1 paper · 1 filter
Jaa-Yeon Lee, Yeobin Hong, Taesung Kwon +1
Diffusion models generate highly realistic images but often struggle with precise text-image alignment. While recent post-training methods improve alignment using external rewards…