It occurred to me that books about famous TV series have been around for decades and usually feature photos or paintings of the actors in the show. Is this part of the actors contract or is this basically a silent understanding? Would this apply to A I usage as well?
If you’re publishing a book, presumably the publisher requires that the copyright to photos be properly licensed (and credited). Presumably, for a TV series or movie, the studio holds the copyright to publicity stills (and the show) and will license those for appropriate distribution. The actors would have signed contracts that allow for that.
There is an exception to copyright for “fair use” and generally analysis and criticism falls into fair use - although (IANAL) I assume there’s a limit to how dense a coffee table book of photos you can produce and claim it’s scholarly analysis and criticism.
Similarly, a person- even a famous person - walking down the street is fodder for the paparazzi who can photograph anyone and anything that is in public view (although apparently there are growing exceptions for artworks and architectural work…?) You can harass a celebrity and put that picture in a magazine - it’s news - but you cannot obviously use it to pretend there’s a commercial endorsement or similar.
But using someone’s image in a transformational way - i.e. photoshopping them into unique situations, or using AI to pose them - now you run into that situation where most actors own the rights to their image.
I guess the current dispute is not about the output - given it’s copyrighted and trademarked up the wazoo… the dispute is using the images (or text, or video) as training: “I’m going to show AI 2,000 pictures of an action hero in an action pose, so it knows what is typical…” Or “I’m going to feed AI 10,000 novels so it figures out what makes a good story”. You are taking something copyright and it presumably becomes 1/2000 or 1/10000 of the output of the commercial AI. AFAIK ocurts ae still grappling with this. (The couner-argument is “well, isn’t that what humans do too? Read 1,000 adventure stories, then write your own based on those…”
The difference between what humans do and what AI can do is that AI can do it in massive bulk, very fast. This could be argued to be only a difference if scale, not of principle, but the orders of magnitude difference in scale makes it a difference of principle - it’s like the difference between one person catching shrimp by hand in a push net vs an industrial trawling operation. They’re doing the same basic thing, but the latter is doing it in a way that destroys the resource for short term gain.
One other difference is that AI can just regurgitate whole pieces of the training data intact and it doesn’t know that it’s doing so, whereas humans typically know when they are plagiarising.
This is a situation where likeness rights would be relevant. To put it very simply: you can’t use someone’s likeness for commercial purposes without permission. You can use it for news or commentary—essentially fair use.
The show will own the copyright to the character, and thus using images of the person playing that character has that additional hurdle. There could also be trademark issues, as the use of the character could imply it is an official product of the show.
So, as stated above, books generally get the rights to any images they use and especially take care with the cover image, as that is basically advertising for the book.
As for how this relates to AI: if they are clearly depicting the person or character, the same rules would clearly apply. It’s when they use the images to create things that appear new that it gets murky.
But I do note that writers and actors have done a lot of negotiating to put in safeguards regarding AI into their contracts.
I think it’s an as yet undecided legal point. The production company owns the Rights to the images (and characters etc) that make up the show. The question is, if they use them to train an AI and make content from it. Is that like deciding to release the show on a new streaming service (which they are allowed to do) or like cropping out bit of the show in a video editor and using it to make a new show without asking the actors permission (which is not)
Its definitely going be answered by the courts in the coming years
I think it becomes a bit more problematic when the likeness (perhaps including such things as appearance, voice performance, mannerisms, etc) of many actors are used to train AI, but the goal is not to reproduce any specific single actor but create artificial performances that are an amalgam of those likenesses and performances - for one thing, it’s just going to be very difficult to prove that the end result output of such a thing is (for example) based on 1/3 each of Scarlett Johansson, Robert Downey Junior and Sir Ian McKellen (in theory, latent spaces in the controls for the thing could allow that sort of explicit selection), or when the output is some far more mixed amalgam of a much more assorted set of styles and pieces.