Friend
Professional
- Messages
- 2,653
- Reaction score
- 851
- Points
- 113
Midjourney, Stability AI and others will now be forced to reveal their secrets.
A group of visual artists is celebrating the first victory in a lawsuit against the largest companies that create image and video generation systems using artificial intelligence. The judge ruled that the copyright infringement suit could move forward.
The case, registered under number 3: 23-cv-00201-WHO, was originally filed in January 2023. Since then, it has been updated several times, and some parts of it have been rejected, including today.
The class-action lawsuit involves prominent artists: Sarah Andersen, Kelly McKernan, Carla Ortiz, Hawke Southworth, Grzegorz Rutkowski, Gregory Manchess, Gerald Brom, Jinna Zhang, Julia Kay and Adam Ellis. They act on behalf of all artists against the companies Midjourney, Runway, Stability AI and DeviantArt.
Organizations are accused of copying plaintiffs work when creating products based on the Stable Diffusion model. According to the plaintiffs, this model was trained on their copyrighted works, which violates the law.
Judge William H. Orrick of the District Court for the Northern District of California, which oversees San Francisco, did not make a final decision on the case. However, he said that "the charges of inciting copyright infringement are sufficiently substantiated" to move to the disclosure stage.
This solution could allow artists lawyers to access the internal documents of companies developing AI-based image generators. In this way, the world can learn more about the training data sets, mechanisms of operation, and internal processes of these systems.
The artists reacted enthusiastically to this decision. Kelly McKernan wrote on social network X: "The judge allowed our copyright claims to be considered, and now we can find out everything that these companies do not want to tell us. This is a huge win for us!"
Carla Ortiz, another artist, noted: "This ruling also means that companies using Stable Diffusion models and / or data sets like LAION can now be held liable for copyright infringement and other misconduct."
Stable Diffusion was supposedly trained on the LAION-5B dataset, which contains more than 5 billion images collected from the Internet by researchers in 2022. However, this database contains only URLs and text descriptions of images. This means that companies developing AI had to separately collect or take screenshots of images for training Stable Diffusion and other derived products.
Judge Orrick also dismissed the artists claims under the Digital Copyright Act of 1998. This law prohibits companies from offering products designed to circumvent copyright protections in the online environment and software.
According to the plaintiffs, Stable Diffusion uses "CLIP-guided diffusion", which relies on hints that include the names of artists, to generate images. CLIP (Contrastive Language-Image Pre-training) is a neural network and AI training technique developed by OpenAI in 2021 that can identify objects in images and label them with text captions.
Source
A group of visual artists is celebrating the first victory in a lawsuit against the largest companies that create image and video generation systems using artificial intelligence. The judge ruled that the copyright infringement suit could move forward.
The case, registered under number 3: 23-cv-00201-WHO, was originally filed in January 2023. Since then, it has been updated several times, and some parts of it have been rejected, including today.
The class-action lawsuit involves prominent artists: Sarah Andersen, Kelly McKernan, Carla Ortiz, Hawke Southworth, Grzegorz Rutkowski, Gregory Manchess, Gerald Brom, Jinna Zhang, Julia Kay and Adam Ellis. They act on behalf of all artists against the companies Midjourney, Runway, Stability AI and DeviantArt.
Organizations are accused of copying plaintiffs work when creating products based on the Stable Diffusion model. According to the plaintiffs, this model was trained on their copyrighted works, which violates the law.
Judge William H. Orrick of the District Court for the Northern District of California, which oversees San Francisco, did not make a final decision on the case. However, he said that "the charges of inciting copyright infringement are sufficiently substantiated" to move to the disclosure stage.
This solution could allow artists lawyers to access the internal documents of companies developing AI-based image generators. In this way, the world can learn more about the training data sets, mechanisms of operation, and internal processes of these systems.
The artists reacted enthusiastically to this decision. Kelly McKernan wrote on social network X: "The judge allowed our copyright claims to be considered, and now we can find out everything that these companies do not want to tell us. This is a huge win for us!"
Carla Ortiz, another artist, noted: "This ruling also means that companies using Stable Diffusion models and / or data sets like LAION can now be held liable for copyright infringement and other misconduct."
Stable Diffusion was supposedly trained on the LAION-5B dataset, which contains more than 5 billion images collected from the Internet by researchers in 2022. However, this database contains only URLs and text descriptions of images. This means that companies developing AI had to separately collect or take screenshots of images for training Stable Diffusion and other derived products.
Judge Orrick also dismissed the artists claims under the Digital Copyright Act of 1998. This law prohibits companies from offering products designed to circumvent copyright protections in the online environment and software.
According to the plaintiffs, Stable Diffusion uses "CLIP-guided diffusion", which relies on hints that include the names of artists, to generate images. CLIP (Contrastive Language-Image Pre-training) is a neural network and AI training technique developed by OpenAI in 2021 that can identify objects in images and label them with text captions.
Source
