Our great sponsors
-
WorkOS
The modern identity platform for B2B SaaS. The APIs are flexible and easy-to-use, supporting authentication, user identity, and complex enterprise features like SSO and SCIM provisioning.
Let's look at this from a perspective of Shannon's Information Theory. Cinema is a double tranmissive system. First, the world has things & shapes: it is information. It transmit / sends information about itself via light, which bounces off it and scatters or bounces. This travels through first an air/liquid/vacuum medium (distorting in some cases) and then the lens's optical medium. Then it impacts either a shutter (blocking the light) or if the shutter is open a frame of film, which is actually a lot of independent little film grains on a transmissive medium. Ok, we have now received the information, and the shutter closes and advances to the next frame, to repeat another reception.
Film is kind of interesting because the process of getting the information isn't done there. We also have to re-broadcast the film out, but honestly, that part is kind of boring: shine light through the developed film and it attenuates some parts of the light more than others, reproducing the information encoded on developed film quite directly & without loss.
So far, this has all been modelled pretty well by this project. We have fancy lens optics, reproducing the light-capture system of a camera. What's missing / un-canny valley so far is that the virtual world is usually a fairly poor facimile of the real world. The modelling straight up isn't as good. How things animate and move lack a subtlty of complex motion that real bodies in motion carry. There's a host of small issues around how light interacts/bounces off subjects that we don't model well in Blender or most systems: subsurface scattering effects aren't as fancy as they could be, the physical based rendering models aren't complex enough, the air itself as as a medium isn't well modelled. There's a huge combo of things the virtual worlds aren't as good at as the real world, and there's so many behaviors and nuances of things in the real world that virtual worlds usually don't capture as well. This largely defines the uncanny valley.
But, just to throw a little more fuel on the fire: this project also is missing another step in cinema that I skipped above. I don't think this is where the uncanny valley problem is, but I think it's a pretty sizable difference between film and digital cinema. Film has another tranmission process that I didn't describe above!
So, we've shot our movie. Now what? Well, we develop the film. What is developing? Well, we emerse the film in an activation bath to develop the exposed silver-halide crystals better known as film grains. There's information trapped in these crystals, they're at a certain state, and we have a chemical process which sends this information out, through a medium. The medium is the chemical developer, which turns the exposure into developed film grain, which is the received information from this system.
One of the really crazy things to me is that developing film is not at all like reading exposure values off a digital sensor. Because the process happens over time chemically, and the process itself is actively consuming the film developer as it works, which creates little local pockets where there's less developer. The process is non-linear. A heavily exposed scene will consume the developer and reduce further development speed not just for that film grain, but for the area around it.
Again, this isn't the uncanny valley problem. But it's still something missing from digital cinema, from this effort, that makes it substantially different from film cinema. There's projects like Filmulator https://filmulator.org/ that I love and adore which can simulate chemical development of film from RAW images. I'd love to see Virtual Blender Camera team up with efforts like these, to create a more genuine film-cinema feel, that models more than just the optical capture systems.