Exploiting the Signal-Leak Bias in Diffusion Models
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.
2309.15842.pdf
preprint
openaccess
n/a
3.13 MB
Adobe PDF
80194b282b4e7da694e1457f2b1022bb
supplementary.pdf
preprint
openaccess
n/a
17.46 MB
Adobe PDF
a8b0d4d4257bf19f19218261ecc45af1