So is your workflow only for reference t2V or I2v as well ?

This isn't something that works with just LoRA —it need to download the node and run it together with that node. Instead of the workflow you provided, I adapted it using Gemini so it could be used in my own workflow. After applying it, the face has become much more similar to the original photo, and I'm really satisfied. I hope upgraded versions of the LoRA keep coming out. Thank you.