There is a bias in the inference pipeline of most diffusion models. This bias arises from a signal leak whose distribution deviates from the noise distribution, creating a discrepancy between training and inference processes. We demonstrate that this signal-leak bias is particularly significant when models are tuned to a specific style, causing sub-optimal style matching. Recent research tries to avoid the signal leakage during training. We instead show how we can exploit this signal-leak bias in existing diffusion models to allow more control over the generated images. This enables us to generate images with more varied brightness, and images that better match a desired style or color. By modeling the distribution of the signal leak in the spatial frequency and pixel domains, and including a signal leak in the initial latent, we generate images that better match expected results without any additional training.
@article{arxiv.2309.15842,
title = {Exploiting the Signal-Leak Bias in Diffusion Models},
author = {Martin Nicolas Everaert and Athanasios Fitsios and Marco Bocchio and Sami Arpa and Sabine Süsstrunk and Radhakrishna Achanta},
journal= {arXiv preprint arXiv:2309.15842},
year = {2023}
}