SANA-Video 2.0用混合注意力实现高效高清视频生成，单卡13秒出720p视频。
SANA-Video 2.0: Hybrid Linear Attention with Attention Residuals for Efficient Video Generation
- 混合线性-软注意力机制，降低计算开销同时保持高质量
- 40步采样下480p视频13.2秒生成，性能超大模型且延迟更低
- 适合需要高速生成高清视频的场景，如实时创作与部署
我们提出SANA-Video 2.0，一种在统一架构下以5B和14B规模实例化的视频扩散变换器，可在单张GPU上生成最高720p的高质量视频。该模型通过混合线性-软注意力（Hybrid Linear-Softmax Attention）避免二次复杂度计算，采用3:1比例的周期性门控软注意力锚点，恢复纯线性注意力缺失的完整秩交互。通过块注意力残差（AttnRes）将块汇总信息传递至深层，提升深层有效秩约12%。从头训练使模型直接学习完整混合结构，而非微调预训练模型；通过低分辨率代理研究确定25%软注意力为最优质量-效率平衡点。40步采样下，480p视频生成耗时13.2秒，VBench得分为84.30，性能远超更大规模的软注意力模型但延迟极低。其编译后的前向传播速度比匹配的全软注意力基线快3.2倍，且随视频时长增加优势扩大。结合完整栈的Sol-Engine优化（内核融合、缓存、稀疏注意力），进一步提速3.58倍，5B模型在720p/5秒下仅需13.06秒，较Wan 2.2-A14B快120倍。整体设计在显著降低代价的同时恢复了软注意力的表达能力，实现可扩展的长时高分辨率视频生成。
原文摘要 · Abstract (English)
We introduce SANA-Video 2.0, a hybrid video diffusion transformer instantiated at 5B and 14B scales under a unified architecture. Designed to generate high-quality video up to 720p on a single GPU, SANA-Video 2.0 matches full-softmax video DiTs in quality while retaining the favorable long-sequence scaling of linear attention. To avoid quadratic attention throughout, Hybrid Linear-Softmax Attention combines gated linear attention for O(N)-dominated mixing with periodic gated-softmax anchors at a 3:1 ratio, restoring the full-rank token interactions that pure linear attention lacks. To propagate these refreshed representations across depth, Block Attention Residuals (AttnRes) route completed block summaries into later linear layers, enabling anchor-feature reuse and boosting deep-layer effective rank by ~12%. Through from-scratch training, SANA-Video 2.0 learns the complete hybrid directly rather than linearizing pretrained models, with reduced-resolution proxy studies establishing 25% softmax as the optimal quality-efficiency trade-off. With 40-step sampling, SANA-Video 2.0 achieves a VBench score of 84.30 in 13.2s at 480p on a single H100, remaining competitive with far larger softmax video DiTs at a fraction of the latency. Its compiled DiT forward pass is 3.2x faster than a matched full-softmax baseline at 720p/60s, a gap that expands with video duration. Furthermore, full-stack Sol-Engine optimization (kernel fusion, caching, and sparse attention) accelerates this hardware-friendly backbone by a further 3.58x, bringing the 5B pipeline to 13.06s at 720p/5s and making it 120x faster than Wan 2.2-A14B on one H100. Overall, our hybrid design recovers softmax-level expressiveness at substantially reduced cost, unlocking scalable long, high resolution video generation.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。