Their comparison with offset noise ( https://www.crosslabs.org//blog/diffusion-with-offset-noise https://www.crosslabs.org//blog/diffusion-with-offset-noise ) i

Their comparison with offset noise ( https://www.crosslabs.org//blog/diffusion-with-offset-noise https://www.crosslabs.org//blog/diffusion-with-offset-noise ) is kind of limited-- essentially dismissed as a hack since it breaks the diffusion conceptual model, but fails to argue that their rigorous approach actually works better.