Generative modeling for images and videos has largely shifted from diffusion models to flow matching models