The growing number of laser welding applications from automobile production to micro mechanics require fast and reliable process control systems. The high process dynamics in time, space and intensity, especially in scanner based remote welding or high speed micro welding, demand extremely fast and spatially resolved in-process control systems to create closed loop control for error prevention and correction. Today's conventional micro processor based image processing architectures (as used for example in [1]) are not able to provide the high frame rates needed for the real-time closed loop control of high speed laser welding. With "Cellular Neural Networks" (CNN) it is possible to implement Single-Instruction-Multiple-Data (SIMD)-architectures in the electronic circuitry of each pixel of the camera chip itself in order to produce a so called Focal Plane Processor (FPP). Such pixel parallel systems provide extremely fast real-time image processing. With these new CNN-cameras it is now possible to implement a camera based high speed in-process control system for laser welding that enables closed loop control of various quality features. With a multi modal diagnostic approach we were able to identify direct and explicit image attributes for a variety of quality features as a base for the process control. It could be shown that closed loop control of the "full- penetration" quality feature is possible with frame rates of 10 kHz and beyond with a CNN-camera system.