如何安装一个comfyui插件？ComfyUI_MiniCPM-V-2_6-int4为例

‍‍

#### 确认插件 requirements 里的依赖包都已经安装成功

Ensure all dependencies listed in the plugin requirements are successfully installed.

#### 启动 ComfyUI 时的报错信息

When starting ComfyUI via cmd, you may encounter error messages.

#### 安装缺少的依赖

Install any missing dependencies using the following command:

```

python_embeded\python.exe -s -m pip install decord

```

注意：`python_embeded\python.exe` 的路径需要是你的 ComfyUI 所使用的 Python 环境。  

Note: The path `python_embeded\python.exe` should be the Python environment used by ComfyUI.

#### 安装成功后，再重新启动 ComfyUI

After successful installation, restart ComfyUI.

#### 模型下载

Model Download:

在仓库的 README.md 里找到下载模型的说明，如果网络正常将自动下载，如果需要手动下载，可以查看代码。  

Refer to the README.md in the repository for model download instructions. It will automatically download if the network is normal. For manual download, check the code.

#### 如果碰到网络问题，可以尝试添加：

If you encounter network issues, you can try adding:

#### 手动下载方式

Manual Download Method:

在代码里找到下载模型的代码，参考所显示的地址，手动到 Hugging Face 下载。  

Find the code for model download in the codebase and manually download it from Hugging Face at:  

[Hugging Face MiniCPM-V-2\_6-int4](https://huggingface.co/openbmb/MiniCPM-V-2_6-int4/tree/main)

#### 下载完成后，尝试运行工作流

After downloading, try running the workflow.

#### 如果发现报错：

If you encounter errors:

需要安装 bitsandbytes  

Install `bitsandbytes`.

需要安装 accelerate  

Install `accelerate`.

安装完成后，再次重启 ComfyUI，运行成功：  

After installation, restart ComfyUI again and it should run successfully.

#### 最后，我做了个优化版本，主要优化的地方：

Finally, I have made an optimized version with the following improvements:

1.  路径读取支持通过配置文件设置。  

    Path reading supports configuration file settings.

2.  精简节点和参数，复用已有的节点。  

    Simplified nodes and parameters, reusing existing nodes.

#### 在 Mixlab-nodes 的 Image 目录下可以找到：

You can find it under the `Image` directory of `Mixlab-nodes`:

MiniCPM\_VQA\_Simple 节点  

`MiniCPM_VQA_Simple` node.

‍