美學品質修飾器 - 完整版 - v1.0 [Illustrious]
推薦提示詞
{{tags}} masterpiece, best quality, very aesthetic
by [lack|fuzichoco|lam|kinta \(distortion\)|wlop|quasarcake|nixeu|happoubi jin|yoneyama mai], shiranui flare, 1girl, hatsune miku, shiranui flare \(1st costume\), racing miku, solo, virtual youtuber, plugsuit, blonde hair, full body, dark-skinned female, knees up, streaked hair, hand between legs, close-up, face focus, floating hair, tree shade, colored eyelashes, long eyelashes, looking at viewer, ginkgo leaf, outdoors, multicolored eyes, sunbeam, jirai kei, makeup, aqua eyes, red eyes, multicolored hair, space station, masterpiece, star \(sky\), best quality, circle formation, very aesthetic, single vertical stripe, absurdres <lora:illustrious_very_aesthetic_v1:1>, spacecraft, black theme
推薦反向提示詞
(worst quality, low quality, sketch:1.1), error, bad anatomy, bad hands, watermark, ugly, distorted, censored, lowres, abstract, signature, bkub
推薦參數
samplers
steps
cfg
clip skip
resolution
other models
推薦高解析度參數
upscaler
upscale
steps
denoising strength
提示
LoRA 最適合搭配其訓練所用的基礎模型使用。
預覽在 Forge 中使用升級與細節修飾器生成。
Noobai V-Pred 有提供包含 DynamicThresholding、升級及 FaceDetailer 的 ComfyUI 工作流程示範。
版本亮點
訓練於 Illustrious-XL
美學品質修飾器 - 完整版
訓練資料是我所有含品質/美學修飾標籤的人工評分資料集的子集,僅包含標註了masterpiece、best quality或very aesthetic的圖像(共7273張)。
是美學品質修飾器集合中模型的超集及最終版本。
ℹ️ LoRA 最適合搭配其訓練所用的基礎模型使用。請閱讀關於此版本來獲取適用基礎模型及工作流程/訓練資訊。
推薦的提示語結構:
正面提示(在提示末尾加入品質標籤):
{{tags}}
masterpiece, best quality, very aesthetic生成設置:
預覽在 Forge 中使用升級與細節修飾器生成。
Noobai V-Pred 的 ComfyUI 工作流程示範,包含 DynamicThresholding、升級及 FaceDetailer,可參考此處:https://civitai.com/posts/11457095
