ByteDance bolsters Seedance 2.0 with watermarking, IP safeguards ahead of global roll-out

Video-generation model Seedance 2.0 is displayed at the Zhongguancun Exhibition Center in Beijing on March 25.Photo: Xinhua ByteDance has bolstered its controversial video-generation model Seedance 2.0 with “advanced” watermarking and intellectual property (IP) protection guardrails ahead of its global roll-out, the TikTok owner has said.

The much-anticipated international release comes amid intense scrutiny from Hollywood studios over alleged IP theft after numerous Seedance-generated videos featuring famous Hollywood actors and characters went viral in February.

On Wednesday, ByteDance said its global safety and intellectual property teams had worked with a third-party “red-teaming” partner to bolster the safety features of Seedance 2.0, which would become available to users of its artificial intelligence content creation platform CapCut in major markets including Europe, Africa, South America and Southeast Asia.

The new features include C2PA watermarking that would allow ByteDance to “identify content made with this model and take action against it when needed, even if it has been shared or altered off our platform”, and “proactive monitoring” of IP violations on the platform.

Launched in 2021, C2PA is an industry standard backed by Microsoft, Google and Adobe that promotes transparency of online content through digital watermarking.

A new watermarking feature allows ByteDance to “identify content made with this model and take action against it when needed, even if it has been shared or altered off our platform”.

Photo: Getty Images ByteDance first partnered with C2PA to automatically label AI-generated content on its flagship app TikTok in 2024, becoming the first video sharing platform to do so.

In the AI industry, “red-teaming” refers to the practice of deliberately simulating attacks on new systems to identify safety and security flaws.

Global CapCut users would also be restricted from generating videos using the faces of real people and copyrighted characters, ByteDance said, in line with restrictions introduced on the mainland market after Chinese users raised concerns about pornographic deepfakes and privacy infringements. “These new safeguards build on the measures CapCut already has that are designed to protect our community, advance AI transparency, and prevent harmful content,” the company said.

The announcement came the same week that OpenAI halted its AI video platform Sora, with ByteDance looking to capitalise on a model widely se

原文链接: 南华早报