Even then, AI models (be it text or image) are generally unethically trained (i.e. without consent of the authors/artists of the training material) and have a significant energy consumption, even for single prompts.

And I do have to ask: To what degree is running your comment through an LLM actually beneficial? You say it improved readability, but how unreadable was your original comment actually, that it would require fixing via external tool?

For books, library genesis would be a better place to look than piratebay though.