MobileWan becomes the first 5B-scale video diffusion model deployable on a commercial mobile device and proposes a learnable attention head pruning method based on binary per-head gates optimized end-to-end using a noise-biased sparsity objective and distillation-based finetuning.
Abstract
Recent advances in video diffusion have been driven by scaling transformer-based architectures to billions of parameters, substantially improving visual fidelity and motion coherence. In contrast, existing mobile video diffusion models remain limited to relatively small parameter budgets, typically 0.4-1.8B, restricting generation quality. In this work, we show that high-quality mobile video generation does not require small models. Instead, we demonstrate that a server-scale 5B-parameter video diffusion transformer can be deployed efficiently on memory-constrained mobile hardware through recurrent reformulation and structured compression. Starting from Wan2.2-5B, we rely on a recurrence distillation framework that converts video generation into a chunk-wise autoregressive process with constant-memory attention computation. Combined with causal linear attention, the model operates as an RNN at inference time while preserving temporal coherence across chunks. We further propose a learnable attention head pruning method based on binary per-head gates optimized end-to-end using a noise-biased sparsity objective and distillation-based finetuning. Together with sampling-step distillation and memory-optimized VAE decoding, MobileWan becomes the first 5B-scale video diffusion model deployable on a commercial mobile device. Our system generates 5-second 480x832 videos at 16 FPS in 20 seconds end-to-end latency, achieving a VBench score of 83.79 and establishing a new state of the art in mobile video generation. Please find the released DiT checkpoint and the sampling code in the project page: https://qualcomm-ai-research.github.io/MobileWan
This work introduces the GenVC, a compression-oriented video diffusion model built on a video diffusion model trained from scratch for compression, and realizes this model directly in pixel space with a global-to-local hierarchy that recovers fine spatio-temporal details, enabling high-quality generative reconstruction from compressed representations.
Naifu Xue, Zhaoyang Jia, Haosen Li et al.· 0 citations
Streaming video diffusion models have made substantial progress toward interactive and dynamic world simulation, but the nested autoregressive and denoising loops of conventional next-frame generation hinder real-time deployment. Recent rolling-window methods pipeline denoising across multiple consecutive frames at different noise levels, improving throughput and long-horizon stability. However, they tokenize every state at the same fine spatial granularity, leaving substantial noise-dependent redundancy in the joint denoising window. We propose Ms.Forcing, an efficient streaming video generation paradigm that adapts spatial granularity to each state's noise level. Its Multi-Scale Patchification (MSP) assigns coarser patches to noisier states, reducing the active-window token count by 45%, while Multi-Scale Self-Attention (MSSA) matches the density of visible non-sink keys and values to each query scale to further reduce attention cost. Because both schedules are fixed by window position, Ms.Forcing retains a static, hardware-friendly computation graph. We further introduce Homogeneous-Noise-Level DMD (H-DMD), which assembles each fake video from clean predictions sharing the same source noise level, thereby reducing the mismatch between DMD training sequences and inference-time rollouts. The multi-scale design helps offset the additional training cost of backpropagating through overlapping windows. We include both quantitative and qualitative experiments to show that Ms.Forcing reaches 22.84 FPS on a single H200 GPU, 39.6% faster than Rolling Forcing, while significantly improving VBench scores in both short video and long video generation setting.
Zekun Li, Xiaoyan Cong, Hongyu Li et al.· 0 citations
High-resolution video diffusion models built on Diffusion Transformers (DiTs) deliver strong fidelity but quickly exhaust the memory budget of a single workstation. A 100 billion-plus parameter DiT easily requires over a terabyte of persistent state, while naive spatiotemporal self-attention grows quadratically in sequence length. These two walls -- parameter memory and activation memory -- prevent researchers from adapting massive generative models without large GPU clusters. We revisit this problem from a systems perspective and introduce MegaSlide-DiT, a prototype that demonstrates how a pre-trained 105B DiT can be adapted on a single H200 GPU with 1.5 TB of host RAM. Our key insight is that the GPU need not own the model state: all persistent weights, master weights and optimizer moments remain in host memory, while only transient shards are streamed to the GPU on demand. Simultaneously, we replace quadratic global attention with 3D Deformable Slide Attention (3D-DSA), a motion-adaptive local attention operator that reduces both memory and computational complexity to linear in the sequence length. We report detailed memory accounting, execution traces and evaluation results to substantiate our design. MegaSlide-DiT does not claim to train a 105B model from scratch on a single GPU, nor does it magically solve bandwidth limits; rather, it offers a pragmatic path for full-parameter adaptation of massive video diffusion models on high-end workstations.
Given the remarkable achievements in image generation using diffusion models, the research community has shown increasing interest in extending these models to video generation. Recent diffusion models for video generation have predominantly utilized attention layers to extract temporal features. However, attention layers are limited by their computational cost, which increases quadratically with sequence length. This limitation poses significant challenges when generating longer video sequences using diffusion models. To overcome these challenges, we propose to leverage state-space models (SSMs) as temporal feature extractors. SSMs (e.g., Mamba) have recently garnered attention as promising alternatives owing to their linear-time memory and time consumption relative to the sequence length. Employing SSMs to capture temporal dependencies in video generation enables significantly higher generative performance at the same computational cost (e.g., memory usage, inference time) compared to attention-based methods, particularly for long-term sequences. For various model sizes, we comprehensively evaluated multiple long-term video datasets: MineRL Navigate, GQN-Mazes, and CARLA-Town01. For 256-frame video sequences, SSM-based models incur lower computational cost to achieve the same Fréchet Video Distance as attention-based models. Furthermore, the ablation study shows that when using SSMs for temporal modeling, incorporating bidirectionality and selective scans enhances video generation performance. Our code is available at https://anonymous.4open.science/r/SSM-Meets-Video-Diffusion-Models-067D/README.md.
Yuta Oshima, Shohei Taniguchi, Masahiro Suzuki et al.· New generation computing· 13 citations
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.
Shanghao Liu, Renze Chen, Size Zheng et al.· 0 citations
This work introduces SANA-Video 2.0, a hybrid video diffusion transformer instantiated at 5B and 14B scales under a unified architecture that recovers softmax-level expressiveness at substantially reduced cost, unlocking scalable long, high resolution video generation.
Junsong Chen, Jincheng Yu, Yitong Li et al.· 0 citations