Xin Jing Bao·2025-09-18 11:59Core Insights - Kuaishou has launched a new digital human feature through its Keling AI, which supports generating videos up to 1 minute long, currently in public beta testing [2] - The digital human technology utilizes a cross-attention mechanism for audio-visual alignment, enhanced lip-sync training strategies, and refined data processing to synchronize voice and lip movements [2] - The model constructs a high-level narrative framework and generates multiple segments of digital human videos in parallel while maintaining identity consistency [2] Performance Testing - The digital human feature was tested with songs in Chinese, English, and Korean, showing better lip-sync performance with Chinese songs compared to English and Korean [4] - In the 8-second video test, certain words in English and Korean songs did not align well with the digital human's lip movements [4]