Energy-Based Cross Attention for Bayesian Context Update in Text-to-Image Diffusion Models Geon Y eong Park 1 Jeongsol Kim 1 Beomsu Kim 2 Sang Wan Lee 1,2,3