At the 2019/2020 year's NHK New Year's event we saw a first truly deepfake artist: Hibari Misora, who died 30 years ago, sang a new song, the singer was 'regular' CGI hologram, the song was a new deepfake-generated song using her old records as training data. I haven't seen that before!
That's interesting. I hadn't heard about deepfake-generated songs before. The main thing stopping deepfakes from becoming a problem that I see is that even though you can swap someone's face onto another person's body and have it mimic their facial expressions, you can't make them sound the same. So if you try to put words in someone else's mouth you have to have a way to generate audio in their voice to match the video. It seems that audio deepfakes are on their way now too, which I think is when we'll see a real problem.
I predict in the next year or so a deepfake will go viral and cause significant harm to an important public person. I bet it will come in the form of a supposed candid shot where the person will speak thinking they are not being recorded and say something damning or inflammatory.
From the article (and where I slowly stopped reading):
"The images produced by Cameos were slightly cartoony, meaning that the digital avatar looked a little bit more like a Bitmoji than the user’s real face. But the effect was essentially the same — it took someone’s likeness and spliced it onto another body or into another scene.
"
This is presented as DeepFake-like technology. However the whole issue with DeepFakes is that they are (or can be) indistinguishable from a real video. Any tech or feature that deliberately or due to limitations in the approach doesn't do that isn't really DeepFaking anything, so I don't quite think the article is on the right track. Once companies actually incorporate DeepFake tech we might be able to make some sort of early assessment, but it's way too early to draw any conclusions yet.
Edit: I guess the author gets this notion from a TechCrunch article they linked, which also essentially calls all these type of filters DeepFaking. It's not. Really, look at the videos of Cameo.
On a side note, we should really have a proper name for this (ultrarealistic fake videos), since the origin of DeepFake is indeed a bit too specific to the technology. Who knows, maybe in the future there are other approaches that will work as well.
10 comments
[ 5.4 ms ] story [ 40.1 ms ] threadI don't know what they intended to convey - that anyone in tech==evil.
Imagine the headline that puts journalism through a similar purity test: "then the journalists monetized them". They'd have you banned on social media
But headlines are often very misleading, and that's all I'm going on here.
https://www.japantimes.co.jp/culture/2020/01/10/films/nhk-ra...
This is presented as DeepFake-like technology. However the whole issue with DeepFakes is that they are (or can be) indistinguishable from a real video. Any tech or feature that deliberately or due to limitations in the approach doesn't do that isn't really DeepFaking anything, so I don't quite think the article is on the right track. Once companies actually incorporate DeepFake tech we might be able to make some sort of early assessment, but it's way too early to draw any conclusions yet.
Edit: I guess the author gets this notion from a TechCrunch article they linked, which also essentially calls all these type of filters DeepFaking. It's not. Really, look at the videos of Cameo.
On a side note, we should really have a proper name for this (ultrarealistic fake videos), since the origin of DeepFake is indeed a bit too specific to the technology. Who knows, maybe in the future there are other approaches that will work as well.