The 3D effects involved were 'characters'. They needed the same attributes that live action needed; size, shape, color, origin, motivation, quirks, even personalities had to be thought of.
After this, the cameracrew needed the data on lighting preparation for the 3D. Audio had to be planned also according to the estimated animation-data. Everybody needed to write down notes, to help CGIcrew later to understand the logic of the 'yet-unseen-3D'.
The shooting was quite straightforward, ironically much like a commercial production. Despite the preplanning, all the shots (which amounted to close 100) needed to be checked with imaginary counteractors (3D). The leading actor had to concentrate on crosses on the walls etc. to 'fake' the interaction with future 3D. He pulled it off magnificiently.
After the laboratory and scanning the film to Beta, all the crucial images were transferred onto Silicons harddisks. The size of storagespace is always a problem, even with PAL resolution. So we started modeling our 3D against just one PAL (720x576) image. The image had letterboxing so the actual imagesize was 697x436 pixels.
The process of modeling was administered by our Visual FX Coordinator. All models tried to maintain the same visual kinship to each other. This artistic choice required great care in detail, so NURBS modeling, Metaballs and patches were utilized to the full. Luckily the project did not force us to do detailed interiors, just the models and animation-data. The lights that were used on the sets were copied to the scenes in animation with good results. Cameras posed some inconsistencies, while being copied. Again, since we had no images that needed tracking, we just searched for the vanishingpoint in each image and pasted the camera interest there. That solved problems there. The lens data, camera position, fov's and dof's came directly from the shooting sheets of the camera assistant.
The 3D animation times are quite short in Touch. The longest single sequence is 440 frames and most fall between 60-120 frames. This tech- nique of slicing the animations (motivated by the original cuts, naturally) helps us to make videopreviws faster. For PAL resolution even Mental Ray can produce shadowed animations reasonably fast. We actually noticed, that with our strategies with lighting, no camera movement and carefully planned cutting we could make highly sophisticated models and animation, which rendered faster that the customary animation. The only bottleneck to videopreviws was compositing (thru alpha or blue_chroma) and ethernet to Accom.
In all, the animatics were done before the deadline. Currently (11.3.97) we are beginning to render the final PAL images, with correct shadow-data and textures on the models. The deadline for the PAL (with mixed sound & music) is second week of April.
Then we'll (hopefully) have already the scanned HiRes (2k) pictures
from the filmnegative to start working with. The technique is
similar that'll we used so far, except the images are quite a bit
larger and therefore more cumbersome to work with. The process
of filmimage rendering and compositing is very difficult to do
accurately, but that is our aim. Since this is the first HiRes
production done in Finland, we have the necessary motivation for it.