---
title: T3-Video 4K
emoji: 🎬
colorFrom: red
colorTo: blue
sdk: gradio
sdk_version: 6.29.1
app_file: app.py
python_version: "3.12"
startup_duration_timeout: 1h
short_description: Native 2176x3840 4K text-to-video with T3-Video-1.3B
---

This Space is the official-style demo of **T3-Video-Wan2.1-T2V-1.3B** — a transformer
retrofitted with multi-scale weight-sharing window attention that generates **native
2176×3840 (4K) video** from text in a single pass, without any upsampling.

- Paper: [Transform Trained Transformer for Accelerating Native 4K Video Generation](https://huggingface.co/papers/2512.13492)
- Code: [github.com/zhangzjn/T3-Video](https://github.com/zhangzjn/T3-Video)
- Weights: [APRIL-AIGC/T3-Video](https://huggingface.co/APRIL-AIGC/T3-Video)

The released checkpoint only supports its native training shape: 2176×3840, 81 frames,
30 fps (the window-attention grid is hard-coded). Generation is compute-heavy; the
slider trades denoising steps for time (the paper's official setting is 50 steps).

Example prompts come from the authors' [4K-World-Vision](https://github.com/zhangzjn/T3-Video)
benchmark (Apache-2.0). Model license: Apache-2.0 with Wan2.1 terms.