Sink‑Sparse Hybrid Attention is a train‑free method that replaces dense self‑attention in Diffusion Transformers with reusable static sink masks and a lightweight pooled‑KV branch, keeping model weights and sampling schedules unchanged. It identifies reusable high‑mass structures from a small offline calibration set and applies them across future generations. Experiments on Wan2.1‑T2V‑1.3B and Z‑Image‑Turbo show 1.54× and 1.57× end‑to‑end speedups while preserving generation quality.
arXiv:2610.08772v1 Announce Type: new Abstract: Diffusion Transformers (DiTs) have achieved strong performance in image and video generation, but the quadratic complexity of full attention makes high...
PISA introduces a training‑free Piecewise Sparse Attention mechanism that preserves full attention span while reducing complexity. It keeps exact computation for critical blocks and approximates non‑critical ones using block‑wise Taylor expansion, avoiding the quality loss seen in conventional sparse attention. Experiments show significant speedups—up to 2.57× on Hunyuan‑Video—while maintaining or improving visual quality across video and image generation tasks.
arXiv:2609.23153v1 Announce Type: new Abstract: Video diffusion transformers are expensive because attention dominates long spatiotemporal token sequences. We identify the \emph{high-sparsity trap}:...
The paper introduces HetA-DiT, a heterogeneous attention mechanism for video diffusion models that allocates computation based on token difficulty. A lightweight uncertainty branch predicts denoising difficulty, routing uncertain tokens through dense global attention while applying efficient local attention to reliable tokens. This adaptive routing retains global context where needed, offers a single parameter to balance quality and efficiency, and achieves competitive generation quality while only about 20% of tokens use dense attention.
arXiv:2608. 12032v1 Announce Type: cross Abstract: Video diffusion transformers are costly to sample: every denoising step applies self-attention over a long 3D token sequence, a quadratic cost that dominates as resolution and duration grow.
arXiv:2609.37001v1 Announce Type: cross Abstract: Diffusion Transformers (DiTs) enable high-quality video generation but suffer from substantial inference latency, primarily attributable to the compu...
The paper introduces Growing Sparsity and Reduced Search (GSRS), a training‑free framework that accelerates Diffusion Transformers for video generation by progressively increasing sparsity and reusing early attention blocks to limit search space. GSRS uses Top‑p selection for adaptive sparsity and a precise evaluation scheme that computes exact attention weights with negligible overhead. Experiments on HunyuanVideo and Wan 2.1 show 1.55×–2.21× speedups while preserving generation quality comparable to full attention.
ToDoPro introduces an early token‑reduction strategy for reference‑based image generation that uses a few denoising steps on 2× downsampled reference images to probe token importance with the existing Diffusion Transformer. Tokens are then hierarchically pruned: low‑importance tokens are discarded, mid‑importance tokens are replaced by their downsampled versions, and high‑importance tokens are retained. Experiments on three base models and two benchmarks show significant inference speedups, reduced peak GPU memory, and improved generation quality by filtering out noisy reference tokens.
arXiv:2610.29441v1 Announce Type: new Abstract: Full 3D attention is costly in video Diffusion Transformers. Block-sparse attention reduces this cost by computing only selected attention blocks. Howe...
arXiv:2601. 11641v3 Announce Type: replace-cross Abstract: While Diffusion Transformers (DiTs) have achieved notable progress in video generation, this long-sequence generation task remains constrained by the quadratic complexity inherent to self-attention mechanisms, creating significant barriers to practical deployment.
arXiv:2610.17552v1 Announce Type: cross Abstract: Recent video super-resolution (VSR) models achieve efficient one-step generation through diffusion distillation. However, high-resolution inference r...
arXiv:2608. 06628v1 Announce Type: new Abstract: Diffusion language models (dLLMs) offer a promising alternative to autoregressive models by accelerating inference through parallel decoding.