Research map / Multi-agent systems and games / LLM agents
PSBench: Editing Image via GUI Agents in Photoshop
Yinuo Zhang, Zian Cheng, Ziya Zhao, Zongyu Li, Bingshuo Liu, Qingbin Liu, Junxian Cai, Xi Chen, Zhiying Tu, Dianhui Chu, Xiaoyan Yu, Dianbo Sui
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces PSBench, a benchmark for GUI agents performing image editing tasks inside Adobe Photoshop, targeting multi-step, layer-based professional workflows.
Problem
Existing GUI agent benchmarks and methods focus on web interfaces and short, low-complexity tasks, and do not model the multi-level menus, fine-grained operations, and layer-based non-destructive editing required by professional image editing software like Photoshop.
Approach
A Photoshop-specific GUI agent benchmark that emphasizes long-horizon, multi-step professional editing with layer-based operations and semantic understanding, rather than web-based short tasks.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use hierarchical agentic framework (GPT-5 planning plus UI execution) instead of a single general MLLM | GPT-4o / Agent S3 · PSBench hard tasks | success rate (%): 18.09 on hard tasks for Agent S3 | Improved |
| use GUI-specialized models instead of general-purpose MLLMs | general-purpose MLLMs · PSBench non-layer-related tasks | success rate (%): 17.46 -> 42.06 | Improved |
| increase task difficulty from easy to hard | GPT-4o · PSBench | success rate (%): easy LR 16.82, hard LR 0.00 | Hurt |
| layer-related instead of non-layer-related tasks | GPT-4o · PSBench overall | success rate (%): 17.46 -> 3.80 | Hurt |
| use specialized GUI models instead of general MLLMs, effect on NDEC | UI-TARS-2-2509 / OpenCUA-7B · PSBench | NDEC (%): 63.22 and 61.78 vs 78.24 for GPT-4o | Hurt |
| increase task difficulty from easy to hard | MLLMs · PSBench | NDEC (%): 95.83 -> 50-67 | Hurt |
| GUI assistant mode (GPT-4o gives instructions, human executes) instead of autonomous GPT-4o agent | GPT-4o · PSBench easy tasks (human-in-loop) | success rate (%): 17.5 -> 100 | Improved |
| GUI assistant mode instead of autonomous GPT-4o agent | GPT-4o · PSBench hard tasks (human-in-loop) | success rate (%): 0 -> 75 | Improved |
| GUI assistant mode instead of unassisted novice | novice user · PSBench easy tasks (human-in-loop) | success rate (%): 85 -> 100 | Improved |
| GUI assistant mode instead of unassisted novice | novice user · PSBench medium tasks (human-in-loop) | success rate (%): 45 -> 85 | Improved |
| GUI assistant mode instead of internet access | novice user · PSBench hard tasks (human-in-loop) | completion time (s): 720 -> 330 | Improved |
| GUI assistant mode instead of unassisted novice | novice user · PSBench easy tasks (human-in-loop) | completion time (s): 180 -> 45 | Improved |
Experimental setup
Benchmarks: PSBench
Closest papers by meaning
- MagicQuill: An Intelligent Interactive Image Editing System (CVPR 2025 · similarity 0.85)
- JarvisArt: Liberating Human Artistic Creativity via an Intelligent Photo Retouching Agent (NeurIPS 2025 · similarity 0.85)
- Envisioning Beyond the Pixels: Benchmarking Reasoning-Informed Visual Editing (NeurIPS 2025 · similarity 0.85)
- ImgEdit: A Unified Image Editing Dataset and Benchmark (NeurIPS 2025 · similarity 0.84)
- UI-Vision: A Desktop-centric GUI Benchmark for Visual Perception and Interaction (ICML 2025 · similarity 0.84)
- RetouchIQ: MLLM Agents for Instruction-Based Image Retouching with Generalist Reward (CVPR 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
