It's come lightyears from the first AI animated photos, still falls well within the uncanny valley for me. The most immediately glaring falsehoods are her hair and skin. Her hair moves like a solid helmet parented to the head, while none of the strands or tendrils bend, wobble, fall, or flex. Especially telling with the bit that kind of peeks out onto the neck. Also telling is the skin of her face and neck. While the animation is very realistic, the skin remains mostly rigid. Her brows move, but her forehead does not. The same is true for the way her skin should stretch and contract around her cheeks when talking. Less noticeable, therefore not one of my main two, is the fact that while her head makes subtle turns, her teeth do not change perspective. The two front teeth always face front and center, like one of those paintings where the eyes follow you around the room.
If you go to the Microsoft page for VASA-1, they demonstrate how this technology can be used with a pencil sketch or to animate the Mona Lisa. It's... unsettling. It does have a very Harry Pottery talking photos effect.
The one thing that will never change is that this is made by Microsoft and thus is a massive security threat from their horrendous programming that will allow for systems to be compromised.
Hi Thomas- If it's any consolation, the "Talking Faces" technology is not out yet. Microsoft announced this technology- meaning they had created it. However, they haven't launched it. If past experience holds out, these tech companies usually wait about six months from the time they launch a white paper to the time they launch the product.
Yes, I found the pausing in the speech to be somewhat strange, especially when I tried to match up to it. I couldn't do it, even though I did seven takes of my video. But, to Microsoft's credit, this is generation one of their technology. It's only going to get better from here.
Kinda sucks they are withholding it to the general public, but I of course understand why.. it's just unfair to those who want to just try it out and have fun with it, I like creating characters and it'd be cool to actually speak to them and see them actually talking
There are other technologies like this out there. Similar lip sync and head movement technology is available from AI company Runway, Nvidia’s Audio2Face AI application, Google’s Vlogger AI launched in March, and Emo AI by China’s Alibaba.
interesting tidbit - however, if I may recommend something? You seem comfortable and interested in exhibiting yourself in the media space to communicate with an audience. Your webcam, however, seems straight out 1999 and makes the whole experience rather unpleasant (it's so bad I really thought _you_ were the talking AI face for the first 10 seconds haha). Even for casual conversations 1x per year with my mom or some interview I make sure to have a decent, modern webcam that can do 1080 and gives a more photographic rendering. Just a tip.
Thanks for the tidbit! I'll work on an upgrade! It's probably the lighting. I was trying to match her background so I just threw up a tablecloth in the background as a mock green screen. I normally don't bother with one. The background you see behind me was just a background I downloaded from Zoom. In the real world, there is a dresser behind me in my office.
I agree! Companies will launch a white paper, and the product will usually come out about six months later. If VASA-1 follows this pattern, we can expect this technology to be available by the end of the year. It occurred to me today that you could use this technology to bring a dead relative back to life. I'm not sure if that's a good thing or not.
@@DeepLearningDaily as I have seen in the Stable Diffusion area and using these tools... projects like this have a weird % rate of actually going from white paper to usable tool, ofc this will become a tool of sorts, but one can't be certain it will be this specific one. Knowing corpospace, it will be kept in a vault with the low quality censored weights sold via subscription services ala chatgpt and other types of platforms like that
It's come lightyears from the first AI animated photos, still falls well within the uncanny valley for me. The most immediately glaring falsehoods are her hair and skin. Her hair moves like a solid helmet parented to the head, while none of the strands or tendrils bend, wobble, fall, or flex. Especially telling with the bit that kind of peeks out onto the neck. Also telling is the skin of her face and neck. While the animation is very realistic, the skin remains mostly rigid. Her brows move, but her forehead does not. The same is true for the way her skin should stretch and contract around her cheeks when talking. Less noticeable, therefore not one of my main two, is the fact that while her head makes subtle turns, her teeth do not change perspective. The two front teeth always face front and center, like one of those paintings where the eyes follow you around the room.
If you go to the Microsoft page for VASA-1, they demonstrate how this technology can be used with a pencil sketch or to animate the Mona Lisa. It's... unsettling. It does have a very Harry Pottery talking photos effect.
The one thing that will never change is that this is made by Microsoft and thus is a massive security threat from their horrendous programming that will allow for systems to be compromised.
Hi Thomas- If it's any consolation, the "Talking Faces" technology is not out yet. Microsoft announced this technology- meaning they had created it. However, they haven't launched it. If past experience holds out, these tech companies usually wait about six months from the time they launch a white paper to the time they launch the product.
The head moves around like it's from a hand puppet and it feels very overperformed, sprinkled with empty phrases.
Yes, I found the pausing in the speech to be somewhat strange, especially when I tried to match up to it. I couldn't do it, even though I did seven takes of my video.
But, to Microsoft's credit, this is generation one of their technology. It's only going to get better from here.
Kinda sucks they are withholding it to the general public, but I of course understand why.. it's just unfair to those who want to just try it out and have fun with it, I like creating characters and it'd be cool to actually speak to them and see them actually talking
There are other technologies like this out there. Similar lip sync and head movement technology is available from AI company Runway, Nvidia’s Audio2Face AI application, Google’s Vlogger AI launched in March, and Emo AI by China’s Alibaba.
interesting tidbit - however, if I may recommend something? You seem comfortable and interested in exhibiting yourself in the media space to communicate with an audience. Your webcam, however, seems straight out 1999 and makes the whole experience rather unpleasant (it's so bad I really thought _you_ were the talking AI face for the first 10 seconds haha). Even for casual conversations 1x per year with my mom or some interview I make sure to have a decent, modern webcam that can do 1080 and gives a more photographic rendering. Just a tip.
Thanks for the tidbit! I'll work on an upgrade!
It's probably the lighting. I was trying to match her background so I just threw up a tablecloth in the background as a mock green screen. I normally don't bother with one. The background you see behind me was just a background I downloaded from Zoom. In the real world, there is a dresser behind me in my office.
Microsoft will launch it eventually just because other companies will. The same can be said about other so called AI tools...
I agree! Companies will launch a white paper, and the product will usually come out about six months later. If VASA-1 follows this pattern, we can expect this technology to be available by the end of the year.
It occurred to me today that you could use this technology to bring a dead relative back to life. I'm not sure if that's a good thing or not.
@@DeepLearningDaily as I have seen in the Stable Diffusion area and using these tools... projects like this have a weird % rate of actually going from white paper to usable tool, ofc this will become a tool of sorts, but one can't be certain it will be this specific one. Knowing corpospace, it will be kept in a vault with the low quality censored weights sold via subscription services ala chatgpt and other types of platforms like that
Your ego got a subscriber.
My ego thanks you!