Research map / Video / Video generation
Preserve Your Own Correlation: A Noise Prior for Video Diffusion Models
Songwei Ge, Seungjun Nah, Guilin Liu, Tyler Poon, Andrew Tao, Bryan Catanzaro, David Jacobs, Jia-Bin Huang, Ming-Yu Liu, Yogesh Balaji
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 |
|---|---|---|---|
| finetune from a pretrained image diffusion model instead of training the video model from scratch (ablation) | video diffusion on UCF-101 · UCF-101 small-scale | FVD: 903.37 -> 566.67 | Improved |
| replace i.i.d. frame noise with mixed correlated noise during finetuning (ablation) | video diffusion finetuning · UCF-101 small-scale | FVD: 566.67 -> 337.40 | Improved |
| replace i.i.d. frame noise with progressive noise during finetuning (ablation) | video diffusion finetuning · UCF-101 small-scale | IS: 41.25 -> 53.52 | Improved |
| use correlated noise to preserve image model quality (ablation) | video diffusion finetuning · UCF-101 small-scale | frame FID: 56.43 -> 31.57 | Improved |
| use progressive instead of mixed noise (ablation) | PYoCo · UCF-101 small-scale | FVD: 337.40 -> 339.67 | No clear effect |
| set correlation ratio alpha to 0 (i.i.d. noise) (ablation) | PYoCo · UCF-101 small (69M) | visual quality and diversity | Hurt |
| set correlation ratio too large (alpha=10) (ablation) | PYoCo · UCF-101 small (69M) | motion quality | Hurt |
| use ensemble expert denoisers on top of base config (ablation) | PYoCo · MSR-VTT zero-shot | FID: 25.39 -> 22.14 | Improved |
| increase model size (69M to 253M) (ablation) | PYoCo · UCF-101 small-scale | FVD: 332 -> 310 (112M to 253M) | Improved |
| use PYoCo against Make-A-Video for zero-shot generation | PYoCo · UCF-101 zero-shot | Inception Score: 33.45 -> 47.76 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Make-A-Video — Make-A-Video: Text-to-Video Generation without Text-Video Data (ICLR 2023) · better on 3 of 3 shared results
- CCVS — CCVS: Context-aware Controllable Video Synthesis (NeurIPS 2021) · better on 1 of 1 shared result
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
- TRIP: Temporal Residual Learning with Image Noise Prior for Image-to-Video Diffusion Models (CVPR 2024 · similarity 0.91)
- Identifying and Solving Conditional Image Leakage in Image-to-Video Diffusion Model (NeurIPS 2024 · similarity 0.91)
- How I Warped Your Noise: a Temporally-Correlated Noise Prior for Diffusion Models (ICLR 2024 · similarity 0.91)
- VideoGuide: Improving Video Diffusion Models without Training Through a Teacher's Guide (CVPR 2025 · similarity 0.91)
- VideoCrafter2: Overcoming Data Limitations for High-Quality Video Diffusion Models (CVPR 2024 · similarity 0.91)
- Spectral Progressive Diffusion for Efficient Image and Video Generation (NeurIPS 2026 · similarity 0.91)
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.
