The intricacies of intellectual property, especially in the context of AI-generated works, are a vital yet often overlooked aspect of an artist's education. Today's artists, spanning various industries and creative disciplines, find themselves at the forefront of a critical dialogue about how AI intersects with copyright, impacting their livelihoods and creative freedoms.
The United States Copyright Office's recent call for public commentary on the intersection of AI and copyright policy has ignited a crucial conversation within the artistic community. With over 20,000 community members, many of whom are professional visual artists, Invoke is deeply committed to advocating for the best interests of those artists.
We are at a turning point. AI policy is rapidly taking shape and artists’ voices need to be heard.
At its core, this is a conversation about the future of creative freedom, expression, and the economic potential for artists in the age of generative AI. We have considered some of the core questions around:
- What rights should creators have with regards to using their own work to train datasets?
- What creations should artists be able to copyright when supported by AI?
- How should the law distinguish between large generalized foundation models, and those that are trained to replicate a specific artists’ style or an individual’s likeness?
As the majority of our community members are visual artists and creative professionals, we have considered these questions from that perspective, but the insights and guidelines discussed here could be broadly applicable to artists in other disciplines as well.
Artists should have inalienable rights to train AI models on their own work, regardless of whether they currently hold copyright or other rights or licenses to the work.
Large companies will often champion copyright as an important foundation for artists and creators to maintain control and ownership of their work - and ideally, copyright would do that. An artist holds the copyright for their creation the moment they create the work, and is granted the legal protections that that copyright provides assuming that it meets the threshold for human authorship defined by the US Copyright Office.
However, we know that in practice, copyright is often forfeited to the large companies that employ these artists, or to the handful of massive corporations who spend billions acquiring the rights of artists’ intellectual property, as is the norm in entertainment and music. Even an artist as powerful as Taylor Swift has had to re-record and re-release all her old music to regain control of her intellectual property, after her record label sold the copyrights of her music to a private equity firm.
If we allow policies to be put into place that require an individual or corporation to have the copyright for all the works used to train a dataset, we essentially solidify a centralization of AI power in the handful of companies that can afford to purchase or license those rights, leaving individual creators unable to compete in the growing economic opportunity that AI presents.
Having access to train and own their own AI model without having to pay licensing or subscription fees to the major tech companies will be crucial to artists’ abilities to remain competitive. Artists should be skeptical about large companies that offer access to closed or proprietary models & datasets through licensing fees, while advocating that this approach is good or somehow more fair for artists.
Invoke AI gives access to our Generative AI Studio for free to anyone to run locally on their computer. Our application currently works with openly licensed foundation models, like SDXL, which allow artists to opt out of having their works included in the foundational training sets. We will continue to work with the open source community to improve features around attribution tracking, bias, and respecting opt-outs.
Artists should be able to protect outputs generated with the support of AI when there is clear human conception & creativity guiding or influencing the output.
Many artists have tried one of the consumer-facing image generation AI tools out there, like DALLE-3. You enter a phrase and out pops an image. You might append a famous artists’ name and out pops an image that is vaguely reminiscent of that artist’s style.
These types of outputs should not be copyrightable. In these examples, the AI is more of a utility, like a search engine, interpreting text and expressing that text into an image.
However, Invoke’s community members and many of the artists using generative AI today aren’t simply entering a single input to generate an image. They have incorporated these tools into their existing creative process and professional workflows. They might ask the tool to provide them with several versions of a work they had previously created using their own style, take a sketch they’ve done and generate a rendered image that follows the same compositional structure and outline, or iterate on specific portions of an image that they’ve created. In many instances, they are almost in conversation with the tool, working through numerous manipulations in order to achieve their creative vision.
In this case, the AI tool serves more like a digital paintbrush or a camera. We argue that these works should be copyrightable, though we believe open-source tooling needs to be developed to identify low-effort content without curbing the creative freedoms of artists trying to protect their original work. Tools from the Content Authenticity Initiatives show some early promise.
If we fail to protect the work that artists and creative professionals produce with these tools, we will see all of the harms that come with the advance of AI capabilities and none of the benefits. By refusing to provide guidance on AI-enabled creativity, the US Copyright Office is abdicating their responsibility to protect the work of creatives. We need to demand clarity, and advocate for a reasonable approach to protecting work which involves human conception and creative inputs in the process of generating content.
Distributing models intended and merchandised as directly copying the likeness of an individual or replicating an individual’s body of work without their permission should be illegal.
In the past few months, we’ve seen an explosion of small, specialized models (sometimes called LoRA models) that can inject foundational models with a completely new concept or style, essentially providing fine-tuned control to the direct the aesthetics of a generated image.
These specialized models are easier to produce and require far fewer images to be trained on. Their narrow focus allows these specialized models to capture the idiosyncrasies of individual artistic styles, making them much more capable of reproducing or even mimicking the unique attributes of specific artists or individuals, unlike foundational models which are not specialized in any one task and lack the nuance to replicate specific style or individual likeness with high fidelity.
Model marketplaces have exploded in popularity by distributing these smaller models with little to no oversight or effort in content moderation, and gamifying the creation of these works to accelerate their business. We think this poses an immediate and pressing concern to artists.
We believe that it should be illegal to distribute these models and the government should act quickly to enact laws around the distribution of these specialized models.
Artists should continue to engage and act.
Copyright law evolves and changes over time.
When the camera and photography first emerged in the early 19th century, many in the art world and the government were skeptical about the idea of a machine capturing reality. It wasn’t until 1865 that the Supreme Court granted copyright protection to photography and it wasn’t until 1940 that the Museum of Modern Art (MoMa) established a Photography Department. Today, photography is a celebrated art form with photographers receiving many of the same protections for their work as the visual artists working in disciplines that came before them.
Not everything captured with a camera is art, nor is it copyrightable, because today, we understand that the camera is both an artist’s tool and a mechanical instrument, depending on how it is used. We believe the same will be true for generative AI tools and the artists that use these tools.
So what can you do? We offer 3 steps forward:
- Stay informed and involved in ongoing discussions about AI and copyright. Though the window for formal commentary has closed, the conversation continues, and your voice is still vital in shaping the future.
- We encourage artists whose only experience with AI tools has been a simple input and output image generator to explore the more advanced AI-powered design applications, like Invoke. We have found that when artists engage meaningfully with AI tools and the growing community of artists that use them, they develop a deeper understanding of how these tools and the policies around them can support the creative expression, freedom, and economic livelihoods of artists.
- We want to hear from you! We are building a foundation for the future of this technology. Please join us on Discord or reach out via firstname.lastname@example.org.
Editor's Note: A version of this article was originally published on September 13th on LinkedIn.