Mind sharing what you have tried? Have you considered training a diffusion model on pixel art, and then conditioning it on a 3D model?