Worldwide - Copyright - Thinking Of Using AI To Make A Music Video? Read This First...

The world's first entirely AI-made music video is here and has sent understandable shockwaves through the industry. Music lawyer Halina Wielogorska takes a closer look at this cutting-edge area where AI meets music law.

The world's first entirely AI-made* music video is here and has sent understandable shockwaves through the industry. The video was created entirely by a generative AI or machine learning solution in response solely to human-created video prompts (known as text-to-video). Released by artist Washed Out, directed by Paul Trillo, and created using OpenAI's new video product Sora.

It's pretty good.

A video in constant motion, following the lives of two AI-generated humans over expanses of time. The concept (in terms of set changes, costumes, extras, and editing) would have been prohibitively expensive to achieve without AI. It was a creative idea netted by Trillo and placed in his TBC bucket for 10 years before it became reality.

Marvel at that achievement for a moment. What other creations might be possible?

In response to written instructions to the AI tool, reportedly 700 generations were created from which Trillo selected the 55 shots that make up the video.

How extraordinary for the creative process to move from prompt to creation in one fell swoop!

I was intrigued enough to dig into the exact narrative given to the software. It's clear that his considerable real-life production experience enabled Trillo to give specific instructions about the type of cameras he wanted from specific periods of history, how he wanted the shots to connect, and where the focus should lie. The AI tool would then convert those prompts to new original content imitating the Director's instructions.

Fascinating. No equipment hire, no studio, no actors, no editors (aside from Trillo's own endeavors), and no budget (TBC re costing - Sora is pre-public launch, though early reports suggest Trillo did not pay to use Sora, nor did Sora pay Trillo to endorse Sora).

But, with my music lawyer hat on, I found myself wondering - who owns it?

A bit about Ownership

Ownership in music videos can already be complex (even without an AI element) as there are often multiple creative assets that attract copyright that co-exist in a music video.

These generally include copyright in the video itself, copyright in the recorded music, and copyright in the composition and lyrics.

Strictly, the creator of a video is the original owner. Simple right? Except, in reality, usually there are several different individuals (or companies) that handle the different specialisms in creating a video.

Where there are differing contributions to the creative work of the video, for instance, a videographer and an editor, then copyright will attach to the individual elements that they created. The videographer will own the copyright in their video creation, and the editor will own the creative elements they have contributed during their editing process. Higher up the food chain, a director and producer are regarded as the two main intellectual property creators and who will generally seek to contract others on a work-for-hire or service contract basis, so as not to dilute their IP.

The video is then synchronised with the recorded audio performance of the releasing artist/s and embodies the underlying copyrights in the song and lyrics (which may have been written by the artist too, but are often composed in collaboration with other writers).

If the video also features the artist performing, the video itself may also attract copyright protection for the artist (or their label) if a new version is recorded, as well as likeness and reputation rights (in some territories).

The artist/s (and writer/s) may have signed their rights on an exclusive basis to various third-party representatives (most commonly record labels and music publishers) who effectively stand in the shoes of the artist in terms of copyright (by way of assignment or exclusive licence) for the length of their deal. On the artist side, the label tends to commission (finance) the videos and control and exploit them during their relationship with the artist. Without a label, the artist would do this themselves directly.

That's a lot of people with a slither of rights.

It's therefore typical for the artist (or their representatives), who sit at the heart of the release of their material, to seek to take ownership of the underlying copyrights in the music video (to the extent they are able to do so). This is usually justified by those entities financing the creation of the work and is typically done on a buy-out basis.

Contracts that require assignment of rights to the artist (and their label for the period of their exclusive arrangement) are common, in exchange for the financial consideration of funding the work. This makes the video rights position clean and makes it more convenient to exploit the video.

Can an AI-created video fit into this model?

Co-ownership of copyright models is generally avoided in the music business where it can be. The introduction of the AI creator (as the notional production company), in addition to the human creatives (in this case, Trillo as director and editor), brings in an uncertain element in terms of copyright ownership.

Here's why:

Questionable legalities around the source of the content:

Whilst the video content and/or images that comprise Sora's (or similar platforms) output, are claimed by the AI companies to be inherently original, we know that the AI platforms have developed their tool's ability by using mass data sets available on the internet to train their tools. In other words, they've been trained by exposure to existing artistic works such as music, film, images etc. Those data sets have yet to be licensed by the platforms, with AI companies relying on exceptions such as the 'fair dealing for the purposes of research' exception in the UK .

Whether that exception applies in this case is yet to be tested, with the tech companies claiming so (or in its place, exceptions in other territories with broader exclusions, like the USA's "fair use" doctrine, and others).

There is uncertainty and therefore a liability issue around AI output at this time. Understandably, the music business and the film industry are adamant that copyright has been infringed where in the training of AI tools the machines have ingested copyright-protected data. Not surprising given the investment these companies have made over the years in creating their content. They state that the AI companies must pay a licence fee for using their content, and without such a licence, maintain that copyright infringement will have occurred. Given how far from resolution this argument is, it would be purely conjecture as to how the matter will be settled.
Existing law is not consistent globally, and developing at an alarming pace:

The way in which existing intellectual property laws tackle the new problem of AI is still developing and is doing so on a territorial basis. Each country, its courts, and legislature are interpreting the existing copyright and other intellectual property law, alongside a large number of micro and macro-economic considerations. This is not a simple case of either siding with the tech companies to foster greater growth and ownership in this space, or siding with the creative industries to protect culture and rights, but moreover a melting pot of geo-political factors.
Assuming we navigate past the other hurdles, is the computer-generated output capable of obtaining copyright protection?

Copyright protection in the UK has several strict requirements. The music video – just as any video – would be regarded as an "artistic work". Current UK law makes clear that the creation of an artistic work solely by computer generation, would not by the manner of its creation, prevent the work from attracting copyright. So, assuming that law remains and until case law tells us otherwise, copyright in the Sora output would be potentially protectable under the UK copyright law.

The CDPA goes on to say: "In the case of a literary, dramatic, musical or artistic work which is computer-generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken" (c.9(s)).

But how this is applied in the context of AI-generated material is, as yet, untested.

In this case, it is not clear who the author or owner is here. Is it the

(i) the platform owner, Sora (or rather the legal entity behind the platform) who made the creational "arrangements" (in which case certain rights would be weakened including length of copyright and no moral rights),
(ii) the prompt writer Paul Trillo, due to his text instructional arrangements, or
(iii) the song artist, having instructed and granted permission to the other two

Arguably there's even a fourth possibility that the output (The Hardest Part music video) of an unlicenced training model (Sora), would be regarded as a derivative of the input training material of multiple third parties. Either way, it all sounds risky to me...!

How Sora/OpenAI approach the question of ownership

As I have discussed, it is common for key stakeholders to make arrangements under contract that override the pure creation (or author ownership) approach to copyright, and indeed in this case it is important to consider the contractual terms between Sora and their users.

Sora's pre-launch info page makes no public statement about their rights position. However, Sora is a part of OpenAI – a company which has its ethos rooted in open-source learning. As time has gone on, OpenAI has developed commercial strands to its business – reportedly out of necessity to keep the company operating. I would expect Sora (as was the case with Chatgpt) to have both a free and premium offering (as well as a research arm – but that's a whole other article topic).

Assuming the terms will be the same or similar – an assumption not a fact - let's dig into the terms and conditions of Chapgpt which state clearly that:

"As between you and OpenAI, and to the extent permitted by applicable law, you (a) retain your ownership rights in Input and (b) own the Output. We hereby assign to you all our right, title, and interest, if any, in and to Output."

So, to the extent that OpenAI will be deemed to own the output of something created in Sora, they will assign it to the user, which presumably here is Paul Trillo.

There are also extensive limitations to liability, exclusions from class actions and mandatory arbitration within the terms. Meaning, it is likely that if a user posts content created by OpenAI, and a third party pursues them for copyright infringement (or anything else for that matter), they are on their own!

Should Trillo seek to licence the video to commercial partners (like the artist and their label), Trillo can expect to receive a contract from them asking that he makes all the warranties and indemnities directly, that OpenAI are not making to him. Again, quite risky!

OpenAI have also retained rights to change their terms and conditions at will in response to changes in the law, and a broad list of other reasons, and can impose new terms on users by notice within 30 days if necessary.

As various legislative AI juggernauts make their way through our international legal systems (the USA and the UK trailing behind the EU on legislation so far), it seems highly likely that OpenAI will seek to rely on this clause, and it is broadly expected that terms of use will tighten. This may leave users in a tricky spot in the future when it comes to content created historically through these platforms.

Artists, take note...

Artists and their record labels should be cautious about any chink in their copyright armour. Not knowing whether an AI-generated creation is infringing copyright from the outset (by virtue of its method of creation where unlicensed training models are used and in dispute) is a huge risk. Then comes the question of who owns the AI-created portion of the work.

As both the film and music businesses are very much rights-rooted industries, investors in those industries are very sensitive to having robust chains of rights where all contributors to copyright are fully cleared, warranted, indemnified and, in many cases, insured.

Artists and their representatives spend a lot of time, money and energy clearing all assets for exploitation, and once cleared, the assets may be used in multiple different formats and licensed to third parties for specialist use. All of this would be deeply risky for an artist and their representatives to do, as they would be taking on the risk of a hole in the chain of title to the rights.

This remains a cutting-edge area of law, in which national and international legislators are scrambling to keep pace with consumer adoption. Existing copyright law raises as many questions as it answers, and case law concerning authorship and ownership has not yet been brought.

The contractual terms by tools like Sora that will apply are, as yet, unknown (and may vary depending on whether the end user is regarded as a consumer or not).

In my view, that's three almighty chinks in the copyright armour to mitigate against and so, in the meantime, I would advise content creators to undertake a thorough risk assessment of their use of AI tools and think carefully and cautiously about their release to the public or their inclusion in output for third parties.

Thinking Of Using AI To Make A Music Video? Read This First...

Contributor