Anime NLP tagging update. Script is ready. It somewhat works but the results are pretty inaccurate. I assume it's because it has not been properly tuned on anime images but that would be hard to do as anime images are not nlp-captioned most of the time. In any case, I am eager to be done tagging 5k images and try to train a lora as a proof of concept to see if some sanity comes to NAI models.
Example of being close but not really:
https://files.catbox.moe/mscbjc.png