Specifically, I have been running into issues with how to properly create high quality emotions from a single source portrait. In my workflow I've created two versions for how I could approach this. Personally, I train my Lora for Flux and I am getting super good results.