Sign In

Florence | CUC_AI_STUDIO

8

33

6

2

Verified:

SafeTensor

Type

LoRA

Stats

33

6

1

Reviews

Published

Jun 9, 2025

Base Model

SD 1.5

Training

Steps: 5,104
Epochs: 8

Usage Tips

Clip Skip: 2

Trigger Words

florence style

Hash

AutoV2
FA0011D58E

🎨 Florence

向震撼人心的独立游戏小品《Florence》致敬的 LoRA 模型,能够生成其简洁温暖的手绘插画风格主要人物形象

模型权重

  • 适配大多数二次元 checkpoint,推荐权重:0.8 – 1.1

  • 若搭配写实风模型,建议降低权重至:0.5

触发词指南

全局触发词:
florence style

女主单人:
florence style, yeoh, solo

男主单人:
florence style, krish, solo

双人合照(情侣):
florence style, fcouple, yeoh, krish

分手后女主(艺术家形象):
florence style, yeoh, solo, very short hair, artist

无人场景推荐使用 NegPiP 插件:
florence style, (no human:1.5), (human:-1.2)

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

  • 纯色背景: simple background

  • 色彩风格控制: limited palette, monochrome, greyscale

出图尺寸

本模型基于 165 张实机截图训练,原始尺寸为:

  • 512×512

  • 512×768

  • 768×512

推荐直出图尺寸 不超过 768×768,避免崩图。

高清放大

《Florence》的美术风格具有低保真(lo-fi)质感,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:4x-AnimeSharpR-ESRGAN 4x+ Anime6B

  • 重绘幅度 ≤ 0.6

  • 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。

使用声明

CUC AI STUDIO 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~
本模型仅供个人学习和非商业创作使用,严禁商用哦!

如果你喜欢这种风格,又还没体验过《Florence》这款独立佳作,强烈推荐支持正版、亲自体验这段温柔又动人的游戏旅程


🎨 Florence

A tribute to the emotionally powerful indie gem Florence, this LoRA model captures its warm, minimalist hand-drawn illustration style and key character designs.

Weights

  • Compatible with most anime-style checkpoints; recommended weight: 0.8 – 1.1

  • When used with realistic-style checkpoints, weight should be lowered to around 0.5.

Trigger Guide

Global Trigger:
florence style

Female protagonist (solo):
florence style, yeoh, solo

Male protagonist (solo):
florence style, krish, solo

Couple:
florence style, fcouple, yeoh, krish

Female protagonist after breakup (artist persona):
florence style, yeoh, solo, very short hair, artist

Empty scenes (recommended to use with NegPiP plugin):
florence style, (no human:1.5), (human:-1.2)

Use the following prompt to achieve

  • Looking at the viewer: looking at viewer

  • Side profile view: from side, profile

  • Solid color background: simple background

  • Color style control: limited palette, monochrome, greyscale

Output Resolution

This model was trained on 165 in-game screenshots with original resolutions of:

  • 512×512

  • 512×768

  • 768×512

It is recommended to keep the output resolution within 768×768 to avoid image distortion or artifacts.

Hires.fix

The art style of Florence has a lo-fi, hand-drawn aesthetic, which may appear too sharp or lose its atmosphere when using Hires.fix

Recommendations:

  • Upscaling algorithms: 4x-AnimeSharp or R-ESRGAN 4x+ Anime6B

  • Denoising strength:0.6

  • Enable: ControlNet Tile for better consistency

You may also try using a pure upscaling algorithm in the Extras tab (without any denoising), which sometimes preserves the original feel more naturally.

Disclaimer

Produced by CUC AI STUDIO
Feel free to use this LoRA for your creative works — we’d love to see what you come up with! 💛
This model is for personal study and non-commercial use only. Commercial use is strictly prohibited.

If you enjoy this art style but haven’t experienced the indie gem Florence yet, we highly recommend supporting the original and diving into this tender, heartfelt journey for yourself.