Answers are grounded in this paper and cite their sources.
Affective virtual humans require facial responses that evolve coherently over time, reflect conversational context, and transfer across heterogeneous character appearances. Existing emotional portrait-animation methods commonly rely on discrete emotion labels, frame-wise motion variables, or latent affect representations. This study formulates AIGC-supported virtual-human animation as a context-conditioned temporal affective-behavior modeling problem and proposes TABI-DH. The framework integrates multimodal speech and context encoding, a Temporal AU Behavior Planner, AU-guided non-rigid facial geometry, externally controlled rigid head pose, and behavior–appearance factorized rendering. On MEAD, TABI-DH achieves 22.74 dB PSNR, 0.71 SSIM, 84.25% emotion accuracy, and an AU-trajectory correlation of 0.89. Cross-dataset evaluation on LRW, GRID, and HDTF demonstrates consistent performance under distribution shift. In session-independent IEMOCAP evaluation, incorporating past conversational information increases four-class emotion agreement to 77.1% and reduces affective-response onset error to 176 ms. Mechanism analyses indicate that temporal AU constraints and upper-face behavior primarily improve affective fidelity rather than lip synchronization. Artistic-portrait experiments further show that a common AU trajectory can transfer across photographic and stylized identities while preserving source appearance. These findings support temporally structured AUs as an interpretable representation connecting affective interaction modeling with controllable AIGC virtual-human animation.
No takes yet. Share an insight, caveat, or question.
Yanshuai Shi (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: