it has to be baked into a machine, and the end user device is just using the trained system. That for instance means, that DAZ will have to think up and train such a system all on their own, or that some bigger or specialized player has to develop such a system and DAZ pays them license fees, to use it in Studio. Meaning, "AI" doesn't initially help DAZ developing machine learning features, so they would have to do that on their own, at their own risk. Maybe that leads to DAZ waiting forever, until it's clear that a specific application can actually work, i.e. others already having implemented such a case. I can't judge that...
 

Very interesting, catching up with this thread as I've seen new changes to Daz3D EULA recently I think. 

WendyLuvsCatz

Minam said:

generalgameplaying said:

For now, the technique is machine learning-based, as opposed to having a "strong AI", the latter of which would be like human intelligence. Using the term "AI" muddles the water somewhat and might lead to people expecting too much, and ads promising too much. Don't expect miracles, it'll rather be cold-blooded killers, like alpha-go or alpha-zero.

 

The current state of things isn't that great, concerning "AI"-assisted content creation, it's just "interesting", maybe on the verge of becoming something annoying. However, specifically as more specialized or cheaper and at the same time better, as well as more parallelized hardware evolves, actual tools that support you, will become better or even just feasible to build at all, meaning tools that employ machine learning (substantially, not just for the buzz-words). It's still a lot of work and research to build useful software, that actually does something with machine learning. So on the one hand, many more machine-learning-based applications pop out of the ground as we speak, specifically in audio processing, but of course also in the headlines with the top-notch stuff at the movies and somewhere in between with image/photo  applications. For Video and rendereing i'd assert, that the complexity is just one or another level higher than with typical audio and photo applications, so the big advances for video and rendering currently are rather with data centers or cloud-applications, at this point, or maybe just up to now. It's just not so easy to build something lasting, that stays useful, like "helping with animation", so not so many players will have succeeded by now, at the edge of the doable. Denoising probably is one of the simpler tasks, in comparison, but it's complex enough to not have been available at randomly low prices on each and every corner, a moment ago. In fact "AI-denoising" actually might be distributed more and more as a part of photo-software, that manufacturers of digital cameras themselves offer their buyers free of charge. So maybe we are there already, at that specific junction for photo, though there still is a lot of higher ground to cover, beyond denoising.

 

I see a problem with ai-assisted art-generating Software taking the jobs too quickly: copyright.

It's premade content but very limited, even if it's been fed "so huge an amount of content". That just for the content, and then it's further going to be limited by policies, like "no religion, no politics, no nudes". Blood and gore and violence likely will be ok on the menu, but probably will not really remain manageable. (Paint him without legs and with tomato juice and upside down, ok not, paint him walking upside down, with his leg taking a swim in the pond, we'll get there....). Let's not divert, the current thing is not enough, it needs more abilities and more content.

So where to train such a system from? There is thinkable sources, of course, but it doesn't look like it could easily work.

1. Copyright against the results. The result might be a copyright infringement, no matter how sophisiticated the system is, no matter if they own all the rights to the assets in use, the result can still infringe on other people's rights. That's probably not even special, BUT it's half a dealbreaker for the "free of charge" cloud-model, that feeds of user-generated content.

2. Deals with other platforms. Maybe platforms like deviantart ;), hopefully not though. So who posts their stuff there, will in future remove their own job by training the ai, or independently: many people train the ai (with what though...). Variations of this. Problem with user-generated content remains the copyright, which may at first seem indetectable within the innerts of the "ai" because you don't see it in the machine learning system, however it's not gone, ... infringing material from training will likely have a chance to generate infringing material as a result, in general.

3. This brings us to the old lie: "the user is safe". Likely they'll have to exclude all liability for generated images, or do it already. With current EU-copyright law there is no way to heal this, AFAIK, not having fair use, and risking high fines for repeated infringement. 

4. Contracts with the biggest organizations that hold rights on videos and images, like "the content industry" :), or in germany the VG-Bild-Kunst (compare to what GEMA is with music) and which again was it then. I don't see them making enough money to pay similar amounts of x*1000000000$ to parts of content industry and the publishing industry in europe, on short to medium term.

5. There could be peculiar alliances (content industry + big tech), which may have some idea of how to pull something off, with using somthing like "the whole internet" for it.

6 . All this will lead to upload-filters and "other kind of restrictions" of all sorts, including showing skin at all, and all the false positives we know from such systems, while the users will have to live with the uncertainty of infringing other people's rights all the time. Can anyone solve this?

 

In essence copyright looks like half a show-stopper for any fast development, but that might not hold true on the medium term. The other very big point is, that for the artist, it probably makes more sense, if the ai-system trains and learns with YOUR very asset library, because you can control the training data this way, and the tool will have to be more of a tool, than a guessing thing. Unfortunately that is a different type of beast "AI-wise". It would have to learn on DAZ-side for instance, on the one hand to train on a lot of assets, but also for retaining assignability for used products, in terms of license-management. Possibly DAZ would free their users from some of the edges, allowing lots of stuff or derived stuff in,  and/or from there it would have to learn detached, just interacting with the user and their differing assets, which may also include assets in different formats or from different vendors, which is where it gets complicated again :), but that's on the user then. The guessing thing may be fun and may be terribly good some day, but it'll likely be "like with smartphone cameras: good enough for a lot, but not everything...".

Then again, if they added a section for adults, nobody knows what'll happen. In that instance, in theory, it could become instant consumption, destroying other markets :). So i am curious if for instance producers of adult movies would license content to such a platform, knowing that it might knock them off the feet the next day. Complexity-wise generating videos (really new ones) is another level, yet to even scratch with less than a super computer.

 

So in the end... what do i think DAZ Studio could profit from, concerning machine learning?

- I think it can :). It usually means a lot of extra work, if it is something substantial, so i don't know how they'll monetize, or if they have already priced it in, or if at all...

- Of course some of the already mentioned algorithms like denoising may come, and maybe helping with animation from poses, helping with muscles from the time line and the posing, helping with dforce cloth, hair, what not, ... all the stuff that's still itchy to do with classic algorithms, but could in theory be mended at a glance. And you might still turn it off, if it can't handle something.

- Might create a learning-capable smart content tool :). Simple things like selecting or advising the best matching categories, e.g. during character setup, after deselecting, reducing clicks. Smarter sorting order with multiple panes, e.g. when i'm looking for shaders for a specific type of item/thing, there might be better choices than showing shaders or morphs from products, for which those are not the primary thing.  Selecting replacment shaders for some contexts. Machine learning might actually help in such instances. Better im- and export of formats, maybe ai-assisted (assigning textures, grouping...). Maybe internal tools to export better to other formats for gaming contexts, e.g. using ai to transform the non-compatible parts like some shaders or rigging into whatever the goal is. However, typically, if it has to run on "some laptop", it has to be baked into a machine, and the end user device is just using the trained system. That for instance means, that DAZ will have to think up and train such a system all on their own, or that some bigger or specialized player has to develop such a system and DAZ pays them license fees, to use it in Studio. Meaning, "AI" doesn't initially help DAZ developing machine learning features, so they would have to do that on their own, at their own risk. Maybe that leads to DAZ waiting forever, until it's clear that a specific application can actually work, i.e. others already having implemented such a case. I can't judge that...
 

Very interesting, catching up with this thread as I've seen new changes to Daz3D EULA recently I think. 

 when I quoted you from the other thread here on this page I didn't realize that is an update to the DAZ EULA

usually they make us sign in again 

it needs clarification if using and training on your own machine for your own 2D renders permitted, I understand not redistributing embeddings or other trained data

TheMysteryIsThePoint

WendyLuvsCatz said:

from the Art Studio forum

Minam said:

Can we even use Stable Diffusion?

I think the EULA was updated? 
 

• License Restrictions. The Content is provided solely for User's use as permitted herein and in conformity for its intended purpose. User does not have the right to provide the Content to others in any form or on any media except as set forth in this Agreement. The Content may be copied in whole or in part solely for User's use as permitted herein. Specifically, User may copy the Content onto the storage device of an unlimited number of computers owned or controlled by User. The Content is for use solely by User and not by any other individual or entity. All individuals or entities must obtain their license to use the Content. Notwithstanding anything to the contrary contained herein, the Content License expressly excludes the use, incorporation, or input of any Content, in whole or in part, in connection with i) any AI engine, program, or system (including, without limitation, Image.ai, Nightcafe, Artbreeder, chatGPT, Shutterstock, DALL-E 2, Deep Dream Generator, Hotpot ai, DeepAI) with capabilities or instructions to auto-generate materials that are derivative, imitative, or otherwise plagiaristic of the Content; and ii) any activity contrary to the intended use of the Content License, or otherwise to circumvent the terms, restrictions, or safeguards applicable to this Content License. For the purposes of this EULA, "AI" means any method of artificial intelligence such as deep learning, neural networks, or any other similar technologies intended to consume and analyze content for the intent of auto-generating new content.

 I am presuming this EULA is only enforcable for Stability's subsequent releases not the various forks made from it when it was OpenSource

oh this is an addition to the DAZ EULA

Does this mean that you can't use DAZ models to train AI to create morphs for better bends, because it would compete with other official JCM products? If so, DAZ is really starting to shoot itself in the foot... really intent on making itself irrelevant.

WendyLuvsCatz

TheMysteryIsThePoint said:

WendyLuvsCatz said:

from the Art Studio forum

Minam said:

Can we even use Stable Diffusion?

I think the EULA was updated? 
 

• License Restrictions. The Content is provided solely for User's use as permitted herein and in conformity for its intended purpose. User does not have the right to provide the Content to others in any form or on any media except as set forth in this Agreement. The Content may be copied in whole or in part solely for User's use as permitted herein. Specifically, User may copy the Content onto the storage device of an unlimited number of computers owned or controlled by User. The Content is for use solely by User and not by any other individual or entity. All individuals or entities must obtain their license to use the Content. Notwithstanding anything to the contrary contained herein, the Content License expressly excludes the use, incorporation, or input of any Content, in whole or in part, in connection with i) any AI engine, program, or system (including, without limitation, Image.ai, Nightcafe, Artbreeder, chatGPT, Shutterstock, DALL-E 2, Deep Dream Generator, Hotpot ai, DeepAI) with capabilities or instructions to auto-generate materials that are derivative, imitative, or otherwise plagiaristic of the Content; and ii) any activity contrary to the intended use of the Content License, or otherwise to circumvent the terms, restrictions, or safeguards applicable to this Content License. For the purposes of this EULA, "AI" means any method of artificial intelligence such as deep learning, neural networks, or any other similar technologies intended to consume and analyze content for the intent of auto-generating new content.

 I am presuming this EULA is only enforcable for Stability's subsequent releases not the various forks made from it when it was OpenSource

oh this is an addition to the DAZ EULA

Does this mean that you can't use DAZ models to train AI to create morphs for better bends, because it would compete with other official JCM products? If so, DAZ is really starting to shoot itself in the foot... really intent on making itself irrelevant.

the whole thing needs clarification and we haven't been prompted to read and accept it yet either  

postwork on 3D renders using it shouldn't be any different to using Photoshop

Training on your 2D renders and redistributing to trained model is something I can understand being iffy but is that what they mean

forbidding using ChatGPT to assist in plugin and script development seems shortsighted 

generalgameplaying

Oh right, i've been stupid. The paragraph with the restrictions is from daz3d.com. (Footer of page: "Licensing Agreement")

"[ai / system] ... with capabilities or instructions to auto-generate materials that are derivative, imitative, or otherwise plagiaristic of the Content"

It would make sense to clarify, if i could train may own ai on my own 2d images, which i created with daz3d assets. Naturally i would have to train with the images, and the result would certainly be imitation, however i own the licenses to the content, and i don't make the resulting system available to anyone else. Just for a scenario.

WendyLuvsCatz

generalgameplaying said:

Oh right, i've been stupid. The paragraph with the restrictions is from daz3d.com. (Footer of page: "Licensing Agreement")

"[ai / system] ... with capabilities or instructions to auto-generate materials that are derivative, imitative, or otherwise plagiaristic of the Content"

It would make sense to clarify, if i could train may own ai on my own 2d images, which i created with daz3d assets. Naturally i would have to train with the images, and the result would certainly be imitation, however i own the licenses to the content, and i don't make the resulting system available to anyone else. Just for a scenario.

I initially didn't realise it was from the DAZ EULA either, just copied the quote from the art studio thread where it didn't belong

I am more annoyed they snuck it in without informing customers than anything 

we should have to reread and agree to the EULA everytime it's amended 

since I load DIM from the sale confirmation page with my receipt it should have happened then

there have been a few DIM updates too where we should have been diverted to that page with our default browser

manual downloads usually trigger it too if it's updated

Artini

They talking about restrictions of use of the content,

but there is nothing about the use of the renders (or videos, which are made from the renders).

The official clarification of this issue would be great.

"the Content License expressly excludes the use, incorporation, or input of any Content, in whole or in part, in connection with i) any AI engine, program, or system (including, without limitation, Image.ai, Nightcafe, Artbreeder, chatGPT, Shutterstock, DALL-E 2, Deep Dream Generator, Hotpot ai, DeepAI) with capabilities or instructions to auto-generate materials that are derivative, imitative, or otherwise plagiaristic of the Content;"

 

generalgameplaying

Artini said:

They talking about restrictions of use of the content,

but there is nothing about the use of the renders (or videos, which are made from the renders).

The official clarification of this issue would be great.

"the Content License expressly excludes the use, incorporation, or input of any Content, in whole or in part, in connection with i) any AI engine, program, or system (including, without limitation, Image.ai, Nightcafe, Artbreeder, chatGPT, Shutterstock, DALL-E 2, Deep Dream Generator, Hotpot ai, DeepAI) with capabilities or instructions to auto-generate materials that are derivative, imitative, or otherwise plagiaristic of the Content;"

Good point. With "whole or in part" we're at pretty elastic formulas, i'd bet. I could imagine stores going overly strict at first, for one scenario, but for another one, that they may be considering their own ai service, just in theory. In the latter case people training their own ones could be seen as some kind of competition, though i think the buyers of DAZ content aren't really the competition here. Never know...

 

Edit: I'm also not 100% sure concerning "input". Put bluntly, using an ai filter on a rendered image from DAZ assets, not training it with them... same with inpainting and modification, e.g. i train some ai on different input to replace faces by my face and apply it to an image made from DAZ assets. I'd assume that's not meant to be restricted, they also allow photoshop brushes and so on, but can't really know. In particularl, that kind of systems are meant to modify given content, which i assume is not intended to be restricted. But i'd still prefer to have such clarified, because i never really manage to get used to broad/elastic formulas.

nonesuch00

Gordig said:

Latest Corridor video:

And the behind the scenes:

Those are pretty slick, thanks for posting.

Artini

Still trying to find out, how to create similar images without AI.

Artini

... or that one ...

Artini

How to not be amazed by such images...

PixelSploiting

There might be more clouds over the horizon for the AI image generators because of some highly smart (not really) folks using it to generate fake news images. It flopped almost immediately, but it seems legal atmosphere is becoming more hostile in general. 

 

tl;dr: One can always count on a few bright individuals breaking it for everyone.

WendyLuvsCatz

are already fake merchandise sites on Facebook using obvious Ai images

clothing, bags etc

they stole actual photos before so not a big leap

 

like this lady's work

Artini

Have fun with the vehicles...

Artini

Another vehicle...

TBorNot

I'd be happy if AI could place a figure's feet on the ground and keep the balance.  You know, basic posing.

RiverMissy

From what I have seen I think we will be replaced by the AI art unfortunately.

wolf359

TBorNot said:

I'd be happy if AI could place a figure's feet on the ground and keep the balance.  You know, basic posing.
 

The Auto posing feature of Cascadeur can do this with AI assist.

Ippotamus

Art, music, writing ... amazing how meaningless it is all becoming for me, and so quickly.

I used to love looking at all the pictures and creations of the digital age.  Now I have to wonder if it is all just AI inspired and/or created.

Soulless... vapid and mass-produced.

Going to pluck a three-stringed banjo on the street corner now.

At least I will feel alive.