To fit product photos into 3d-surroundings requires a number of specific operations. To begin with, a rough 3d model of the respective product needs to be built which aids in determining the photo's perspective. If the product's dimensions are not known — as was the case here —, they have to be assessed on the basis of the photo.
Next, the virtual 3d camera needs to be positioned in the exact same angle as the photo camera had been towards the real object. In film productions, this problem is solved by digital motion capture. In photo productions, such complex data is usually not available.
Therefore, in order to perspectively adapt the virtual to the real camera positon, one has to rely on estimation, just as in freehand drawing.
Once the 3d model is congruent with the photographed product, the objects and spaces surrounding it can be modeled. Regardless of how the additional virtual objects are shaped or arranged — perspectively, they will always perfectly fit the camera angle which has previously been adapted to the product photo.
The product's 3d model doesn't just aid in assessing the correct perspective. Its second function is to project shadows and reflexes exactly matching the product photo into the virtual space, thereby emulating the photo's lighting situation.
Even by advertising standards, the Audi accessories project had to be executed under extraordinary pressure of time. More than 150 pictures were to be produced in under 50 days. That meant: three times a day objects had to be modeled, 3d cameras positioned, surroundings designed, modeled, lighted, and, finally, renderings and photos composited.
Even though designs and renderings may appear a bit dated by now — the project dates back to the year 2004 — in principle, the production process would still be the same today.