Release Wan-Alpha-I2V model weights ? Take a look at H3
Time flies—it’s been nearly a year since your last release.
In the meantime, MiniMax-H3 has become a mainstream model.
It would be fantastic if you could release a transparent video model based on H3!
Really looking forward to your next project!
Thank you for your interest and support!
Our image-to-video updates are available in Wan-Alpha++:
https://github.com/WeChatCV/Wan-Alpha_PlusPlus
We do not plan to release or support a MiniMax-H3-based version. However, we have open-sourced our training dataset here:
https://huggingface.co/datasets/htdong/Wan-Alpha-dataset
The complete organization of the VAE training data is also documented in our CVPR 2026 paper:
https://openaccess.thecvf.com/content/CVPR2026/html/Dong_Video_Generation_with_Stable_Transparency_via_Shiftable_RGB-A_Distribution_Learner_CVPR_2026_paper.html
We hope these resources will be helpful to the community!