In some of the images on the 500-image subsample we can see checkerboard patterns. We could remove them using this trick.
Not as high a priority as getting generated images across all images / just seeing if it goes away after enough iterations. But could potentially lead to better images faster.
In some of the images on the 500-image subsample we can see checkerboard patterns. We could remove them using this trick.
Not as high a priority as getting generated images across all images / just seeing if it goes away after enough iterations. But could potentially lead to better images faster.