New in Labs: AI Avatars

I just used the block to generate a video 49 seconds long. It took about 10 minutes to generate. I used the provided Avery avatar with Jessica Bogart's voice. Even though I'm highly perceptive, I think anyone would agree the result was goofy. The lip-syncing was off most of the time, she opened her mouth wide several times, and the mouth, teeth, and eyes changed in size and sharpness frequently--feeling more like Elastigirl. The teeth morphed together a couple times (something AI has had a particularly hard time with).

As my favorite YouTube home inspector likes to say, "If you squint, it's mint."

All that said, it is very close to passably human. There are moments when the avatar is sharp and in sync, and that gives me hope. This is a really cool idea and I hope it succeeds. I agree with all of the feature requests my colleagues have already stated.

With or without avatar video, I also want the ability for AI voice to narrate an entire course, much like Microsoft Edge's "Read Aloud" feature, with a floating set of audio controls which follows you as you scroll.