https://arxiv.org/abs/2210.00939
Demos:
Bilinear interpolation: Attention mask is interpolated using bilinear method, resulting sharper image
Attention target: Choose the block Attention mask would apply to, dynamic means depending on noise sigma value
Base resolution: Change attention resolution scaling, set it to 0 to ignore this setting
Smooth Vectors: inspired by Smooth Energy Guidance, we smoothed and applied median blur on QKV vectors first to make latents cleaner, set it to 0 to disable it
Perturbed Vectors: inspired by Perturbed-Attention Guidance, keep it at low scale (around 0.05~0.6)
Perturbed Tokens: inspired by Token-Perturbation-Guidance, keep it at low scale (around 0.05~0.6)
Tried PLADIS guidance but it need pass tensor through entmax and it either consume a lot memory or time, also it change the result bigly