They spend 2 millions on character model plus rigging now? I'm shocked. That's 10 times more than i would have assumed.
But even with such budgets, it's still not 'easy' to improve, i'm sure.
Modeling the interior of a body is much more work than modeling just the surface. Like 10 or 100 times more.
But the larger problem is at first to develop a solution practical for realtime. To see if a potential method works, you need to create the muscle rig for it.
But we don't know yet which method might work, so we also don't know what the muscle rig should be.
Many open questions, e.g.: Can we model all or most muscles with straight lines, or do we need to simulate / fake how muscles slide over bones?
How do we model the volume of a muscle? It must be something very simple, so skin collision keeps cheap.
How do we model fascia connecting adjacent muscle strands, and do we even have to?
Do we need some form of simulation, or can we get away without it?
That's all open questions, and expensive methods from offline rendering are barely more worth than some inspiration.
So we need to try out a lot of options, and for each we need artwork, including the training time artists need to get used to the experimental method. Also artists need to optimize and figure out how to get the most out of it.
We rely on good feedback between devs and artists, bringing in personal bias and misconceptions.
Ideally, the dev researching the methods should be an artist too, to avoid this. But to get anatomy right, you need to be a very skilled artist.
I guess that's the primary reason why we have not seen any technological progress at all. The only news i remember was the introduction of dual quaternion skinning, which - contrary to some beliefs - does nothing towards better anatomy.
No wonder people now look at ML. Maybe that's the future standard. But the fact Epic used a muscular mans back to show it off, which is not even a typical failure case, does not make me too optimistic.
Till yet all progress came from the art side. They set up extra bones driven from simple procedural constraints, and within the last decade results became really pretty good. (Interestingly there is no presentations or talks about this topic - seems industry secrets built on decades of personal artist experiences.)
