Beyond Single-Concept Personalization: A Systematic Survey of Multi-Concept Methods in Text-to-Image Diffusion Models

DOI

10.2991/978-94-6239-828-3_2How to use a DOI?

Keywords

Multi-concept personalization; Text-to-image synthesis; Diffusion models; Concept disentanglement; Identity preservation

Abstract

Text-to-image (T2I) generative diffusion models have made remark- able progress in single-concept personalization. However, extending customization to multiple concepts simultaneously introduces critical challenges, including identity bleeding (where visual features of distinct subjects merge) and attribute leakage (where properties bind to incorrect subjects). These failures stem from global cross-attention mechanisms that lack spatial constraints. Despite the rapid proliferation of methods addressing multi-concept personalization, a comprehensive survey that systematically categorizes these approaches is still missing in the literature. This paper presents a systematic survey of the state-of-the-art in multi-concept personalization. We propose a unified taxonomy that classifies existing methodologies along four primary dimensions: (1) personalization paradigm and concept representation space (optimization-based, tuning-free, and hybrid), (2) disentanglement and generation strategy (weight merging, attention routing, layout control, and score fusion), (3) target personalization granularity (full-object, part-level, relation-aware, and identity-specific), and (4) generative backbone architecture (U-Net, Diffusion Transformer, and Autoregressive models). Furthermore, we systematically examine how these methods integrate with modern backbones and granularities, highlighting key trade-offs between parameter tuning efficiency and zero-shot generalization. We analyze standard evaluation protocols, covering benchmark datasets and quantitative metrics for subject fidelity, text alignment, and spatial composition. Finally, we identify key open challenges such as strict spatial isolation, hierarchical subject-object binding, and zero-shot scaling, offering a clear, structured roadmap for future research in this rapidly evolving field.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Muhammad Rifqi Fauzi
AU  - Gembong Edhi Setyawan
AU  - Novanto Yudistira
PY  - 2026
DA  - 2026/11/04
TI  - Beyond Single-Concept Personalization: A Systematic Survey of Multi-Concept Methods in Text-to-Image Diffusion Models
BT  - Proceedings of the 2026 11th International Conference on Sustainable Information Engineering and Technology (SIET 2026)
PB  - Atlantis Press
SP  - 7
EP  - 17
SN  - 2589-4900
UR  - https://doi.org/10.2991/978-94-6239-828-3_2
DO  - 10.2991/978-94-6239-828-3_2
ID  - Fauzi2026
ER  -