PhotoAgent: Agentic Photo Editing with Exploratory Visual Aesthetic Planning
Mingde Yao, Zhiyuan You, King-Man Tam, Menglu Wang, Tianfan Xue · 2026
Read PhotoAgent: Agentic Photo Editing with Exploratory Visual Aesthetic Planning by Mingde Yao, Zhiyuan You, King-Man Tam, Menglu Wang, Tianfan Xue (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-22809.txt.
Description
Mingde Yao and four co-authors let users skip step-by-step photo-editing instructions. Editing quality today hinges on carefully crafted prompts, dumping task breakdown and sequencing on the user. PhotoAgent instead treats autonomous editing as long-horizon decision making.
It reasons over aesthetic intent, plans multi-step edits through tree search, and refines results in a closed loop with memory and visual feedback, needing no per-step prompts. For reliable real-world evaluation, the team built UGC-Edit, a benchmark of 7,000 photos with a learned aesthetic reward model, plus a test set of 1,017 photos. Experiments show PhotoAgent consistently improves both instruction adherence and visual quality over baselines.
Catalog record
| CREATOR | Mingde Yao, Zhiyuan You, King-Man Tam, Menglu Wang, Tianfan Xue |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-22809