Hacker News Reader

Comment by extension | Hacker News Reader

Do we even know how to render reality? If we can do it at 1/2000th of real time then somebody should have made a prerendered animation that looks real by now, but I've never seen one. And if we don't know how to do it, how do we know what GPU we will need?

View on HN

We know the rough bounds of the technical parameters of a GPU good enough to fool a human. We don't yet know, for a large variety of cases, how to generate artwork that good in any better way than taking a snapshot of something real.

That being said:

> somebody should have made a prerendered animation that looks real by now, but I've never seen one

How do you know? A friend of mine works on visual effects, and his show-reel has quite a few shots where, if you didn't have before and after shots side by side, you wouldn't know that a computer had been involved at all.

> How do you know? A friend of mine works on visual effects

Any chance there's a link to it online? I'd love to see an example :)

Anyhow, the problem for me is less with still renderings, and more with natural motion, like a human walking or talking. Also, I'd love to see a realistic splash as something falls into the water.

I can't find the effects reel for that modern CSI+disasters show I was looking for... You'll have to settle for this: http://youtu.be/aFHKwaW4Um8

Thanks. I enjoyed watching that.

It looks like we're slowly getting there. See the Luminous engine, for example:

http://www.digitaltrends.com/gaming/square-enixs-new-gaming-...

Thanks for the video link. While definitely an improvement over what we currently have, it's an empty environment with literally 3 patterns and 1 object. Color me unimpressed.

Some film CGI (which renders much less than realtime) is getting pretty close to realism, to the extent that for certain snippets you can probably fool a large proportion of people, especially people who aren't computer graphics experts, and therefore don't know what specifically to look for. Most work seems to be focusing on very specific tough cases, like cloth animation. It'd be interesting to see an actual survey of what people can tell apart, though.

With film CGI one also has to take into account that it's highly polished by humans. To be able to reproduce that quality in an automated fashion will likely be much more taxing.

Reply on news.ycombinator.com