Image generation improves training data for graphical interface agents
AutoGUIWorld: Image Generators as Visual World Models for GUI Agent
Computer Vision and Pattern Recognition
Summary
Training computer agents to interact with software GUIs usually needs lots of data showing how those interfaces respond to actions. The authors created a way to make this training data by combining task planners with image generators, so they don’t have to run or install the actual software. This approach generates many example interactions for different operating systems, helping agents learn better how to complete tasks. Testing showed that using this artificial data improved agents’ success at real-world tasks on desktop and scientific software.
What this means in practice
- •For software developers: Create diverse training datasets for agents that interact with different software GUIs without needing to run the apps themselves.
- •For automation engineers: Improve robotic or AI automation of complex multi-step workflows across desktop and scientific applications by using generated training data.
Authors
Cheng Yang, Yifan Wu, Yutao Huang, Zhaohua Zhang, Beiduo Chen, Muxi Chen, Chenchen Zhao, Hexuan Deng, Haolin Yang, Geyuan Zhu, Sa Zhu, Jianhuan Zhuo, Qiuyong Xiao, Jianhao Ruan, Yiran Peng, Jiayi Zhang, Tian Ye, Xinlei Yu, Tianwen Jiang, Jihong Zhang, Yuyu Luo
Abstract
GUI agents require high-quality interaction trajectories to learn how software environments respond to actions, maintain state, and support multi-step workflows. However, the diversity of available trajectories is constrained by the applications, interface states, and workflows accessible in the underlying environments. Expanding this coverage requires deploying increasingly diverse and complex software, with specialized applications imposing additional installation, configuration, and runtime costs. We introduce AutoGUIWorld, a data generation framework that combines the visual priors of image generators with the task knowledge of a planner to synthesize GUI interaction trajectories without deploying or running the corresponding software environments. AutoGUIWorld samples initial GUI scenes from structured specifications of operating-system context, visual appearance, and interface state, and generates tasks conditioned on those scenes. A planner then specifies atomic actions and their intended visual consequences, while an image generator iteratively edits the current screenshot to produce subsequent observations. Action grounding and transition-level quality filtering yield 79,266 spatially annotated step-level training samples across Ubuntu, Windows, macOS, and Chrome. Fine-tuning Qwen3.5-35B-A3B on AutoGUIWorld trajectories improves the mean task score on OSWorld from 33.0% to 40.8% and the task success rate on ScienceBoard from 14.0% to 32.2%. These results show that generated trajectories improve GUI-agent performance on real desktop and scientific tasks.