The tags in that page are surely correct, you can use them of course. But what's more important is you understand the concepts you want to train. And I suggest you to train a Lora, not a full model. Looking forward to your results.

>>1780
If you want to tag images for AI training you should consider using wd14 to tag images for you. It'll tag most of them correctly and you can simply look through to make sure they are correct/make changes.

For Automatic1111: https://github.com/toriato/stable-diffusion-webui-wd14-tagger
For ComfyUI: https://github.com/pythongosssss/ComfyUI-WD14-Tagger

If you are deadset of tagging them yourself using gelbooru would be a good idea. I'm pretty sure ponyXL is trained using mostly gelbooru tags. If you're looking to train a ponyXL model it would be a good idea to look there. https://gelbooru.com/index.php https://gelbooru.com/index.php?page=tags&s=list

thanks!

Will try to follow. I wan tto make artist LoRA and they are usually a mix of all guro concepts

>>1796
When you tag images it's important you tag everything, even the non-guro aspects of the image. wd-14 will do that for you which is why I recommend you use it. If you take tags out it will bake them into your trigger word. So for example if you're training on a character, by removing clothing tags the clothing becomes a part of the character's trigger word and will always generate with that keyword. The trigger word you choose by the way should ideally be something the model is already trained on so that it gets a head start on training. Not everything works with it however. Hanging for example is notorious for being a trigger word you cant train hanging on since on ponyXL, hanging will become a bdsm tied up image of the person being strung up. In such a case either create your own trigger word or use another adjacent trigger word that wont conflict. ex, noose, rope_around_neck etc. etc. When in doubt just check gelbooru.

Good luck can't wait to see what you create.

>>1780
>>1801

Not sure if it will apply to what you're trying to do but I figure I should at least mention it.

If you are planning on training a PonyXL lora you only need about 20 images per concept. More is not always better. Quality > Quantity. You will get much better results if each image clearly shows what you are trying to replicate. Pruning images that aren't as useful is also a part of Lora training. Having more is better but make sure they are all quality images.