Implementation of Removing/inserting low-level traces in images in the paper.

[Mwzz]

Thanks for a lot of valuable work presented in the paper Raising the Bar of AI-generated Image Detection with CLIP.
I find section 6. Beyond low-level forensic traces is very interesting, can you release the code of the relevant experiments. In particular, how to reconstruct an image using the VAE of SDXL, and inject low-level traces, is implemented?