There are no trigger words for this model. Simply put the model in your prompt to activate it.Depicts the style of Aeba Futi (also known as Aeba Fuchi or 饗庭淵.)Link to his Pixiv: <a target="_blank" rel="ugc" href="https://www.pixiv.net/en/users/20046">https://www.pixiv.net/en/users/20046</a>His works typically depict very large-breasted women, mostly in japanese clothing, although other clothing is seen as well. This model is very flexible in terms of clothing and breast sizes.Unfortunately, due to a high variance in how the eyes are drawn, quality in the eyes tends to vary somewhat. In particular, purple eyes tend to display this behaviour the most. This is due to the auto-tagger incorrectly tagging <code>purple eyes</code> for <code>pink eyes</code> at times.P.S.The dataset consists of 1,524 images obtained from the author's pixiv, IMHentai entries, game CGs as well as screenshots I've personally made. This is my second LoRA. It took an unbelievably long time to make this. I'd wager it took me hours split among roughly two weeks of working to make this model.I've manually cropped, edited and upscaled many, many images.Many crops had to be made to remove any letterboxing/pillarboxing and parts of the image that were unnecessary.For the images that included japanese text, I made great efforts into cleaning up most of them. (Now you can prompt without having to put <code>japanese text</code> in your positive prompt! Yay!) For the images where it was simply too hard to remove text, I made sure to properly tag it as: <code>japanese text</code>. So, if you somehow still manage to get text in your output images, please put <code>japanese text</code> in your negative prompt. Note that I haven't gotten any text in my images, so I'd be really surprised if this somehow slipped through the cracks.The auto-tagger made plenty of incorrect captions of a majority of the images, leading me to manually looking over and fixing each individual image. This was a great pain. Perhaps the artist style was too confusing for the auto-tagger? I suspect the confidence threshold for tagging an image with a particular tag was too high. Oh well, practice comes with learning.Some of the images were in low resolution, mostly character face portraits from game CGs. I've upscaled these and removed the bad ones. I used <a target="_blank" rel="ugc" href="https://github.com/chaiNNer-org/chaiNNer">chaiNNer </a>to upscale the images along with the <a target="_blank" rel="ugc" href="https://huggingface.co/uwg/upscaler/blob/main/ESRGAN/lollypop.pth">lollypop.pth</a> upscaler model.Thanks for taking a look at this model! Despite the many hours I've put into this, I only managed to push through because of my love for the author's work. I hope you'll enjoy it.A special thank you to <a target="_blank" rel="ugc" href="https://civarchive.com/user/justTNP">justTNP</a>'s for his LoRA <a target="_blank" rel="ugc" href="https://civarchive.com/articles/680/characters-clothing-poses-among-other-things-a-guide">Guide</a>. It worked wonders!Toodles.

Aeba Futi / Aeba Fuchi / Aeba no Mori / 饗庭淵

Files with SHA256: 21409bc6664ba74b9a37e3991926104d04ad6b4e3f9412c934aaec6a1a5462a4