Hacker News new | past | comments | ask | show | jobs | submit login
Software from Disney Research Seamlessly Blends Faces from Different Takes (disneyresearch.com)
67 points by protomyth on Dec 15, 2015 | hide | past | favorite | 26 comments



Given the history of cinema, I predict that this will be used excessively by some directors, who will fall into an "uncanny valley" the perceptive and intelligent will find disturbing while others will find it to be a "super-stimulus."

Then the industry will mature, and this will be used with greater subtlety.


s/the perceptive and intelligent/people who share my taste/

Yeah, but otherwise I guess you're right.


I'm not sure why the intelligent would be immune from finding that particular thing a super-stimulus.


Laugh tracks? The Transformers movies?


Yes, lots of people don't like those things (though the groups for each are distinct but overlapping.)

I'm not convinced that there is much correlation with intelligence.


Is the intention to use it in real life cinema? I made the assumption it would be used more in animation (especially mo-cap) - which will still fall into the uncanny valley traps, but it should make for more smooth transitions between emotions for these types of outputs.


Does Zemeckis have kids?


Odd, not long ago French movie theaters were shown this https://www.youtube.com/watch?v=AhYynZMwzYs

a blend of hand made compositing and simple warping. Technology is catching up too fast.


This reminds me of the automatic pitch correction (auto-tuning) in the recent pop music. Hopefully this will not signify the dusk of the art of acting.


I don't think so, the actor is supplying the 'raw materials', this is another tool for the director to achieve the truest representation of their vision, actors might appreciate not having to do an entire retake due to one duff line, they still get paid :)


And the next step will be to show different variations to multiple focus groups, then train a neural network to automatically select the set of blending weights for the film that are projected to maximize revenue, thereby automating the post production as well.


maybe, that sounds pretty dire though, how about interactive gaming where your input affects the mood of the conversation without the start/stop clip style used now.


Imagine the two of these technologies used together.


So this can be used to "auto-happy" me on webcam meetings when I'm feeling grumpy...


That's a wrap - don't worry we'll emote it in post.


Faces might be blending but the audio isn't exactly something I enjoy, pretty wild that's how it sounds. Could surely pick a favorite audio track, or, a la Max Martin, comp all the takes for material knowing the digital warp will be passable. Slick technique, I'm definitely impressed with the the show of tech and capability.


This was more focused on the video blending (and how it's synchronized with the audio cues). The clipping is from a naïve audio speed algorithm (the other naïve approach results in pitch shifting).

One thing this probably does give, though, is the necessary curve to output a well-formed audio blend.


Most likely the directors / editors who use this would overdub the audio once they get the visual take they want. They already overdub audio pretty often anyway, due to set noise.


It's easier to have the actor redub a scene, than to have him nail a take.


The mad + sad takes seemed to give a lot more emotional nuance than either of them separate. I like this, but hope it doesn't become as over-abused as autotune is.


This is really impressive. I can't see any seams in the merged output, though I do see the actors' eyes creepily going out of alignment (a fixable problem).


This is really interesting, but seeing the woman's right eye move down slightly in the angry/sad scene had freaked me out a bit.


I sure hope that audio clicking and popping would also smooth out.


So now we have Autotune for movies.

Oh boy.


The autotune of acting!!!





Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: