CLIP ViT-L

StitchVM framework improves diffusion model alignment efficiency

Researchers have developed StitchVM, a novel framework for aligning diffusion models with specific rewards like prompt fidelity. This method efficiently transfers reward models trained on clean images to handle noisy in…