README.md

metadata

title: AAD-1
emoji: 🎬
colorFrom: red
colorTo: gray
sdk: gradio
sdk_version: 6.15.1
app_file: app.py
short_description: One-step autoregressive image-to-video generation
python_version: '3.12'
startup_duration_timeout: 1h

AAD-1: Asymmetric Adversarial Distillation for One-Step Autoregressive Video Generation

This Space demonstrates AAD-1, a one-step causal video world model that generates short video clips from a reference image and a text prompt. AAD-1 is distilled from Wan2.1-T2V-14B using asymmetric adversarial distillation, enabling single-step generation per causal chunk.

• Paper: arXiv:2606.03972
• Code: github.com/AutoLab-SAI-SJTU/AAD-1
• Model: Watay/AAD-1