PosterCopilot: Toward Layout Reasoning and Controllable Editing for Professional Graphic Design
Jiazhe Wei, Ken Li, Tianyu Lao, Haofan Wang, Liang Wang, Caifeng Shan, Chenyang Si
2025-12-04
Summary
This paper introduces PosterCopilot, a new system designed to help computers create professional-quality graphic designs, like posters, automatically.
What's the problem?
Current AI systems that try to do graphic design aren't very good at getting the details right. They often mess up the placement of elements, making things look geometrically incorrect, and they don't allow designers to easily make specific changes to individual parts of the design – like moving one image without affecting everything else – the way professional designers need to.
What's the solution?
The researchers developed PosterCopilot using a three-step training process for the AI. First, they showed it lots of examples and slightly messed them up to help it understand shapes and positioning. Then, they used a reward system to teach it to make designs that look realistic. Finally, they had people give feedback on the designs, which the AI used to improve its aesthetic sense. They also combined this AI with other tools that let users edit individual layers of the design, making it much more controllable.
Why it matters?
This work is important because it brings AI closer to being a truly useful tool for graphic designers. It allows for more accurate and visually appealing designs, and gives designers a level of control over the process that wasn't possible before, potentially speeding up workflows and opening up new creative possibilities.
Abstract
Graphic design forms the cornerstone of modern visual communication, serving as a vital medium for promoting cultural and commercial events. Recent advances have explored automating this process using Large Multimodal Models (LMMs), yet existing methods often produce geometrically inaccurate layouts and lack the iterative, layer-specific editing required in professional workflows. To address these limitations, we present PosterCopilot, a framework that advances layout reasoning and controllable editing for professional graphic design. Specifically, we introduce a progressive three-stage training strategy that equips LMMs with geometric understanding and aesthetic reasoning for layout design, consisting of Perturbed Supervised Fine-Tuning, Reinforcement Learning for Visual-Reality Alignment, and Reinforcement Learning from Aesthetic Feedback. Furthermore, we develop a complete workflow that couples the trained LMM-based design model with generative models, enabling layer-controllable, iterative editing for precise element refinement while maintaining global visual consistency. Extensive experiments demonstrate that PosterCopilot achieves geometrically accurate and aesthetically superior layouts, offering unprecedented controllability for professional iterative design.