The model I think is fairly self explanatory.
Works well at a strength of 1.
Previews made with novaFurryXL_illustriousV40.
Description
FAQ
Comments (14)
Thank you! This is just what I needed.
I'm having trouble using it well with WAI 14. I used other models before and there this lora worked perfectly. Do you happen to know why it's giving me trouble with WAI 14?
Not entirely sure, I think this was done a while ago before I had some better understanding of how to models better over the last few months, so I'll redo it in a little bit to see if that helps that issue and the overall quality of it
Updated version is out, hopefully it works better
@Malebolgia You're a legend, I'll check it out later. Do you by chance have a dataset for the Hentaikey girl from the zone flash games?
@Lopodelas I don't have the dataset, but I do have a sort of model for it, https://civitai.com/models/1601342/zone-hentai-key-girl-character-illustrious it apparently works well for the character at a strength of 1.2-1.3 from what I've heard of the person who wanted it. But it would be pretty simple to recreate the dataset if you'd want, all I really did was take screenshots of the flash animations.
@Malebolgia Did you take the screenshots of the entire screen? I wanted to train a model of her at some point which recognizes the different styles of the those flash games, tho the main focus is the fifth one. Sadly my pc is too weak to handle sdxl training. I tried it like 5 times already for some style loras, but the end result is not good at all due to low training settings
@Lopodelas I mostly took screenshots of that and the screencaps for this LoRA as well of most of the screen with a little bit of an attempt to crop out UI elements, but for the most part the whole screen I'd say. It does seem as though from a search on the site that a person did a LoRA of the hentai-key girl other than me that seems a lot more representative of the character. I'll likely have to spend a bit of time trying to figure out how to make characters again because what I'm doing doesn't seem to work too well for characters, but works well for styles.
How I'd suggest tagging a LoRA of a 'same character/artist' but 'different variant' LoRA would be sort of how I did my 'omnibus' LoRA's a few months ago around December or so. How I did that for some Yaoi stuff was to tag each individual character / doujin / style within the lora with it's own unique identifier, and then have one tag that linked them all as a sort of 'omnibus' tag that applied to every image to sort of loop them together, so you could use everything separate, or in interesting combinations. It's a little tedious to set up which is why I almost never do it anymore haha, too lazy. But you can try something like that I think.
@Malebolgia Man, you're so real replying in such detail. Thank you so much for the information and of course the lora. I managed to check it out and it works really well, but also seems to look differently than v1. Either way, it looks really nice :D
@Lopodelas Hm. The difference in style could be from that I additionally realized that Zone had done 'animations' in the timeframe that I additionally took screengrabs of, specifically the Raven Teen Titans animations from that era. I'll always be a sucker for Raven, so that could explain a bit of it, and additionally why the LoRA is functionally made with double the sample images.
@Malebolgia I see. That animation is a timeless classic
@Malebolgia I totally forgot to mention that the Hentai Key Girl lora from someone else you most likely were mentioning doesn't work. The training data even only consists of 3 very small images. I tried it, but I literally only saw an extremely small change in the image and it made little to no difference no matter how much I cranked up the strength.
@Lopodelas Hm. I guess they must have been experimenting or new to this whole thing. I sort of figured off their preview image that it worked very well, but I guess they used a preview image that was one of their sample images.