As mentioned (somewhere, I think…), the base mesh is based on front and profile composite photos, in an effort to arrive at something of a neutral average. Inevitably something was lost in the translation to 3D, and this has been throwing off my ability to discern whether the textures and surface details are holding up.
Since I don’t have access to scanners or a multitude of willing subjects to scan, the only alternative I could think of was to make a multitude of individual likenesses, each hand-matched to photos from multiple angles, and blended together into a true 3D composite. To get a relatively unbiased result, I determined this would require at least three males and three females for each of the three races. Without access to camera information, pupillary distance or any other measurements, even one likeness would be a chore let alone 18 of them. Only a nutcase would put themselves through that kind of gruelling ordeal.
Several weeks later…
Despite models and celebrities being the most logical candidates due to the abundance of photos available, I spent half the time hunting for decent close-ups (profiles in particular) with the same relaxed, neutral expression. A preference for ones with close to “ideal” proportions reduced my options even further.
Once I had found enough photos to proceed with a likeness, I attached them as translucent Image empties to cameras, and positioned them in the ballpark of where the photo might have been taken, using the eyes for calibration. Then I’d start sculpting, moving from camera to camera as matching progressed, correcting their positions and focal lengths as the accuracy improved.
Once they were all done I had to figure out a formula for blending them, which turned out to be as simple as mixing the first two 50/50, then each consecutive one by the amount they contribute to the overall pool (0.333 for the third, 0.25 for the fourth and so on).
It was indeed an arduous, frustrating task, but for movie/fashion projects in particular I think this might actually have been an even better idea than using scans, since the base mesh now has top actors and models in its DNA…