RB-Modulation is a novel, training-free personalized diffusion model solution based on stochastic optimal control released by Google. It achieves precise extraction and control of style and content by terminal cost encoding, enabling the generation of images with a consistent style to the reference image and following given text prompts without additional training. This technology, without requiring training, maintains high fidelity to the reference image through its novel Attention Feature Aggregation (AFA) module and follows given prompts, holding significant research and application value.