- AI For Humans - The Newsletter
- Posts
- Presenting: The Daily Drip
Hello dearest internet friends.
You signed up to receive something from us, then our newsletter went out for more Juul-Pods and Monster (Zyns and Four-Loko?), and never came back…
So, WELCOME TO THE DAILY DRIP:
A delightful whisper of AI delivered to your inbox whenever we don’t forget forget to send it!
DALLE Prompt: The letters "AI" but coated in thick, drippy, multi-colored paints. As if the letters on on the wall and the layers of paint are dripping down/over the letters...
You see, every day Gavin and I collect interesting AI tidbits which sometimes make it into the show, and often end up covered in cobwebs inside abandoned GoogleDocs.
And today we bring you…
EMO: EMOTE PORTRAIT ALIVE
Explicit language/headphone warning: Eminem’s “Rap God” is used in the paper’s official examples. And you’l Don’t blame us, we’re just the messengers here.
This technique was developed by the Institute for Intelligent Computing at Alibaba Group. With only a single source image, Emote Portrait Alive uses a two-step process to generate a super emotive performance.
It combines audio embeddings with visual information to create dynamic, expressive facial animations and head poses that correspond with the audio input. As you can see in the examples above, it pretty much nails COHERENT AND VARIED head poses, with expressions, across different languages and styles. And it supports both singing and speaking!
Remember, this is all from ONE IMAGE. Imagine a near-future where an approach like this supports multiple source images, to help with accurately reproducing a character from all angles, or with extreme expressions.
“Prompt to Hollywood” is happening, and on the timeline we predicted!
Unrelated, a gentle reminder that Episode 46 of AI FOR HUMANS just dropped. Links to all the things can be found here:
Kisses, hugs and belly-rubs. See you all tomorrow, maybe?
-Kevin and Gavin