The way HeyGen trains an avatar on uploaded pics is quite amazing. A lot of good results, but it would be a nice option if you could work further on one result. Sometimes I get the perfect result in an environment, but I would love to get variations...
