Wan 2.2 Wink LoRA by MQ Lab
This LoRA from MQ Lab was generated by selecting an appropriate training method, curating many datasets to avoid bias, and manually writing captions.
MQ Lab's LoRA will continue to evolve. I will continue to add more datasets, improve quality, and provide new versions.
Prompt ex
A woman winks once.
Tips
From my perspective, she can only wink with her left eye. I tried various captions during training, but nothing worked.
The output video is blurry because there aren't enough low-noise processing steps. Please increase the number of steps.
If the movement is too fast, set the FPS to 16.
I have moved my main activities to Discord. I will post information on Discord and publish my work on MEGA. There is a contact link for MEGA in Discord. https://discord.gg/5vwmMNCZ
I will be creating various LoRA versions of Wan2.2 in the future, which will require a powerful server-grade GPU (such as the H200). I plan to use a cloud GPU service to run this, which will require funding. If you are willing to help with funding, you can do so on this site. Thank you.