Generate Subgoal Images Before Act: Unlocking the Chain-of-Thought Reasoning in Diffusion Model for Robot Manipulation with Multimodal Prompts | IEEE Conference Publication | IEEE Xplore