We propose a novel method to mitigate memorization in text-to-image diffusion models by adjusting the initial noise.