andyfilms1

I would imagine that performance is coming from essentially dedicating 3x context windows to one task, rather than any particular inter-agent magic.

That's not really what we're talking about though. Compiling verilog from simulink or whatever is totally different than an agent writing code.

If you design a custom piece of furniture, then hire someone to build it, that's totally fine! But if you then tell everyone "I constructed this cool table," well, that's just not true.

I love living in the future!

It's an unfortunate truth that there is the right way to do things, and then there is the way they have to be...

I was working on a project recently where I wanted to express a relationship (that I knew existed, but didn't know how to express) between four measured scalar values. Astra insisted there was no relationship, and that any correlation wouldn't make sense.

Eventually, by walking through them, it proposed an additional fifth value and from there was able to tie everything together.

Sometimes you just gotta hit the machine until it works again.

We finally have PADDs from Star Trek!

It feels like 50% of AI "progress" is just finding different ways to say "tell the model things in English."

> Stagecraft is just one implementation but you don't think that the Unreal 3D worlds that are used behind it are beginning to be generated with AI?

They are not. Source: I have run one of these LED volume studios for 4 years.

> Color management and maintaining context across multiple renders is actually the conversation we were talking about and the software that he is using has mechanisms to define context across a rendered shots and even across multiple projects. He just wasn't using them as it's an alien concept to him. My point is all of that is possible and solved and will become easier as the market matures.

These problems are solved...ACES, DRT, FDL, etc. exist, are open, and AI companies are welcome to utilize them. But for some reason, they don't (or can't.)

> I also have a friend who was significantly involved in Odyssey and I think the reason why so many films are made with film at the moment is because the industry is clinging on to what it believes is authentic when being pushed towards a commoditized race to the bottom.

The reason people are still shooting on film is because it forces you to slow down and think about every decision, because mistakes become very costly. The decision to shoot on film objectively does not make technical sense, and studios would much prefer people shoot on cheaper digital cameras. But filmmakers still use it because of the specific creative mindset it requires of you.

> The end goal of all of this is actually customized content where the film that I see and the film that you see is different because it's rendered to our tastes. There's a whole school of thought there around a collective experience versus an individual experience, but the technology will allow it if it's completely digitally created and one would assume that will be where it will go.

The end goal...for who? Totally fine if this is something you want and interests you, but unless you have a time machine, this is just speculation.

With a few notable exceptions (James Cameron) the film industry is largely unconcerned with "possibility." I think this is often why there is such a big misunderstanding from tech folks about how movies are made, since tech is very feature and tool driven whereas filmmaking is very much process-driven.

The Odyssey was shot on IMAX cameras half the size of a car that weighed 400lbs. It's easy to call that out as being ridiculous, surely a much smaller camera system could have achieved 99% the same results, etc. but that totally misses the point. Nolan could have shot it on anything he wanted, and he chose the less convenient way, because that is his process. If he was forced to use a more reasonable camera, it would not have been a Christopher Nolan film.

Sure, I'm down with that mix of generation + animation + mocap workflow. But at some point, people will going to look around at all the equipment, the people in mocap suits, the thousands of hours of generation and the millions of dollars of compute, and just wonder...wouldn't it be easier to just film someone with a real camera?

Sure, but saying that AI is the magic bullet to solve this is completely speculative. If and when a fully GenAI film is released and performs well at the box office, than I'm sure the industry will align itself to the new direction. But pretending like that situation is an inevitability is just wishful thinking...any number of things could, or could not happen.

Man, this is so misguided....

1. Stagecraft/virtual production is a thing, yes, but the backgrounds are not "AI rendered," they're 3D scenes built in Unreal. They have to be, because they need to dynamically react to the camera moving to create parallax.

2. Yes, some people are using AI to plan scenes. With respect to your relative, those people are not the ones getting hired on large projects. When someone is hired in the film industry, they're hired for their specific taste, creative eye, and opinions. That contribution is literally their whole job. Outsourcing that task to AI is, within the industry, seen as an indication that someone is inexperienced or simply bad at their job. Crews don't want to work for those people, and audiences don't want to watch content created that way.

3. It's easy to "expect" that "one day" AI might become the way movies are captured if you don't actually understand the limitations of AI video generation versus the demands of filmmakers and the expectations of audiences. Putting aside the obvious stuff like continuity, shot duration, and resolution...GenAI is not color managed, it's not physically accurate, the results can't be copyrighted and it struggles with anything not in its training data (I.E., genuinely new ideas.)

This is NOT to say that this technology has no future in the film industry, obviously it does. But half the movies at the Oscars last year were shot on film, so the notion that all filmmaking is just going to turn into pixel soup being generated from prompts is just hilarious to everyone actually within the industry.

This is good general advice. If your software needs a "tour," that's a good indicator you should work on your UX instead. If you're just excited to tell the user about a new feature, they'll discover it when they need it.

Oh, so mass theft is okay as long as American companies are doing it