The geometric buffer (G-buffer) is a well-known approach to implement image-based rendering algorithms. We propose a framework that maps the G-buffer concept and associated image-space operations to the graphics processing unit (GPU). This GPU-G-buffer (G²-buffer) framework consists of two major components: first, a texture-based representation of the G-buffer attributes; second, an implementation of the image-space operations by fragment programs. The G-buffer setup and the actual rendering algorithm are described in a plain text file, whose syntax is an extended version of the Effect Files in DirectX or CgFX. Our approach provides a fast and easy method to implement G-buffer algorithms that automatically exploit the GPU with its high memory bandwidth and processing power. We demonstrate the usage of the G²-buffer for a couple of applications in non-photorealistic rendering.