Image Latent Auto Sort

This GUI will auto sort images into three category based of the following:

1. Latent Vector Magnitude (L2 norm) - relative to the black image

2. Variance in Latent Space (relative to the black image) - This should likely be a batch calculation rather then just a single calculation but this would require batch size scripting

3. Distance to the All-Black Latent Vector

It will move the images into one of three folders Low, Medium, High - It will move any .txt files and rename appropriately if they exist (If you have a large captioned dataset)

Use SD 1.5 Vae for images 512x512 and 768x768, and SDXL VAE for larger images.

It could use some safety checks for disk space and other cosmetic improvements, but to my knowledge it is the first tool of its kind.

Note: Low would not necessarily mean low image quality.

The Low images may converge faster then high and actually require less steps.

One thought is to use the High images for high batch, low res training, and the low images for high resolution "finetuning" of a model or lora.

Example for SD 1.5: High Folder set to 256x256 batch size of 16, Medium Folder 512x512 Batch size 4, Low Folder 768x768 batch size 2