US Copyright Office Issues Rules For Generative AI Part 2: More Nuance! 

You did an amazing job explaining something that is so complicated and confusing but also very relevant right now in both videos. That’s why it got so popular (even if you’re not a lawyer 🤓). Very good job 👏

I did a project on this topic back when the first video was released and it was the most extensive and accessible resource at the time! Thanks for the effort you put in :)

Regarding character images being characters: "A drawing, picture, depiction, or written description of a character can be registered for copyright. Protection does not, however, extend to... the general idea or name for characters depicted, or their intangible attributes." -United States Copyright Office

I'm just hoping there will be way to make it a fair playing field for independent artists and not just those in a company.

Interesting follow-up video. While this is of course only relevant to USA copyright and other countries work completely differently, especially since America is one of the few countries where copyright actually has to be actively registered, it is interesting and relevant to how AI-generated imagery is seen in other places too. As to "if it isn't public, it's not public domain", one problem with AI generated images is that you still can't copyright an exact copy of it without significant creative changes. Why? Because someone else could technically and unbeknownst to you generate the exact same image that is (currently) not copyrightable. If a copyright was granted to an image that had no significant changes made, and it was found it was based on an AI image, that copyright should be revoked. Importantly, AI image generators don't create anything or add anything creative. They can only generate an image from a finite set of possible latent images that are based on statistical models created from the training set. Even if you use your own image in a prompt, it is actually not using that image itself or including it in its mathematical model, except as a replacement for the random noise cloud that it uses by default to generate an image from. It is the same process as when you select an image to seed a new generation of images. The resulting image is still one of the latent images purely derived from the mathematical and statistical model of the training set images. It does not add any creativity that generates copyright even if the image you use as a prompt was yours. The non-copyrightability is akin to zooming in on the mandelbrot set, and though you may have found a piece of it no one else has looked at before, you still can't copyright the elements in the image that are reproducible by someone else using the same mathematical formula. The set of possible images in a particular model is finite. No image is unique but infinitely reproducible. The prompt is not what creates an image that otherwise would not have existed, but what, combined with random seeds, simply finds a coordinate in the multidimensional system of latent images. Any image can be recreated once you know the coordinates. But, and this is of course going to be the big topic in upcoming lawsuits, since AI image generators can _only_ generate 100% derivative images of the training set, doesn't that mean that they are also, by the very definition of how it works, infringing the copyright of all the copyrighted images that went into the training set? Certainly there are many latent images an AI can generate that will be infringing on trademarks, character copyrights, or that will have such a similarity to an original copyrighted image that a court would usually rule infringement if they had a case in isolation where those images were compared, regardless of the method that went into their creation, but since you can argue that _all_ images are derivative, the only possible legal argument would be fair use, and that is going to be a difficult argument to win with if the court actually understands how the images are generated. Since even most users of AI's don't understand how it works, they could just as easily base their verdicts on misunderstandings. It will be interesting to see how legislation and case law deals with these issues, and how the current batch of lawsuits pan out.

Thanks for covering a part 2, even as a non expert the video is very insightful and interesting to listen to. You really called out some unique questions in this one that I have never considered and some insight on existing policies that I never even knew 😆. I suppose at the end of the day the boundaries will be pushed and tested by those more adventurous, I'm very curious to see how it all plays out. Great video! Happy to see more of your educational art videos again.

It's about time that they protected artists. Thanks for this! It's raising awareness. I've been to Bluewillow and I saw that some of the arts used was from some artist that I know!

Playing devil's advocate to your suggestion that an AI generated image with a character would necessarily have the character public domain, I think there is room for that not to be the case and that it likely won't be. My reasoning is based around the idea that the design of the character could be argued to be a human undertaking in the design of the prompt. While the guidance suggests that the AI generated image can't be copyrighted due to most of the creative undertakings by non-human and that the prompt is insufficient input I do not imagine that would hold for the character. After all as you state yourself in this video, "[a] written description of a character can be registered for copyright." I think there is a fair argument to be made that the initial prompt used may be sufficient to copyright the character even if the resultant image is not copyrightable as a piece of art. In order to hammer home my points on this let me give a few examples to illustrate. Firstly suppose we had a written description of a character which met the criteria to be copyrightable, but before copyrighting the text we used it to generate an AI image in order to check that it lined up with our mental conception of the character. Do you suppose that the existence of the AI art should then invalidate the copyrightability of the initial text description? Alternately suppose we were in the same situation, but instead you were commissioning an artist to draw the piece. Typically you would expect that you would retain the character copyright, while the artist would retain the copyright for the composition of the piece, no? If yes, how would using AI rather than an artist fundamentally alter this dynamic such as to invalidate the character copyright?

in terms of money for testing the copyright office, you could always start a crowdfunding project. Come up with a budget based on how many images you'd be submitting. But since it's a case-by-case basis, even then, we might not get a definitive answer, only an approximate probability.

One thing people often miss is that while the Mona Lisa may not be copyrighted - the photograph itself of the Mona Lisa may be.

I think this is a situation where the Copyright Office needs to create strong affirmative examples of what IS copyrightable, because if people want to invest their time and effort into building up a brand that involves some AI contributions, and then once they are really rolling and invested the Copyright office goes "nah, that's not Copyrightable," that would be disastrous.

Thanks you Neil, as always amazing deep researches! Always inspired and learning from your workflows since back years ago 👏

There is a phrase it goes like this -"Easy comes easy goes. Learn to make your own images with no AI, making art is a spiritual exercise and can be an enlightening experience. AI art is just a path to nowhere when it's gone you will stand there naked in the cold, there is no golden pot at the end of that rainbow.

Wait, you have lawyer friends? My whole estimation of you... ;-)

I still find it important to steer use from generators such as midjourney and SB to ethical generators soon to come. And somehow compensate as many artists as possible if their work is found within the unethical generators.

If you generate and sell A.I. Art there is no way for anyone to know it’s A.I. Art. So, the chances that someone takes it and uses it themselves is very slim.

on your issue #2, a simple solution for people being dishonest about this is to force the underlying company that offers the software to submit every image that gets generated to a repository which the copyright office can check against.

Thanks for doing these. I'm curious how businesses will react to this rather than individual artists like us. For example imagine a game Dev company makes a game and uses generative ai all through the concept process to make 2D concepts but the game that is released is a 3D game and the public never see the 2D concepts that helped inform the creation of the 3D art. So is that still allowed? Basically can people make other art based purely on AI generated concepts? I think this is the big question for video game concept artists.

Hi Neil: For next time, you’re encouraged to invite one or more of your copyright/IP attorney (friends) to sit for a live Q&A session, as that would provide you and your subscribers more definitive (non-legal) answers. Go Habs & DDO!

If a person can paint exactly same quality as the AI generated image, then it kind of defeat the purpose of making things faster using AI. Plus copying one to one is also a skill high enough that he doesn't even need to use AI. People use AI cause either they don't have enough skill or just want to get result faster.

Here's one: what if I take my original painting, use it to generate 15 variations of it, and then photobash elements from those 15 variations back into the original painting? what now??!

Hey man I'm curious if you had an opinion on this particular scenario: If you write a book and create a distinctive character through publishing your book (like harry potter). That character would be a copyrightable character correct? So then if you generate an AI image in the likeness of that character use exact descriptive language from your book would that then make them copyrightable? Thank you!

The reason your original video was so popular is because you're the only person who seems to care about the legality of the issue. Many have spoken about the economic, ethical and moral implications, but they seem to take for granted that paying for software that essentially makes "some version" of what you're going for makes you an artist, even in the eyes of the law, which interestingly, seems contrary to reality

I am curious how they would handle the new MJ “describe”. If I use MJ descriptions of my original image to generate new images, is that still considered a variation like your robot example or is it now considered just a regularly prompted generated image?

i think there should be some way to limit use of copyrighted material in training as well as prompting. but i dont know how realistic that is, to achieve or police without doing it totalitarian. even trained on non copyrighted content ai is super powerful and imo should be embraced by all artists as there are a lot of tools that are not related to generating completely new content, but rather detail tweaking and smart masking

So I have an image that I edited together from 3 seperate ai generated images (all generated based on the same prompt image) and then I've painted over parts of it to fix errors (such as the hands, the eyes, hair, and neck) What I'm understanding is, I can't copyright the original images I used to edit together this final image, but I CAN copyright this final image because of the editing and painting over I did?

How would this affect Video created with AI? For example An animated series with original non- trademarked characters? I developed a technique that enables me generate rather coherent animation from even a single prompt. It's an industry game changer...83% complete

Neil, can I include this is a blog article with credit of course? Thank you.

@6:08 again you can copyright things about a character, but the character itself is only copyrightable as a derivative copyright. By itself, a character can't be copyrighted.

Nice job!

So basically, anything we make on AI generators cant be copyrighted, which means people can use your work and profit from it

So all the AI created stuff people create , we can steal and sell. On my way to make big $$$ of off these fools

Ok big question, the ONLY THING I WANT TO KNOW. If i create a colored drawing using shit cheap markers and pens, use ai to turn it into a 3d model (much like ANY 3D SOFTWARE) then sell it on etsy, is this OK. Im holding my breath on this i have a LOT of ideas

Apparently the Copyright Office's opinion is obsolete.

As long as copyright exists it's bad.

For Public Domain Works... Because American Courts are not always about the law (litigation Hell being a thing). Disney is a prime example of how grey some Public Domain Works are. But over all, Public Domain is as you explained it. Personally, AI Art shouldn't be Public Domain because that's assuming every AI Tech was both designed and to function in the same "prompting" fashion. Still... This Policy is as I stated before. It's not well articulate and leaves too much room for interpretation rather than for a legal procedure. (I guess mankind will learn this the hard way. I'm already eternities beyond this cosmic infinitum.)

thank you the copyright claims board began 2022 as help for illustration artists.. .