How can we save pre-processed images as png - what is used during training

[FurkanGozukara]

I know --debug option but it is very very slow

Are there any option to what is being used in the training? Like when pre-processing images and prompts we can save final used image and the prompt?

Thank you @kohya-ss

[coudys]

that would be great to double check what is being used in the traiining, or perhaps an output of each image's resolution scaled down or cropped. I am gonna try --debug but to be able to save processed png files or "see" insode of pnz files to understand the latent representaion better

[kohya-ss]

It's certainly important to see the data that is actually being trained.

You said that the --debug (--debug_dataset) option is very slow, but how slow is that? Does it take a while for the first image to appear?