AI models acting as 'digital directors'? Seedance 2.0 take up the challenge

ByteDance released its AI video-generation model, Seedance 2.0, on Monday, marking a step toward large AI models acting as "digital directors."

Seedance 2.0 can create cinematic-style videos from text or image input and generate video and audio simultaneously. By providing a detailed prompt or uploading a single image, users can generate a multi-shot sequence with native audio in about 60 seconds.

A key feature of the model is cinematic storytelling. It can automatically generate multiple interconnected scenes from a single prompt.

Pan Tianhong, founder of the Chinese film and video channel Mediastorm, said in a review that when generating fight scenes, the model showed clear variations in shot composition, with frequent shifts in camera angles, similar to how a human director would move the camera.

Feng Ji, producer of the video game Black Myth: Wukong, said that after using the model, the video content industry could see an unprecedented surge in production capacity in the future and a reshaping of industry production processes.

However, the model has also raised concerns.

Pan found in his test that the system could generate voice characteristics closely matching his own, based solely on facial imagery, without user authorization. "I didn't provide any prompts, any text, or any information. I didn't upload my voice either. I only uploaded my face, and the AI still knew what voice belonged to that face," he said.

To prevent misuse, ByteDance said it has restricted some functions of the model, including allowing the generation of videos featuring real people only after identity verification, and disabling the use of real-person images or videos as references.

AI video-generation models are developing rapidly in China. On February 5, Kling AI launched its 3.0 model series, with cinematic storytelling also highlighted as a core feature.