Does Synthetic Layered Design Data Benefit Layered Design Decomposition?

Hugging Face Daily Papers Papers
synthetic-data layer-decomposition graphic-design data-centric research

Summary

Investigates whether synthetic layered data can improve graphic design decomposition, finding that synthetic data outperforms non-scalable datasets and enables balanced layer-count distributions.

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.
Original Article
View Cached Full Text

Cached at: 05/15/26, 04:23 AM

Paper page - Does Synthetic Layered Design Data Benefit Layered Design Decomposition?

Source: https://huggingface.co/papers/2605.15167

Abstract

Synthetic layered image data improves graphic design decomposition by enabling scalable training and better layer distribution control compared to traditional methods.

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 partiallysynthetic datafrom 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 onCLD baseline, which is a state-of-the-artlayer decompositionframework. Based on the baseline, we construct our ownsynthetic dataset, SynLayers, generate textual supervision usingvision language models, and automate inference inputs with VLM-predicted bounding boxes. Our study reveals three key findings: (1) even training with purelysynthetic datacan 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 dataenables 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 ofsynthetic dataas a practical foundation forlayered design editingsystems.

View arXiv pageView PDFGitHub3Add to collection

Models citing this paper1

#### SynLayers/Bbox-caption-8b Image-Text-to-Text• 9B• Updatedabout 2 hours ago • 530 • 4

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2605.15167 in a dataset README.md to link it from this page.

Spaces citing this paper1

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

An Information-Theoretic Criterion for Efficient Data Synthesis

arXiv cs.LG

This paper provides an information-theoretic account of when synthetic data improves or degrades LLM training, distinguishing between information-open and information-closed generation loops and explaining collapse via the data processing inequality.

ReDesign: Recovering Editable Design Structures from Images via Agentic Decomposition

Hugging Face Daily Papers

ReDesign is an agentic framework that recovers editable layer hierarchies from raster images by selecting and composing specialized tools across modalities, introducing graceful verification to prevent error accumulation. It also introduces the FigmaEditReplay Benchmark for evaluating editability at scale, achieving high visual fidelity and superior editability over baselines.