Људске фигуре у архитектонској визуализацији: методе, реалистичност и изазови инкорпорирања људи у рендерима помоћу АI алата (истраживање из 2024. године)

Станислава  Вучетић

doi:10.24867/34FA05Vucetic

Architecture

Vol. 41 No. 04 (2026): Proceedings of the Faculty of Technical Sciences

Human Figures in Architectural Visualization: Methods, Realism and Challenges of Incorporating People in Renderings Using AI Tools (2024 Research)

Станислава Вучетић

34FA05Vucetic.pdf (Serbian)

DOI:: https://doi.org/10.24867/34FA05Vucetic
Submitted: March 26, 2026
Published: 2026-04-21

Abstract

This Master’s thesis explores the use of artificial intelligence in the incorporation of human figures in architectural visualization. As a part of the research, traditional methods (3D modeling and postproduction) are being compared with contemporary AI approaches based on the Stable Diffusion model. The research includes multiple types of scenes (daytime and nighttime scenes, interior and exterior scenes) and different application methods (with a base picture, with a base in the form of human silhouettes, and without a base), with an assessment of the realism, efficiency and technical limitations of the obtained results. Research shows that AI tools can significantly speed up the visualization process and improve the aesthetic quality of the display, with some limitations in certain aspects of human anatomy and lighting.

References

[1] Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep Learning. MIT Press.
[2] Nichol, A. Q., & Dhariwal, P. (2021). Improved Denoising Diffusion Probabilistic Models. Advances in Neural Information Processing Systems (NeurIPS).
[3] Maestri, G. (2006). Digital Character Animation 3. New Riders Press.
[4] Rombach, R., Blattmann, A., Lorenz, D., Esser, P., & Ommer, B. (2022). High-Resolution Image Synthesis with Latent Diffusion Models. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 10684–10695.
[5] https://github.com/CompVis/stable-diffusion (приступљено у октобру 2025.)
[6] https://huggingface.co/docs/diffusers/api/pipelines/
stable_diffusion/inpaint (приступљено у октобру 2025.)
[7] https://archicgi.com/cgi-services/people-in-architectural-
renderings-options/ (приступљено у октобру 2025.)

[1] [1] Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep Learning. MIT Press.

[2] [2] Nichol, A. Q., & Dhariwal, P. (2021). Improved Denoising Diffusion Probabilistic Models. Advances in Neural Information Processing Systems (NeurIPS).

[3] [3] Maestri, G. (2006). Digital Character Animation 3. New Riders Press.

[4] [4] Rombach, R., Blattmann, A., Lorenz, D., Esser, P., & Ommer, B. (2022). High-Resolution Image Synthesis with Latent Diffusion Models. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 10684–10695.

[5] [5] https://github.com/CompVis/stable-diffusion (приступљено у октобру 2025.)

[6] [6] https://huggingface.co/docs/diffusers/api/pipelines/

[7] stable_diffusion/inpaint (приступљено у октобру 2025.)

[8] [7] https://archicgi.com/cgi-services/people-in-architectural-

[9] renderings-options/ (приступљено у октобру 2025.)