Painting with Light is a user-guided interactive evolutionary system for daylighting design. It allows architects to use color to specify desired light levels in spaces, and searches for solutions that bring building geometry and materials close to performance targets desired by the architect. The proposed interface directly addresses a main limitation of generative design systems based on building performance metrics, by allowing the user to specify daylight spatial patterns with a high degree of granularity. This development poses new challenges for both objective function and penalty function definitions, addressed in this paper. Painting with Light automatically computes and displays statistical indexes that inform the user on the deviation error between the performance of the solution found by the system and the desired targets. The system was implemented in Python on top of a popular Computer-Aided Design software, Rhinoceros, as an add-on to its Visual Programming Language, Grasshopper. This choice of implementation allows the use of Grasshopper's built-in functions and methods for 3D parametric modeling, of tools that provide direct access to Radiance, a lighting simulation software, and of different types of genetic algorithms. Five experiments were conducted on a freeform parametric model for progressive system calibration, which encompassed four steps: 1) adjustment of painted targets to fit the problem feasible solution space; 2) devise appropriate weights and penalty factors for the fitness function; 3) test two different evolutionary solvers; 4) test the system's capability to find a predefined solution where the optimal values were known. After calibration, the system was able to produce solutions that closely approximate the painted goals.