Research map / Video / Video generation
ViBe: Ultra-High-Resolution Video Synthesis Born from Pure Images
Yunfeng Wu, Hongying Cheng, Zihao He, Songhua 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 Relay LoRA (low-res image stage, then high-res stage) (ablation) | ViBe · VBench 3840x2176 | overall score: 66.0% -> 74.4% | Improved |
| add Relay LoRA (ablation) | ViBe · VBench 3840x2176 | imaging quality: 37.4% -> 66.1% | Improved |
| add GCLFA attention (ablation) | ViBe · VBench 3840x2176 | overall score: 69.1% -> 74.4% | Improved |
| add GCLFA attention (ablation) | ViBe · VBench 3840x2176 | aesthetic quality: 52.2% -> 61.4% | Improved |
| add HFATO (degradation plus x0 reconstruction) (ablation) | ViBe · VBench 3840x2176 | overall score: 70.8% -> 74.4% | Improved |
| add x0 reconstruction loss on top of latent degradation (ablation) | ViBe · VBench 3840x2176 | overall score: 72.3% -> 74.4% | Improved |
| downsample ratio 1/4 instead of 1/2 in GCLFA (ablation) | ViBe · VBench 3840x2176 | overall score: 74.4% -> 74.2% | No clear effect |
| use ViBe versus training on real 4K video (T3-Video) | ViBe · VBench 3840x2176 | overall score: 72.8% -> 74.4% | Improved |
| use ViBe at 1080P versus Wan2.2 native | Wan2.2 · VBench 1920x1088 | overall score: 72.5% -> 74.3% | Improved |
| add ViBe to Wan2.2 at 4K | Wan2.2 · VBench 3840x2176 | imaging quality: 33.9% -> 66.1% | Improved |
| replace global coarse branch with local attention (ablation) | GCLFA · qualitative | visual quality | Hurt |
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.
- Upscale-A-Video — Upscale-A-Video: Temporal-Consistent Diffusion Model for Real-World Video Super-Resolution (CVPR 2024) · better on 4 of 4 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
- Hierarchical Patch Diffusion Models for High-Resolution Video Generation (CVPR 2024 · similarity 0.89)
- BIVDiff: A Training-Free Framework for General-Purpose Video Synthesis via Bridging Image and Video Diffusion Models (CVPR 2024 · similarity 0.89)
- One-Step Diffusion for Detail-Rich and Temporally Consistent Video Super-Resolution (NeurIPS 2025 · similarity 0.88)
- Towards Precise Scaling Laws for Video Diffusion Transformers (CVPR 2025 · similarity 0.88)
- Upscale-A-Video: Temporal-Consistent Diffusion Model for Real-World Video Super-Resolution (CVPR 2024 · similarity 0.88)
- Learning Spatial Adaptation and Temporal Coherence in Diffusion Models for Video Super-Resolution (CVPR 2024 · similarity 0.88)
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.
