Research map / Video / Video generation
Video Generation Models Are Inherent Lighting Estimators
Ziqi Cai, Shuchen Weng, Kaiqi Liu, Zifeng Wang, Zhiquan Zhang, Minggui Teng, Han Jiang, Boxin Shi
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 HDR-aware VAE and LoRA fine-tuning to the LDR baseline (ablation) | V-LITE · Editable Indoor | AER: 5.45 -> 4.77 | Improved |
| add LoRA fine-tuning after inserting the HDR VAE (ablation) | V-LITE · Editable Indoor | MSE: 3.00 -> 0.10 | Improved |
| add static HDR images (mixed-data training) to video-only training (ablation) | V-LITE · Editable Indoor | AER: 4.82 -> 4.77 | Improved |
| add static HDR images (mixed-data training) to video-only training (ablation) | V-LITE · EnvMapNet | AED: 41.96 -> 41.63 | Improved |
| add static HDR images to video-only training (ablation) | V-LITE · Editable Indoor | MSE / SI-MSE: 0.10 -> 0.10 MSE, 0.03 -> 0.03 SI-MSE | No clear effect |
| V-LITE versus DiffusionLight on 10-video subset | V-LITE · Editable Indoor / EnvMapNet subset | AER, AED: AER 4.58 -> 4.68; AED 40.07 -> 42.53; AS 16.76 -> 15.96 | Mixed |
| V-LITE versus DiffusionLightTurbo on AED | V-LITE · EnvMapNet | AED: 37.74 -> 41.63 | Hurt |
| use V-LITE instead of per-frame StyleLight | lighting estimation · Editable Indoor | MSE: 0.13 -> 0.10 | Improved |
| use video inpainting with chrome ball instead of per-frame image methods and diffusion runtime | V-LITE · Efficiency | seconds: 145,800 -> 80 | Improved |
Closest papers by meaning
- LuxDiT: Lighting Estimation with Video Diffusion Transformer (NeurIPS 2025 · similarity 0.91)
- IllumiCraft: Unified Geometry and Illumination Diffusion for Controllable Video Generation (NeurIPS 2025 · similarity 0.87)
- UniRelight: Learning Joint Decomposition and Synthesis for Video Relighting (NeurIPS 2025 · similarity 0.87)
- DiffusionLight: Light Probes for Free by Painting a Chrome Ball (CVPR 2024 · similarity 0.87)
- Lux Post Facto: Learning Portrait Performance Relighting with Conditional Video Diffusion and a Hybrid Dataset (CVPR 2025 · similarity 0.86)
- Do Image Editing Models Understand Lighting? (NeurIPS 2026 · 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.
