Google, Alphabet Stole Visual Artists’ Copyrighted Images to Train Imagen AI, Class Action Lawsuit Alleges

New to ClassAction.org? Read our Newswire Disclaimer

Several visual artists allege in a new proposed class action lawsuit that Google and parent company Alphabet have violated federal copyright law by copying an enormous number of protected digital images to train the text-to-image Imagen artificial intelligence software.

Get class action lawsuit news sent to your inbox – sign up for ClassAction.org’s free weekly newsletter here.

The 16-page AI lawsuit explains that Imagen is a text-to-image diffusion model, which takes a short text description of an image and then, via the machine-learning technique called diffusion, generates an image in response to the prompt. Rather than be programmed by humans, a diffusion model is trained by copying vast amounts of digital images with text captions, the filing shares.

According to the suit, Google’s Imagen AI was trained on massive datasets containing millions of images paired with descriptive captions, and as the model copies and ingests billions of training images, it progressively develops the capability to generate outputs that “mimic the protected expression copied from the dataset,” the case relays.

The plaintiffs say neither they nor proposed class members gave Google permission to use their copyrighted digital works as training material for Imagen, which is “itself an infringing derivative work” because it contains large sets of numbers, called weights, that represent a transformation of the copyrighted work in the training dataset, according to the suit.

“Alphabet, as the corporate parent of Google, also commercially benefits from these acts of massive copyright infringement,” the case alleges.

A plaintiff in the proposed class action was among three artists who filed the first lawsuit in the United States challenging the legality of training text-to-image diffusion models on copyrighted material without consent, credit or compensation, the filing relays. Both of the diffusion models at the center of the plaintiff’s initial AI lawsuit, called Stable Diffusion and Midjourney, were trained on the LAION-400M dataset, which was also utilized by Google for Imagen, the lawsuit says.

The complaint alleges that for Imagen 2, the successor to Imagen that Google released in December 2023, the tech giant did not disclose details about the training dataset for the updated AI because it was aware of the plaintiff’s prior diffusion model class action and “hoped to avoid being named as a defendant in a lawsuit over the legality of training on mass quantities of copyrighted works without consent, credit, or compensation.”

The Google AI lawsuit looks to cover all persons or entities residing in the United States that own a U.S. copyright in any work that Google used as a training image for the Google-LAION models at any time since April 26, 2021.

Are you owed unclaimed settlement money? Check out our class action rebates page full of open class action settlements.