PixelART transforms images into editable layers without heavy pretrained models
PixelART: Image-to-Layer Decomposition without Latents or Text-to-Image Pretraining
Computer Vision and Pattern Recognition
Summary
Editing specific parts of an image is easier if the image is split into layers with transparency. Existing tools to do this rely on large pretrained models or complex components, which can be slow and resource-heavy. The authors introduce PixelART, a new method that learns from scratch to break down images into layers using a simpler approach. It works faster and uses less memory while still accurately separating image layers. This method avoids relying on big image-generation models and focuses on improving how layers are assigned during training.
What this means in practice
- •For graphic designers: Generate editable layers from flat design images quickly and efficiently for precise element-level editing without heavy hardware requirements.
- •For game developers: Convert flat artwork into layered assets for flexible in-game editing and manipulation while reducing latency and memory use.
Authors
Zelin Jia, Zhao Zhang, Zhicong Tang, Yuhui Yuan, Shixia Liu
Abstract
Image-to-layer decomposition converts a flattened image into editable RGBA layers, enabling element-level editing in design workflows. Existing diffusion-based systems typically adapt large pretrained text-to-image (T2I) models and introduce RGBA autoencoders or variable-layer architectural modules. We revisit this design choice and ask whether layer decomposition truly requires these heavyweight components. We introduce PixelART, a pixel-space rectified-flow Transformer trained from scratch for image-to-layer (I2L) decomposition. PixelART directly denoises regional RGBA pixel patches using a single-stream multi-modal diffusion Transformer, avoiding RGBA-VAEs, pretrained T2I backbones, and layer-specific decoders. We identify a key property of the task: high-noise timesteps determine layer assignment and coarse layer organization, while low-noise timesteps mainly refine color, alpha, texture, and boundaries. Based on this observation, we propose a terminal-boosted timestep sampling strategy to increase training coverage in the high-noise layer assignment regime. Trained on 4M multi-layer design templates, PixelART achieves state-of-the-art layer decomposition and composite reconstruction on Design-Multi-Layer-Bench and LICA with over 80% fewer parameters, 98% lower latency, and 85% lower memory than the recent Qwen-Image-Layered model. Ablation experiments show that pixel-space $\mathbf{x}$-prediction, terminal-boosted timestep sampling, and data/model scaling are critical, while T2I pretraining brings marginal benefits to the I2L task.