Research map / Image generation / Text-to-image diffusion
Masked Region Transformer for Layered Image Generation and Editing at Scale
Zhicong Tang, Jingye Chen, Zhao Zhang, Mohan Zhou, Yuchi Liu, Yifan Pu, Yalong Bai, Ethan Smith, Yuhui Yuan
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 |
|---|---|---|---|
| scale backbone from FLUX.1 dev 13B to Qwen-Image 20B (ablation) | MRT text-to-layers · 0.5M subset | FID_merged: 17.79 -> 16.15 | Improved |
| scale training data from 0.5M to 10M samples (ablation) | MRT text-to-layers · T2L | FID_merged: 16.15 -> 15.63 | Improved |
| add overflow data and canvas layer (ablation) | MRT text-to-layers · T2L 0.5M subset | FID_merged: 15.68 -> 16.15 FID; PSNR 21.81 -> 22.75; SSIM 0.8543 -> 0.8711 | Mixed |
| add image-to-layers task to joint training (80/20/0) (ablation) | MRT · T2L metrics | FID_merged: 16.15 -> 15.68 | Improved |
| add layers-to-layers task to joint training (70/15/15) (ablation) | MRT · T2L metrics | FID_merged: 15.68 -> 17.06 | Hurt |
| add global text caption condition to image-to-layers (ablation) | MRT image-to-layers · Design-Multi-Layer-Bench | PSNR_layer: 26.03 -> 27.24 | Improved |
| add layer grouping augmentation (ablation) | MRT image-to-layers · Design-Multi-Layer-Bench | PSNR_merged: 21.65 -> 21.97 | Improved |
| add layer grouping augmentation (ablation) | MRT image-to-layers · Design-Multi-Layer-Bench | PSNR_layer: 27.24 -> 26.96 | No clear effect |
| distill with DMD2 from 50 to 16 steps (ablation) | MRT text-to-layers · T2L | FID_merged: 16.02 -> 16.21 | No clear effect |
| distill with DMD2 from 50 to 8 steps (ablation) | MRT text-to-layers · T2L | FID_merged: 16.02 -> 18.58 | Hurt |
| train on mixed short and long captions instead of one type (ablation) | MRT text-to-layers · T2L | FID_merged: short-only on long eval 18.56 -> mixed 15.93 | Improved |
| use MRT instead of Qwen-Image-Layered for image-to-layers | image-to-layers · OOD 100 designs, 16-32 layers | PSNR_merged: 22.1828 -> 25.7229 | Improved |
Closest papers by meaning
- ART: Anonymous Region Transformer for Variable Multi-Layer Transparent Image Generation (CVPR 2025 · similarity 0.91)
- Mask^2DiT: Dual Mask-based Diffusion Transformer for Multi-Scene Long Video Generation (CVPR 2025 · similarity 0.89)
- LayerCraft: Enhancing Text-to-Image Generation with CoT Reasoning and Layered Object Integration (NeurIPS 2025 · similarity 0.89)
- Click2Mask: Local Editing with Dynamic Mask Generation (AAAI 2025 · similarity 0.89)
- Stable Flow: Vital Layers for Training-Free Image Editing (CVPR 2025 · similarity 0.88)
- Generative Image Layer Decomposition with Visual Effects (CVPR 2025 · similarity 0.88)
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.
