On February 9, the well-known tech blogger "Film Hurricane" Tim released a new video, offering an in-depth review of ByteDance's latest AI video model Seedance 2.0. Although Tim gave high praise for the model's "industry-level" performance in terms of generation accuracy, camera movement continuity, and audio-visual synchronization, two details he discovered during testing have triggered deep concerns within the industry about AI data ethics.
"Foreseeing the Future" Spatial Modeling and Voice Cloning
In the video, Tim demonstrated two phenomena that left him exclaiming "terrifying":
Precise Reconstruction of Spatial Blind Spots: Even when only the front photo of a building was uploaded without any background information, Seedance 2.0 generated a camera movement around the back of the building that accurately recreated the real structure behind it.
Unperceived Voice Simulation: The model generated a voice highly similar to Tim's own tone and manner of speaking, based solely on his face photo without any reference audio.
The Mystery of Data Authorization: Your Video Has Become "Nutrition"?
Based on these findings, Tim admitted that it is basically certain that the large amount of high-definition materials previously published by Film Hurricane on the cloud have been included in ByteDance's training set. Although he has never received any authorization request or compensation, he suspects that the relevant authorization might have already been hidden in complex user agreement clauses.
Tim further found through testing that the model also has a high level of accuracy in reproducing several other bloggers, including "He Tongxue." He warned that if all forms and voices of a person are simulated by AI at 100%, the authenticity of the generated content will reach a level where even family members may not be able to distinguish it, and society urgently needs to be vigilant about the potential risks brought by such technology.
