That’s how Wix got involved in a class action lawsuit against AI-based image generators

by time news

When Avishai Abrahami, the CEO of Wix, decided to purchase the DeviantArt website, a website that allows creators and digital artists to store and display their works, he certainly did not imagine that a few years later it would find itself at the center of a legal storm alongside the largest companies in the field The creative artificial intelligence: Midjourney and stability.

Deviant Art, which is managed by the Israeli Moti Levy and is under Wix’s control, was joined by a number of artists and programmers in a class-action lawsuit filed by the Joseph Sabry law firm, which specializes, among other things, in class-action lawsuits in the field of artificial intelligence. According to the plaintiffs – artists, programmers and lawyers – Deviant Art allowed the image generators to draw the works from its database, which allowed them to “train” the software and thus improve their abilities to produce better images themselves using artificial intelligence.

The image generators based on Generative AI such as Stable Diffusion (Stable Diffusion) or DALL-E from the Open AI company have become a sensation in recent months with their ability to produce images in seconds using a simple written instruction that includes the necessary objects, the setting and the drawing style. For example: “Create a picture of a polar bear walking in the desert, comic style”, or “Astronaut in a space suit, impressionistic style”.

Artificial intelligence training is a process during which software learns reality by scanning millions of images or lines of text. An autonomous car, for example, “trains” by driving on roads, photographing the road and avoiding obstacles. The same goes for the AI-based image generators who practice imitating images created by human creators by scanning millions of them.

What sets Stable Diffusion apart from other AI engines?

Stable Diffusion, which was launched by the company Stability last August, differs from DALL-E and other engines in its openness to many databases and in being an open source database that companies can combine in order to enhance their artificial intelligence capabilities. The prosecutors estimate that this openness helped her “practice” on billions of images, most of which are protected under copyright and without the knowledge of the artists who created them. “Even if we assume that the damage caused for each image is one dollar, the cumulative damage can be estimated at at least 5 billion dollars.”

“Even though the product may be similar to the original images with which the machine was trained, it may compete with the original image on the market,” the plaintiffs write in the announcement. “Stable Diffusion may, therefore, flood the market with an unsupported numberof copyright infringement of images, which will cause damage to the artists and the market in which they operate.” It should be noted that both in Israel and in the US, training artificial intelligence engines on images protected by copyright is legal without the need to pay the rights holders, as long as the product itself does not constitute Clearly a violation of copyright.

Deviant Art was founded in 2000, and over the years has become the leading website among digital art creators. Creators usually open an account there, upload their works and comment on each other’s works. BeVix paid $36 million for the company in part to increase their offering to artists and to encourage many of them to open their own site on Wix or sell works through its store.

According to the class action, millions of images and works on the database were copied and scanned from Deviant Art in order to train the image generator Stable Diffusion using the LAION training model. “Instead of protecting the community of artists from being trained by a machine, Deviant Art launched its own engine called DreamUp – a paid application based on Stable Diffusion’s open source code,” reads a statement accompanying the filing of the lawsuit. “Afterwards, a surge of AI-based art promoted Deviant Art, but pushed out the creators. Management led by CEO Motti Levy failed to explain why they betrayed the artist community by adopting the Stable Diffusion engine for themselves, violating the terms of The use and privacy of the platform they themselves manage.”

It should be noted that at the time DreamUp was launched, Deviant Art allowed its registered artists to protect their works from being scanned by image generators. “Technology for the automatic creation of images is a huge force that cannot be ignored, it will be impossible for Deviant Art to block or censor it,” Tal said at the launch. “We see that many times it helps artists to express themselves better than before. At the same time, we believe that we have a responsibility for all artists, and we must create tools that are fair and allow protection in this area.”

According to the writ of the class action, it appears that Moti Tal bragged about bringing the open code of Stable Diffusion to the site under his management. “It was the only option to integrate open source – the other platforms didn’t allow it, and by the way, it was my decision,” he said.

This is not the first time that Tal wants to integrate innovative tools on the creator community website. Previously, the company published a tool that allows creators to receive notifications about NFT works that infringe their copyrights.

The plaintiffs claim: the Lyon organization maintains databases of protected images without the consent of their creators

Stability (Stability AI) was founded by the entrepreneur Imad Mustak. Behind it is the German organization Lyon (LAION) which enables the use of its technology as open source for other companies. According to the claim of the plaintiffs, the Lyon organization maintains databases of protected images without the consent of their creators.

The lawsuit was filed by the Savory Law Firm, on behalf of Matthew Butrick – a programmer, designer and copyright activist who also recently filed a class action lawsuit against Microsoft-owned GitHub for development code copyrights. He was joined by artists Sarah Andersen, Kelly McKernan and Carla Oritz. The claim of the four has not yet been accepted by the US court, and only if it is accepted will it be moved to the stage of proving the amount of the damage.

No response to the news was received from Wix. And yet, the chances of the class action being accepted or continuing to later stages in court are low. Legal experts are not impressed by the detailed claim letter. “The class action against Stable Diffusion, Midgerani and Deviant Art will oblige the US courts to deal for the first time in the world directly with the question of whether training generative models on a data set that includes works that are protected by copyright constitutes fair use,” a lawyer tells Globes Rod Khoresh who specializes in artificial intelligence. “Assuming that this lawsuit will be managed until a verdict and will not end in a compromise, which is certainly likely, it does an important service to the field. Legal certainty in any case is better than the current situation.”

“On the other hand, as far as the artists are concerned, this lawsuit suffers from two fundamental problems that may hinder obtaining a legal result that benefits creators.” Horesh says. “The first problem relates to the fact that generative models are described in the lawsuit as algorithms that produce a collage of images in the training data. Models don’t really work like that and it’s very easy to disprove it. When this happens in the framework of the procedure, it may lead the judges to doubt the entire argument of the plaintiffs.”

“The second problem is more legal: the lawsuit should have focused on the claim that the very training of the models constitutes a violation of copyright. However, the plaintiffs also make many claims regarding the products of the models. In the legal world there is no dispute that if, according to the existing legal tests, a product violates copyright or other rights , the fact that it was created through artificial intelligence is irrelevant, and the creator whose rights have been violated has a legal remedy. Therefore, the class action should have focused on the issue of training the models only. Mixing the claims weakens the strength of the plaintiffs’ argument and may harm their chances of getting a verdict that serves the community of artists.”

You may also like

Leave a Comment