Say Anything with Any Style

arXiv:2403.06363v1 Announce Type: new
Abstract: Generating stylized talking head with diverse head motions is crucial for achieving natural-looking videos but still remains challenging. Previous works either adopt a regressive method to capture the speaking style, resulting in a coarse style that is averaged across all training data, or employ a universal network to synthesize videos with different styles which causes suboptimal performance. To address these, we propose a novel dynamic-weight method, namely Say Anything withAny Style (SAAS), which queries the …

abstract arxiv cs.cv data diverse head natural network speaking style talking head training training data type universal videos