README.md

metadata

title: Echo-Infinity
emoji: 🎬
colorFrom: red
colorTo: gray
sdk: gradio
sdk_version: 6.15.1
app_file: app.py
short_description: Infinite video generation with learnable evolving memory
python_version: '3.12'
startup_duration_timeout: 1h

Echo-Infinity: Real-Time Infinite Video Generation

This Space demonstrates Echo-Infinity, a method for real-time, constant-cost infinite video generation using a learnable evolving memory mechanism. It is built on top of the Wan2.1-T2V-1.3B text-to-video model.

How it works

Echo-Infinity introduces a learnable memory that filters, abstracts, and compresses any-length video history at constant computational cost. This enables generation of arbitrarily long videos without the quadratic memory growth typical of autoregressive video models.

This demo

This Space generates short video clips (21 frames ≈ 1.3 seconds at 16fps) from text prompts, capped to fit within ZeroGPU time limits. The full model supports hour-scale and even 24-hour continuous video generation.

Links

• 📄 Paper (arXiv:2606.04527)
• 🌐 Project Page
• 🤗 Model
• 💻 Code

Citation

@article{bian2026echoinfinity,
  title={Echo-Infinity: Learnable Evolving Memory for Real-Time Infinite Video Generation},
  author={Bian, Yuxuan and Xue, Zeyue and Zhang, Songchun and Zhang, Shiyi and Jin, Weiyang and Li, Yaowei and Zhuang, Junhao and Li, Haoran and Huang, Jie and Huang, Haoyang and Duan, Nan and Xu, Qiang},
  journal={arXiv preprint arXiv:2606.04527},
  year={2026}
}