digitizing
modelling
texturing
programming
sound
setup
base mesh
creating the mesh involved several steps. based on the photographs and the 3d scan of the head a basic mesh was created in softimage|xsiusing the subdivision modelling procedure. this process was on the one hand constrained by the maximum number of vertices (~ 30.000) we could handle during realtime rendering. on the other hand it was very important to distribute the vertices well, keeping in mind all of the other poses of the face that would be created out of the base mesh. since we are using the different meshes for morphing it was important that the resulting subdivision models not only had the same number of vertices but they also needed to be in the same order for all morph targets.
where the 3d scan did not provide enough information (ears, open eyelids, nostrils, facial features) we had to fit the geometry on photographs via rotoscoping.
eyes
initially we didn't want to waste a morph target for making the eyes blink because we thought that we would be limited in the total number of morph targets we could handle. therefore we modelled the eye portion including lashes as four separate morph targets. these had to fit seamlessly onto the base mesh regarding geometry and texturing. of course now blinking was only possible with poses that didn't change the eye portion. when the eye portion was changed our extra modelled eyes would no longer fit. so in that case it was just hidden and blinking was no longer possible. in retrospect an additional morph target with closed eyes would have been simpler to integrate. in the end we had 14 poses and one more for the eyes would not have made a difference.
our original plan was to further refine the geometry for each pose individually and apply the additional geometrical information as normalmaps on the lower resolution poses. so we wouldn't have to exceed the number of vertices of our base mesh, and at the same time we would have not been restricted in using different geometries for higher resolution versions of each pose. but in the end we abandoned this plan for performance reasons (loading a high resolution normalmap for the 14 poses was not feasible) and for the benefit of other effects. after texturing the base mesh it looked detailed enough anyway, and there were still other problems that we had to face.
> texturing
creating the mesh involved several steps. based on the photographs and the 3d scan of the head a basic mesh was created in softimage|xsiusing the subdivision modelling procedure. this process was on the one hand constrained by the maximum number of vertices (~ 30.000) we could handle during realtime rendering. on the other hand it was very important to distribute the vertices well, keeping in mind all of the other poses of the face that would be created out of the base mesh. since we are using the different meshes for morphing it was important that the resulting subdivision models not only had the same number of vertices but they also needed to be in the same order for all morph targets.
where the 3d scan did not provide enough information (ears, open eyelids, nostrils, facial features) we had to fit the geometry on photographs via rotoscoping.
eyes
initially we didn't want to waste a morph target for making the eyes blink because we thought that we would be limited in the total number of morph targets we could handle. therefore we modelled the eye portion including lashes as four separate morph targets. these had to fit seamlessly onto the base mesh regarding geometry and texturing. of course now blinking was only possible with poses that didn't change the eye portion. when the eye portion was changed our extra modelled eyes would no longer fit. so in that case it was just hidden and blinking was no longer possible. in retrospect an additional morph target with closed eyes would have been simpler to integrate. in the end we had 14 poses and one more for the eyes would not have made a difference.
our original plan was to further refine the geometry for each pose individually and apply the additional geometrical information as normalmaps on the lower resolution poses. so we wouldn't have to exceed the number of vertices of our base mesh, and at the same time we would have not been restricted in using different geometries for higher resolution versions of each pose. but in the end we abandoned this plan for performance reasons (loading a high resolution normalmap for the 14 poses was not feasible) and for the benefit of other effects. after texturing the base mesh it looked detailed enough anyway, and there were still other problems that we had to face.
> texturing