Try size: 512 x 768
Weights: 0.6 (0.3~1.0 are within the valid range)
Clip skip: 2
trigger words: head high (Not necessary, It is recommended to improve by lora weights)
Pictures from majicmixRealistic_v6 (Sample pictures only select successful ones)
Note: If you often use pony or sdxl asian models, you may also be tired of the frequent head down of SD1.5 asian models, so you can consider using this slider to improve it. Sometimes it may be ineffective, sometimes it may be worse, you can make your own decision. If the weight is too high, although the head up effect is better, but the prompt will be difficult to control, so you can start with 0.6.
1.0 Plus版本可以取代1.0版本,所以不用同時下載。如果你希望麥橘女孩像flux的女孩一樣展現自信,這個lora是適合你的;如果你希望她時常羞怯地低頭看著你,這個lora不適合你。我不希望你們用動漫模型或半寫實去測試lora,真有需要,我會用anything v5去訓練,而不是用majicmix v6去訓練。但是我真的不覺得需要去練一個動漫專用的,我使用動漫模型時沒有這個困擾,動漫女孩低頭看你時很可愛。
Description
FAQ
Comments (5)
Hmm...I don't recommend using it on anime models and semi-realistic models...
脖子长度?头的角度?
脖子沒影響,頭抬起的角度會略高一些。Hold your "head high" 昂首、抬頭挺胸的意思,因為我測試的結果是這個詞比 "head up"有效,所以這樣命名。而looking up的效果又太強,looking the sky也可以,但會出現天空,視線也不看鏡頭了。其實不需要觸發詞,靠lora權重就可以調整。再來就是,是不是有使用MajicMix或融合比例偏高的亞洲寫實模型的習慣。假如沒有,這個lora"完全沒有使用"的必要,動漫低頭、抬頭角度略小都算自然,chilloutmix也不太需要,直視較多。head up應該是無效的,danbooru沒有收錄進去。head high雖然也沒收錄,但在半寫實跟亞洲寫實模型中它是有效的。如果使用flux、pony跟sdxl的寫實模型出圖,會感覺很自然的原因之一,是因為女孩大多是大方直視你的,而不是低頭看著你,臉上陰影更不會這麼多。
@ghostpaint 好复杂啊
@ghostpaint 大哥你这么认真的吗?我好感动❤
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.



















