🎸 Hitori Gotoh (Bocchi) | 后藤一里

作者发现市面上很多波奇模型在更换服装或改变画风时不够灵活，于是决定自己训练了这个模型。目前这个版本完美还原了“波奇酱”的神韵，无论是社恐表情还是日常装扮。

模型权重
基于 Anything 训练，适配各种二次元/动漫风格的 Checkpoint（推荐配合 Anything, Counterfeit, AOM 等使用）。
推荐权重：0.7 - 0.9
如果生成的波奇面部不够还原，可适当提高权重。

触发词指南
全局触发词：

1girl, bochi, gotoh hitori

推荐结构：

1girl, bochi, gotoh hitori, 服装, 姿势, 表情, 环境, 风格, 质量

推荐服装 (Outfits)：

• 经典运动服： pink tracksuit, track jacket, zipper

• 学校制服： school uniform, sailor collar, serafuku, grey skirt

• 特征物品： cube hair ornament, guitar case

表情建议：

• nervous, sweat (紧张流汗)

• screaming (尖叫/崩坏脸)

• blush (脸红)

数据集比较有限，使用 Anything 系列底模效果会最好。

出图尺寸
本模型训练集包含动画截图及高质量同人图，数据集尺寸为：

512×512
只推荐直出 512×512，避免出现多人或肢体崩坏。

高清放大
由于原图较小，必须使用高清修复。
推荐：

• 放大算法： R-ESRGAN 4x+ Anime6B

• 重绘幅度： ≤ 0.5

• 建议启用 ControlNet Tile 增加细节

使用声明
[你的名字] 出品

欢迎大家使用这款 LoRA 自由创作！我们很期待看到你的返图作品~ 🎸

本 LoRA 训练所使用的数据集由以下内容混合构成：

• 来自《孤独摇滚！》的动画截图

• 网络公开的高质量同人插画

训练数据仅用于学习角色“后藤一里”的一般性视觉特征，并不旨在侵犯原角色的版权。

本 LoRA 可用于个人及商业用途。
使用者需自行确保其生成内容的使用方式符合相关法律法规、平台政策及第三方权利要求。

除作者本人创作的内容外，作者不对训练图像主张任何所有权，亦不对生成内容中可能涉及的第三方权利作出任何保证。

🎸 Hitori Gotoh (Bocchi) | 后藤一里

I discovered that many existing Bocchi models lack flexibility when changing outfits or art styles, so I decided to train this model myself. This version perfectly captures the charm of "Bocchi-chan," from her social anxiety expressions to her daily outfits.

Weights
Trained based on Anything, compatible with various anime-style checkpoints (Recommended for use with Anything, Counterfeit, AOM, etc.).
Recommended weight: 0.7 - 0.9
If the generated face does not look enough like Bocchi, you can slightly increase the weight.

Trigger Guide
Global Trigger:

1girl, bochi, gotoh hitori

Recommended structure:

1girl, bochi, gotoh hitori, outfit, pose, expression, background, style, quality

Outfits:

• Classic Tracksuit: pink tracksuit, track jacket, zipper

• School Uniform: school uniform, sailor collar, serafuku, grey skirt

• Key Items: cube hair ornament, guitar case

Expressions:

• nervous, sweat

• screaming

• blush

Due to the dataset characteristics, using Anything-based checkpoints produces the best results.

Output Resolution
The model was trained on anime screenshots and high-quality fan art.

Dataset resolution:

512 × 512
Only direct output at 512 × 512 is recommended to avoid multiple subjects or limb distortions.

Hires.fix
Since the base resolution is small, Hires.fix is highly recommended.
Recommended settings:

• Upscaler: R-ESRGAN 4x+ Anime6B

• Denoising strength: ≤ 0.5

• ControlNet Tile is recommended for adding details

Disclaimer
Produced by [Your Name]

Feel free to use this LoRA for your creative works — we’d love to see what you come up with! 🎸

This LoRA was trained on a mixed dataset consisting of:

• Screenshots from the anime Bocchi the Rock!

• Publicly available high-quality fan art

The training data was used solely to learn the general visual features of the character "Hitori Gotoh" and does not intend to infringe on the original character's copyright.

This LoRA is released for both personal and commercial use.
Users are responsible for ensuring that their use of the generated outputs complies with applicable laws, platform policies, and third-party rights.

The author makes no claims of ownership over the training images beyond those personally created, and does not guarantee the absence of third-party rights in generated content.