Brief
Lora model of Furina. Trained with 1024x1024 resolution for SD1.5 and Ver 2.0 and below, 1536x1536 for SDXL,Pony and IL ver 3.0 and above. Models are enhanced with 3D model for best accuracy.The training dataset is categorized to further improve the flexibility, making my model easy to remove/change the costume.
Usage
Version 3 and above, for Pneuma Furina (White): furina \(genshin impact\), short hair, heterochromia, mismatched gloves, top hat, blue jacket, waist cape, ascot, white waistcoat, waist bow, white shorts, thigh strap, block heels, frilled socks
Version 3 and above, for Ousia Furina (Black): furina \(genshin impact\), short hair, heterochromia, mismatched gloves, top hat, blue jacket, waist cape, ascot, black waistcoat, waist bow, black shorts, thigh strap, block heels, frilled socks
Version 3 and above, for Focalors: focalors \(genshin impact\), heterochromia, anklet, evening gown, long dress, juliet sleeves, bridal gauntlets, barefoot
Version 3 and below:
Additional prompts for her default costume, add/remove as you needed: heterochromia, mismatched gloves, top hat, blue jacket, ascot, waist bow, short shorts, thigh strap, high heels, frilled socks, waist cape
Trigger prompt is: focalors \(genshin impact\)
Additional prompts for her default costume, add/remove as you needed: heterochromia, anklet, evening gown, long dress, juliet sleeves, bridal gauntlets, barefoot
The model was well trained that the default costume can be removed/changed easily even with high lora weight, recommended weight is 0.9
Description
Improved thigh straps
Improved heterochromia
FAQ
Comments (7)
😭😭😭🦀🦀💢💢 Uohhhh Cunny😭
just wow! I like the way it is!
😭😭😭
感谢你在抱脸分享的训练经验,其中我很好奇的是furina使用了非常多的插画素材但都被干净利落地剔除了背景,这是使用inpaint anything,或者ps工具做到的吗?有没有其他更好的方法推荐?
可以考虑使用segment anything. 但实际情况是,只要标签打得好,没必要专门去做剔除背景的工作
@okingjo 感谢回复,此外我还有个疑问,如果我想达到你的参数设置:1024*1024分辨率,最大桶2048,每次抽8张图训练。对配置的要求如何,是否需要24GB显存加4090,一个丹大概多久时间呢?
@Sammian 1024*1024的话24G也只能batch2左右,你要batch8的话我试过,只能是H100那种了。当然这都是不开xformer的前提下












