backpropagation

The paper presents AlignProp, a method for aligning text-to-image diffusion models to desired reward functions via backpropagation. These models are useful in image generation but controlling their behavior for specific…