Try size: 768 x 1024 or 768 x 1280

Weights: 0.5~1.0 (ILXLx3)

Weights: 2.0~3.0 (anime and asian base model: 1.0~2.0)

Clip skip: 2

Pictures from realisticBlendXL_v50 and Z-image turbo

Note: First, trigger words are essential; otherwise, satisfactory results are almost impossible. For the Caucasian realistic model, which is particular about skin texture but lacks training on absurdly long hair, a LoRa weight of 2.0~3.0 is used; for the anime-based model, which has a solid training foundation for absurdly long hair, and the Asian realistic model with a high proportion of anime elements, a LoRa weight of 1.0~2.0 is used. Why is such a high LoRa weight necessary? Because slider LoRa has a lower level of visual clutter. However, many platforms have a default LoRa weight of 1, with a maximum of 2. Therefore, if you cannot increase the weight, please abandon its use. / 首先，觸發詞是必要的，否則幾乎達不到滿意的效果。對於皮膚較講究，但荒謬長髮缺乏訓練的高加索寫實模型採用的lora權重為2.0~3.0；動漫基礎模型對荒謬長髮的訓練較扎實，以及混合動漫比例較高的亞洲寫實模型採用的lora權重為1.0~2.0。為何lora權重會需要怎麼高？因為slider lora對於畫面的汙染程度較低。但是很多平台的lora權重預設值為1，最高只能達到2，所以如果你不能提高權重，請放棄使用它。(ILXL)

Note: Trigger words are also necessary; otherwise, the desired effect will not be achieved. Although Z-Image Turbo has a training prompts for absurdly long hair, its absurdly long hair doesn't look very good, so I recommend using the Illustrious version. / 觸發詞也是必要的，否則同樣達不到預期的效果。Z-Image Turbo雖然有訓練荒謬長髮的提詞，但它的荒謬長髮並不是很好看，所以我較推薦使用Illustrious版本。(ZIDT)

Search AI Models & Addons

very few results