DetailMaster: Can Your Text-to-Image Model Handle Long Prompts?

Qirui Jiao, Daoyuan Chen, Yilun Huang, Xika Lin, Ying Shen, Yaliang Li
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:53641-53676, 2026.

Abstract

While recent Text-to-Image (T2I) models show impressive capabilities in synthesizing images from brief descriptions, they struggle with the long, detailed prompts required for professional applications. We present DetailMaster, a comprehensive benchmark for evaluating T2I capabilities on long prompts with complex compositional requirements, accompanied by an automated data construction pipeline and an evaluation workflow. Comprising expert-validated prompts averaging 284.89 tokens, our benchmark introduces four critical evaluation dimensions: Character Attributes, Structured Character Locations, Multi-Dimensional Scene Attributes, and Spatial/Interactive Relationships. Evaluations on various general-purpose and long-prompt-optimized models reveal critical performance limitations, showing that weak encoders struggle to preserve syntactic dependencies within prompts and diffusion models suffer from attribute leakage under detail-intensive conditions. Through a controlled ablation study under varying constraints, we further show that high-fidelity generation requires a synergistic combination of expanded prompt limits and long-prompt training. We open-source our dataset and code to foster progress in long-prompt-driven T2I generation.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-jiao26b, title = {{D}etail{M}aster: Can Your Text-to-Image Model Handle Long Prompts?}, author = {Jiao, Qirui and Chen, Daoyuan and Huang, Yilun and Lin, Xika and Shen, Ying and Li, Yaliang}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {53641--53676}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/jiao26b/jiao26b.pdf}, url = {https://proceedings.mlr.press/v306/jiao26b.html}, abstract = {While recent Text-to-Image (T2I) models show impressive capabilities in synthesizing images from brief descriptions, they struggle with the long, detailed prompts required for professional applications. We present DetailMaster, a comprehensive benchmark for evaluating T2I capabilities on long prompts with complex compositional requirements, accompanied by an automated data construction pipeline and an evaluation workflow. Comprising expert-validated prompts averaging 284.89 tokens, our benchmark introduces four critical evaluation dimensions: Character Attributes, Structured Character Locations, Multi-Dimensional Scene Attributes, and Spatial/Interactive Relationships. Evaluations on various general-purpose and long-prompt-optimized models reveal critical performance limitations, showing that weak encoders struggle to preserve syntactic dependencies within prompts and diffusion models suffer from attribute leakage under detail-intensive conditions. Through a controlled ablation study under varying constraints, we further show that high-fidelity generation requires a synergistic combination of expanded prompt limits and long-prompt training. We open-source our dataset and code to foster progress in long-prompt-driven T2I generation.} }
Endnote
%0 Conference Paper %T DetailMaster: Can Your Text-to-Image Model Handle Long Prompts? %A Qirui Jiao %A Daoyuan Chen %A Yilun Huang %A Xika Lin %A Ying Shen %A Yaliang Li %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-jiao26b %I PMLR %P 53641--53676 %U https://proceedings.mlr.press/v306/jiao26b.html %V 306 %X While recent Text-to-Image (T2I) models show impressive capabilities in synthesizing images from brief descriptions, they struggle with the long, detailed prompts required for professional applications. We present DetailMaster, a comprehensive benchmark for evaluating T2I capabilities on long prompts with complex compositional requirements, accompanied by an automated data construction pipeline and an evaluation workflow. Comprising expert-validated prompts averaging 284.89 tokens, our benchmark introduces four critical evaluation dimensions: Character Attributes, Structured Character Locations, Multi-Dimensional Scene Attributes, and Spatial/Interactive Relationships. Evaluations on various general-purpose and long-prompt-optimized models reveal critical performance limitations, showing that weak encoders struggle to preserve syntactic dependencies within prompts and diffusion models suffer from attribute leakage under detail-intensive conditions. Through a controlled ablation study under varying constraints, we further show that high-fidelity generation requires a synergistic combination of expanded prompt limits and long-prompt training. We open-source our dataset and code to foster progress in long-prompt-driven T2I generation.
APA
Jiao, Q., Chen, D., Huang, Y., Lin, X., Shen, Y. & Li, Y.. (2026). DetailMaster: Can Your Text-to-Image Model Handle Long Prompts?. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:53641-53676 Available from https://proceedings.mlr.press/v306/jiao26b.html.

Related Material