Research map / Image generation / Editing and style transfer
Qwen-Image-Layered: Towards Inherent Editability via Layer Decomposition
Shengming Yin, Zekai Zhang, Zecheng Tang, Kaiyuan Gao, Xiao Xu, Kun Yan, Jiahao Li, Yilei Chen, Yuxiang Chen, Heung-Yeung Shum, Lionel M. Ni, Junyang Lin and 1 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Layer3D RoPE (vs standard 2D RoPE) (ablation) | Qwen-Image-Layered-I2L · Crello (max layer merge 0) | Alpha soft IoU: 0.3725 -> 0.5844 | Improved |
| add RGBA-VAE encoding for the input image (vs RGB VAE) (ablation) | Qwen-Image-Layered-I2L · Crello (max layer merge 0) | Alpha soft IoU: 0.5844 -> 0.6504 | Improved |
| add RGBA-VAE encoding for the input image (vs RGB VAE) (ablation) | Qwen-Image-Layered-I2L · Crello (max layer merge 0) | RGB L1: 0.1894 -> 0.1649 | Improved |
| add multi-stage training (vs init directly from text-to-image weights) (ablation) | Qwen-Image-Layered-I2L · Crello (max layer merge 0) | Alpha soft IoU: 0.6504 -> 0.8705 | Improved |
| add multi-stage training (vs init directly from text-to-image weights) (ablation) | Qwen-Image-Layered-I2L · Crello (max layer merge 0) | RGB L1: 0.1649 -> 0.0594 | Improved |
| use end-to-end Qwen-Image-Layered-I2L instead of LayerD | image decomposition · Crello (max layer merge 0) | Alpha soft IoU: 0.7520 -> 0.8705 | Improved |
| use end-to-end Qwen-Image-Layered-I2L instead of LayerD | image decomposition · Crello (max layer merge 5) | RGB L1: 0.0396 -> 0.0363 | Improved |
| use RGBA-VAE instead of AlphaVAE (FLUX) | RGBA reconstruction · AIM-500 | PSNR: 36.9439 -> 38.8252 | Improved |
| use RGBA-VAE instead of AlphaVAE (FLUX) | RGBA reconstruction · AIM-500 | rFID: 11.7884 -> 5.3132 | Improved |
| fine-tune on Crello training set before evaluation | Qwen-Image-Layered-I2L · Crello | Alpha soft IoU | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LayerD — LayerD: Decomposing Raster Graphic Designs into Layers (ICCV 2025) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Generative Image Layer Decomposition with Visual Effects (CVPR 2025 · similarity 0.89)
- PixelART: Image-to-Layer Decomposition without Latents or Text-to-Image Pretraining (NeurIPS 2026 · similarity 0.88)
- Decomposition of Graphic Design with Unified Multimodal Model (ICML 2025 · similarity 0.87)
- FluxSpace: Disentangled Semantic Editing in Rectified Flow Models (CVPR 2025 · similarity 0.87)
- Stable Flow: Vital Layers for Training-Free Image Editing (CVPR 2025 · similarity 0.86)
- EditInfinity: Image Editing with Binary-Quantized Generative Models (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
