Research map / Video / Video generation
Mobile Video Diffusion
Haitam Ben Yahia, Denis Korzhenkov, Ioannis Lelekas, Amir Ghodrati, Amirhossein Habibian
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 |
|---|---|---|---|
| low-resolution finetuning of SVD at 512x256 (ablation) | SVD · UCF-101 7 FPS | FVD: 476 -> 196 | Improved |
| add one-step adversarial finetuning (ablation) | SVD at 512x256 · UCF-101 7 FPS | FVD: 196 -> 168 | Improved |
| remove no-op query/key similarity in cross-attention (ablation) | SVD UNet · Xiaomi-14 Pro | phone latency: OOM -> 3630 ms | Improved |
| use temporal multiscaling (ablation) | MobileVD UNet · UCF-101 7 FPS | phone latency: 3630 -> 2590 ms; FVD 168 -> 156 | Improved |
| use spatial instead of temporal multiscaling (ablation) | MobileVD UNet · UCF-101 | FVD: 145 vs 138 FVD | Hurt |
| combine spatial and temporal multiscaling (ablation) | MobileVD UNet · UCF-101 | FVD: 133 -> 163 | Hurt |
| add learned temporal block pruning (70%) (ablation) | MobileVD · UCF-101 7 FPS | phone latency: 2590 -> 2100 ms; FVD 156 -> 150 | Improved |
| prune temporal blocks to 80% instead of 70% (ablation) | MobileVD · UCF-101 | FVD: 127 -> 245 | Hurt |
| use proposed learned pruning instead of L1 regularization (ablation) | temporal block pruning at 70% · UCF-101 | FVD: 207 -> 127 | Improved |
| add channel funnels (fun-factor 0.5) (ablation) | MobileVD · UCF-101 | phone latency: 2100 -> 1780 ms; FVD 150 -> 171 | Improved |
| use coupled singular initialization instead of He init for funnels (ablation) | channel funnels · UCF-101 | FVD: 332 -> 132 | Improved |
| reduce fun-factor to 0.25 (ablation) | channel funnels · UCF-101 | FVD: 133 -> 155 | Hurt |
Closest papers by meaning
- MobileWan: Closing the Quality Gap for Mobile Video Diffusion (NeurIPS 2026 · similarity 0.91)
- OSV: One Step is Enough for High-Quality Image to Video Generation (CVPR 2025 · similarity 0.89)
- SF-V: Single Forward Video Generation Model (NeurIPS 2024 · similarity 0.88)
- SnapGen-V: Generating a Five-Second Video within Five Seconds on a Mobile Device (CVPR 2025 · similarity 0.88)
- Efficient Video Diffusion Models via Content-Frame Motion-Latent Decomposition (ICLR 2024 · similarity 0.88)
- Identifying and Solving Conditional Image Leakage in Image-to-Video Diffusion Model (NeurIPS 2024 · similarity 0.87)
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.
