MobileWan: Closing the Quality Gap for Mobile Video Diffusion

2026-07-07

AI Summary

Existing mobile video diffusion models, constrained by small parameter counts (0.4–1.8B), struggle to simultaneously achieve high generation quality and temporal coherence. This work presents the first efficient deployment of a 5B-parameter video diffusion Transformer on commercial mobile devices, introducing a chunked autoregressive recursive distillation framework alongside a learnable attention head pruning method based on binary gating. The approach further integrates causal linear attention, a noise-biased sparse objective, and memory-optimized VAE decoding. The resulting system generates 5-second, 480×832 videos at 16 FPS within 20 seconds end-to-end on-device, achieving a VBench score of 83.79 and substantially advancing the state of the art in mobile video generation.

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. Project page: https://qualcomm-ai-research.github.io/mobilewan

Problem

• mobile video generation
• video diffusion models
• model compression
• parameter budget
• generation quality

Innovation

• recurrent reformulation
• structured compression
• attention head pruning
• distillation-based finetuning
• mobile video diffusion