Describe, Don't Dictate: Semantic Image Editing with Natural Language Intent

Banner Image

Abstract

Method Overview

Banner Image

Overview of the DescriptiveEdit framework. We propose a novel image editing framework...

Experiment

Quantitative Results

Quantitative Example

Quantitative Comparison on the EMU-Edit test-set...

Qualitative Results

Qualitative Chart

Qualitative comparison...

Ablation Studies

Ablation Result 1
Qualitative comparison...
Ablation Result 2
Effect of λI...

Extensions

Community Model Example 1
Cross-architecture compatibility...
Community Model Example 2
Showcases...

Conclusion

We proposed a description-based method...

BibTeX

@InProceedings{Ci_2025_ICCV,
    author    = {Ci, En and Guan, Shanyan and Ge, Yanhao and Zhang, Yilin and Li, Wei and Zhang, Zhenyu and Yang, Jian and Tai, Ying},
    title     = {Describe, Don't Dictate: Semantic Image Editing with Natural Language Intent},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2025},
    pages     = {19185-19194}
}