Research map / Video / Video generation
ZoLA: Zero-Shot Creative Long Animation Generation with Short Video Model
Fu-Yun Wang, Zhaoyang Huang, Qiang Ma, Guanglu Song, Xudong LU, Weikang Bian, Yijin Li, Yu Liu, Hongsheng Li
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 spatiotemporal length expansion on top of temporal co-denoising (ablation) | ZoLA · 64-frame ablation | LPIPS: 0.533 -> 0.423 | Improved |
| add noise init strategy (ablation) | ZoLA (LE) · 64-frame ablation | LPIPS: 0.423 -> 0.212 | Improved |
| add noise init strategy (ablation) | ZoLA (LE) · 64-frame ablation | user study motion quality (%): 23.8% -> 18.4% | Hurt |
| add noise travel (ablation) | ZoLA (LE+NI) · 64-frame ablation | user study motion quality (%): 18.4% -> 35.6% | Improved |
| add noise travel (ablation) | ZoLA (LE+NI) · 64-frame ablation | CLIPSIM / LPIPS: CLIPSIM 0.961 -> 0.953; LPIPS 0.212 -> 0.249 | Hurt |
| noise init mask ratio 0.5 instead of 0.0 or 1.0 (ablation) | ZoLA noise init · visual ablation | qualitative consistency and motion | Improved |
| replace Conv-Attn / Gen-L with ZoLA for 256-frame generation | short video models · 256-frame long animation | FVD: 440.3 -> 387.1 vs Gen-L | Improved |
| replace Gen-L with ZoLA | long animation generation · user study consistency | consistency (%): 21.5% -> 68.3% | Improved |
| combine ZoLA with a video consistency model | ZoLA · 32-second animation | time (s): 934.5 -> 82.3 | Improved |
| ZoLA versus Gen-L on memory | ZoLA · 32-second animation | VRAM (GB): 5.1 (Gen-L) -> 11.8 | Hurt |
| ZoLA output versus short-video upper bound | ZoLA · 256-frame animation | CLIP-SIM: 0.9693 -> 0.9443 | Hurt |
Closest papers by meaning
- FreeLong: Training-Free Long Video Generation with SpectralBlend Temporal Attention (NeurIPS 2024 · similarity 0.86)
- IM-Zero: Instance-level Motion Controllable Video Generation in a Zero-shot Manner (CVPR 2025 · similarity 0.86)
- PEEKABOO: Interactive Video Generation via Masked-Diffusion (CVPR 2024 · similarity 0.86)
- LongDiff: Training-Free Long Video Generation in One Go (CVPR 2025 · similarity 0.86)
- VideoPoet: A Large Language Model for Zero-Shot Video Generation (ICML 2024 · similarity 0.86)
- ControlVideo: Training-free Controllable Text-to-video Generation (ICLR 2024 · 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.
