Core Viewpoint - The article discusses the development of EasyCache, a new framework for accelerating video diffusion models without requiring training or structural changes to the model, significantly improving inference efficiency while maintaining video quality [7][27]. Group 1: Research Background and Motivation - The application of diffusion models and diffusion Transformers in video generation has led to significant improvements in the quality and coherence of AI-generated videos, transforming digital content creation and multimedia entertainment [3]. - However, issues such as slow inference and high computational costs have emerged, with examples like HunyuanVideo taking 2 hours to generate a 5-second video at 720P resolution, limiting the technology's application in real-time and large-scale scenarios [4][5]. Group 2: Methodology and Innovations - EasyCache operates by dynamically detecting the "stable period" of model outputs during inference, allowing for the reuse of historical computation results to reduce redundant inference steps [7][16]. - The framework measures the "transformation rate" during the diffusion process, which indicates the sensitivity of current outputs to inputs, revealing that outputs can be approximated using previous results in later stages of the process [8][12][15]. - EasyCache is designed to be plug-and-play, functioning entirely during the inference phase without the need for model retraining or structural modifications [16]. Group 3: Experimental Results and Visual Analysis - Systematic experiments on mainstream video generation models like OpenSora, Wan2.1, and HunyuanVideo demonstrated that EasyCache achieves a speedup of 2.2 times on HunyuanVideo, with a 36% increase in PSNR and a 14% increase in SSIM, while maintaining video quality [20][26]. - In image generation tasks, EasyCache also provided a 4.6 times speedup, improving FID scores, indicating its effectiveness across different applications [21][22]. - Visual comparisons showed that EasyCache retains high visual fidelity, with generated videos closely matching the original model outputs, unlike other methods that exhibited varying degrees of quality loss [24][25]. Group 4: Conclusion and Future Outlook - EasyCache presents a minimalistic and efficient paradigm for accelerating inference in video diffusion models, laying a solid foundation for practical applications of diffusion models [27]. - The expectation is to further approach the goal of "real-time video generation" as models and acceleration technologies continue to evolve [27].

Core Insights - The article discusses the challenges and advancements in video generation using diffusion models, particularly focusing on the computational bottlenecks associated with attention mechanisms in the Diffusion Transformer (DiT) model [1][6][14] - A new method called DraftAttention is introduced, which significantly reduces the computational overhead of attention mechanisms while maintaining high generation quality, achieving up to 2x end-to-end inference acceleration on GPUs [3][22][46] Group 1: Background and Challenges - Diffusion models have become mainstream for high-quality video generation, but the computational load of attention mechanisms increases dramatically with video length and resolution, leading to inefficiencies [1][6] - In models like HunyuanVideo, attention computation can account for over 80% of the total processing time, with generating an 8-second 720p video taking nearly an hour [1][5] - The complexity of attention mechanisms grows quadratically with the number of tokens, which is directly proportional to video frame count and resolution, causing significant slowdowns in inference speed [6][7] Group 2: Existing Solutions and Limitations - Current acceleration methods, such as Sparse VideoGen and AdaSpa, utilize sparse attention mechanisms for some level of end-to-end acceleration on GPUs, but their effectiveness is limited due to insufficient sparsity and rigid design [2][3] - These methods often rely on fixed sparse operators and lack dynamic adaptability to input content, making it difficult to achieve fine-grained, content-aware sparse pattern control [2][7] Group 3: DraftAttention Methodology - DraftAttention is a plug-and-play, dynamic sparse attention mechanism that does not require training, designed to reduce the computational burden of attention mechanisms while preserving generation quality [3][11][46] - The method involves creating a low-resolution attention map to estimate token importance, guiding the selection of sparse patterns for high-resolution attention calculations [11][12] - A token rearrangement strategy is introduced to enhance the execution efficiency of sparse computations on GPUs, making the approach hardware-friendly [13][22] Group 4: Theoretical Foundations and Experimental Results - The effectiveness of DraftAttention is supported by theoretical analyses demonstrating that the approximation error between the low-resolution and high-resolution attention maps is bounded [15][17] - Experimental evaluations show that DraftAttention outperforms existing sparse attention methods like Sparse VideoGen across multiple metrics, including PSNR and SSIM, particularly at high sparsity rates [20][21] - On NVIDIA H100 and A100 GPUs, DraftAttention achieves up to 1.75x end-to-end inference acceleration, with performance improvements scaling with video length, resolution, and sparsity [22][46] Group 5: Future Directions - The authors plan to further optimize efficiency bottlenecks in long video generation by integrating techniques such as quantization and distillation, aiming to extend high-quality video generation capabilities to resource-constrained environments like mobile and edge devices [46]