Popular GitHub Repositories for Alignprop

AlignProp uses direct reward backpropogation for the alignment of large-scale text-to-image diffusion models. Our method is 25x more sample and compute efficient than reinforcement learning methods (PPO) for finetuning Stable Diffusion

Promptable, training-free, plug-and-play differentiable rewards from Vision-Language Models (VLMs) to guide image generation. Specify what/how to reward via prompts; backprop through the VLM with no fine-tuning. Includes anti-periodic loss functions that suppress the 16x16 pattern from image-token patching for cleaner textures.