Amateur AI seems to improve by the month

Amateur AI has a ways to go to get facial expressions and human body language right, but this kind of amateur video is improving in a hurry.  Seemingly by the month. 

How long can it be before professional unlimited-budget AI will look close to live action? Or will it be artificially held back to preserve acting and stunt roles for humans?

I used to think it was going to take decades more, but now I'm thinking a few years. Might happen in time for me to see it, or so I hope.

When it comes to our genre, lifelike AI will open the floodgates to an immense amount of fantastic video creations, which given our settings and characters and their superhuman abilities could be pretty wild. Limited only by imagination, and not the real world. 

Then, combine that with decent VR... the 2030's could be pretty wild.



Shadar

Friends and I have this discussion a lot Shadar. We've been following the development of AI since the beginning.

There are two main problems with AI and images/movies (text and writing is WAY easier and will almost certainly be perfected within a year or two. You can conceivably write a novel with AI now).

With drawings, its a little trickier. The first issue is creating consistent characters. The second is creating correct anatomy and interactions between characters. 

If you look at AI art, it tends to be fantastic at creating "Portrait" images. But trying to get it to create "action scenes" (in the sense of two people "fighting" as an example), tends to be way trickier.
Its good at one character, but not interactions between characters. 

The "creating consistent characters" is getting solved. Midjourney and SD can sort of do it with some trickery. But its not 100% yet as far as I can tell.

It will get there very soon.

The anatomy issue, there's a paper on how to correctly draw anatomy using AI published recently. This seems to fix the "six fingers" problem AI art encounters. Interactions between characters in an action type scene, I haven't seen that yet.

But I think 5 years and we will see huge leaps in AI art and movies.

Just my thoughts.

Peace.

/K

kikass2014 wrote:

Friends and I have this discussion a lot Shadar. We've been following the development of AI since the beginning.

There are two main problems with AI and images/movies (text and writing is WAY easier and will almost certainly be perfected within a year or two. You can conceivably write a novel with AI now).

With drawings, its a little trickier. The first issue is creating consistent characters. The second is creating correct anatomy and interactions between characters. 

If you look at AI art, it tends to be fantastic at creating "Portrait" images. But trying to get it to create "action scenes" (in the sense of two people "fighting" as an example), tends to be way trickier.
Its good at one character, but not interactions between characters. 

The "creating consistent characters" is getting solved. Midjourney and SD can sort of do it with some trickery. But its not 100% yet as far as I can tell.

It will get there very soon.

The anatomy issue, there's a paper on how to correctly draw anatomy using AI published recently. This seems to fix the "six fingers" problem AI art encounters. Interactions between characters in an action type scene, I haven't seen that yet.

But I think 5 years and we will see huge leaps in AI art and movies.

Just my thoughts.

Peace.

/K

Someone in the industry once described it to me as the 'body language problem'.  Humans who can read the nuances of body language had an evolutionary advantage, so modern humans are very, very good at it. Our brains are super-optimized for spotting falseness -- but only for other humans. We aren't particularly good at seeing subtle falseness in animals or SciFi or fantasy characters.

The trick is for an AI to learn body language using the same cues that our minds use -- and basically as seen through the limitations of our eyes. But it's hard to manage that kind of a deep learning process, given that we humans do most of that processing unconsciously. Most of us would have great trouble explaining someone's body language at the physical and kinetic level, and the linkages between face, eyes and body. 

Shadar