Research map / Video / Video generation
Temporally Stable Generative Illumination with a One-Step Diffusion Model
Harish Anand, Alexandr Kuznetsov, Sungye Kim, Wojciech Uss, Wojciech Kaliński, Rama Harihara
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 albedo demodulation and modulation (ablation) | Ours · Staircase | CGVQM: 60.94 -> 69.80 | Improved |
| add albedo demodulation and modulation (ablation) | Ours · White Room | CGVQM: 75.18 -> 80.36 | Improved |
| add albedo demodulation and modulation (ablation) | Ours · Shadow Room | CGVQM: 87.66 -> 87.23 | No clear effect |
| add Temporal VAE (TVAE) instead of SD-VAE (ablation) | Ours · Kitchen | CGVQM: 74.74 -> 75.97 | Improved |
| add Temporal VAE (TVAE) instead of SD-VAE (ablation) | Ours · White Room | CGVQM: 80.60 -> 80.36 | No clear effect |
| inject conditions via Adapters instead of concatenated VAE latents (ablation) | Ours · Synthetic test scenes | reflection quality (visual) | Improved |
| encode noisy GI through the adapter instead of the VAE encoder (ablation) | Ours · Synthetic test scenes | glossy reflections (visual) | Improved |
| add direct lighting as the initial hint (ablation) | Ours · Synthetic test scenes | detail in geometry and illumination (visual) | Improved |
| use two adapters instead of one (ablation) | Ours · Synthetic test scenes | indirect lighting (visual) | Improved |
| use optical flow instead of motion vectors in TVAE (ablation) | Ours · Real-world data (real6) | FovVDP: 7.5901 -> 7.1617 | Hurt |
| use low-quality intrinsics instead of high-quality (ablation) | Ours-C · Real-world data (real1) | PSNR: 25.64 -> 25.85 | No clear effect |
| single-step diffusion instead of multi-step or video diffusion (ablation) | Ours · 512x512 runtime | Runtime (s): 52.09 -> 0.29 | Improved |
Closest papers by meaning
- IllumiCraft: Unified Geometry and Illumination Diffusion for Controllable Video Generation (NeurIPS 2025 · similarity 0.89)
- Diffusion Renderer: Neural Inverse and Forward Rendering with Video Diffusion Models (CVPR 2025 · similarity 0.88)
- UniRelight: Learning Joint Decomposition and Synthesis for Video Relighting (NeurIPS 2025 · similarity 0.88)
- TC-Light: Temporally Coherent Generative Rendering for Realistic World Transfer (NeurIPS 2025 · similarity 0.87)
- LightIt: Illumination Modeling and Control for Diffusion Models (CVPR 2024 · similarity 0.87)
- Generative Rendering: Controllable 4D-Guided Video Generation with 2D Diffusion Models (CVPR 2024 · 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.
