Article: The stable diffusion model incorporating group feature constraints and interactive guidance, and a high-fidelity generation method for public group portraits Journal: International Journal of Reasoning-based Intelligent Systems (IJRIS) 2026 Vol.18 No.14 pp.1 - 21 Abstract: This study constructs a multi-dimensional conditional control strategy and a lightweight human-machine collaborative generation framework. The model is trained based on the Person Group Portrait 10K (PPR10K), Flickr-Faces-HQ (FFHQ), and WikiArt datasets (a total of 161,181 images). Verification is conducted using 1,116 test samples from the PPR10K dataset, and independent-samples t-test is adopted (P < 0.05). Experimental results show that the improved model achieves a peak signal-to-noise ratio of 32.40 decibels and a structural similarity index of 0.943. It represents improvements of 28.06% and 10.16% compared with the original SD model. The customised group consistency index and pose coordination index reach 0.91 and 0.89 (an increase of over 26%). In complex scenes such as strong lighting and large-scale groups (2-15), the model exhibits excellent stability and generalisation. This study provides a proposed framework for feature modelling and style adaptation of public group portrait generation. Inderscience Publishers - linking academia, business and industry through research

Title: The stable diffusion model incorporating group feature constraints and interactive guidance, and a high-fidelity generation method for public group portraits

Authors: Binghao Qu; Min Shi

Addresses: Department of Architectural Environment Art, Xi 'an Academy of Fine Arts, Xi'an, 710065, China ' AVIC Research Institute for Special Structures of Aeronautical Composites, Aviation Key Lab of Science and Technology on High Performance Electromagnetic Windows, Tsinan, 250014, China

Abstract: This study constructs a multi-dimensional conditional control strategy and a lightweight human-machine collaborative generation framework. The model is trained based on the Person Group Portrait 10K (PPR10K), Flickr-Faces-HQ (FFHQ), and WikiArt datasets (a total of 161,181 images). Verification is conducted using 1,116 test samples from the PPR10K dataset, and independent-samples t-test is adopted (P < 0.05). Experimental results show that the improved model achieves a peak signal-to-noise ratio of 32.40 decibels and a structural similarity index of 0.943. It represents improvements of 28.06% and 10.16% compared with the original SD model. The customised group consistency index and pose coordination index reach 0.91 and 0.89 (an increase of over 26%). In complex scenes such as strong lighting and large-scale groups (2-15), the model exhibits excellent stability and generalisation. This study provides a proposed framework for feature modelling and style adaptation of public group portrait generation.

Keywords: public group portrait; stable diffusion model; group feature constraint; high-fidelity generation; interactive guidance; AIGC art creation.

DOI: 10.1504/IJRIS.2026.154241

International Journal of Reasoning-based Intelligent Systems, 2026 Vol.18 No.14, pp.1 - 21

Received: 13 Jan 2026
Accepted: 01 Apr 2026
Published online: 17 Jun 2026 *

Title: The stable diffusion model incorporating group feature constraints and interactive guidance, and a high-fidelity generation method for public group portraits

Keep up-to-date