Hacker News Reader

Comment by ChuckNorris89 | Hacker News Reader

If you follow the 'VFX artists react' series on YouTube you learn what give CGI humans away pretty easily. Getting CGI humans right is very difficult and few VFX houses, with decades of experience, get it right even with post processing, let alone doing it in real time.

CGI Jensen looks like he came out of a 2009 videogame. Nvidia may have the tech but you can tell it lacks the artistic supervision of a proper VFX house so the end result looks like something their engineers put together in a last minute crunch (which is probably what happened).

View on HN

I can't think of a time when I've been fooled by CGI (though, of course, I wouldn't be aware of being fooled if I was, so it's a catch-22).

I really hate when movies throw in a CGI character expecting it to come off as real. It pulls you out of the movie in a way that normal over-the-top action CGI doesn't. I remember watching the recent Star Wars movies and being jerked out of the flow by CGI Princess Leah and some other characters (whoever the villainous old man was in that movie).

I'd be interested to see any videos of CGI faces that can fool experts.

I get your point, but "I can't think of a time when I've been fooled by CGI" is a tautology: you remember = you weren't fooled. If you had been fooled, chances are you wouldn't remember.

There are videos from CGI artists, e.g. Corridor Crew, that break down really good CGI, which is almost entirely invisible to the non-expert eye. That is in fact the best kind of CGI: the one that enhances reality so seamlessly that it's almost impossible to pick apart. Quite entertaining and informative videos.

There are plenty of possibilities to be fooled and then find out later. At which point you would remember later.

Although if you're fooled, and find out later, I think there's a strong human tendency to re-write the internal narrative.... "Really, I could kind of tell all along." In which case it's not so clear how you'd "remember" things later.

You laid out how: if you don't re-write your internal narrative

You're right that theres a possibility that someone is fooled and never finds out, and another possibility that someone is fooled, finds out later, and claims they never thought it was real. However, it is very clear that it's possible to be fooled and find out later.

CGI humans are pretty weak, but most mainstream TV and movies have a lot of background CGI that is very well done.

Objects mostly just sit there, or move in more predictable ways. But when it comes to faces and bodies, we're super-tuned to facial detail, and are heavily primed to read emotions and context from facial details and from posture.

So if any part of that falls into uncanny valley, the whole experience looks wrong.

Back when animations were done by hand, Disney's animators handled this by rotoscoping (drawing over...) live action, which created very convincing results even when the movements were exaggerated for effect.

Doing the same with CGI without a template is an unbelievably difficult challenge. It's much easier to create cartoonish exaggerations than to get spot-on perfect realism.

The best results I've seen are analogous to a modern version of rotoscoping. It involves creating a fully animated 3d model of a person and then using a well trained deep fake of the actor's face in combination.

Youtuber Shamook did just that with the finale of the Mandalorian season 2. https://www.youtube.com/watch?v=wrHXA2cSpNU

He actually got a job with ILM out of it. The future of CGI actors will probably involve a fusion of techniques like this in order to get it right.

I'm not so sure that it's easy to tell the difference between CGI and reality. Sometimes scenes in movies look really fake, but aren't actually CGI. The way scenes are lit and the heavy makeup actors end up wearing can make it seem like a real scene looks fake.

It would be interesting if costume people could figure out how to give the reverse effect, where they dress someone up like they are a bad CGI. That would be pretty funny! (also it would make a great Halloween costume)

> [Max Headroom] was called "the first computer-generated TV personality", although the computer-generated appearance was achieved with an actor in prosthetic make-up and harsh lighting, in front of a blue screen

https://en.wikipedia.org/wiki/Max_Headroom

Some episodes of 'Love Death + Robots' (Netflix) have caused me to double-take as I didn't realise at first they were CGI.

There's a BTS video[1] about the animation on one episode, appears that it was a blend of both CGI + real faces.

[1]: https://www.youtube.com/watch?v=BWB507yaBXI

I have one: Ping Pong, the Japanese film from 2002. After watching the bonus features about how they developed the actors' individual playing styles, they revealed that the balls were entirely CGI animated. That let the actors focus on the precision of their movements, without worrying about actually volleying!

The balls looked a bit odd in the film, but I figured it was simply an artifact of 23fps film capturing a small plastic ball moving at high speed. I never would have guessed it was all CGI. I think I'd seen the movie three times before buying the DVD and watching the special features, years later. So I was fooled in the mid-00s and again in the late 10s.

Slightly off-topic, but still anecdotal how far CGI has come, I recently showed my dad who hasn't played any video games for at least 20 years this [0] video of the current MS Flight Simulator. It is produced in a documentary style, so there are no real cues to it being a game. After watching it together, my dad said: "Yeah that's a nice documentary about this difficult airport, but I thought you wanted to show me that new game?". You can imagine his face when I told him that we've in fact watched just that.

[0] https://www.youtube.com/watch?v=UtAF1I_H01w

With respect to your dad, how did he explain the shots outside of the airplane? Did he interpret that as a helicopter flying around the airplanes with a video camera?

Also, I'd like to point out that Microsoft has skillfully choosen scenes that are "easy" to create: airplanes with smooth shapes, buildings from long distances and low detail, landscapes that don't include movement (no grass moving in the wind or waves lapping on the shore). MS set them selves up for success here.

He most likely just didn't pay enough close attention or thought about it I guess.

And I agree that MS set themselves up for success with this type of scenery, it is fascinating how well the end result looks overall.

> If you follow the 'VFX artists react' series on YouTube you learn what give CGI humans away pretty easily.

For those who don’t want to watch all of it (even though it is quite excelent) the points I remember are:

- subsurface scattering of skin

- do the pores stretch? (As the face is moving)

- movement, especially movement of the upper lip as the character speaks

In the case of the nVidia presentation, it's quite obvious from the video where it's CGI Ceo: Awkward standing position, stiff arms and fairly inarticulate fingers. He looks fairly suspended, as if there's nothing to convey weight/physics in his movements... however brief.

> CGI Jensen looks like he came out of a 2009 videogame. Nvidia may have the tech but you can tell it lacks the artistic supervision of a proper VFX hous

That's part of the point. If computer software alone can replicate state of the art human effort from 12 years ago, that's what they're showing off. It's not "look how great this rendering is", it's "look how good the AI that made this is".

I don't think you have to watch anything to learn what gives CGI humans away. I don't think anyone has actually managed a truly convincing one yet. This is the closest I've seen:

https://www.youtube.com/watch?v=wrHXA2cSpNU

The coolest thing about that video is how good it is, but the second coolest thing about that video is the fact that Lucasfilm recently decided to hire the guy that made it.

https://www.cnet.com/news/lucasfilm-hires-deepfake-youtuber-...

I feel like Deepfakes really count as CGI human though. At least not in the sense being thrown around by Nvidia here

nVidia used deep fakes for this.

Rachel's appearance in Blade Runner 2049 was shockingly good to me. (https://www.youtube.com/watch?v=rDx3WwWR6y4) On the other hand, I though Tarkin looked pretty good in Rogue One, and for some people he's a corpse. I don't think the uncanny valley is as consistent among people as folks here are suggesting.

Yeah I thought Tarkin was not convincing (though still really impressive, especially for the time). I totally forgot about Blade Runner - I agree that was really good - better than Star Wars.

> so the end result looks like something their engineers put together in a last minute crunch (which is probably what happened)

They actually say that’s what happened in the video on it.

I'm pretty sure that's SOP for Nvidia conference demos.

Was in a breakout after the main Jensen presentation and the engineer was talking about how they were fiddling with K8s to get the demo to work right, the night before. And this was the money demo in the middle of the presentation.

On the other hand... I actually kind of like a company that half-asses demos. It suggests they're spending their time on more important things.

> If you follow the 'VFX artists react' series on YouTube

They should have given the work to those VFX guys, Corridor crew would have done something much better in just couple of days.

Reply on news.ycombinator.com