You define your scene in terms of nested frames. Each frame is defined with normalized coordinates inside its parent. Initially you render from the top-level fr

You define your scene in terms of nested frames. Each frame is defined with normalized coordinates inside its parent. Initially you render from the top-level frame, but as you zoom into some other frame, you switch to rendering from that frame. This avoids numerical issues, no matter how far you zoom, because the view coordinates are reset each time you switch to a new render root. When rendering descendant frames, you should stop when something becomes smaller than a pixel. Normally your frame graph would be a tree. But you can add cycles to your frame graph, to achieve weird effects, like getting back to the macro scale after zooming in to nano scale, etc.