Research map / Image generation / Text-to-image diffusion
Hybrid Layout Control for Diffusion Transformer: Fewer Annotations, Superior Aesthetics
Keming Wu, Junwen Chen, Zhanhao Liang, Yinuo Wang, Ji Li, Chao Zhang, Bin Wang, 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 |
|---|---|---|---|
| add semantic-layout stage on top of anonymous-layout training (hybrid 0.8:0.2) (ablation) | Anonymous layout · LayoutSAM-Eval Spatial | Spatial: 83.78 -> 97.50 | Improved |
| use hybrid 0.8:0.2 compute instead of semantic layout only (ablation) | Layout control · LayoutSAM-Eval Spatial | Spatial: 97.13 -> 97.50 | No clear effect |
| add quality-tuning with 1K semantic layout images (ablation) | Hybrid Layout · LayoutSAM-Eval | HPSv2: 26.99 -> 28.47 | Improved |
| add quality-tuning with 50K anonymous images (ablation) | Hybrid Layout · LayoutSAM-Eval | HPSv2: 26.99 -> 29.30 | Mixed |
| downsample clean regional tokens instead of noisy ones (ablation) | Regional DiT · LayoutSAM-Eval | Spatial: 93.79 -> 97.18 | Improved |
| larger downsample ratio (1 -> 1/2) (ablation) | Regional DiT · LayoutSAM-Eval | Spatial and time: 97.50 -> 97.18; 103.01s -> 72.13s | Mixed |
| share noise map between regional and global tokens (ablation) | Hybrid Layout · LayoutSAM-Eval | ImageReward: 53.22 -> 63.84 | Improved |
| use smaller logit-normal mean (-1.0 instead of 1.5) (ablation) | Anonymous layout · LayoutSAM-Eval Spatial | Spatial: 78.14 -> 83.78 | Improved |
| scale anonymous layout data from 20K to 200K (ablation) | Anonymous layout · LayoutSAM-Eval Spatial | Spatial: 77.00 -> 83.78 | Improved |
| keep regional prompts in SiamLayout (ablation) | SiamLayout · LayoutSAM-Eval Spatial | Spatial: 85.13 -> 92.23 | Improved |
| switch backbone from SD3 to FLUX | Ours · COCO | AP: 34.6 -> 55.2 | Improved |
| hybrid layout vs SiamLayout on FLUX numeracy | Ours w/ FLUX · T2I-CompBench | Numeracy: 54.67 -> 62.74 | Improved |
Closest papers by meaning
- Build-A-Scene: Interactive 3D Layout Control for Diffusion-Based Image Generation (ICLR 2025 · similarity 0.91)
- SSMG: Spatial-Semantic Map Guided Diffusion Model for Free-Form Layout-to-Image Generation (AAAI 2024 · similarity 0.90)
- R&B: Region and Boundary Aware Zero-shot Grounded Text-to-image Generation (ICLR 2024 · similarity 0.90)
- Towards Aligned Layout Generation via Diffusion Model with Aesthetic Constraints (ICLR 2024 · similarity 0.89)
- Adversarial Supervision Makes Layout-to-Image Diffusion Models Thrive (ICLR 2024 · similarity 0.89)
- Control and Realism: Best of Both Worlds in Layout-to-Image without Training (ICML 2025 · similarity 0.89)
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.
