Tim and I have two ideas. First, we have a small box containing sand, clay, blocks, or some other familiar, plastic medium. A Kinect watches the box from overhead and gets a heightmap of whatever the participant creates. A program then renders a natural-looking landscape from the terrain, possibly including flora and fauna, which is projected near the sandbox.

Second, I Kinect watches a crowd of people and a program generates thought balloons above their heads. Thoughts would be selected based on observable qualities of each person, like height, gesture, and proximity to other people. Thoughts could be humorous or invasive.

Click for large.