Applies to SD version: Trained on Anything V5, 12 images. Now, I'll complain about TLDR: Hands generate "horror" results more often than my other LoRa, but you'll still get good results in an acceptable amount of time. You can skip the complaining, but I want to warn you about issues I experienced.
Also applies to SD version, DiT V2 is fine: If this wasn't a major character in the anime I wouldn't probably even publish it. Quite rough LoRa, you can definitely get results by brute force, I did. If you want to generate Triage X characters, then I suggest giving a try to my other LoRas from the series, unless you are looking for this character specifically.
Again, applies to the SD version: What's the problem with this LoRa? Well, hands, hands generate worse than usual on average, you will get good hands but it is going to be a bit of a grind. Don't get discouraged okay? Despite my essay, it won't take 100 tries to get a good result, not even 50 tries no. I give you this warning because I noticed it takes longer to get good results and the hands generate more often in the "inhuman" lengths or shapes.
Still applies to SD version: Still want to use it? Okay, this LoRa used more closeup images, you might experience an influx of closeup shots, and you might have to use prompts to get "full body".
I used mostly Dark Sushi Mix 大颗寿司Mix, Coffee NSFW, Dark Sushi 2.5D 大颗寿司2.5D, and
SakushiMix-Hentai v3.0 ( I suggest using it only for closeup shots, hands don't like to generate "human").
I wanted to use ThorneMix_v1, but it got removed from PixAI. Feel free to try other models and share your positive results, maybe I'll update this bio too.
I use Euler A
23 sampling steps ( I use this little because of testing, feel free to use more)
7 CFG
Thumbnail Model: Tsubaki V2
