@NeogenArt Link to the tool is attached as hyperlink in message above, but here is the direct link just in case - https://civitai.com/models/249721 . Regarding model, when you train model using base Illustrious XL it doesn't mean that you will need to use it with that model, I personally haven't generated any images with base Illustrious XL even though all my models trained using it, you will still use your LORA with WAI NSFW, but it will work better with other checkpoints too, because all checkpoints based on Illustrious XL 0.1.

Also this info might be helpful to you, It is the message I sent to person who asked me for advice with LORA training:

"By the way, these are good resources I used to learn how to train LORAs: https://civitai.com/articles/3522/valstrixs-crash-course-guide-to-lora-and-lycoris-training https://civitai.com/articles/138/making-a-lora-is-like-baking-a-cake https://civitai.com/articles/3105/essential-to-advanced-guide-to-training-a-lora https://rentry.org/59xed3 These are the programs I use for dataset creation: https://github.com/mifi/lossless-cut - for capturing frames from anime https://www.xnview.com/en/xnviewmp/ - to quickly crop dozens/hundreds of images. I also use it to look at my generated images, it's also far better at browsing folders with thousands of images than default Windows Explorer (Explorer sometimes takes several minutes before I can sort images, with such big folders) https://github.com/HaW-Tagger/HWtagger - dataset tagging, relatively big learning curve, but far better than anything else, I don't think I would be able to tag big datasets without it. Also uses hardware accelerated tagging, so it can tag hundreds of images under a minute. (I recommend WD EVA02-Large Tagger v3 tagger it's current SOTA), you can skip CUDA toolkit and CUDNN instillation, https://github.com/1038lab/ComfyUI-RMBG - currently the best background remover, useful if your dataset has the same annoying background to give some diversity "Remove tool" in Photoshop is pretty good at removing unwanted details."

Also what I've recently discovered, but used only in my last LORA is that you can combine booru tags into single one and it appears to improve performance of LORA and simplify prompting (I think this method works with Illustrious because it works with natural language), for example tags "cropped jacket" and "blue jacket" can be combined into "cropped blue jacket"

@NeogenArt Also you might not know, but when you tag images for LORA using tag like "2b (nier:automata)" you keep it as is, but when you're using this tag for image generation you add escape parentheses to tag like this - 2b \(nier:automata\). Be careful not to use "/" instead accidentally. It is because "(" and ")" symbols treated as tag strength in image generation programs. There is more detailed explanation in manual I attached in the first message in "Escape parentheses if necessary " part. Can't past it here because of broken formatting in this document.

By the way don't look to much into this manual if you aren't using NoobAI, cause some things there are exclusive to NoobAI, and won't work elsewhere