Copyrighting AI-Assisted Film and TV Could Get Complicated
For Hollywood and other creative industries, most of the debate about generative AI and copyright has emphasized the unlicensed use of copyrighted content to train AI models. Yet some independent studio teams are examining how copyright will apply to AI outputs as they begin to create AI-assisted content, which uses generative AI in production workflows for film and TV projects.
To understand how producers are evaluating copyrightability for AI-assisted motion picture works, VIP+ spoke with creative leads at AI studios and entertainment lawyers advising clients on copyright registration.
Copyrighting AI-assisted creative works is still a relatively nascent and untested activity. Independent productions are likely to act as the initial proving grounds for copyrighting films that use generative AI and incorporate AI elements. Major studios still distrust using generative video and image models for onscreen material, including doubt about being able to copyright AI outputs, as VIP+ previously reported.
Despite recent guidance from the U.S. Copyright Office, copyrightability for AI-assisted works has ambiguities. Bright-line answers aren’t possible because a copyright officer needs to analyze each work case by case, based on the circumstances of its creation — a fact-based analysis that considers the controllability of the AI tools that were used and how and to what effect they were used by a human to create the work. Subsequent changes in the tech or case law could also change the view of the copyright office over time.
Producers who use generative AI in production will need to prove to the copyright office that sufficient human authorship went into the work. This means rigorously documenting their use of AI tools, clearly indicating human creative contributions versus those by AI, which will need to be disclaimed in a copyright application.
Sources in AI studios said they were tasking creatives with documenting their work and tracking the AI tools, prompts and discrete processes used to produce and edit any given shot. Gen AI filmmaking workflows often happen in ComfyUI, a modular node-based creative platform that displays the entire workflow. Filmmakers at AI studio Asteria have also created videos and Figma boards showing the series of steps in their workflows and exactly how human-authored art is used and where AI assists.
“We’ve been keeping good track of everything from a standpoint of human authorship because it’s very important for the copyright office to show the impact we’ve had on the work and that we are really driving it and not just pressing a button,” said Nik Kleverov, CCO at AI studio Native Foreign, which is producing the animated feature “Critterz” based on an earlier short film.
Typically, only cleared third-party or public domain creative elements are disclaimed and excluded in the copyright registration for a film, which now could add on AI-generated elements incorporated into the work. “The chain of title is much more complex when you’re using generative AI, so we have to get very clear about the materials inside the shots,” said Tom Paton, CEO at U.K.-based AI studio Pigeon Shrine.
As I wrote in the June 2024 VIP+ report, “Parsing human from machine authorship in an AI-assisted work is likely to become ever more convoluted, particularly when creative processes involve more sophisticated tools and techniques than simply text-prompting an image model to receive an output. The burden of that ‘parsing’ will fall to media companies because registration applications must disclaim any contributions by AI. For audiovisual media, that burden could be extreme, particularly if gen AI is used in numerous assets and points in the pipeline.”
AI-assisted films are likely to be protectable as a whole, along with any human-generated elements, but specific AI-generated elements in a film may not be and may need to be excluded under the registration. That was argued by the U.S. Copyright Office in its January report on AI and copyrightability: “The inclusion of elements of AI-generated content in a larger human-authored work does not affect the copyrightability of the larger human-authored work as a whole. For example, a film that includes AI-generated special effects or background artwork is copyrightable, even if the AI effects and artwork separately are not.”
Producers at AI studios argued a similar stance — that the complexity of the audiovisual medium and their workflows required a large amount of human creative interaction that would be sufficient to qualify for protection, as they were selecting, modifying and arranging numerous creative elements using multiple different tools and processes, even if those elements were AI-generated.
Sources using AI video and image models further described applying substantive edits to AI outputs they believed would be enough to rise to a minimum standard of originality. “We can show the copyright office, here’s where it ended up after hundreds of revisions,” said Kleverov.
“A frame of a film is not just a single image prompted from an AI model. It contains After Effects work, where we’re manipulating the initial material to get it where we want but also voice work, music, foley,” said Paton. “I think the argument [against copyrighting AI-generated material] immediately gets lost when you’re using six or seven different platforms, generative tools, prompt skills and Premiere Pro just to assemble a 30-second clip. I think you’re already in a space of being able to argue that you were substantially involved in the work’s creation.”
But it’s possible that AI-generated elements used in the making of a film would be excluded from the registration of the work.
The AI-assisted film short “Critterz” was registered as a whole, but the initial versions of the AI-generated 2D images of the film’s woodland creature characters and background images, which had resulted from text-prompting OpenAI’s DALL-E image model, were excluded from the registration “to the extent not copyrightable,” said Lisa Oratz, IP entertainment and tech senior counsel at Perkins Coie, who is advising Native Foreign and worked on the copyright registration for the “Critterz” short.
Those initial images were substantively edited (such as changing characters’ eyes, hands, expressions) before being used in the film and animated in Unreal Engine to create the final short, as described in a making-of video, and the modifications were covered by the registration.
“We only excluded the 2D renderings of the images that came out of the AI tool ‘to the extent not copyrightable,’ in order to reserve the right to argue that the initial 2D AI outputs should be included in the registration,” said Oratz. Oratz explained that given the uncertainty in the law, they didn’t want to limit the scope of their rights if case law finds later on that AI-generated images from text prompts are protectable.
Even so, those initial DALL-E images of the “Critterz” characters are in limbo: While it can’t be said they’re excluded, they’re also not included and may never be, unless law or guidance changes. The film as a whole and the edited versions of the characters as they appear in the final film are protected in the registration, but the characters as they appear in the initial image outputs from DALL-E are exposed.
Theoretically, if someone found and used those initial images, it could be difficult to assert an infringement claim, if the Copyright Office view prevails. Oratz confirmed to VIP+ that the initial images of the characters aren’t public information.
By contrast, several AI studios, such as Asteria, Native Foreign and Invisible Universe, are starting the AI production process with human-authored art assets, such as original character designs, which are then used as a dataset to train custom models called LoRAs, in turn used to produce new, similar derivative outputs.
While the copyright office doesn’t directly address this type of use, the closest analogy it provides is “Rose Enigma,” for which an original hand-drawn image was filled in with AI elements. The drawing was protectable, as were the specific expressive elements that remained “perceptible” in the AI version, but anything new the AI supplied in addition was excluded. This casts some doubt on the copyrightability of unmodified AI outputs that don’t visually correspond with original artwork, such as an original character an AI model generates in novel poses or scenes.
It’s these kinds of ambiguous outcomes that major studios wouldn’t be able to accept. Given the massive investment that goes into studio productions, ironclad copyright protection is imperative, particularly on core creative elements like character design.
Even if an AI-assisted film as a whole is copyrighted, any unprotectable production assets contributed by AI and listed as exceptions in the registration are sitting ducks, exposed to others taking and copying them, meanwhile limiting a studio’s ability to defend against infringement. Having many unprotectable AI elements within a film could even devalue the IP in the eyes of a distributor or licensor whose usual interest is content exclusivity.
“The concern is the studio or distributor would pay a lot of money acquiring, marketing and distributing a film only to find itself facing legal challenges to the copyright protectability of that film or certain of its elements, and being limited in any way in its legal remedies,” said Tom Ara, transactional entertainment partner at DLA Piper, who has advised Hollywood clients on the implications of AI use on the exploitation of film and TV rights.
VIP+ Unearths Gen AI Insights From All Angles — Pick a Story
To understand how producers are evaluating copyrightability for AI-assisted motion picture works, VIP+ spoke with creative leads at AI studios and entertainment lawyers advising clients on copyright registration.
Copyrighting AI-assisted creative works is still a relatively nascent and untested activity. Independent productions are likely to act as the initial proving grounds for copyrighting films that use generative AI and incorporate AI elements. Major studios still distrust using generative video and image models for onscreen material, including doubt about being able to copyright AI outputs, as VIP+ previously reported.
Despite recent guidance from the U.S. Copyright Office, copyrightability for AI-assisted works has ambiguities. Bright-line answers aren’t possible because a copyright officer needs to analyze each work case by case, based on the circumstances of its creation — a fact-based analysis that considers the controllability of the AI tools that were used and how and to what effect they were used by a human to create the work. Subsequent changes in the tech or case law could also change the view of the copyright office over time.
Producers who use generative AI in production will need to prove to the copyright office that sufficient human authorship went into the work. This means rigorously documenting their use of AI tools, clearly indicating human creative contributions versus those by AI, which will need to be disclaimed in a copyright application.
Sources in AI studios said they were tasking creatives with documenting their work and tracking the AI tools, prompts and discrete processes used to produce and edit any given shot. Gen AI filmmaking workflows often happen in ComfyUI, a modular node-based creative platform that displays the entire workflow. Filmmakers at AI studio Asteria have also created videos and Figma boards showing the series of steps in their workflows and exactly how human-authored art is used and where AI assists.
“We’ve been keeping good track of everything from a standpoint of human authorship because it’s very important for the copyright office to show the impact we’ve had on the work and that we are really driving it and not just pressing a button,” said Nik Kleverov, CCO at AI studio Native Foreign, which is producing the animated feature “Critterz” based on an earlier short film.
Typically, only cleared third-party or public domain creative elements are disclaimed and excluded in the copyright registration for a film, which now could add on AI-generated elements incorporated into the work. “The chain of title is much more complex when you’re using generative AI, so we have to get very clear about the materials inside the shots,” said Tom Paton, CEO at U.K.-based AI studio Pigeon Shrine.
As I wrote in the June 2024 VIP+ report, “Parsing human from machine authorship in an AI-assisted work is likely to become ever more convoluted, particularly when creative processes involve more sophisticated tools and techniques than simply text-prompting an image model to receive an output. The burden of that ‘parsing’ will fall to media companies because registration applications must disclaim any contributions by AI. For audiovisual media, that burden could be extreme, particularly if gen AI is used in numerous assets and points in the pipeline.”
AI-assisted films are likely to be protectable as a whole, along with any human-generated elements, but specific AI-generated elements in a film may not be and may need to be excluded under the registration. That was argued by the U.S. Copyright Office in its January report on AI and copyrightability: “The inclusion of elements of AI-generated content in a larger human-authored work does not affect the copyrightability of the larger human-authored work as a whole. For example, a film that includes AI-generated special effects or background artwork is copyrightable, even if the AI effects and artwork separately are not.”
Producers at AI studios argued a similar stance — that the complexity of the audiovisual medium and their workflows required a large amount of human creative interaction that would be sufficient to qualify for protection, as they were selecting, modifying and arranging numerous creative elements using multiple different tools and processes, even if those elements were AI-generated.
Sources using AI video and image models further described applying substantive edits to AI outputs they believed would be enough to rise to a minimum standard of originality. “We can show the copyright office, here’s where it ended up after hundreds of revisions,” said Kleverov.
“A frame of a film is not just a single image prompted from an AI model. It contains After Effects work, where we’re manipulating the initial material to get it where we want but also voice work, music, foley,” said Paton. “I think the argument [against copyrighting AI-generated material] immediately gets lost when you’re using six or seven different platforms, generative tools, prompt skills and Premiere Pro just to assemble a 30-second clip. I think you’re already in a space of being able to argue that you were substantially involved in the work’s creation.”
But it’s possible that AI-generated elements used in the making of a film would be excluded from the registration of the work.
The AI-assisted film short “Critterz” was registered as a whole, but the initial versions of the AI-generated 2D images of the film’s woodland creature characters and background images, which had resulted from text-prompting OpenAI’s DALL-E image model, were excluded from the registration “to the extent not copyrightable,” said Lisa Oratz, IP entertainment and tech senior counsel at Perkins Coie, who is advising Native Foreign and worked on the copyright registration for the “Critterz” short.
Those initial images were substantively edited (such as changing characters’ eyes, hands, expressions) before being used in the film and animated in Unreal Engine to create the final short, as described in a making-of video, and the modifications were covered by the registration.
“We only excluded the 2D renderings of the images that came out of the AI tool ‘to the extent not copyrightable,’ in order to reserve the right to argue that the initial 2D AI outputs should be included in the registration,” said Oratz. Oratz explained that given the uncertainty in the law, they didn’t want to limit the scope of their rights if case law finds later on that AI-generated images from text prompts are protectable.
Even so, those initial DALL-E images of the “Critterz” characters are in limbo: While it can’t be said they’re excluded, they’re also not included and may never be, unless law or guidance changes. The film as a whole and the edited versions of the characters as they appear in the final film are protected in the registration, but the characters as they appear in the initial image outputs from DALL-E are exposed.
Theoretically, if someone found and used those initial images, it could be difficult to assert an infringement claim, if the Copyright Office view prevails. Oratz confirmed to VIP+ that the initial images of the characters aren’t public information.
By contrast, several AI studios, such as Asteria, Native Foreign and Invisible Universe, are starting the AI production process with human-authored art assets, such as original character designs, which are then used as a dataset to train custom models called LoRAs, in turn used to produce new, similar derivative outputs.
While the copyright office doesn’t directly address this type of use, the closest analogy it provides is “Rose Enigma,” for which an original hand-drawn image was filled in with AI elements. The drawing was protectable, as were the specific expressive elements that remained “perceptible” in the AI version, but anything new the AI supplied in addition was excluded. This casts some doubt on the copyrightability of unmodified AI outputs that don’t visually correspond with original artwork, such as an original character an AI model generates in novel poses or scenes.
It’s these kinds of ambiguous outcomes that major studios wouldn’t be able to accept. Given the massive investment that goes into studio productions, ironclad copyright protection is imperative, particularly on core creative elements like character design.
Even if an AI-assisted film as a whole is copyrighted, any unprotectable production assets contributed by AI and listed as exceptions in the registration are sitting ducks, exposed to others taking and copying them, meanwhile limiting a studio’s ability to defend against infringement. Having many unprotectable AI elements within a film could even devalue the IP in the eyes of a distributor or licensor whose usual interest is content exclusivity.
“The concern is the studio or distributor would pay a lot of money acquiring, marketing and distributing a film only to find itself facing legal challenges to the copyright protectability of that film or certain of its elements, and being limited in any way in its legal remedies,” said Tom Ara, transactional entertainment partner at DLA Piper, who has advised Hollywood clients on the implications of AI use on the exploitation of film and TV rights.
VIP+ Unearths Gen AI Insights From All Angles — Pick a Story