AI summaryⓘ
The authors explore how purely synthetic layered data can help break down graphic designs into separate parts, like layers, which makes editing easier. They found that in graphic design, elements are often separate enough that exact modeling of layer relationships isn't as necessary as in natural images. Their experiments show that training with synthetic data can work better than some existing real-world datasets and that having more data generally improves performance up to a point. They also demonstrate that synthetic data allows better control over the number of layers, avoiding imbalances seen in real designs. This work suggests synthetic data could be a practical way to improve tools for editing layered graphic designs.
image generationlayer decompositiongraphic designsynthetic datavision language modelsbounding boxesdata-centric studyCLD baselinelayer-count distribution
Authors
Kam Man Wu, Haolin Yang, Qingyu Chen, Yihu Tang, Jingye Chen, Qifeng Chen
Abstract
Recent advances in image generation have made it easy to produce high-quality images. However, these outputs are inherently flattened, entangling foreground elements, background, and text within a fixed canvas. As a result, flexible post-generation editing remains challenging, revealing a clear last-mile gap toward practical usability. Existing approaches either rely on scarce proprietary layered assets or construct partially synthetic data from limited structural priors. However, both strategies face fundamental challenges in scalability. In this work, we investigate whether pure synthetic layered data can improve graphic design decomposition. We make the assumption that, in graphic design, effective decomposition does not require modeling inter-layer dependencies as precisely as in natural-image composition, since design elements are often intentionally arranged as modular and semantically separable components. Concretely, we conduct a data-centric study based on CLD baseline, which is a state-of-the-art layer decomposition framework. Based on the baseline, we construct our own synthetic dataset, SynLayers, generate textual supervision using vision language models, and automate inference inputs with VLM-predicted bounding boxes. Our study reveals three key findings: (1) even training with purely synthetic data can outperform non-scalable alternatives such as the widely used PrismLayersPro dataset, demonstrating its viability as a scalable and effective substitute; (2) performance consistently improves with increased training data scale, while gains begin to saturate at around 50K samples; and (3) synthetic data enables balanced control over layer-count distributions, avoiding the layer-count imbalance commonly observed in real-world datasets. We hope this data-centric study encourages broader adoption of synthetic data as a practical foundation for layered design editing systems.