In my evaluation of various AI avatar generation tools, they all are striving for “photo-realistic” or “hyper-realistic” avatars – synthetic or cloned. I’m finding that all of them have moved firmly into the “uncanny valley” and I don’t have hope that a computer generated avatar would not be off putting or just plain creepy.
As I’m thinking about this race to perfection, I’m reminded of a time during my work in animation when I was disappointed about a similar race for human like renditions of animated characters and their environments. To me, animation is an artistic interpretation of reality, not a mirror. Hair renderers and water simulation doesn’t represent a human’s vision of what they are seeing – it is just trying to be a perfect imitation.
Using avatars in production have great advantages – similar to those of using cloned voices. Production doesn’t take the instructor’s time and when changes inevitably have to be implemented, it makes updating simple and seamless. If the words are created (or curated) by the instructor, is it critical that the instructor be on camera to recite his work? I don’t know.
So, what if we allowed an instructor to create an avatar, either of himself or really anything, that they feel represents them and the topic which they are discussing? That seems an approach worth investigating.
I asked ChatGPT to take a photo of me and make me an avatar in the “Marvel Comic” style. Results are above. While it absolutely nailed my physique, the face seems a bit off – but, it is a stylized version of myself, so a great starting point.
I then took that avatar and loaded it into “D-ID” to create a video avatar. This is similar to the test I did with Colossyan – it allows you to upload an image or take a video. I typed a 1 line script and grabbed a generic male voice since I’m not interested in testing the tool’s voice cloning functionality.
The results are absolutely hysterical. It reminds me of the old Quiznos commercials where they superimposed a human mouth on a hamster.
It would appear that they are just using a generic male mouth for lip sync rather than cloning the the mouth from the image. Maybe if it had a video source it would have worked better?
I’m going to reach out to the folks at D-ID to see if they are able/interested to do animated lipsync of an illustrated avatar. I may put on some lipstick to see they are not cloning lips when doing a photo or video reference for their lipsync. Stay tuned for that!
