Paper 2607.25565
ReDesign: Recovering Editable Design Structures from Images via Agentic Decomposition
- Published
- Jul 2026
- Research lab
- Independent
- Citations
- 1
- GitHub
- 177 stars
01 In brief
Summary
ReDesign is an agentic framework that recovers editable design structures (e.g., Figma files) from raster images by growing a layer hierarchy through tool composition.
A VLM controller selects actions from a fixed set (text extraction, multi-layer decomposition, connected component labeling, detection/segmentation, vectorization) to expand nodes, with graceful verification at each step providing accept, prune, or retry feedback to prevent error accumulation.
The method enables parallel tree expansion, achieving up to 7.1x speedup over serial tool use.
The authors introduce the Figma Edit Replay Benchmark with 909 raw Figma files and 14,796 edit instructions.
ReDesign outperforms baselines (VTracer, LayerD, Qwen-Image-Layered, Tool Agent) in editability across layout, color, and text edits, and achieves strong visual fidelity on both Figma and Crello datasets.
Key results include best SSIM after edit replay, highest text recall, and superior reconstruction metrics (e.g., PQ 45.37 on Figma).
The system supports user-in-the-loop granularity control and is robust to different VLM backbones.
02 From the paper
Abstract
Recovering an editable design file from a raster image is a common and costly bottleneck in modern design workflows, yet remains challenging since editability depends on recovering multi-modal attributes, such as typography, vector geometry, colors, grouping, and layer ordering. We present ReDesign, an agentic framework that grows an editable layer hierarchy by selecting and composing specialized tools across modalities. To keep this long decision process reliable despite imperfect tool outputs, we introduce graceful verification at each expansion, which provides local accept, prune, or retry feedback that prevents error accumulation and avoids large scale reruns. To evaluate editability at scale, we introduce the Figma Edit Replay Benchmark, consisting of 909 raw Figma files and 14,796 controlled edit instructions that replay edits on reconstructed outputs. Across this benchmark and standard reconstruction metrics, ReDesign achieves strong visual fidelity while delivering the highest editability across layout, color, and text edits, outperforming layered decomposition baselines and serial tool use pipelines.