Summary
PSDesigner is an automated graphic design system from researchers including Xincheng Shuai, Song Tang, Yutong Huang, Henghui Ding, and Dacheng Tao (arXiv:2603.25738) that emulates the creative workflow of professional human designers. Graphic design is central to e-commerce and advertising, but existing systems built on text-to-image models and multimodal large language models tend to oversimplify professional workflows, limiting flexibility and intuitiveness. PSDesigner addresses this by collecting theme-related assets based on user instructions and autonomously inferring and executing tool calls to manipulate editable design files—for example, integrating new assets or refining low-quality elements. To equip the system with robust tool-use capability, the authors built CreativePSD, a dataset of high-quality PSD design files annotated with operation trajectories covering diverse design scenarios and artistic styles. Experiments show PSDesigner outperforms existing methods across various graphic design tasks, enabling non-professionals to easily create production-quality designs.
Paper Overview
- Field: Computer Vision (CV)
- Authors: Xincheng Shuai, Song Tang, Yutong Huang, Henghui Ding, Dacheng Tao
- Published: 2026-03-26
- arXiv: 2603.25738
Abstract
Graphic design is a creative and innovative process that plays a crucial role in applications such as e-commerce and advertising. However, developing an automated design system that can faithfully translate user intentions into editable design files remains an open challenge.
Although recent studies have leveraged powerful text-to-image models and multimodal large language models (MLLMs) to assist graphic design, they typically simplify professional workflows, resulting in limited flexibility and intuitiveness.
To address these limitations, the authors propose PSDesigner, an automated graphic design system that emulates the creative workflow of human designers. Key aspects:
- Built upon multiple specialized components
- Collects theme-related assets based on user instructions
- Autonomously infers and executes tool calls to manipulate design files, such as integrating new assets or optimizing low-quality elements
To enable strong tool-use capability, the team constructed
CreativePSD, a design dataset containing a large collection of high-quality PSD design files annotated with operation trajectories that cover a wide range of design scenarios and artistic styles, allowing the model to learn professional design workflows.
Extensive experiments demonstrate that PSDesigner outperforms existing methods across various graphic design tasks, enabling non-professionals to conveniently create production-quality designs.
---
*Source: arXiv preprint 2603.25738*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/177169357