---
title: ZZZ-Style-QwenImage-LoRA
canonical_url: "https://www.modelscope.cn/models/wesjos/ZZZ-Style-QwenImage-LoRA"
md_url: "https://www.modelscope.cn/models/wesjos/ZZZ-Style-QwenImage-LoRA.md"
repository: wesjos/ZZZ-Style-QwenImage-LoRA
chinese_name: "Qwen Image 2.1 绝区零风格LoRA"
last_updated: 2026-09-29
license: apache-2.0
pipeline_tag: text-to-image-synthesis
tasks:
  - text-to-image-synthesis
base_model:
  - Qwen/Qwen-Image-2.1
base_model_relation: adapter
parameters: 318.8M
tensor_type:
  - F16
library_name:
  - lora
  - safetensors
language:
  - en
  - zh
supports_inference: txt2img
downloads: 78
stars: 1
tags:
  - text-to-image
  - lora
  - qwen-image
  - style-lora
  - zenless-zone-zero
  - game-style
  - comfyui
---

# ZZZ-Style-QwenImage-LoRA

> ZZZ-Style-QwenImage-LoRA - wesjos 在 ModelScope 开源的模型。these amre lora adapters for zzz game style.

wesjos/ZZZ-Style-QwenImage-LoRA 是 ModelScope 魔搭社区上的 318.8M 参数text-to-image-synthesis模型，采用 apache-2.0 许可，基于 Qwen/Qwen-Image-2.1 构建，并支持在线推理（txt2img）。

- **Repository**: wesjos/ZZZ-Style-QwenImage-LoRA
- **License**: apache-2.0
- **Tasks**: text-to-image-synthesis
- **Parameters**: 318.8M
- **Base model**: Qwen/Qwen-Image-2.1
- **Online inference**: txt2img
- **Tags**: text-to-image, lora, qwen-image, style-lora, zenless-zone-zero, game-style, comfyui
- **Downloads**: 78
- **Stars**: 1
- **Last updated**: 2026-09-29

Source: https://www.modelscope.cn/models/wesjos/ZZZ-Style-QwenImage-LoRA

---

# ZZZ Style LoRA (Qwen-Image-2.1) — zzz_style 系列

> 绝区零(Zenless Zone Zero)风格 LoRA,基于 **Qwen-Image-2.1** 训练。触发词 `zzz_style`,还原 ZZZ。
>

## 📋 模型简介

| 属性 | 值 |
|---|---|
| **架构** | Qwen-Image-2.1 (qwen_image_2 arch) |
| **类型** | 画风 Style LoRA(优化手脚) |
| **触发词** | `zzz_style`或者不用 |
| **Rank / Alpha** | 32 / 64 |
| **底模量化** | Comfy-Org int8 convrot + convrot8 训练 |
| **训练框架** | ai-toolkit (flowmatch / shift) |

## 📦 Checkpoint 选择

- step 2000/3000 首选, step 4000 有点过拟合
- cfg 建议 4-5, step 25-40, lora scale 0.8

## 🚀 使用方法

### Prompt 模板

```
zzz_style, furina from genshin impact,waving hands
```

### 推荐 Scale

| Scale | 效果 |
|---|---|
| 0.6 ~ 0.8 | 风格轻盈,内容自由度高 |
| **0.8** | **实测甜点起点,风格完整且不压构图** |
| 1.0 ~ 1.1 | 风格最强,可能影响构图/姿势 |

### ComfyUI

2. 底模:`qwen_image_2.1_int8_convrot.safetensors`(diffusion_models)+ `qwen3vl_8b_int8_convrot.safetensors`(text_encoder)+ `qwen_image_2.1_vae_bf16.safetensors`(vae)
3. KSampler:25 steps / CFG 4.3 / euler+simple / shift 3.1

### 负面提示

不需要专用负面词,留空即可;如遇质量波动可用 `blurry, low quality, jpeg artifacts`。

## ⚠️ 局限

1. 训练集以 **1girl 单人像**为主,双人/多人场景与男性角色泛化较弱
2. 底模为 int8 量化训练(convrot8)

## 🙏 致谢

- 底模:[Qwen-Image-2.1](https://www.modelscope.cn/models/Qwen/Qwen-Image-2.1)(阿里 Qwen 团队)
- 训练框架:[ai-toolkit](https://github.com/ostris/ai-toolkit)(ostris)
- 原作美术:米哈游《绝区零》美术团队

---