Research map / Video / Video generation
Minute-Long Videos with Dual Parallelisms
Zeqing Wang, Bowen Zheng, Xingyi Yang, Zhenxiong Tan, Yuecong Xu, Xinchao Wang
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 feature cache (reuse KV of previous block) (ablation) | DualParal · 129 frames, 8x4090 | latency: 182.71 s -> 142.62 s | Improved |
| add feature cache (ablation) | DualParal · 129 frames, 8x4090 | peak memory: 16.76 GB -> 15.52 GB | Improved |
| combine queue with device pipeline instead of queue only (ablation) | DualParal · 129 frames, 8x4090 | latency: 621.37 s -> 182.71 s (without cache) | Improved |
| add queue to device pipeline (ablation) | DualParal · 129 frames | infinite-length generation: device pipeline only: not supported; with queue: supported | Improved |
| add coordinated noise initialization with neighboring blocks (ablation) | DualParal · qualitative frames (Figure 5) | temporal consistency | Improved |
| use DualParal instead of Ring Attention for 1025-frame videos | long video generation · 1025 frames, 8x4090 | latency: 3907.5 s -> 596.9 s | Improved |
| use DualParal instead of FIFO | long video generation · 513 frames, 8x4090 | latency: 565.3 s -> 309.3 s | Improved |
| use DualParal instead of DeepSpeed-Ulysses | video quality · VBench, 129 frames | overall score: 81.85% -> 80.93% | Hurt |
| use DualParal instead of DeepSpeed-Ulysses | video quality · VBench, 257 frames | overall score: 74.05% -> 78.17% | Improved |
Closest papers by meaning
- DistriFusion: Distributed Parallel Inference for High-Resolution Diffusion Models (CVPR 2024 · similarity 0.90)
- AsyncDiff: Parallelizing Diffusion Models by Asynchronous Denoising (NeurIPS 2024 · similarity 0.89)
- Fast and Memory-Efficient Video Diffusion Using Streamlined Inference (NeurIPS 2024 · similarity 0.89)
- PipeFusion: Patch-level Pipeline Parallelism for Diffusion Transformers Inference (NeurIPS 2025 · similarity 0.88)
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.88)
- AsymRnR: Video Diffusion Transformers Acceleration with Asymmetric Reduction and Restoration (ICML 2025 · 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.
