Research map / Video / Video generation
FlowPortal: Residual-Corrected Flow for Training-Free Video Relighting and Background Replacement
Wenshuo Gao, Junyi Fan, Jiangyue Zeng, Shuai Yang
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 Residual-Corrected Flow (ablation) | FlowPortal · Video relighting test set (69 clips) | structural consistency: 0.8516 -> 0.8804 | Improved |
| add Residual-Corrected Flow (ablation) | FlowPortal · Video relighting test set (69 clips) | identity consistency: 0.4153 -> 0.7328 | Improved |
| add Residual-Corrected Flow (ablation) | FlowPortal · Video relighting test set (69 clips) | CLIP-T: 0.3310 vs 0.3271 | No clear effect |
| add High-Frequency Transfer (ablation) | FlowPortal · Video relighting test set (69 clips) | identity consistency: 0.5527 -> 0.7328 | Improved |
| add High-Frequency Transfer (ablation) | FlowPortal · Video relighting test set (69 clips) | detail consistency: 40.4852 -> 41.2044 | Improved |
| add foreground masking mechanism (ablation) | FlowPortal · Video relighting test set (69 clips) | CLIP-T: 0.2809 -> 0.3271 | Improved |
| add foreground masking mechanism (ablation) | FlowPortal · Video relighting test set (69 clips) | structural consistency: 0.8649 -> 0.8804 | Improved |
| use FlowPortal vs Lumen (training-based) | FlowPortal · Video relighting test set (69 clips) | identity consistency: 0.7392 vs 0.7328 | Hurt |
| use FlowPortal vs AnyPortal/Light-A-Video | FlowPortal · Video relighting test set (69 clips) | CLIP-T: 0.3196 -> 0.3271 | Improved |
| combine text, structure and reference frame conditions (Decoupled Condition Design) (ablation) | FlowPortal · Video relighting test set (69 clips) | qualitative | Improved |
| reuse consistency residual velocity every r=10 steps (ablation) | FlowPortal · Video relighting test set (69 clips) | running time / quality: T=50 -> 55 steps | No clear effect |
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.
- AnyPortal — AnyPortal: Zero-Shot Consistent Video Background Replacement (ICCV 2025) · better on 6 of 6 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
- UniLumos: Fast and Unified Image and Video Relighting with Physics-Plausible Feedback (NeurIPS 2025 · similarity 0.89)
- UniRelight: Learning Joint Decomposition and Synthesis for Video Relighting (NeurIPS 2025 · similarity 0.87)
- Lux Post Facto: Learning Portrait Performance Relighting with Conditional Video Diffusion and a Hybrid Dataset (CVPR 2025 · similarity 0.86)
- IllumiCraft: Unified Geometry and Illumination Diffusion for Controllable Video Generation (NeurIPS 2025 · similarity 0.86)
- DreamLight: Towards Harmonious and Consistent Image Relighting (NeurIPS 2025 · similarity 0.85)
- SyncLight: Single-Edit Multi-View Relighting (NeurIPS 2026 · 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.
