If you're following me, you probably like the images I make, I bid on the illustrious checkpoint I use the most and now it's available for onsite generation: miruki~ by r_ain.
Taking inspiration from reijlita, I would like to share some of my favorite tags that I've found illustrious to be responsive to on images generated by this checkpoint. Probably not that useful for those that are accustomed to using tag-based models, but more for the people that are used to natural language based models.
While I have a custom GPT for helping to prompt with illustrious, and I did extract a list of >900K danbooru tags to have it reference, that amount proved too much for a custom GPT to process and it hallucinates constantly. However, it is still more effective than attempting to use natural language prompting.
A good way to learn how to start prompting for illustrious is to read prompts from other generations you like, and to search for tags on danbooru, generally, if a tag has >1,000 posts, it is likely that illustrious will be responsive to it. It is useful to go look, as the tag you are looking for could be called something else. For example, if you search for "cringe", it will redirect you to "wince". When using "cringe", the model seems to think that you misspelled , "crying", and you will get an output that more resembles that. While "wince" is closer, it isn't what I think of cringe, which is why I made a lora for cringing.
Tag(s): eyewear on head / shutter shads | Link to post
Tag(s): dynamic pose / unsheathing / drawing sword | Link to post
Tag(s): midair | Link to post
Tag(s): fisheye/foreshortening | Link to post
Tag(s): eye focus / contrast lighting | Link to post
Tag(s): blurry background | Link to post
Tag(s): half-closed eyes | Link to post
