Hu Tao

模型简介 | Model Overview

本模型以《原神》角色胡桃为蓝本，采用多阶段采样、结构保持与风格迁移算法训练，确保角色面部特征、姿态、服饰与光影风格高度还原，表现力极强。  

This LoRA model is based on the character Hu Tao from Genshin Impact. It was trained with a multi-stage sampling pipeline, structure-preserving optimizations, and style-transfer techniques to faithfully recreate facial features, poses, clothing, and lighting styles with impressive expressiveness.

本项目旨在实现艺术表达与技术实现的高度融合，特别适用于角色复现、动漫风格插图创作以及二次元艺术生成。  

The project strives to merge artistic representation with technical excellence, making it ideal for character restoration, anime-style illustration, and 2D content generation.

使用说明 | Usage Instructions

| 推荐权重 :`0.6 ~ 0.8` | Recommended weight: 0.6 ~ 0.8

| 适配基础模型 | 基于 Stable Diffusion 1.5，推荐结合动漫风格模型（如 Anything、Counterfeit） | Built for SD 1.5 base models (preferably anime-style models like Anything or Counterfeit) |

| 正向提示词示例 | hu tao, red eyes, twin braids, black dress, hat, flame| Positive prompt: hu tao, red eyes, twin braids, black dress, hat, flame |

| 反向提示词示例 | nsfw, lowres, blurry, extra arms, deformed | Negative prompt: nsfw, lowres, blurry, extra arms, deformed|

| 采样建议 | DPM++ 2M Karras，步数 20~28，CFG 7~9，分辨率 512x768 最佳 | Sampler: DPM++ 2M Karras, Steps: 20~28, CFG: 7~9, Resolution: 512x768 optimal |

模型测试与表现 | Testing & Performance

我们进行了广泛的多采样器、多提示词、多模型适配测试，模型表现出极强的稳定性与风格一致性。  

The model has been thoroughly tested across various samplers, prompt combinations, and base models, showing excellent stability and stylistic consistency.

- 多次渲染中胡桃的发型、帽子与眼神表现清晰准确  

  > Hu Tao’s hairstyle, hat, and eye expression are consistently accurate across generations  

- 与不同动漫模型（例如 AOM3、Counterfeit）兼容性良好  

  > Compatible with various anime checkpoints like AOM3 and Counterfeit  

- 可应对复杂背景、动作、光影变化场景  

  > Handles complex backgrounds, dynamic poses, and lighting variations well  

声明 | Disclaimer

- 模型生成内容仅供学习与研究用途  

  > Generated content is for research and educational purposes only  

- 请勿上传含有冒犯性内容的图片至公共平台  

  > Do not upload offensive outputs to public platforms  

- 本项目不承担因不当使用本模型所导致的任何责任  

  > The authors are not responsible for any misuse of this model