OpenAI's video generation artificial intelligence (AI) 'Sora' is reportedly learning from actual game content, leading to copyright infringement controversies. Experts believe that the outcome of impending copyright lawsuits involving ChatGPT will also impact Sora.
According to OpenAI on the 12th, the company launched the video generation AI 'Sora' on the 9th (local time). Sora generates the desired scenes when users input commands. The videos that can be generated by Sora can be up to 20 seconds long with a maximum resolution of 1080 pixels (p). Sora also provides the ability to create videos based on images, expanding existing videos or filling in missing frames. OpenAI stated, 'Through Sora, users can visualize their imagination beyond text or images with videos.'
However, there are claims in the industry that Sora appears to have learned from actual game content. When OpenAI first showcased Sora in February, it hinted at learning the model from Minecraft videos. OpenAI has not clearly disclosed what data was used to train Sora, only mentioning that it utilized public data and licensed data. Users who have used Sora directly assert that its data seems to come from actual games or Twitch streaming.
According to the IT media outlet TechCrunch, users who have actually tried Sora could generate videos in the style of the Nintendo game 'Super Mario Brothers' or the well-known first-person shooter game 'Call of Duty.' The videos created by Sora feature pixel art-style characters that closely resemble Super Mario, jumping to collect coins. Another video created a game where users shoot guns and move forward, similar to Call of Duty. Additionally, Sora generated characters reflecting the appearance of popular Twitch streamers, even replicating the tattoos on those streamers' arms.
Experts warned that OpenAI could face copyright infringement lawsuits if it misused game play videos. They particularly noted that game content involves not only the developers but also various stakeholders such as streamers, making the legal risks substantial.
Kim Kyung-hwan, a representative attorney at the law firm Minhoo, said, 'Currently, OpenAI is in litigation with media organizations, image production companies, etc., regarding the training data of ChatGPT.' He added, 'It seems likely that the copyright risk of Sora will be linked to the ruling of that lawsuit, and under current laws, copyright protection is deemed necessary concerning the reproduction or derivative works of games, indicating that copyright infringement factors exist.'
Choi Byung-ho, a professor at Korea University's AI Research Institute, explained, 'In the case of videos, there is not as much data available as there is for text, and Sora likely learned from limited content to provide high-quality videos.' He added, 'In this case, game content becomes a useful learning resource.'
He further noted, 'Currently, the most important thing for OpenAI is market capture,' stating, 'Although they are well aware of the copyright infringement controversies, they prefer to negotiate after lawsuits arise rather than consulting with copyright-holding companies beforehand.'