Since its initial public outing, AI imagery has faced many challenges and criticisms. Chief among them is the topic of copyright. Essentially, many AI models today have been trained using copyrighted content.
Now, Israeli startup Bria AI has teamed up with Getty Images, Alamy and others to take the “responsible” approach to AI imagery. Their AI models are trained on legally licensed stock images.
[Related reading: Getty Images sues Stable Diffusion makers over copyright violations]
Many of the current major AI players, including OpenAI, Google and Microsoft, are facing legal issues regarding the origins of the data on which their models were trained and the data those AI models produce.
Bria AI’s Ethical Approach to AI-Generated Images
Bria says that they’re taking a different approach, though. A more “responsible” approach. The company has stated they’re going to be using only permissible content from day one. That means everything will be legally licensed and above board.
This ethically sourced content is the key differentiator between Bria and many other AI platforms available today. Not all platforms use copyrighted content without permission to train their models, but it seems to be the majority if reports are to be believed.
To aid in this, Bria has teamed up with Getty Images, bringing not only legitimacy to the company’s name and products but also a massive treasure trove of legally licensed images on which to train the company’s AI models.
Getty’s controversial maze and AI
Getty might want to come across as the saviour in this story, but they’ve faced their share of legal issues and criticism in the past, too. In April 2019, Getty was sued for licensing public domain images. In July 2020, it was discovered that they were licensing screenshots from video games as “photos”.
Last year, Getty banned AI-generated images from its platform. This was seen as a great move. Getty taking a stand for copyright owners! Right? Well, no, they just didn’t want to find themselves in the legal hot water of AI-generated images from models trained on copyrighted material.
It’s not an unreasonable position to take, though, even if their motives were purely those of self-preservation. They even sued the creators of Stable Diffusion for illegally using images represented by Getty to train their models.
Most recently, Getty, along with others, penned an open letter to the AI community. They want more transparency with AI-generated images and content. They want to know what source material the AI used and whether or not it was legally licensed.
In short, though, Getty’s focus seems to be on keeping itself out of legal issues and making a profit. If photographers and other content creators benefit from this, too, it’s simply an unintended byproduct of their own motives.
Bria and Getty want to pay you
Bria’s move to partner with Getty is a step in the right direction. Getty is reported to have bought a $2 million minority stake in Bria AI, although its total investment has not been revealed. But one of the goals of the system is to be able to pay the creators whose content was used to generate AI content.
While Getty brings the images, Bria AI brings the attribution model. It keeps track of each individual image used in the training data sets for the AI models. When something is used to create an AI-generated image, that something’s creator is paid – bringing that transparency the open letter demanded.
The proposal is said to work similarly in concept to Spotify. Artists have their music put on some playlist that randomly comes up for a user’s listening pleasure. Each time that happens, Spotify pays the artists a sum of money. It’s a teeny-tiny, almost non-existent sum of money, but it’s some.
I expect that it won’t be a lot of money at all. It may not even be enough for you to buy yourself a beer at the end of each month. With the number of images used to train AI models, any revenue is likely going to be split between a lot of people.
[Related reading: Photo agencies band together in open letter pleading for AI rules]
Even if the AI were basing an image solely on images from a single photographer, the sheer volume of people creating AI-generated images makes the figure the original creator would receive extremely small per new image generated.
When you’re potentially splitting that across 50 or 100 different authors of original works, “extremely small” doesn’t even begin to describe it. Not to mention the fact that this figure likely won’t be split evenly, as each image will offer a different amount of influence in the generated result.
So, this isn’t going to be an easy road to travel, by any stretch of the imagination. But, as I said, it’s a step in the right direction.
It will be interesting to see how this plays out and if Bria and Getty can pull off their goal of actually paying creators.