Our research proposes a step towards generating levels mainly focused on player's experience, to help reach a high level of perceived diversity in procedural level generation. For a given gameplay, we want an algorithm that generates levels with enough structural quality, while optimizing mainly for a specific play experience. Then, diversity should be reached by searching for levels that are far apart in play experience space. To do so, we propose Genetic WFC, a way to mix genetic optimisation with the Wave Function Collapse algorithm, and describe the synthetic player used to evaluate each level's fitness.