If it's an actual person's face that you've shot, reshoot. You could try to fake it, but that would require way more time/money than a reshoot.
You would be getting a 3d model of a head (that's at least fairly similar to your talent's), motion tracking the talent's movement exactly in 3d (using PF Track or similar), projecting light onto the 3d model, and compositing that over his/her face in the footage.
I mean, depending on how the movement is in your shot, you might be able to fake it with some distortion effects on a precomp to simulate the lights, but dang.
Oooh! Alternate thought, you might be able to fake it by using the face to drive a displacement map effect on the precomp with the "light".
Kubrick's effects team, Con Pederson, Wally Veevers, and Douglas Trumbull, knew a few things. Projecting 8mm and 16mm films into the set required creating the films of the readouts AND synching the projectors with the 70mm shooting cameras. There is an apocryphal story about fights over whether or not the projections should be readable or reversed overshadowed by the fact that such projections of the readouts probably would not even be visible in a real world cockpit like that; they wold just be glows.
We tried this is high school with an 8mm projector but, back then, we did not have a Plumbicon studio camera with enough sensitivity to register the images. Looked cool though and I used the techniques later in my careers as a stage lighting designer and light show purveyor for rock bands.
Chase Chick wrote:
"I realize that I could have done this practically, but too late now. Anybody have any ideas?"
Sure. let's start with a question:
Which would you rather do?
Spend weeks -- literally weeks -- learning the techniques necessary to salvage the shot, plus days -- literally days -- executing those techniques..... or re-shooting and getting it right to begin with?
I'm not joking about those time estimates, either. You're looking at creating a dead-solid-perfect 3D model of the subject's face. Then animating that face to precisely match the motion of the subject you already shot. From a distance, position and lens focal length of the footage you already shot. Oh, and precisely matching any camera movement inherent in the footage you already shot. Then projecting the graphics onto said face and rendering it to a footage file. Once you've got it, it's a simple matter to match the two pieces of footage and use an AE blend mode for compositing.
Suddenly, "it's too late now" is starting to look like a lame excuse.
We await your response with breathless anticipation.
Welp, I just ended up tracking my eyeball and projecting that way. A couple times I just let it get sloppy and let it sit where it sat. I ended up making a spider web and luma matting it over my face (or one of the mattes, can't remember)
Since you guys were so awesome and are awaiting with breathless anticipation and all, here is the finished.
Fortunately this wasn't for a client, it was a quickie deal for a contest. I didn't win, but had fun doing it anyway.
Thank you for following up! Too bad you didn't win, but that looked like a fun project to work on.