Article: A virtual human generation method combining user emotional preferences with implicit reconstruction Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.50 pp.95 - 110 Abstract: Combining personalised expression with high-fidelity geometric reconstruction has been a challenging task. To generate realistic virtual humans, this paper proposes an effective new framework. By combining users' emotional preferences with implicit neural radiance fields, personalised virtual human bodies are generated. This method encodes multi-modal user input into structured conditional variables, and then guides the conditional neural radiance field model to generate facial images with emotional expressiveness. The innovative learnable user-specific embeddings can capture individual expression styles. Additionally, the attention-based fusion module ensures precise alignment between emotional semantics and facial details. Through experiments on standard datasets, the proposed method achieved a fréchet inception distance score of 15.38 and an emotion recognition accuracy of 0.892, significantly outperforming three baseline approaches. These results demonstrate its substantial advantages in emotional accuracy, identity preservation, and overall visual quality. Inderscience Publishers - linking academia, business and industry through research

Title: A virtual human generation method combining user emotional preferences with implicit reconstruction

Authors: Liang Cao; Shuoqiu Zhao

Addresses: School of Animation and Digital Arts, Communication University of China, Nanjing, 211172, China ' School of Animation and Digital Arts, Communication University of China, Nanjing, 211172, China

Abstract: Combining personalised expression with high-fidelity geometric reconstruction has been a challenging task. To generate realistic virtual humans, this paper proposes an effective new framework. By combining users' emotional preferences with implicit neural radiance fields, personalised virtual human bodies are generated. This method encodes multi-modal user input into structured conditional variables, and then guides the conditional neural radiance field model to generate facial images with emotional expressiveness. The innovative learnable user-specific embeddings can capture individual expression styles. Additionally, the attention-based fusion module ensures precise alignment between emotional semantics and facial details. Through experiments on standard datasets, the proposed method achieved a fréchet inception distance score of 15.38 and an emotion recognition accuracy of 0.892, significantly outperforming three baseline approaches. These results demonstrate its substantial advantages in emotional accuracy, identity preservation, and overall visual quality.

Keywords: virtual human generation; implicit neural radiance fields; affective computing; conditional generation.

DOI: 10.1504/IJICT.2026.153625

International Journal of Information and Communication Technology, 2026 Vol.27 No.50, pp.95 - 110

Received: 28 Jan 2026
Accepted: 03 Mar 2026
Published online: 18 May 2026 *

Title: A virtual human generation method combining user emotional preferences with implicit reconstruction

Keep up-to-date