1980 年代，日本科學家在參觀米酒釀造廠時，意外發現釀酒工人的雙手皮膚極為細緻，進而研究發現發酵過程中產生的天然酵母具備顯著護膚功效，並成功提煉出核心成分「Pitera」。該系列的產品最初命名為「Secret Key To Beautiful Skin」（美之匙），自第二代產品起簡稱為「SK-II」。
In the 1980s, Japanese scientists visiting a sake brewery were struck by the remarkably smooth and youthful skin of the elderly brewers' hands. Their research revealed that a natural yeast byproduct from the fermentation process possessed significant skincare benefits, leading to the isolation of the core ingredient "Pitera." The product line was originally named "Secret Key To Beautiful Skin," and was later shortened to SK-II starting from the second generation.
[Project Objective]
本工作流旨在透過配置兩組不同的模型與 KSampler，精準還原晶瑩透亮且具真實感的皮膚質感，故命名為「ppkkmoon Rice Wine Style」。本套方案中選用的模型、採樣器、LoRA、排程器及 LLM 皆經過本人長時間的交叉比對與測試，若您希望省去繁瑣的調試過程，建議直接採用此預設方案。
This workflow is designed to reproduce the translucent, glowing, and realistic skin texture through a dual-model and KSampler configuration, which I have dubbed the "ppkkmoon Rice Wine Style." The models, samplers, LoRAs, schedulers, and LLMs selected for this setup have undergone extensive testing and comparative analysis. If you prefer to bypass the tedious trial-and-error process, I recommend utilizing this preset configuration directly.
[Technical Details & Insights]
模型性能與選擇： 經實測發現，Anima LLM LayerWise FP32 在角色構圖與理解能力上雖略遜於 QWEN3_06B，但在擬真皮膚紋理的細膩度表現上更具優勢。因此，我在第二階段的 KSampler 與 FaceDetailer 中優先採用了 Anima LLM LayerWise FP32。當然，若全程使用 QWEN3_06B 亦可獲得不錯的效果。
Model Performance & Selection: Testing shows that while the Anima LLM LayerWise FP32 is slightly less capable in composition and comprehension compared to QWEN3_06B, it excels in rendering hyper-realistic skin textures. Therefore, I prioritize using Anima LLM LayerWise FP32 for the second KSampler and FaceDetailer stages. However, using QWEN3_06B throughout the entire process remains a perfectly viable option.
「2D 污染」問題： 針對 Anima 模型在繪製角色時易出現的「2D 污染」問題，本工作流已拆分為「角色 (Character)」與「NPC」兩個獨立版本。實測顯示，NPC 工作流在視覺表現上更為自然。
The "2D Contamination" Issue: To address the "2D contamination" frequently encountered when generating characters with the Anima model, I have split the workflow into separate "Character" and "NPC" versions. Empirical testing shows that the NPC workflow produces more natural visual results.
臉部修繕： 由於目前在角色臉部的穩定度上仍有挑戰，本工作流內建了四次不同的 FaceDetailer 修繕步驟，建議產出後手動篩選最佳成果。若各位對「提升角色臉部穩定度」有更好的優化方案，懇請指教，不勝感激。
Face Restoration: Given the current challenges in maintaining consistent facial stability for characters, I have integrated four distinct FaceDetailer refinement steps into this workflow. I recommend manually curating the best outputs. If you have any superior optimization strategies for enhancing facial stability, your suggestions would be greatly appreciated.
[Instructions for Use]
模型資源請參考「Suggested Resources」連結。
Please refer to the "Suggested Resources" section for model downloads.
未來我將陸續上傳更多非此工作流的相關案例圖，您可以直接將這些圖檔拖曳至 ComfyUI 中，即可自動解析並匯入對應的工作流。
I will be uploading more sample images unrelated to this specific workflow in the future. You can simply drag and drop these image files into ComfyUI to automatically parse and import the corresponding workflow.
工作流產出的圖片將依據角色名稱命名，相關 Wildcards 設定已同步封裝於檔案中。
The output images are named based on the character names, and the relevant Wildcards settings have been packaged within the files.
Description
1980 年代，日本科學家在參觀米酒釀造廠時，意外發現釀酒工人的雙手皮膚極為細緻，進而研究發現發酵過程中產生的天然酵母具備顯著護膚功效，並成功提煉出核心成分「Pitera」。該系列的產品最初命名為「Secret Key To Beautiful Skin」（美之匙），自第二代產品起簡稱為「SK-II」。
In the 1980s, Japanese scientists visiting a sake brewery were struck by the remarkably smooth and youthful skin of the elderly brewers' hands. Their research revealed that a natural yeast byproduct from the fermentation process possessed significant skincare benefits, leading to the isolation of the core ingredient "Pitera." The product line was originally named "Secret Key To Beautiful Skin," and was later shortened to SK-II starting from the second generation.
[Project Objective]
本工作流旨在透過配置兩組不同的模型與 KSampler，精準還原晶瑩透亮且具真實感的皮膚質感，故命名為「ppkkmoon Rice Wine Style」。本套方案中選用的模型、採樣器、LoRA、排程器及 LLM 皆經過本人長時間的交叉比對與測試，若您希望省去繁瑣的調試過程，建議直接採用此預設方案。
This workflow is designed to reproduce the translucent, glowing, and realistic skin texture through a dual-model and KSampler configuration, which I have dubbed the "ppkkmoon Rice Wine Style." The models, samplers, LoRAs, schedulers, and LLMs selected for this setup have undergone extensive testing and comparative analysis. If you prefer to bypass the tedious trial-and-error process, I recommend utilizing this preset configuration directly.
[Technical Details & Insights]
模型性能與選擇： 經實測發現，Anima LLM LayerWise FP32 在角色構圖與理解能力上雖略遜於 QWEN3_06B，但在擬真皮膚紋理的細膩度表現上更具優勢。因此，我在第二階段的 KSampler 與 FaceDetailer 中優先採用了 Anima LLM LayerWise FP32。當然，若全程使用 QWEN3_06B 亦可獲得不錯的效果。
Model Performance & Selection: Testing shows that while the Anima LLM LayerWise FP32 is slightly less capable in composition and comprehension compared to QWEN3_06B, it excels in rendering hyper-realistic skin textures. Therefore, I prioritize using Anima LLM LayerWise FP32 for the second KSampler and FaceDetailer stages. However, using QWEN3_06B throughout the entire process remains a perfectly viable option.
「2D 污染」問題： 針對 Anima 模型在繪製角色時易出現的「2D 污染」問題，本工作流已拆分為「角色 (Character)」與「NPC」兩個獨立版本。實測顯示，NPC 工作流在視覺表現上更為自然。
The "2D Contamination" Issue: To address the "2D contamination" frequently encountered when generating characters with the Anima model, I have split the workflow into separate "Character" and "NPC" versions. Empirical testing shows that the NPC workflow produces more natural visual results.
臉部修繕： 由於目前在角色臉部的穩定度上仍有挑戰，本工作流內建了四次不同的 FaceDetailer 修繕步驟，建議產出後手動篩選最佳成果。若各位對「提升角色臉部穩定度」有更好的優化方案，懇請指教，不勝感激。
Face Restoration: Given the current challenges in maintaining consistent facial stability for characters, I have integrated four distinct FaceDetailer refinement steps into this workflow. I recommend manually curating the best outputs. If you have any superior optimization strategies for enhancing facial stability, your suggestions would be greatly appreciated.
[Instructions for Use]
模型資源請參考「Suggested Resources」連結。
Please refer to the "Suggested Resources" section for model downloads.
未來我將陸續上傳更多非此工作流的相關案例圖，您可以直接將這些圖檔拖曳至 ComfyUI 中，即可自動解析並匯入對應的工作流。
I will be uploading more sample images unrelated to this specific workflow in the future. You can simply drag and drop these image files into ComfyUI to automatically parse and import the corresponding workflow.
工作流產出的圖片將依據角色名稱命名，相關 Wildcards 設定已同步封裝於檔案中。
The output images are named based on the character names, and the relevant Wildcards settings have been packaged within the files.