Spark-Wan: Few-Step Video Diffusion Transformer via Hybrid Adversarial Post-Training

Zongjian Li, Shenghai Yuan, Li Yuan,

If you like our project, please give us a star ⭐ on GitHub for latest update.

📰 News

• [2026.3.08] 👋👋👋 We introduce Helios, a breakthrough video generation model that achieves minute-scale, high-quality video synthesis at 19.5 FPS on a single H100 GPU — without relying on conventional long video anti-drifting strategies or standard video acceleration techniques. Welcome to check Technical Report!
• [2025.9.26] 🔥🔥🔥 Release 4-steps merged checkpoint in huggingface. Click to download.
• [2025.9.26] 🔥🔥🔥 We open-source the distillation weights for 8-steps and 4-steps based on Wan2.1 14B, as well as the training code.

🚀 Main Results

Step4
0.mp4	1.mp4	2.mp4
3.mp4	4.mp4	5.mp4
6.mp4	7.mp4	8.mp4

Step8
0.mp4	1.mp4	2.mp4
3.mp4	5.mp4	4.mp4
6.mp4	7.mp4	8.mp4

🔒 License

This project is released under the Apache 2.0 license as found in the LICENSE file.