Yes we are edging into tool-land. Mostly still toys though. That’s my conclusion after playing with Gen-3 alpha a little.
However I can now conceive of using this technology to create certain types of shots. To find out which, read on.
Mostly, while we are looking at yet another leap forward in quality and consistency there are still many artifacts that scream ‘AI-generated’ and make it unusable for professional production purposes.
Also in alpha it isn’t directable as it is in Gen-2 (no motion brush or camera controls though better prompting). Obviously shows a lot of promise but… well it’s still a toy.
Some examples.
Well for starters I tried some establishing shots. These seem to me to be the first point of entry for synthetic video. We are so close.
‘An establishing shot of the Denver Museum of Art at Christmas time.’
You have to look for the artifacts, but when you see them you see them…
‘An establishing shot of the Island of Santorini from the perspective of a fishing boat.’
We’re almost at the point where there are few enough artifacts that one might be able to salvage the shot with some infilling. Or perhaps with enough iterations something actually usable might get created. This is 90%, Super impressive to be sure. But until it is close to 100%…. Getting to 100% is hard. I know.
That water isn’t bad though. I wonder…
‘a tsunami careening through a small Parisian street’
Not bad actually. I feel like with enough iterating you might be able to put together a respectable disaster movie destruction sequence (as long as it doesn’t include people).
Speaking of people and action, let’s try for something which is guaranteed to be a disaster.
A human… skateboarding…
‘An action shot of a skateboarding teenage hacker girl desperately holding on to a futuristic taxi as it rounds a bend, her skateboard wheels shooting sparks as she grinds the turn. Anime style.’
Yuck.
How about a shot from a period battle sequence?
‘An aerial shot of a massive melee As Spartans fight Persians in hand to hand combat.’
Not there yet but man this is going to be so fun in a year or two.
Maybe photorealism is a lot to ask. What if we try animated styles?
‘A CU of a teenage homeless girl crying, overwhelmed, CAMERA PULLS BACK DRAMATICALLY TO REVEAL she is in a vast futuristic cyberpunk Vancouver teeming with people, traffic, bots, and drones. Animated hand-drawn anime style.’
Interesting…. When this gets controllable we may be onto something here. Especially if we can solve character/wardrobe/set consistency. Probably less than a year.
And then there’s other creative visualizations. Things we don’t really know what they’re supposed to look like so we can get away with stuff.
For instance this cool title card.
Or how about this black hole?
Or this visualization of mitochondria inside a human cell.
I think we may have encountered some actual use cases guys. We have begun to make our way across that uncanny valley. And what an amazing land lies there.