✨ Prompt Guide (提示词指南)
1. 🟢 Core Trigger (核心触发)
Trigger Word:
sampo_koskiExpression (推荐表情):
smirk(坏笑) 或winking(眨眼)Note: Facial features are baked in. 核心面部特征(蓝紫发、绿眼等)已深度绑定,无需额外输入发色/瞳色提示词。
2. 🛡️ Default Costume (默认原皮服饰)
Prompts:
sampo_koski, sampo_default_costume, 1boy, smirk, red_jacket, vest, pants, boots, belt, fingerless_gloves, shoulder_armor, asymmetrical_clothes, thigh_strapNote: 极高精度还原游戏原画。请保留
shoulder_armor和asymmetrical_clothes以稳定他标志性的不对称肩甲设计。
3. 👕 Custom Outfit Mode (自由换装模式)
How to use: Remove
sampo_default_costumeand gear tags to easily change clothes! 得益于严格的打标解耦,删掉sampo_default_costume及原皮细节词,即可完美换装。Example (西装):
sampo_koski, 1boy, smirk, formal, black_suit, tie, holding_wine_glass
4. 💣 Special Props (专属道具隔离)
Note: 道具已在训练时做过特征隔离,可干净召唤,避免污染人物服饰。
Bomb (炸弹):
holding, bomb, heart_printMoney (钞票):
money, banknote, floating_moneyCards (卡牌):
playing_card, floating_cards
⚙️ Recommended Parameters (推荐生成参数)
LoRA Weight (权重): 0.8 - 1.0 (服饰已完全解耦,高权重不影响换装)
Resolution (分辨率): 1024 x 1024 / 832 x 1216
Sampler (采样器): Euler a / DPM++ 2M Karras
Steps (步数): 28 - 35
CFG Scale: 5 - 7
⚠️ Negative Prompts (核心负面词):
lowres, bad anatomy, bad hands, missing fingers, worst quality, (3d, 3d render, cgi, polygones:1.2)👉 Important: 训练集包含高质量多视角 3D 建模图。为了生成纯正的 2D 插画质感,请务必在负面词中加入
3d, 3d render, cgi!
💬 Feedback (反馈)
If you notice any issues (e.g., asymmetrical armor on the wrong side), please share your images in the comments! 如果你在出图时发现任何瑕疵(如不对称肩甲跑到了另一边、炸弹融合错误等),请务必带图在评论区告诉我。大家的返图将作为下个版本 (V2.0) 优化的最重要依据!Let's make it better together!
