Wan
Alibaba's open-source AI video model family that generates up to 15-second clips with native audio-visual sync, reference-to-video character consistency and multi-shot storytelling.
✅ Our verdict
Wan is the standout open-source choice for developers and studios that want full control over an AI video model, including weights and code. It rivals closed tools on audio sync and character consistency while staying free to self-host.
👍 Pros
- +Fully open-source (Apache 2.0)
- +Native audio with accurate lip-sync
- +Reference-to-video character consistency
- +Up to 15-second clips in one run
- +Text-to-video and image-to-video
👎 Cons
- −Self-hosting needs strong GPUs
- −Setup more technical than closed tools
- −Hosted compute costs for heavy use
🎯 Use cases
ℹ️ Key facts
Last updated: Jun 2026
🔀 Compare Wan with
Try Wan Now
Alibaba's open-source AI video model family that generates up to 15-second clips with native audio-visual sync, reference-to-video character consistency and multi-shot storytelling.
❓ Frequently Asked Questions
Is Wan really open source?
Yes. Wan is a family of open-source video models from Alibaba's Tongyi Lab, released with weights, training code and inference scripts under the Apache 2.0 license. You can self-host it for free if you have suitable GPUs, or use hosted plans on wan.video without managing infrastructure.
What can Wan generate?
Wan supports both text-to-video and image-to-video, generating clips up to about 15 seconds with native audio-visual synchronization, including speech with lip-sync and sound effects. Its reference-to-video feature keeps a character's appearance consistent across multiple clips for multi-shot storytelling.