Few-shot spatial control for Text-to-Image Diffusion models by leveraging the analogy between query and support spatial conditions to construct task-specific control features.