A data-driven model that predicatively generates photorealistic RGB images of dough surface browning is proposed. This model was validated in a practical application using a CO2 laser dough browning pipeline, thus confirming that it can be employed to characterize visual appearance of browned samples, such as surface color and patterns. A supervised deep generative network takes laser speed, laser energy flux, and dough moisture as an input and outputs an image (of 64 x 64 pixel size) of laser-browned dough. Image generation is achieved by nonlinearly interpolating high-dimensional training data. The proposed prediction framework contributes to the development of computer-aided design (CAD) software for food processing techniques by creating more accurate photorealistic models.