mlsanityData to 7 Oct 2026

SPADE: An Input-Adaptive Sparse Attention Engine for Fast Video Diffusion Models Inference

arXiv 2608.03335 ↗First posted 4 Aug 2026cs.CVcited by 0

On the map: Diffusion models → Video generation

Abstract (arXiv)

Video diffusion transformers (vDiTs) generate high quality but pay quadratic self-attention cost, making inference prohibitive at video-token scales. The challenge is input-adaptive sparsity: selecting critical Q/K/V tokens with negligible overhead and executing them for end-to-end gains. We present SPADE, a training-free sparse-attention engine of three parts: (i) vDiT-SSR, a specification defining 3D blocking candidates and formalizing dynamic masks via Summarizer/Estimator expressions; (ii) runtime scheme generation using SICS and a head-wise policy; and (iii) an executor with low-overhead index search, flash block-sparse attention, and kernel grouping. Across Hunyuan-Video and Wan 2.1/2.2 for text-to-video and image-to-video generation, SPADE raises sparsity and speed while preserving quality, accelerating attention by 2.26x-3.40x and end-to-end inference by 1.49x-1.80x. Our code is open-sourced at https://github.com/6somehow/DAC-SPADE.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Video generation, one of 14 directions in Diffusion models. Builds on 22 of its 33 references that are in the map. First posted on arXiv 4 Aug 2026.

Builds on

22 of 33 references.

Closest ancestors · 6

Same direction, newest first

  1. 2025Sparse VideoGen2: Accelerate Video Generation with Sparse Attention via Semantic-Aware Permutation24 May
  2. 2025VSA: Faster Video Diffusion with Trainable Sparse Attention19 May
  3. 2025DiTFastAttnV2: Head-wise Attention Compression for Multi-Modality Diffusion Transformers28 Mar
  4. 2025Fast Video Generation with Sliding Tile Attention6 Feb
  5. 2024FasterCache: Training-Free Video Diffusion Model Acceleration with High Quality25 Oct
  6. 2024DiTFastAttn: Attention Compression for Diffusion Transformer Models12 Jun

Other references · 16

  1. 2022FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awarenesscited by 3,611
  2. 2023FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioningcited by 2,551
  3. 2025Wan: Open and Advanced Large-Scale Video Generative Modelscited by 2,536
  4. 2023Efficient Streaming Language Models with Attention Sinkscited by 2,115
  5. 2023VBench: Comprehensive Benchmark Suite for Video Generative Modelscited by 1,755
  6. 2018Group Normalizationcited by 1,514
  7. 2024SnapKV: LLM Knows What You are Looking for Before Generationcited by 848
  8. 2024PyramidKV: Dynamic KV Cache Compression based on Pyramidal Information Funnelingcited by 420
  9. 2025VBench-2.0: Advancing Video Generation Benchmark Suite for Intrinsic Faithfulnesscited by 289
  10. 2024DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Headscited by 283
  11. 2025XAttention: Block Sparse Attention with Antidiagonal Scoringcited by 131
  12. 2024SeerAttention: Learning Intrinsic Sparse Attention in Your LLMscited by 121
4 more
  1. 2025FlexPrefill: A Context-Aware Sparse Attention Mechanism for Efficient Long-Sequence Inferencecited by 121
  2. 2024DistriFusion: Distributed Parallel Inference for High-Resolution Diffusion Modelscited by 98
  3. 2024ThunderKittens: Simple, Fast, and Adorable AI Kernelscited by 34
  4. 2024FlashMask: Efficient and Rich Mask Extension of FlashAttentioncited by 18

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Sparse-vDiT: Unleashing the Power of Sparse Attention to Accelerate Video Diffusion Transformers
  2. 2026SALAD: Achieve High-Sparsity Attention via Efficient Linear Attention Tuning for Video Diffusion Transformer
  3. 2026Sol-Attn: Accelerating Video Generation Inference via On-the-Fly Attention Sparsification
  4. 2026LVSA: Training-Free Sparse Attention for Long Video Diffusion
  5. 2026Ride the Wave: Precision-Allocated Sparse Attention for Smooth Video Generation
  6. 2026LoSA: Near-Lossless Sparse Attention for Training-Free Video Diffusion Acceleration
  7. 2025BLADE: Block-Sparse Attention Meets Step Distillation for Efficient Video Generation
  8. 2025Bidirectional Sparse Attention for Faster Video Diffusion Training