模型权重
适配大多数刘备历史人物形象的 checkpoint，推荐权重：0.7-0.9
若搭配写古典类人物模型，建议升高至至：1.0以上
触发词指南
全局触发词：
Chinese history style
男主单人：
Chinese history style, krish, solo
无人场景推荐使用 NegPiP 插件：
Chinese history style, (no human:1.5), (human:-1.2)
军营场景
Chinese history style，krish，solo，military camp，soldiers
用这些提示词达到指定的效果
看向镜头： looking at viewer
侧脸视角： from side, profile
手里拿起书来：take the book
暖色调：warm color palette
望向远方：gazing into the distance
丰富的胡须：beard with fine details
出图尺寸
本模型基于 25 张实机截图训练，原始尺寸为：
512×512
推荐直出图尺寸 512×512，避免崩图。
高清放大
《Chinese history》的美术风格如果在过低分辨率的时候可能会出现人物形象不饱满等问题，在重绘幅度适中时生成的人物形象较为相似，在放大时可能出现“过于锐利”或“破坏气氛”的情况。
推荐：
放大算法： R-ESRGAN 4x+ Anime6B
重绘幅度 0.5≤幅度≤ 0.8
建议启用 ControlNet Tile
另可尝试在 Extras 中使用纯放大算法（不带重绘），有时反而保留得更自然。
Model Weight
• Compatible with most checkpoints for the historical figure Liu Bei, recommended weight: 0.7–0.9
• When paired with classical-style character models, it is advisable to increase the weight to above 1.0
Trigger Word Guide
• Global Trigger Word:
Chinese history style
• Male Lead (Solo):
Chinese history style, krish, solo
• No-human Scenes (NegPiP plugin recommended):
Chinese history style, (no human:1.5), (human:-1.2)
• Military Camp Scenes:
Chinese history style, krish, solo, military camp, soldiers
Prompt Words for Specific Effects
• Looking at the camera: looking at viewer
• Profile view: from side, profile
• Holding a book in hand: take the book
• Warm color palette: warm color palette
• Gazing into the distance: gazing into the distance
• Detailed beard: beard with fine details
Image Output Size
• This model is trained on 25 actual screenshots, with the original training size:
512×512
• It is recommended to generate images directly at 512×512 to avoid image distortion.
HD Upscaling
• For the Chinese history art style, excessively low resolutions may lead to underdeveloped character features. Generating images with a moderate redraw strength will result in more consistent character likenesses, while upscaling may cause issues such as over-sharpness or atmosphere disruption.
• Recommendations:
◦ Upscaling algorithm: R-ESRGAN 4x+ Anime6B
◦ Redraw strength: 0.5 ≤ strength ≤ 0.8
◦ It is advisable to enable ControlNet Tile.
◦ Alternatively, you can try using a pure upscaling algorithm (without redraw) in the Extras tab; this sometimes preserves the original style more naturally.