Fair Pay for Creators: Compensating Artists in the Age of AI Image Generation
AI image generators such as Dall-E and Midjourney have captivated the masses in the past months. Although this technology is both exciting and disruptive for the visual arts industry, there has been little discussion about how to fairly compensate artists for their contributions to the training sets of these AI image generators.
As you read on, you'll learn about the importance of compensating artists, and how to make that happen.
Why should we compensate artists?
Currently, many AI image generators use datasets of existing images found on the internet that are downloaded en masse by their developers using automated spiders. These images are used as part of the training process for the AI, and without them, we wouldn't have AI image generators like Stable Diffusion or Dall-E.
It's likely that if you've ever uploaded any images online, your work has been used to train an AI model. You can check if your art has been used to train an AI model on the Have I Been Trained website here.
Given that these models process hundreds of millions of images, it's safe to say that there are thousands of artists whose work has been used to train them, if not millions. However, many of the companies that profit from the use and access to these AI image generators don't compensate the artists whose work has contributed to their tools. This needs to change.
Compensating artists for the use of their work to train AI models is simply the right thing to do. We spend a lifetime learning our craft and perfecting our art, spending significant amounts of money on equipment and training. It's not fair for others to profit off our imagination and dreams without giving us anything in return.
I propose to adopt a royalty-based model. This is already common in other creative industries, or with patents, where a company licenses its technology to anyone who is willing to pay for it. This is the fair thing to do, and a royalty-based model ensures that artists are properly recognized and compensated for their contributions. By adopting this approach, we can create a more equitable and sustainable ecosystem for AI image generation.
This proposed framework aims to ensure that artists are fairly compensated for their work used to train AI models. AI model creators would be required to disclose the list of artists in their training database and distribute a certain percentage of proceeds from the image generation platform as royalties to these artists. If the end-user specifies an artist whose work is under copyright in the image generator prompt, that artist would receive half of the royalties from generating that image. The rest would be distributed evenly among the other artists.
In detailAI model creators would first need to review their databases and identify artists whose work is included. They would then make a best effort to contact these artists and inform them that their work is being used to train AI models. They would then add the artists to their royalty database. Artists should also be provided with a tool like the "Have I Been Trained" -site where they can check if their art has been used in this way and tell developers to add them to the royalty recipient list.
To ensure fair and consistent compensation for these artists, a global industry-wide standard for disbursing royalties would need to be developed. This standard could be based on existing schemes in the recording and movie industries.
Each image generated by the end-user would be valued by the AI image generator provider, using either metered usage or per-image approach. The appropriate royalty rate would then be applied to this value, and the artist compensated on a regular basis. However, payments would only be made after a certain materiality threshold is reached to avoid excessive overhead and fees.
The royalty would be distributed evenly among the thousands of artists who contributed to the particular model. The general rule would be that "one contributed image equals one share of the total royalty paid." However, if the model weighs some images higher than others, this would increase or decrease the share of each image accordingly. For example, a model designed for photorealistic output might weigh photographic images more heavily, while still utilizing inputs from oil painters.
If the end-user specifies an artist whose work is protected by copyright in the image generator prompt (e.g. "in the style of Damien Hirst"), that artist would receive half of the royalties from generating the image. Rest would be distributed evenly among the other artists. This recognizes the value of the artist who inspired the user's request, while still compensating for the contributions of the other artists.
How much should the compensation be?
Determining the appropriate level of compensation for artists whose work is used to train AI models is a complex issue that will require input from a range of stakeholders. Standard royalty rates for this type of use vary widely, so it will be important to have an open and transparent discussion among artists, businesses, and society as a whole to determine a fair rate.
As a starting point, I put forth that artists should be compensated 20% of the revenue from the AI models that use their work. This is higher than many artists currently receive, but lower than the 30-40% royalties and fees that are common in the tech industry. The goal should be to find a balance between encouraging innovation and rewarding artists for their contributions to the technology.
One way to determine a fair compensation rate for artists is to consider the value of their work to the AI model. If an artist's work is critical to the training set and is used extensively by the AI model, then they should be compensated more than an artist whose work is used less frequently. This approach could be implemented by tracking the usage of each artist's work in the training set, and using that information to calculate their share.
Ultimately, the level of compensation for artists should be determined through a collaborative process that involves input from all key stakeholders. By considering the value of the artist's work to the AI model and the potential benefits to the artist, we can arrive at a fair and equitable solution.
What are the benefits?
AI image generation is a revolutionary technology. To ensure that this technology is developed and used in a fair and ethical manner it is important to find a balance between encouraging innovation and rewarding those who contribute to the development of these models, often without their knowledge or consent.
Implementing a framework for compensating artists whose work is used to train AI models would have several benefits. The artists themselves would directly benefit from the use of their work, and the developers of AI model generators would benefit from increased transparency in the creation of their models. This would help to ensure that the developers of these models reward the artists whose work they use, and encourage the creation of more images for their models.
End-users would also benefit in several ways.
First, end-users would have more information about how their "creations" are generated, which could help them understand and appreciate the role of the artists whose work is used to train the models. This could also help to prevent misunderstandings or disputes about the ownership or authorship of generated images.
Second, disclosing the stylistic source of each generated image could encourage end-users to explore the work of these artists and schools of art and potentially create new revenue streams for them. This would encourage end-users to engage more deeply with the art and creativity that is the foundation of AI image generation technology.
Society as a whole would also benefit from this framework. As AI image generation becomes more pervasive, the transition to a creative industry that relies on this technology would be smoother and more equitable when we pay the artists who helped create the tools.
By providing a financial incentive for artists to create and share their work, this framework would encourage more people to engage in artistic and creative endeavors. This could lead to a creation of new and diverse visual content which would be used to train AI models and support the industry.
What are the challenges?
Implementing a global framework for compensating artists whose work is used to train AI models would be a complex and challenging endeavor. There are several potential challenges and obstacles that would need to be overcome in order to create a system that is fair, equitable, and effective.
One challenge would be to ensure that artists from all over the world have equal access to the royalties generated by the use of their work. This could be particularly difficult for artists who are not part of the formal financial system, such as a graffiti artist in the slums of Rio de Janeiro. It would also be important to address issues of fairness and equity, such as whether famous or well-known artists with impactful work should receive a higher royalty rate than less established but prolific artists.
Another challenge would be to determine how to distribute the royalties in a way that is fair and equitable. The proposed "one image = one share" rule may not be adequate, as some images may be weighted more heavily by the algorithms used to train the AI models. In this case, it would be important to disclose this information and ensure that artists receive a fair share of the royalties based on the contribution of their work.
Some well-known artists may also object to the idea of having their work used to train AI models, or to receiving lower royalty rates than they are accustomed to. It will be important to address these concerns and explain the benefits of this framework to all artists, including those who are already successful.
While at the other end of the spectrum, lesser-known artists should also have access to royalties if their work is used to train AI models - no matter how small their contribution is.
Finally, the rise of AI image generation technology may pose challenges for upcoming artists, who may wonder why they should invest time and effort in creating art when anyone with a computer can generate new images with just a few words. It will be important to address this issue and explain the value of original, human-created art, as well as the potential opportunities for artists in a world where AI image generation is becoming increasingly common.
As AI image generation technology continues to advance and become more widely used, it is important to consider the implications for artists whose work is used to train these models. The rise of AI image generation presents both challenges and opportunities for the visual arts industry, and it will be important to develop a framework for compensating artists whose work is used in this way.
Creating a global, universal framework for compensating artists whose work is used to train AI models will be complex and challenging, but it is necessary in order to ensure that artists are fairly compensated for their work and to promote the ethical development of AI technology. By addressing these issues and engaging in open and civil public debate, it may be possible to minimize the disruption caused by AI image generation and maximize its potential for innovation and growth.
Thank you for reading this far! On the right a link to my first (very) short story Perspective, illustrated using Midjourney v4!
If you'd like to stay up-to-date on my new blog posts, the progress of my AI image projects, or my photography, please consider subscribing to my mailing list here. I promise I won't spam you, sell or give away your information, and I only send occasional updates so you won't be overwhelmed. Join now to be the first to know about new developments and updates from me!
Full disclosure: my photography has been used to train some of these AI models.
I used a chat-based AI tool called ChatGTP to help me generate the outline and structure of this post, as well as to assist with copy-editing. This includes the title of the post, which was generated by the AI tool. I believe it is important to be transparent about the role of AI in the creation of this content, and to acknowledge the contributions of both human and machine. It even added the last sentence by itself, and I agree wholeheartedly!