Research map / Image generation / Text-to-image diffusion
LaRender: Training-Free Occlusion Control in Image Generation via Latent Rendering
Xiaohang Zhan, Dingming Liu
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 LaRender latent rendering in place of cross-attention on SDXL | SDXL · T2I-CompBench++ 3D Spatial (val) | UniDet: 0.357 -> 0.416 | Improved |
| add LaRender latent rendering on SDXL | SDXL · RealOcc | User study HPSR: 0.396 ± 0.045 -> 0.767 ± 0.094 | Improved |
| add LaRender latent rendering on SDXL | SDXL · T2I-CompBench++ 3D Spatial (val) | User study AUR: 2.36 ± 0.49 -> 4.94 ± 0.11 | Improved |
| add LaRender latent rendering on SDXL | SDXL · RealOcc | CLIP score: 29.42 -> 28.30 | Hurt |
| add LaRender latent rendering on SDXL | SDXL · T2I-CompBench++ 3D Spatial (val) | CLIP score: 31.12 -> 30.98 | No clear effect |
| add LaRender latent rendering on SDXL | SDXL · T2I-CompBench++ / RealOcc | inference time (s): 7.44 -> 7.46 | No clear effect |
| add inverse-proportional density schedule instead of fixed opaque density (ablation) | LaRender · T2I-CompBench++ 3D Spatial (val) | UniDet / CLIP score | Improved |
| add inverse-proportional density schedule instead of fixed normal density (ablation) | LaRender · T2I-CompBench++ 3D Spatial (val) | UniDet / CLIP score | Improved |
| add cross-attention maps to shape transmittance maps beyond bounding box masks (ablation) | LaRender · T2I-CompBench++ 3D Spatial (val) | occlusion control | Improved |
| use LaRender instead of layout-to-image MIGC | MIGC · T2I-CompBench++ 3D Spatial (val) | User study HPSR: 0.448 ± 0.068 -> 0.767 ± 0.070 | Improved |
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.
- MIGC — MIGC: Multi-Instance Generation Controller for Text-to-Image Synthesis (CVPR 2024) · better on 7 of 8 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
- OccFusion: Rendering Occluded Humans with Generative Diffusion Priors (NeurIPS 2024 · similarity 0.86)
- UMAMI: Unifying Masked Autoregressive Models and Deterministic Rendering for View Synthesis (NeurIPS 2025 · similarity 0.86)
- A Diffusion-Based Framework for Occluded Object Movement (AAAI 2025 · similarity 0.86)
- Denoising Diffusion via Image-Based Rendering (ICLR 2024 · similarity 0.86)
- Global Latent Neural Rendering (CVPR 2024 · similarity 0.86)
- Move Anything with Layered Scene Diffusion (CVPR 2024 · similarity 0.85)
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.
