Purpose of this code is to try to capture cross-attention maps for SD with the intention of being able to replicate Alibaba's Cross Attention Map injection technique whereby we are able to directly edit images through the stable diffusion process, enabling for spatial regions to remain consistent while being edited.