我们推出 SANA-Video 2.0,这是一个在统一架构下以 5B 和 14B 规模实例化的混合视频扩散 Transformer。SANA-Video 2.0 专为在单张 GPU 上生成高达 720p 的高质量视频而设计,其质量与全 softmax 视频 DiT 相当,同时保留了线性注意力在长序列上的优异扩展能力。为避免全程使用二次复杂度注意力,混合线性-softmax 注意力机制将门控线性注意力(用于 O(N) 主导的混合)与周期性门控 softmax 锚点以 3:1 的比例结合,恢复了纯线性注意力所缺失的全秩 token 交互。为了将这些刷新后的表征沿深度方向传播,块注意力残差(AttnRes)将已完成的块摘要路由到后续线性层中,实现了锚点特征复用,并将深层有效秩提升了约 12%。通过从头开始训练,SANA-Video 2.0 直接学习完整的混合架构,而非对预训练模型进行线性化处理,并通过降分辨率代理研究确定 25% softmax 为质量与效率的最佳平衡点。在 40 步采样下,SANA-Video 2.0 在单张 H100 上以 480p 分辨率、13.2 秒内取得了 84.30 的 VBench 分数,其性能与规模大得多的 softmax 视频 DiT 相当,而延迟却低得多。在 720p/60 秒条件下,其编译后的 DiT 前向传播速度比匹配的全 softmax 基线快 3.2 倍,且这一差距随视频时长增加而扩大。此外,全栈 Sol-Engine 优化(内核融合、缓存和稀疏注意力)进一步将这一硬件友好的骨干网络加速了 3.58 倍,使得 5B 模型在 720p/5 秒条件下仅需 13.06 秒,在单张 H100 上比 Wan 2.2-A14B 快 120 倍。总体而言,我们的混合设计以大幅降低的成本恢复了 softmax 级别的表现力,为实现可扩展的长视频、高分辨率视频生成开辟了道路。
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.