Seedance 2.0 Photo: VCG

Chinese tech giant ByteDance on Thursday officially launched its artificial intelligent (AI) video generation model Seedance 2.0 after several days of beta test, fully integrating the model into its Doubao and Dreamina AI product lines, while introducing restrictions on the upload of real-person images and videos, according to a statement sent from ByteDance to the Global Times on Thursday.

ByteDance said in a post that Seedance 2.0 has made notable progress in addressing key technological challenges, including adherence to physical laws and long-sequence consistency, significantly enhancing controllability and output quality to meet professional production standards. The model is designed to offer creators greater flexibility while maintaining stability across complex, multi-shot video generation tasks.

Under the updated policy framework, Seedance 2.0 now restricts the use of real-person images or videos as primary reference inputs. Users seeking to generate content based on real individuals must complete identity verification or obtain proper authorization, reflecting tighter compliance safeguards amid growing regulatory scrutiny of generative AI applications, the statement said. 

Seedance 2.0 went through close beta test during the last weekend, which sparked widespread discussions on social media at home and abroad featuring powerful video-generation capabilities.

Bloomberg in a report noted that the app "surprised observers with the quality of generated clips." Technology-focused media outlet The Information reported that some called Seedance 2.0 a "game changer," adding that it is the latest indication of how Chinese companies are playing central roles in the advancements of AI video technology.

To comprehensively evaluate the model's performance across multimodal scenarios, ByteDance collaborated with film and television industry experts to establish a dedicated benchmarking system. According to the company, Seedance 2.0 achieved industry-leading results under the new evaluation standards.

ByteDance noted, however, that further improvements remain underway in areas including fine-grained stability, multi-speaker lip synchronization, multi-subject consistency, text rendering accuracy and advanced editing effects, as the company continues iterative optimization of the model, according to the statement. 

Global Times