Long video sequences make attention a substantial cost in video diffusion models. Block sparse attention (BSA) reduces this cost by computing selected blocks exactly while compressing the remaining blocks. However, fixed token grouping can place tokens with dissimilar attention preferences in the same group, while uniform key averaging can underestimate the attention mass of compressed blocks. We introduce Spark-Attn, a training-free BSA method comprising Spark-Reblock and Spark-Reweight. Spark-Reblock recursively groups tokens by attention preference while preserving regular block shapes for efficient GPU execution. Spark-Reweight uses the global mean query as an anchor to construct attention-weighted key-value representations and logit biases for the compressed blocks, calibrating their attention mass. We instantiate Spark-Attn in MiniMax-H3 as Spark-H3. At 10% exact-block density, Spark-H3 delivers a 2.49× DiT speedup and reaches 24.81 dB PSNR against the corresponding dense outputs on MiniMax-H3 (20 denoising steps, 345 frames, 1344×768). Increasing the density to 20% further improves PSNR to 26.76 dB while still achieving a 2.26× DiT speedup. These results demonstrate that Spark-H3 achieves a state-of-the-art quality-efficiency trade-off for training-free BSA on video diffusion models.
Figures & tables
Figure 2: BSA, CSA, and Spark-Reblock. Colors denote attention preferences, and N denotes the number of video tokens. Spark-Reblock uses recursive balanced splits to construct preference-aligned, uniformly sized tiles in O(NlogN) .
Figure 3: Spark-Attn overview. (1) Block routing: Spark-Reblock groups query and key–value (KV) tokens into preference-aware, uniformly sized tiles; tile means guide selection of a fraction ρ of KV tiles per query tile. (2) Two-branch attention: Selected blocks are computed exactly, while the rest are approximated using Spark-Reweight’s weighted KV representatives and logit biases. Both branches share one softmax normalization, with compressed entries for selected blocks masked to avoid double counting. N , M , and ρ denote the number of video tokens, the number of KV tiles, and the exact-block density, respectively.
Method
Category
Density
DiT latency (s) ↓
Speedup ↑
PSNR (dB) ↑
SSIM ↑
LPIPS ↓
VBench ↑
Dense
–
100%
4061.8
1.00 ×
∞
1.00
0.00
84.38
SVG2 [ 27 ]
CSA
≈17.8%
1876.1
2.16 ×
23.70
0.79
0.15
84.39
SVG-EAR [ 38 ]
CSA
≈17.8%
1929.3
2.11 ×
24.88
0.82
0.13
84.50
SVOO [ 14 ]
CSA
≈19.7%
1847.6
2.20 ×
23.73
0.79
0.14
84.56
PISA [ 11 ]
BSA
10%
1643.4
2.47 ×
20.20
0.71
0.23
84.60
Sol-H3 [ 10 ]
BSA
≈20.8%
1781.9
2.28 ×
21.26
0.74
0.19
84.47
Table 1: Quantitative Results on MiniMax-H3. Green shading marks the first , second , and third distinct scores among methods. † uses separate query and KV partitions.
Method
Category
Density
DiT latency (s) ↓
Speedup ↑
PSNR (dB) ↑
SSIM ↑
LPIPS ↓
VBench ↑
Dense
–
100%
1625.0
1.00 ×
∞
1.00
0.00
84.32
SVG2
CSA
≈17.6%
846.7
1.92 ×
24.01
0.78
0.15
84.32
SVG-EAR
CSA
≈17.6%
864.8
1.88 ×
25.01
0.81
0.13
84.31
SVOO
CSA
≈19.1%
829.0
1.96 ×
24.33
0.80
0.14
84.54
PISA
BSA
10%
718.7
2.26 ×
21.28
0.71
0.22
84.39
Sol-H3
BSA
≈20.9%
775.6
2.10 ×
22.44
0.75
0.17
84.43
Table 3: Quantitative Results on MiniMax-H3-Turbo. Notation and shading follow Table 2 .
Spark-Reblock
Density
DiT Latency (s) ↓
Recall (%) ↑
PSNR (dB) ↑
10%
1598.6
61.36±2.71
20.30
✓
10%
1629.6
79.69±0.91
24.81
20%
1763.2
74.05±2.12
21.65
✓
20%
1796.1
88.63±0.61
26.76
Table 4: Ablation of token grouping methods . Attention-mass recall is the fraction of dense-softmax attention mass captured by selected blocks (mean ± sample SD).
Spark-Reblock
Spark-Reweight
DiT Latency (s) ↓
MAE ↓
PSNR (dB) ↑
1765.0
3.38
21.24
✓
1763.2
2.52
21.65
✓
1796.0
1.39
26.24
✓
✓
1796.1
1.16
26.76
Table 5: Ablation of compression methods . Log-mass MAE is the mean absolute error between predicted and dense block log-masses.
Figure 4: Motion incoherency. Zoom in to compare body poses between adjacent frames and hair details.
Figure 5: Graphic inconsistency. Cyan and yellow boxes mark the hoodie regions enlarged below; zoom in to compare the graphics.
Video diffusion transformers generate high-quality videos, but the quadratic cost of 3D full attention dominates their inference. Dynamic block-sparse attention reduces this cost by computing exact attention only on key/value blocks selected from current activations. However, its fidelity degrades under aggressive sparsity: blocks are scored from coarse summaries or formed by costly global clustering, pruned blocks vanish from the softmax, and whole tiles are kept or dropped even when only a few of their keys matter. To address these limitations, we propose Accurate Block Attention (AB-Attention), a training-free method that accurately approximates dense attention at high sparsity. Its Intra-Block Clustering Selector (IBCS) estimates block importance faithfully yet cheaply by clustering queries only within each query block, without permuting tokens. Sampling with Inclusion-Probability Weighting (SIP) then samples blocks in proportion to these estimates and reweights them by inverse inclusion probability, so that one pass estimates dense attention rather than truncating it. Finally, its kernel gathers key/value blocks much smaller than a tile into full tensor-core tiles, refining selection at no extra tensor-core cost. On Wan2.1 and HunyuanVideo, AB-Attention skips 89--92% of attention, whereas the baselines compute 1.8--3.6× as much, yet it still improves PSNR by up to 4.4~dB over the strongest baseline and achieves up to 2.23× end-to-end speedup. Our code is available at https://github.com/InfRec-Cardinal/AB-Attention.
Bin Xu, Mo Zhang, Zhengfeiyang Zhang +6
InfRec · The Chinese University of Hong Kong, Shenzhen · Shandong University +5
Diffusion transformers have achieved remarkable success in high-quality video generation, yet their reliance on spatiotemporal 3D full attention incurs prohibitive computational cost due to the quadratic complexity of attention. Block sparse attention is a common approach to mitigate this by focusing computation on important regions. However, attention maps in DiTs exhibit inherently dynamic and fine-grained sparsity, which causes existing block sparse attention methods to degrade significantly in quality, especially at high sparsity ratios. In this paper, we revisit block sparse attention and derive a theoretical lower bound on attention recall to characterize the key factors governing its effectiveness. Guided by these insights, we propose DFSAttn, a training-free sparse attention framework that enables dynamic, fine-grained sparsification efficiently. DFSAttn incorporates three core designs: Hilbert curve-based token reordering to achieve fine-grained sparsity while preserving efficient GPU execution, hierarchical block scoring for accurate block importance estimation, and sparse mask caching with adaptive ratios to balance accuracy and efficiency. Experimental results demonstrate that DFSAttn consistently outperforms prior methods under high sparsity, achieving up to 2.1× end-to-end speedup while maintaining high generation quality. Our code is open-sourced and available at https://github.com/jessica-hujie/DFSAttn.
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.