Latent inversion for consistent identity preservation in character animation

doi:10.1007/s00371-025-03956-z

	Latent inversion for consistent identity preservation in character animation
	Li, Haochen 1,2; Tang, Sheng 1,2; Wan, Zhang 1,2; Cao, Juan 1,2; Li, Jintao 1,2
	2025-05-22
发表期刊	VISUAL COMPUTER
ISSN	0178-2789
页码	14
摘要	This paper presents InvLatents, a novel framework for character animation that leverages latent inversion diffusion models to ensure consistent identity preservation across frames. Existing diffusion-based character animation methods often struggle with maintaining identity consistency due to the inherent randomness in the generation process. To address this issue, InvLatents introduces a latent inversion technique that incorporates target identity and pose guidance into the inference stage. By controlling different injection ratios in different branches, the method obtains richer identity information from the reference image. Additionally, a lightweight pose integration module is introduced to compensate for potential missing pose guidance. Experimental results on the TikTok dataset demonstrate that InvLatents achieves competitive performance compared to state-of-the-art approaches, effectively maintaining both identity and pose consistency without requiring additional training. The proposed method can be integrated as a plugin into other diffusion models, offering a promising solution for generating temporally coherent motion videos with consistent identity. Project page: https://github.com/SodaLee/InvLatents.
关键词	Diffusion model Character animation Human dance generation DDIM inversion
DOI	10.1007/s00371-025-03956-z
收录类别	SCI
语种	英语
资助项目	Beijing Science and Technology Plan Project[Z231100005923033]
WOS研究方向	Computer Science
WOS类目	Computer Science, Software Engineering
WOS记录号	WOS:001492947600001
出版者	SPRINGER
引用统计
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/42406
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Tang, Sheng
作者单位	1.Chinese Acad Sci, Inst Comp Technol, Beijing 100190, Peoples R China 2.Univ Chinese Acad Sci, Beijing 100049, Peoples R China
推荐引用方式 GB/T 7714	Li, Haochen,Tang, Sheng,Wan, Zhang,et al. Latent inversion for consistent identity preservation in character animation[J]. VISUAL COMPUTER,2025:14.
APA	Li, Haochen,Tang, Sheng,Wan, Zhang,Cao, Juan,&Li, Jintao.(2025).Latent inversion for consistent identity preservation in character animation.VISUAL COMPUTER,14.
MLA	Li, Haochen,et al."Latent inversion for consistent identity preservation in character animation".VISUAL COMPUTER (2025):14.