Game level design is a time-consuming process that often requires multiple iterations ofplaytesting and refinement. This can pose challenges for smaller development teams withlimited resources. This thesis investigates how a Conditional Generative Adversarial Network(CGAN) combined with a Human-In-The-Loop (HITL) framework can support game leveldevelopment within the Godot game engine, reducing manual effort required in level design. A HITL framework is designed and implemented, in which developers can iterativelyguide the system by selecting generated candidate levels, which are incorporated into thetraining data for subsequent generations. The framework generates two-dimensional platformer levels by conditioning the model on previously generated level segments, producing locally coherent and structurally connected levels. Automated evaluation methods forplayability, difficulty and linearity are implemented to filter generated levels and assess thegenerative capabilities of the system across iterations.The systems ability to follow feedback was tested by an automated process, whichemulated developer feedback in selecting levels for iterations of the system. The tests weredivided into four different cases, where the generated levels linearity and difficulty metricwas requested to be either high or low. Two different datasets were used, each with adifferent set of tiles and levels, resulting in eight tests in total. Only playable levels wereconsidered during this process. The results indicate that the system can adapt to feedbackrelating to the linearity and difficulty, but with a high dependency on the used dataset.