The action migration workflow of wan2.1_fun is built based on the official workflow.

wan2.1_fun has been tinkered with for quite a long time. Both KJ and the official flow were tested, but issues like image distortion, smudged faces, poor consistency, and many others occurred. After trying various methods, none performed as well as mimimotion. Thus, the workflow wasn't released until today, when a sudden idea struck me: using mimimotion's openpose for action processing, then blending it with depth processing after blurring—this unexpectedly achieved better control effects. At the same time, facial expression synchronization became more reasonable. Sharing it here.

Here's an overall explanation:

① Compared to mimimotion, this version of the wan2.1_fun workflow offers more reasonable dynamic control of the overall video scene. For example, objects in the background have better dynamic effects. The downside is that low-resolution generation appears somewhat blurry, so 720*1280 is recommended.

② The generation efficiency is not as good as mimimotion.

③ As the generation duration increases, the face gradually deviates more from the reference image (mimimotion also has face distortion issues). Post-generation, face swapping is still needed for repair.

④ Hand and foot generation is poor (a common AI issue). Currently, no optimal solution for handling hands and feet has been found. Looking forward to new technologies.

Summary: Overall, wan2.1_fun and mimimotion each have their pros and cons. Feel free to try both workflows to find a solution that suits your specific needs and scenario.

Creating content isn’t easy. If this has helped you, please bookmark, execute the application or workflow online. Your support is my motivation for creating. Thank you for your encouragement! 

If you have any questions, please leave a comment. I'll reply promptly when I see it. Welcome to exchange ideas, learn together, and make progress together!