AI Created images (generated by Midjourney) lose registered copyrights

[porina]

48 minutes ago, YellowJersey said:

But the law is what the law is... until it isn't anymore.

Agree, but what gets applied now is what we have now. That is what's being discussed. It doesn't matter today what may or may not change in the distant future.

[LAwLz]

12 hours ago, goodtofufriday said:

This is what I mean by disingenuous. 

You want to argue now how a pen and paper aren't an exact 1:1 to a digital pen. 

The point is both instruments draw a line. Draw enough types of lines and you get an image. 

 

You want to argue how blender and clay aren't 1:1. 

The point is both require you to be able to mold a form into something. The creative process of it all.

I feel like you are being very vague and loose with your definitions right now but okay... How does a camera fit into this equation?

What physical tool is a camera like? What creative process is involved with taking a picture?

 

 

12 hours ago, goodtofufriday said:

You either miss or entirely ignore the artistic and creative process. 

Please define "artistic and creative process", because in my mind there is most certainly a creative and artistic process to Photoshopping an image, taking a picture or generating an picture using an AI art program.

 

 

12 hours ago, goodtofufriday said:

A photo of the grand canyon involves where the sun is hitting, what angle and elevation you're standing at, type of lens, zoom levels. Etc. 

All those elements create a scene that was made by a human. Intentional or not. And thats what makes a photo copyrightable. It's not because it's simply a photo.

Okay, but in that case why not apply the exact same logic to AI art generation and say that the prompts were made by humans, intentionally or not, and that makes the output copyrightable just like the output from a camera is copyrightable?

 

 

12 hours ago, goodtofufriday said:

And if I'm not mistaken didn't taren create a lot of those short cuts himself? Short cuts that are definable? A series process that human made? 

Doesn't sound like a ai generative process there to me.

So if I make a script that runs the AI art program, then I should have the copyright for the outputs?

Because Taran did not write the functions that were called by his scripts himself. He just made a script that called code someone else wrote and bound those scripts to keys on his keyboard.

 

 

I feel like you are doing mental gymnastics to try and carve out very particular requirements that makes it so that anything you are okay with fits the definitions, but not AI art.

In what way is pressing a button on a keyboard to execute a preprogrammed function in Adobe After Effect any different than pressing a button on a keyboard to execute a preprogrammed function in Stable Diffusion?

 

Both of them are done using a keyboard, a very distinctly computer-related item that has no older roots like a paint brush or pencil. It does not exist in the natural world.

Both of them are using commands inputted from a human to the keyboard that then gets interpreted by a computer to execute preprogrammed functions that are mapped to fairly high level concepts (things like "apply blur", "create a sphere", "zoom in", "replace all green with a picture of space", etc). When someone tells After Effect to replace all green screen in a video with an image, they are in practice doing the same thing as someone telling Stable Diffusion to generate an image. The functions the computer runs work differently, but from the humans perspective they are just giving vague commands to a computer to achieve a result they hope they are satisfied with.

Both processes involves quite a lot of trial and error as well. I think anyone who has worked with green screen will tell you that they have to tweak settings to get the desired results. They won't be able to tell you exactly how the result looks like just from the input they gave the computer. They have to look at the result and then change settings, just like someone using an AI art program would.

 

 

 

12 hours ago, goodtofufriday said:

Actively typing in an idea to a program that does the creative process for you does not rise to the level copyrightable work. Because said person is not doing the creative process work themselves. 

I would argue that the "creative process work" when taking a picture is done entirely by the camera. Ever opened a JPEG that was generated by a camera? Let me tell you, the human who asked the camera to generate would have no idea how to crate that by themselves using let's say a pen and paper. They 100% rely on a computer to generate the image for them. The human might give the computer some rather vague inputs by pointing the camera a certain way, maybe changing some parameters, but as soon as the human presses the shutter button their input stops mattering and it is 100% up to the camera to do all of the actual computing which is what results in an image being generated.

 

When someone takes a picture with a camera, the human gives the computer (camera) some inputs and then tells it to generate the output for them. The human might have some idea what they want the picture to look like and therefore change the input by moving the camera for example.

 

When someone creates an image using an AI program, the human gives the computer some inputs and then tells it to generate the output for them. The human might have some idea what they want the picture to look like and therefore change the input by changing the prompts for example.

 

 

If we are going to argue that the "creative process" is what's important then I genuinely do not see how AI art generation is any different than taking a picture with a camera. Especially since modern cameras use AI to enhance the images in ways a human can't do.

[Kisai]

2 hours ago, porina said:

Agree, but what gets applied now is what we have now. That is what's being discussed. It doesn't matter today what may or may not change in the distant future.

Ultimately I feel that lobbyists might push to allow "AI inventions/copyrights" but any such inventions would require the AI reach a level where it actually has thought and intent behind it's composition, which right now, it doesn't.

 

The AI art generators are essentially going "what do you want to see?" and then it pulls the keywords and weights from it's database, finds the relationship between them, and arranges it by the weights. That's not a creative decision, that's getting 10 dog "puzzle"'s for each keyword and haphazardly shuffling the pieces around without any regard for how they fit together. It can't put any pieces in that it doesn't have. You can't tell it to do anything post that initial prompt to change the inference of that initial prompt. It's basically just changing the weights, so going from "in front of" to "behind a" will end up with two very different images, where as a human artist would probably change the layers in photoshop.

 

Like when they're being compared to auto-complete, that's very much an accurate assessment. It's not considering the subject of the work as individual symbols, but as a single symbol.

 

[TOMPPIX]

Human artists unconsciously draw inspiration from other artists. Isn't the ai also doing the same?

[Kisai]

40 minutes ago, TOMPPIX said:

Human artists unconsciously draw inspiration from other artists. Isn't the ai also doing the same?

Nope.

 

If I ask a human artist to draw a dog, they will draw a dog, and I know it's a dog.

If I ask an AI to create an image of a dog, it can only use what was been tagged as "dog" in it's training corpus that left a latent impression in the model. Idiots on the internet go "AI steals my artwork, rawrgarble..." when that's not the case. 

 

The AI doesn't know to use the same dog. It just gives you an auto-complete of "dog" based on what the most things tagged "dog" are. It has no idea what a dog is.

 

That's why AI art generators suck at hands, and making eyes point in the same direction, because these are details that are rarely tagged. The AI has no reference for what "two human hands with 4 fingers and a thumb" is.

 

Like trying to get the AI to create something that doesn't already exist is difficult because the AI has no imagination.  It can only use what it's seen before, and it only recalls a latent image, if anything. A human goes "I think this fellow needs a hat" and if they have no idea what kind of hat they should have, they look at references of hats. They don't copy pieces of 10 different styles of hats and assemble it on their drawing. a human knows that a ballcap and a fedora are not a trillby, but the AI doesn't know that, and the AI doesn't know that trillby's are frequently mislabeled as fedora's.

 

The AI has no means of being inspired. A human might go "well maybe I want a white fedora like that Indiana Jones guy, but with a different color ribbon" the AI, will not understand the parts of the hat, because they're not typically labeled.

 

 

[Kinda Bottlenecked]

4 hours ago, LAwLz said:

What creative process is involved with taking a picture?

 

Many actually. You'll find that many of the artistic concepts that you can apply on traditional mediums like pen and paper, can also apply to photography. Lighting, values, colors, framing, shapes, line etc. All of which you can use to help you compose the image. The camera would just be a tool to help you capture your composition. 

 

Very much unlike how you'll interface with AI art generators of today.

[wanderingfool2]

18 hours ago, LAwLz said:

The line should be drawn somewhere, and right now it seems like the line is drawn at "if the program mentions AI, it's not copyrightable" because the people who make these rules don't understand how it works.

Well I mean in all fairness to the current set of rules, I don't think anyone would have imagined AI would get to the point where it could "draw" or be "creative" in any sense of the matter...and the law in itself sort of makes sense.

 

The major case, and one similar to this is when PETA sued over the monkey selfie (claim the animal should be granted the rights).  If the rule was in place it would have caused a whole lot of issues, as then if animals are involved in a copyrighted work then other groups could claim issues with the copyright.  With that said as well, that also meant the selfie was in public domain, which I don't agree with.  In my opinion the  owner of the equipment should get the copyright on it, but at the same time they only should get the copyright on the image as it exists (so people could make their own drawings of it, or their own versions...just not copying the image directly).

[divito]

18 minutes ago, Kinda Bottlenecked said:

Many actually. You'll find that many of the artistic concepts that you can apply on traditional mediums like pen and paper, can also apply to photography. Lighting, values, colors, framing, shapes, line etc. All of which you can use to help you compose the image. The camera would just be a tool to help you capture your composition. 

 

Very much unlike how you'll interface with AI art generators of today.

All of those concepts can be programmatically built into AI, and you can weight them as well.

And while photographers can use some of those concepts when they try and capture something, that doesn't negate those that don't consider those concepts and take good pictures, and utilize AI and other filters to alter their composition.

[Kisai]

23 minutes ago, wanderingfool2 said:

  With that said as well, that also meant the selfie was in public domain, which I don't agree with.  In my opinion the  owner of the equipment should get the copyright on it, but at the same time they only should get the copyright on the image as it exists (so people could make their own drawings of it, or their own versions...just not copying the image directly).

No, that's a very dangerous idea. Because if that's the case, then all productions would no longer come on physical media. No posters, no dvd's, no blueray's, no t-shirts, because this would be saying the person or company who owned the printer has the right to make infinite copies of the thing had them made, an even sell them without your permission.

 

The right outcome was reached. No copyright for any image that is not initially created by a human. If the AI creates it, or a fractal algorithm creates it, or any other algorithm-generated (such as a rogue-like characters or dungeons) asset is no longer eligible for copyright.

 

Like imagine someone spending an hour creating a character in a game that lets you have 100 parameters to pick from (eg like in VROID) who owns the copyright? The developer of the software? Or the person who sat there and picked all the various presets?

 

Because if you create a character in a game, at present the game developer claims 100% ownership over your character, but when you create one in VRoid, there is NO claim of ownership, and you can further export that model into blender or unity and make even further changes. How is that different from "prompts" ?

 

So the theory goes, if you directed that creation of that model, composed it, made all the changes to the textures and so forth, because it didn't originally come from your mind, it's not copyrightable?

 

Like as much as this ruling kinda says that anything generated by midjourney is basically PD, it also opens the door to claim any character made by a character creator in a game is too if it did not first exist as a drawing by a human.

 

[Brooksie359]

18 hours ago, LAwLz said:

I mean, isn't that what we do all the time with computers?

When I apply an effect using After Effect, I tell a computer what I want it to do, and it does all the computation for me automatically in the background. I don't think it would make sense to attribute copyright of a video I edited in After Effect to Adobe just because they wrote the program which did the work for me.

If you took what the AI did and modified it I could see getting copyright to it but when it's just you asking the AI to make something I do not see how that is anywhere near the same as someone using after effect. 

[wanderingfool2]

10 minutes ago, Kisai said:

No, that's a very dangerous idea. Because if that's the case, then all productions would no longer come on physical media. No posters, no dvd's, no blueray's, no t-shirts, because this would be saying the person or company who owned the printer has the right to make infinite copies of the thing had them made, an even sell them without your permission.

You are misconstruing what I was saying.  The way I was saying, printers wouldn't get the rights.  It would the the owner of the equipment that created the copyright, which at that stage a copyright is attributed to someone (so reprints/etc wouldn't be counted as there is already a copyright holder, even if that wasn't the strict case though physical media creation would be under contract so it would then be the person who order the printing of it who would own the copyright).

 

14 minutes ago, Kisai said:

The right outcome was reached. No copyright for any image that is not initially created by a human. If the AI creates it, or a fractal algorithm creates it, or any other algorithm-generated (such as a rogue-like characters or dungeons) asset is no longer eligible for copyright.

Then you start getting into the argument that I was making before though.  If you generate artwork that was trained off of your current work, you are not no longer capable of copyrighting it based on what you are saying.

 

You can have varying levels of copyrighted work, and it was designed like that so that if someone felt their copyright was infringed they could try making the argument (i.e. no black and white border which they have now set).  As an example of drawing a stickman.  I own the copyright to the stickman picture I drew, but if someone else draws pretty much an identical one I wouldn't be successful in a lawsuit (as images of a stickman don't hold much creative work so it's very minimal in what can be claimed).  If you were to directly copy the stickman though, then it would be infringement.

 

My suggestion is that AI generated works should be granted, and essentially held that that sort of standard (where the input required and uniqueness/usage comes into play).  The person who sues for infringement would need to show what was used to generate such and image, and then it would be up to the judge/jury to determine whether or not the infringement had occurred.

 

As an example, I lack art skills but could program up an AI that takes copyright expired mickey mouse pictures that could redraw Mickey (ignoring trademark here).  If I use the AI along with a skeleton structure to draw up scenes for an animated movie idea, your claim is that I don't own the copyright to it, as no "art" is drawn by myself...I just provided the direction in what I wanted to see.

 

Would you consider the above example as being copyrightable?  How is it different than if I tasked a human to do it (given the same movements I wanted).

 

If you do a full scan of a 3D environment, and make a movie where you sweep through the environment, should that be null and void?  After all, the lidar equipment (not you) is doing the scan and calculations and generating a point cloud and the computer is doing the work of a point cloud to polygon...then you just setting a few keyframes on where the camera will be.  The computer itself is doing the Cubic Hermite splines, bezier or whatever to generate all the points in between.

 

In the above scenario the computer is pretty much doing all the work.  The only thing you did was minimal input (placing the equipment), then settings a few waypoints.  Under the current system one could argue that AI is doing it all and thus it shouldn't be copyrightable then.

[Kisai]

7 minutes ago, wanderingfool2 said:

 

 

Then you start getting into the argument that I was making before though.  If you generate artwork that was trained off of your current work, you are not no longer capable of copyrighting it based on what you are saying.

 

If you are the sole source of the input images, then yes, you should be able to copyright that, as long as the model remains private. As soon as you sell or let someone else use that model, then you've established that someone else is able to duplicate your work, and thus get to the other half the problem where it's a transformative work and not a new one. You can never claim ownership of a transformed or derivative work.  Someone can take that model, add their own art to it, yet still be able to produce things that are somewhere in between theirs and yours.

 

The point that people seem to want to ignore here is where the initial creation took place. A model based on a billion photos by 10,000 people, is not transforming 1 photo into another. It's not inspired by anything in the training data. All it has are keywords and some weights associated with that keyword. Those weights may form a full fledged symbol, or it may form just specific deltas with other weights.

 

Like if you train any machine learning, you'll know the term "overfit", which is when the same keywords encounter the same training data repeatedly. Even if that's from a different sample. So the consequences of having a limited input (eg only your own art) is that most of the works will end up overfit, even if you add someone elses works to it. The base model has already overfit. To avoid overfitting you have to start the training from scratch and reduce the amount of times the overfit data shows up.

 

That will be the Achilles heel of machine learning. It doesn't know jack. But if you feed it the same training data, repeatedly, it will continue to adjust the weights until that IS the only thing in the model. Training requires a diverse amount of input with almost no repeated data to actually be efficient and have nothing overfit. 

 

That can be accomplished by curating the data or that can be accomplished by having datasets opt'd in from collections that have already been curated correctly and accurately.  But you can't copyright something that training on these collections results in because the inference is not actually thinking creatively on how to assemble an image. 

 

So again, unless you're Disney, you're unlikely to have the art resources to actually have an AI based entirely on artwork you own to do this. 

[Lightwreather]

7 hours ago, LAwLz said:

n what way is pressing a button on a keyboard to execute a preprogrammed function in Adobe After Effect any different than pressing a button on a keyboard to execute a preprogrammed function in Stable Diffusion?

Just to add to this, but Adobe has been including a lot of ML (ie. AI) based features in Photoshop and After-effects. eg. Include the Roto-brush 2.0, the Sky replacement features, Neural Filters, etc. Honestly, Some of these reimplements features that have existed but would take hours to do (especially rotoscoping, bidoof that was pain) and makes it happen in a few minutes instead.

[wanderingfool2]

1 hour ago, Kisai said:

If you are the sole source of the input images, then yes, you should be able to copyright that, as long as the model remains private. As soon as you sell or let someone else use that model, then you've established that someone else is able to duplicate your work, and thus get to the other half the problem where it's a transformative work and not a new one.

Under that logic, I could argue that all compressed works sold would not be copyrightable.  Or by that logic any software under the MIT license is void of copyright...which isn't the case.

 

The ability to duplicate ones work doesn't make it not copyrightable.  If you take a picture of a mountain, you OWN the copyright for that specific photo.  If a news outlet sees your photo and uses it in their news piece they would be infringing.  Now if they were at the exact same location and took the exact same shot, then they could use their photo (despite it being exactly the same).

 

1 hour ago, Kisai said:

You can never claim ownership of a transformed or derivative work.  Someone can take that model, add their own art to it, yet still be able to produce things that are somewhere in between theirs and yours.

Actually you can claim ownership, but it's going to depend on the rights surrounding what you originally derived the work from.  As an example, there are some pictures that are reference pictures and if you had purchased the reference pictures and copied it manually pixel by pixel it would be allowed (and used in your work), while just using the jpg image itself would not be allowed.

 

It's a similar concept that the "MIT-0" license in software spawns.  If I were to make an image software and put the source code online under the "MIT-0" license, I would own the copyrights to the source code.  If someone comes along, forks it makes it closed source and makes changes to the software then yes that derived work is their copyright.  While they wouldn't be able to pursue people who still use the original source they could pursue people who utilized their software.

 

1 hour ago, Kisai said:

So again, unless you're Disney, you're unlikely to have the art resources to actually have an AI based entirely on artwork you own to do this. 

Early Disney works are no longer under copyright claim, and while you are saying that AI doesn't have certain concepts...things are drastically changing still and the models are changing.  There could easily be a time where a specific model is trained to lets say "understand" enough of the concept of fingers to draw them.  There will be models eventually that can see an image and "understand" the concepts of it in one look (not current models, as it's highly dependent on back prop).

 

There's enough stuff out there that one probably could make an AI that could draw Mickey Mouse  in all perspectives and poses.  So again, it leads me back to what I said, if I make a movie using AI as the artist of Mickey Mouse under what was stated I cannot hold the copyright on any individual frames (and thus the movie), as I did nothing but tell it what I wanted to see.  This is despite it being publicly available images.

[Kisai]

1 hour ago, wanderingfool2 said:

 

 

There's enough stuff out there that one probably could make an AI that could draw Mickey Mouse  in all perspectives and poses.  So again, it leads me back to what I said, if I make a movie using AI as the artist of Mickey Mouse under what was stated I cannot hold the copyright on any individual frames (and thus the movie), as I did nothing but tell it what I wanted to see.  This is despite it being publicly available images.

Have you not seen behind-the-scenes videos from Disney? They retain all the hand-painted backgrounds and reference materials for every production. It's entirely possible that Disney could train a model on their house style, from any time period, because they just need to feed it the materials they already have.

 

Warner Bros owns a lot of IP, but they don't have a "house style", so if they were to do the same, they would have to make a separate model per style to get something like that, and unless they're going to make additional seasons to something they've already canceled, that's not a help.

 

Anyway the point stands that you *can not* copyright something that didn't start as a drawing you had control over. Hence Disney could, because they have the original materials in a vault. 

[wanderingfool2]

19 hours ago, Kisai said:

Have you not seen behind-the-scenes videos from Disney? They retain all the hand-painted backgrounds and reference materials for every production. It's entirely possible that Disney could train a model on their house style, from any time period, because they just need to feed it the materials they already have.

Are you trying to ignore my point?  It doesn't matter if Disney has all that stuff still...I'm saying that there is a lot of material that is now classified as part of public domain.  You are 100% allowed to use public domain works and create your own versions.

 

19 hours ago, Kisai said:

Anyway the point stands that you *can not* copyright something that didn't start as a drawing you had control over. Hence Disney could, because they have the original materials in a vault. 

That's not true.

 

As a more straight forward example, since you can't seem to grasp it.  If you take Moonlight Sonata 3rd movement (written over 200 years ago, so 100% in public domain), you are 100% allowed to copyright the playing of the piece.  You own the copyrights to that performance even though you did nothing but play it.

 

If you use a program and put in all the notes and create your own MP3 of it, you own the copyright that.  It won't have much pull in the sense that in the case you would have to literally prove they copied your MP3 itself but you still own the copyright.

 

According to how things are being ruled, even if Disney created their own AI that could draw everything (based on the older works) it wouldn't be copyrightable.

[LAwLz]

6 hours ago, Kinda Bottlenecked said:

Many actually. You'll find that many of the artistic concepts that you can apply on traditional mediums like pen and paper, can also apply to photography. Lighting, values, colors, framing, shapes, line etc. All of which you can use to help you compose the image. The camera would just be a tool to help you capture your composition. 

 

Very much unlike how you'll interface with AI art generators of today.

Please describe how it is different from AI art generators.

Lighting, values, colors, framing, shapes, lines, etc are all things you can dictate and control in AI art generator programs. 

 

 

I also think you are overestimating how much control a human has over a modern camera.

What is being captured is not what the human composed. The camera will "fix" the composition by itself in the background with no input from the human. For example in the case of Google's camera, it uses AI to analyze the composition and then alter the colors based on what it thinks the image should look like.

When I point my camera around, it automatically alters where the lens is located in physical space because it counteracts my hand movements it thinks are unintentional. If I move my camera 10mm to the left, my camera might think "no, the image looks better 9mm to the left" and then physically move the lens elements by 1mm in the opposite direction of my hand movement.

When I click my shutter button, my camera is not even taking one image. It takes several images then stitches them together in a way it thinks makes the most sense.

 

The image I see in my viewfinder is very, very different from the image that gets outputted by the imaging pipeline. This is even more true for images taking in automatic mode, which is like 99% of pictures these days.

Long gone are the days of photography being a manual thing that humans control. Most of the work is being done by the camera trying to figure out what the human operating the camera wants. There is a reason why photography used to require a lot of skill and knowledge, and nowadays pretty much everyone can take good looking pictures in seconds. What used to be a profession that people dedicated their lives to is now something my 8 year old niece can do, because it has been automated through computers.

 

 

 

  

5 hours ago, Brooksie359 said:

If you took what the AI did and modified it I could see getting copyright to it but when it's just you asking the AI to make something I do not see how that is anywhere near the same as someone using after effect. 

Why do you not think it is the same?

Is someone using After Effect to apply a green screen effect not "just asking the computer to make something"? It's not like the human sitting at the computer is manually calculating the mathematics necessary to make that work. They just push a button and the computer does it for them. I doubt most editors even understand how it works. If someone asked an editor like Taran to explain how the computer actually computes the green screen effect they asked it to apply I doubt they would be able to explain it in terms of how the code functions. They 100% rely on someone else doing all the math and programming the function for them, so that they can push a button and have it applied. And there is nothing wrong with that.

 

These programs are just tools that people can use to express themselves. What annoys me is when people try and say "I think this tool is okay to use but not that one". To me, saying that using a camera is art and has a human element to it is like a carpenter saying a nail gun is cheating and you are only allowed to call yourself a carpenter if you use a hammer, and then trying to justify it like "carpenters don't rely on electricity", meanwhile they stand there operating a power drill and then goes "this doesn't count...".

[Kisai]

49 minutes ago, wanderingfool2 said:

Are you trying to ignore my point?  It doesn't matter if Disney has all that stuff still...I'm saying that there is a lot of material that is now classified as part of public domain.  You are 100% allowed to use public domain works and create your own versions.

Good luck getting a version that's in the public domain. Finding a 1920's-era film that you have to scan, ain't gonna happen.

 

And before you go "you don't need that", yes you so, as that's the version that's in PD, not a DVD/BD version printed later, and this is the same criteria we use for dumping floppy discs, cd-roms and game roms. You have to do it yourself from the original media.

 

49 minutes ago, wanderingfool2 said:

That's not true.

 

As a more straight forward example, since you can't seem to grasp it.  If you take Moonlight Sonata 3rd movement (written over 200 years ago, so 100% in public domain), you are 100% allowed to copyright the playing of the piece.  You own the copyrights to that performance even though you did nothing but play it.

No, you can only have a copyright on the physical recording. That's why people keep getting nailed for copyrights on public domain music, even when they play it live on Youtube. The bot doesn't know the difference between a 1920 recording and a 2020 recording.

 

49 minutes ago, wanderingfool2 said:

According to how things are being ruled, even if Disney created their own AI that could draw everything (based on the older works) it wouldn't be copyrightable.

They probably could given that they could name everything in the training corpus, therefor they can draw a line directly from "this is what we trained it on" to "this is what we had it create for us". 

 

But if we're going to nitpick things in that way

https://www.cined.com/netflix-uses-ai-to-generate-anime-short-film-reactions-follow/

 

Then which parts of this film can't be copyrighted? The few panels in the end credits that shows the process?

 

 

 

[YellowJersey]

Copyright law also requires a degree of effort to be put into the work in order for it to be protected by copyright. You can't just load up a book, find/replace all the names, and all it a day. My guess is that since the AI does most of the work and the user just enters prompts doesn't meet the threshold for skill and judgment put into a work.

[TOMPPIX]

16 hours ago, Kisai said:

Nope.

 

If I ask a human artist to draw a dog, they will draw a dog, and I know it's a dog.

If I ask an AI to create an image of a dog, it can only use what was been tagged as "dog" in it's training corpus that left a latent impression in the model. Idiots on the internet go "AI steals my artwork, rawrgarble..." when that's not the case. 

 

The AI doesn't know to use the same dog. It just gives you an auto-complete of "dog" based on what the most things tagged "dog" are. It has no idea what a dog is.

 

That's why AI art generators suck at hands, and making eyes point in the same direction, because these are details that are rarely tagged. The AI has no reference for what "two human hands with 4 fingers and a thumb" is.

 

Like trying to get the AI to create something that doesn't already exist is difficult because the AI has no imagination.  It can only use what it's seen before, and it only recalls a latent image, if anything. A human goes "I think this fellow needs a hat" and if they have no idea what kind of hat they should have, they look at references of hats. They don't copy pieces of 10 different styles of hats and assemble it on their drawing. a human knows that a ballcap and a fedora are not a trillby, but the AI doesn't know that, and the AI doesn't know that trillby's are frequently mislabeled as fedora's.

 

The AI has no means of being inspired. A human might go "well maybe I want a white fedora like that Indiana Jones guy, but with a different color ribbon" the AI, will not understand the parts of the hat, because they're not typically labeled.

 

 

Well then the tech is not ai it's just good coding, mixing images together. 

[05032-Mendicant-Bias]

19 hours ago, Kisai said:

The AI art generators are essentially going "what do you want to see?" and then it pulls the keywords and weights from it's database, finds the relationship between them, and arranges it by the weights.

I mean, yes. Who knows that human creativity isn't just a more refined version of this core process? Human artists do spend time looking at art made by others, practicing the skills needed to replicate/create art, and eventually after much training they make their own style.

I'm much better as coder than as artist, and ChatGPT process of looking at textbooks, stack overflow queries and github snippet of codes to glean for their intent (keywords), than adjusting and arranging them to fit a query IS what I'm doing. I think I'm more proficient, especially with weird quirky algorithms, but ChatGPT does take me 90% of the way there often.

 

ChatGPT beats me hands down in research. The question: "do you know a library to let me pilot a delta robot?" yields me with very relevant answers, much quicker than a deep google search dive into the topic. ChatGPT can't  understand math (yet), so it cannot solve the inverse matricies for a four motor delta, but neither can google.

 

My point is that I don't claim AI generators are perfect. I claim they do something no tool that came before was capable of doing, and they add value in a fundamental new way. I believe AI Generators are part of the next wave of automation, and automation always win. AI Generators cannot be stopped, we can only adapt to the fact that we live in a world where AI Generators exist.

8 hours ago, Kisai said:

Then which parts of this film can't be copyrighted? The few panels in the end credits that shows the process?

I suspect that since AI Generators work, law will be eventually redesigned to accomodate their existence, one way or the other.

I do hope that law will be made to favour small artists and coders, rather than favour large owners of IP (Disney).

What I Love about ChatGPT and DallE is that they are accessible to everyone. No matter what. I couldn't make images. DallE came. Now I can make images. For how flawed those tools are, you can access them, and have a pipe to partially refined human knowledge. It lowers significantly the barrier of entry to technical fields. Currently there is a large chance that the output is a bold lie, but that just requires the users to be trained in mistrusting the output.

[Arika]

10 hours ago, YellowJersey said:

Copyright law also requires a degree of effort to be put into the work in order for it to be protected by copyright. You can't just load up a book, find/replace all the names, and all it a day. My guess is that since the AI does most of the work and the user just enters prompts doesn't meet the threshold for skill and judgment put into a work.

well ignoring the prompt part i train my own models, create hypernetworks and embeddings. all in all i would say i've spent over 100 hours fine tuning my models to get them how i want them to be. is that not sufficient effort?

 

Spoiler

i don't actually super care about copyrighting AI art, i'm mainly playing devil's advocate.

 

[Kinda Bottlenecked]

14 hours ago, LAwLz said:

Please describe how it is different from AI art generators.

Lighting, values, colors, framing, shapes, lines, etc are all things you can dictate and control in AI art generator programs. 

 

On 2/24/2023 at 9:03 PM, Kisai said:

If I ask a human artist to draw a dog, they will draw a dog, and I know it's a dog.

If I ask an AI to create an image of a dog, it can only use what was been tagged as "dog" in it's training corpus that left a latent impression in the model. Idiots on the internet go "AI steals my artwork, rawrgarble..." when that's not the case. 

 

The AI doesn't know to use the same dog. It just gives you an auto-complete of "dog" based on what the most things tagged "dog" are. It has no idea what a dog is.

 

That's why AI art generators suck at hands, and making eyes point in the same direction, because these are details that are rarely tagged. The AI has no reference for what "two human hands with 4 fingers and a thumb" is.

 

Like trying to get the AI to create something that doesn't already exist is difficult because the AI has no imagination.  It can only use what it's seen before, and it only recalls a latent image, if anything. A human goes "I think this fellow needs a hat" and if they have no idea what kind of hat they should have, they look at references of hats. They don't copy pieces of 10 different styles of hats and assemble it on their drawing. a human knows that a ballcap and a fedora are not a trillby, but the AI doesn't know that, and the AI doesn't know that trillby's are frequently mislabeled as fedora's.

 

The AI has no means of being inspired. A human might go "well maybe I want a white fedora like that Indiana Jones guy, but with a different color ribbon" the AI, will not understand the parts of the hat, because they're not typically labeled.\

 

Kisai's post describe it pretty well. These models aren't context aware yet. Lighting, values, colors, framing, shapes, guiding lines aren't learned in the model. 

 

14 hours ago, LAwLz said:

I also think you are overestimating how much control a human has over a modern camera.

What is being captured is not what the human composed. The camera will "fix" the composition by itself in the background with no input from the human. For example in the case of Google's camera, it uses AI to analyze the composition and then alter the colors based on what it thinks the image should look like.

When I point my camera around, it automatically alters where the lens is located in physical space because it counteracts my hand movements it thinks are unintentional. If I move my camera 10mm to the left, my camera might think "no, the image looks better 9mm to the left" and then physically move the lens elements by 1mm in the opposite direction of my hand movement.

When I click my shutter button, my camera is not even taking one image. It takes several images then stitches them together in a way it thinks makes the most sense.

 

The image I see in my viewfinder is very, very different from the image that gets outputted by the imaging pipeline. This is even more true for images taking in automatic mode, which is like 99% of pictures these days.

Long gone are the days of photography being a manual thing that humans control. Most of the work is being done by the camera trying to figure out what the human operating the camera wants. There is a reason why photography used to require a lot of skill and knowledge, and nowadays pretty much everyone can take good looking pictures in seconds. What used to be a profession that people dedicated their lives to is now something my 8 year old niece can do, because it has been automated through computers.

 

And yet will you ask your 8 year old niece to be a a wedding photographer? You probably would not because a skilled photographer has the timing and framing skills that you'll want. All artistic concepts I've described is not something the camera can manipulate while capturing the image. Yes you can add prost processing to adjust the image after the capture but it also does not manipulate the overall composition of the scene for you. 

 

Can the camera give instructions to your subjects in your scene? It does not. Does it rearrange the subject in your scene so that you can have your desired composition? It does not. Can it adjust scene lighting to add rim lights to all the subjects in your scene? It does not. It works within the boundaries it is in. If you ask me, I think you're underestimating the control the human has over their scene.

 

I think the overall issue in this case is that we have a human assisted machine creation rather than a machine assisted human creation. None of which are alike the examples you've given. 

[wanderingfool2]

17 hours ago, Kisai said:

Good luck getting a version that's in the public domain. Finding a 1920's-era film that you have to scan, ain't gonna happen.

 

And before you go "you don't need that", yes you so, as that's the version that's in PD, not a DVD/BD version printed later, and this is the same criteria we use for dumping floppy discs, cd-roms and game roms. You have to do it yourself from the original media.

IT'S AN HYPOTHETICAL EXAMPLE.  The fact it "might" be hard to get your hands on it literally doesn't matter.  It's a hypothetical example, you assume you can.  As you are arguing about the bit that doesn't have much to do with the point being put across, which is that if you had you had the early drawings of public domain work and use an AI to draw it for you in the movie that you are describing then you wouldn't get copyright over it (in this extreme example)

 

17 hours ago, Kisai said:

No, you can only have a copyright on the physical recording. That's why people keep getting nailed for copyrights on public domain music, even when they play it live on Youtube. The bot doesn't know the difference between a 1920 recording and a 2020 recording.

Don't conflate what YouTube does with copyright law.  YouTube does the over-sensitive approach where they assume things sounding similar (not even exact) will be copyright infringement due to not wanting to get sued/tied up in lawsuits.

 

It's like I said, you own the rights to the performance.  If you burn 10 CD's, people are not allowed to use that CD and put it in their movies.  People aren't allowed to create copies direct of it.  They are allowed to make their own performance (even if it sounds identical), and at that stage they can do what they want.  (Like I said, it will have a reduced pulling power in terms of what you can claim...like in this case you would have to prove they used your performance, which would be pretty hard to do)

 

18 hours ago, Kisai said:

But if we're going to nitpick things in that way

https://www.cined.com/netflix-uses-ai-to-generate-anime-short-film-reactions-follow/

 

Then which parts of this film can't be copyrighted? The few panels in the end credits that shows the process?

That is the whole reason why we are having this discussion.  The courts essentially at least set one line in the sand (where I think it should be more blurry).

 

Like in that case the AI is essentially just doing shading and such, which seems to be upheld...for now (as no one has tested it in court).

 

It gets back to the monkey case though, the guy literally setup the scenario for that "selfie" to happen going as far as even helping the tripod a bit...but he lost his copyright.

 

If I describe a full scene in a book, that is pretty distinct/unique and the output image that I get from the AI will also be pretty unique.

[lostcattears]

All I can tell is AI is going to kill a ton of white collar jobs.