Apppublic

Qualcomm-AI-Research/mobilewan-t2v

sourceHugging Faceupdated 2mo agoView on Hugging Face

8likes

App README

MobileWan T2V

This Space runs MobileWan — Qualcomm AI Research's lightweight, mobile-optimized adaptation of Wan2.2-5B for text-to-video generation.

—Paper: arXiv:2607.06173
—Model: Qualcomm-AI-Research/mobilewan
—Code: Qualcomm-AI-research/mobilewan

The model applies hybrid recurrent attention and self-attention head pruning to the Wan2.2 transformer, then distills the pipeline to 3 denoising steps. It produces 5-second videos at 480×832 resolution.