Heian period Sukuna from Jujutsu Kaisen. This LoRA was trained without tags like "Jujutsu Kaisen" "Ryomen Sukuna" to avoid data contamination. This Lora is a mix of Anime/Manga style, or the closest it can be without Fanart heavily impacting the style. IMPORTANT: I accidentally misspelled the trigger word in the older versions of this Lora. It was "heian sakuna" without the underscore. This is fixed in VPRED 2.0 and Epsilon, which have new trigger words. You can see them down below: Trigger words: heian_sukuna, comic_sukuna, anime screencap, Appearance tags: Mask, closed stomach mouth, stomach mouth, open stomach mouth, extra eyes, pink hair, extra arms, 4 arms, tattoo, facial tattoo, white pants, baggy pants, jacket on shoulders, japanese clothes, haori heian_sukuna triggers all images in the dataset. comic_sukuna specifically targets artwork directly from the comic or fanart mimicking it. anime screencap mimics the anime style but is fan artist interpretation. ⚠️ I currently do not plan to retrain this for LoRA for Pony as this particular character is a bit of a pain in the ass to train due to his unique features. NoobAI has given really good results, so I'm kind of done with it for now. OTHER ISSUES: This LoRA might run into issues with ADetailer as it sometimes does not recognize Sukuna's face. Most of the time it works, don't think there's a way to fix this.

Ryomen Sukuna [Heian period version]

<ul><li>Experimental LoRA trained on Noobai's VPRED Model</li><li>Trained with Adafactor because Prodigy was giving bad results, might be overfitting but I find it very difficult to tell with vpred.</li><li>Better at concepts such as posing with weapons, Sukuna's unique features, etc.</li><li>Must use Euler sampler to work properly, results from on-site generator might not look the same as ReForge extensions were used. I like using Euler Ancestral CFG++ with SGM Uniform at 1.5 or 2 CFG scale.</li></ul>

Files with SHA256: 8971dacca39b908e8ab8f0c94f75b92e903dec67bd435f3e2e6d913752a4d4ad