Research map / Video / Video generation
ScrollScape: Unlocking 32K Image Generation With Video Diffusion Priors
Haodong Yu, Yabo Zhang, Donglin Di, Ruyi Zhang, Wangmeng Zuo
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 fine-tuning on the 3K multi-ratio dataset (ablation) | ScrollScape · 8:1 panorama | FID: 272.0 -> 214.7 | Improved |
| add fine-tuning on the 3K multi-ratio dataset (ablation) | ScrollScape · 8:1 panorama | Style-L (x10^-3): 7.6 -> 4.0 | Improved |
| add TAP (training-free decoder anchoring) (ablation) | ScrollScape · 8:1 panorama | FID: 231.6 -> 214.7 | Improved |
| add TAP (ablation) | ScrollScape · 8:1 panorama | GSD (LPIPS): 0.469 -> 0.674 | Improved |
| add ScrollSR (ablation) | ScrollScape · 8:1 panorama | FID: 218.8 -> 214.7 | Improved |
| add ScrollSR (ablation) | ScrollScape · 8:1 panorama | CLIP: 26.5 -> 30.0 | Improved |
| add ScrollSR (ablation) | ScrollScape · 8:1 panorama | GSD (LPIPS): 0.668 -> 0.674 | No clear effect |
| add ScanPE and fine-tuning to vanilla Wan2.1 (ablation) | ScrollScape · 8:1 panorama | GSD (LPIPS): 0.303 -> 0.674 | Improved |
| add ScanPE and fine-tuning to vanilla Wan2.1 (ablation) | ScrollScape · 8:1 panorama | Style-L (x10^-3): 2.0 -> 4.0 | Hurt |
| use ScrollScape instead of DyPE | ScrollScape vs DyPE · 8:1 panorama | FID: 248.1 -> 214.7 | Improved |
| use ScrollScape instead of Tiled Diffusion | ScrollScape vs Tiled Diffusion · 8:1 panorama | KID (x10^-2): 3.0 -> 2.0 | Improved |
| use ScrollScape instead of DyPE | ScrollScape vs DyPE · user study | Structural coherence preference: 92% | Improved |
Closest papers by meaning
- ScaleCrafter: Tuning-free Higher-Resolution Visual Generation with Diffusion Models (ICLR 2024 · similarity 0.87)
- PatchVSR: Breaking Video Diffusion Resolution Limits with Patch-wise Video Super-Resolution (CVPR 2025 · similarity 0.87)
- FrameBridge: Improving Image-to-Video Generation with Bridge Models (ICML 2025 · similarity 0.86)
- Hierarchical Patch Diffusion Models for High-Resolution Video Generation (CVPR 2024 · similarity 0.86)
- LongScape: Advancing Long-Horizon Embodied World Models with Context-Aware MoE (NeurIPS 2026 · similarity 0.86)
- Everything at Every Scale: Scale-Invariant Diffusion with Continuous Super-Resolution (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.
