1,本模型所有数据全来视频截图

2,最大出图大小建议不要超过 1280 x 1280

3,建议比重 在0.6-0.8左右脸和衣服还原不够提高比重,

4,采样方式 DPM++2Mkarras   步数  20-30  重绘-0.3-0.5

5,算法 4×-UltraShrt   高分步数 15-20  

展示图描述词

(1girl:1.1),(looking at viewer:1.4),solo,(hair ornament:1.6),

hime cut,bangs,(black hair:1.4),(long hair:1.6),(hair_bun:1.6),(jewelry:1.6),(crystal:1.4),(hairpin:1.4),(forehead pendant:1.6),(forehead imprint:1.6),(water_eyes:1.2),(decorations on the neck:1.6),(dress:1.6),(chinese clothing:1.6),(white dress:1.4),(shoulder:1.4),(detached_sleeves:1.4),(skirt:1.4),(exposed legs:1.4),(fluffy:1.3),(frilled:1.4),(lace:1.3),high_heels,boots,(earrings:1.4),(necklace:1.5),(translucent ribbon:1.4),hair_ribbon,(ribbon:1.3),(clothes fluttering in the air:1.5),(waterfall:1.5),(castle:1.1),(cityscape:1.6),(sky:1.4),(white clouds:1.6),(cherry_blossoms:1.3),(sakura tree:1.6),(river:1.4),(flowers and trees:1.3),(petals:1.5),(water:1.6),