<p>Additive manufacturing is increasingly adopted in industry, employing powder-based processes, resin curing, and material extrusion. Extrusion-based methods are usually less precise, however, the advantages are lower material and acquisition costs as well as a greater variety of materials used. In known extrusion processes, the material flow is set by the motion of the extrusion mechanism, such as an extruder screw, a filament drive gear, or a piston in syringe-based systems. Material flow and the resulting strand width are critical process variables, as they influence directly dimensional accuracy, surface quality, and mechanical performance of additively manufactured parts. To improve flow consistency and dimensional accuracy, a feedback controller is required to actively regulate the material flow and compensate for disturbances and nonlinearities. As sensors required to measure the volume flow are either too expensive or not technically feasible, control strategies based on direct feedback are not practical. Model-based control methods require either enough data for parameter estimation or the training of AI-based models. This work presents a low-cost approach to measuring and controlling material flow in FFF. Optical images and AI-supported image processing are used to determine the width of the extruded strand as a function of process parameters. The data obtained using this method is used for parameter identification of a lumped model that describes the FFF process. This model is later used by a Model Predictive Controller to calculate optimal speed profiles that compensate for the nonlinear material properties during extrusion. The advantage of this method is that it eliminates the need for complex parameterization in the slicer, such as print speeds and retraction parameters. Materials with nonlinear viscosity and difficult extrusion properties can benefit from this in particular. This method should be transferable to any extrusion process by adapting the model structure.</p>