Getty Images has filed a lawsuit against Stability AI for allegedly replicating over 12 million images, captions, and information illegally and without compensation. In addition, three graphic artists launched a class-action complaint against Stability AI for copyright infringement in January.
Stability Steady Diffusion, AI’s program, was trained on photos from diverse online sources. Several of the photographs were not in the public domain or had been distributed under open licenses such as Creative Commons. Artists and photographers throughout the world are concerned that their photographs are being used without their consent.
Law experts have spoken in, pointing out that using generative AI in this fashion is a unique technique, and courts have not before addressed its copyright consequences. While some say that Stability AI can use the photographs under the copyright law’s fair use concept, others argue the reverse.
According to the experts, there is a good chance that the courts will decide that Stable AI committed a massive copyright violation. If this happens, it might have a significant influence on the industry. Developing powerful generative AI will need getting permissions from hundreds, if not millions, of copyright holders, making the process time-consuming and costly.
The claim made by the plaintiffs in the class-action lawsuit is that Stable Diffusion is a “complex collage tool” that has “compressed copies” of its training images. Nevertheless, experts contend that this assertion is false. Stable Diffusion operates by transforming a user’s input into a latent representation, identifying images based on their “coordinates” in the “picture space,” and then transforming this latent representation into an image.
Stability AI is likely to compare Stable Diffusion to services such as Google Book Search and TurnItIn, arguing that computer programs only “view” the training images and not human beings.
The sources for this piece include an article in ArsTechnica.