So here it is the result from ZRT Face Trainer and below are my thoughts:
Looks like its all blend-shape driven process with a base FACS based logic driven face rig setup. Your base model which is termed as “target mesh” gets converted to match the topology of ZIVA base model (termed as “source model”) which I believe already has all the necessary face setups and connections. Once you upload your file to ZRT Face Trainer portal, it takes a few hour for the preview result to be published. During the process of training in my opinion ZIVA Rapid Trainer essentially swap the face through some blend-shape algorithm and possibly face swapping AI similar to deep fake to get the result! This workflow is good for fast output for non real (living) human or CGI generated hyper real model and may not be suitable for CGI digital double. In our case we used our in-house 3D scanned digital double and the result doesn’t match the real character in terms of facial expression. For CGI digital double face rig based on actual 3d scans of facial expression of the actor is essential. But anyways this looks promising and in my opinion a GAME CHANGER! Finally the Face Trainer just provides you with a video preview, there is no mesh file, rig file or animation file provided with this early access which is a bummer! As I could have dig deeper and also do a nice little render with custom animation 🙂