ByteDance Suspends Seedance 2.0 Voice Feature Over Privacy Risks
ByteDance has urgently suspended a controversial feature in its Seedance 2.0 AI video generation model, known as Jimeng in China, after it was discovered that the system could generate highly accurate personal voice clones using only facial photographs. The feature, which leverages a dual-branch diffusion transformer architecture to create 2K resolution videos with native audio, raised significant privacy and ethical concerns regarding unauthorized identity forgery. The issue came to light when Pan Tianhong, founder of MediaStorm, demonstrated that uploading a static facial image resulted in audio nearly identical to his real voice without any prior voice samples or consent. In response to public backlash and potential misuse for deepfakes or scams, ByteDance announced immediate changes, prohibiting the use of real-human-like photos as reference subjects. Additionally, the company introduced live verification steps requiring users to record their own image and voice before creating digital avatars. This incident highlights the growing challenge of balancing rapid AI innovation with regulatory compliance and data security. While Seedance 2.0 remains in internal testing, ByteDance emphasizes its commitment to responsible AI development by strengthening content reviews and limiting high-risk functionalities to ensure a safe creative environment.
Wire timeline
ByteDance Suspends Seedance 2.0 Voice Feature Over Privacy Risks
ByteDance has urgently suspended a controversial feature in its Seedance 2.0 AI video generation model, known as Jimeng in China, after it was discovered that the system could generate highly accurate personal voice clones using only facial photographs. The feature, which leverages a dual-branch diffusion transformer architecture to create 2K resolution videos with native audio, raised significant privacy and ethical concerns regarding unauthorized identity forgery. The issue came to light when Pan Tianhong, founder of MediaStorm, demonstrated that uploading a static facial image resulted in audio nearly identical to his real voice without any prior voice samples or consent. In response to public backlash and potential misuse for deepfakes or scams, ByteDance announced immediate changes, prohibiting the use of real-human-like photos as reference subjects. Additionally, the company introduced live verification steps requiring users to record their own image and voice before creating digital avatars. This incident highlights the growing challenge of balancing rapid AI innovation with regulatory compliance and data security. While Seedance 2.0 remains in internal testing, ByteDance emphasizes its commitment to responsible AI development by strengthening content reviews and limiting high-risk functionalities to ensure a safe creative environment.
TechNode