🇬🇧 English
A robust and versatile Vision Language Model (LLM) API node for ComfyUI. It is designed to provide a unified interface for various OpenAI-compatible services, supporting multi-image inputs, and featuring intelligent, automatic GPU cleanup for local Ollama users.
✨ Key Features
- Flexible Image Inputs: Supports zero, one, or two image inputs, making it perfect for tasks from text generation to creating complex prompts for video transitions (e.g., start frame to end frame).
- Universal API Compatibility: Works with any service compatible with the OpenAI API V1
chat/completionsendpoint, such as Ollama, DeepSeek, Groq, and official OpenAI APIs. - Smart GPU Cleanup: Automatically detects local Ollama instances and executes
ollama stop <model_name>after each run to free up VRAM. - Advanced Seed Control: Offers
fixed,increment,decrement, andrandomizeoptions for the seed value after each generation, enabling creative automation. - Robust & User-Friendly: Features secure seed handling to prevent errors, robust image conversion for better compatibility, adjustable
max_tokens, and helpful error messages for easier debugging.
🔧 Installation
-
Clone the Repository
- Open your terminal or command prompt.
- Navigate to your ComfyUI
custom_nodesdirectory:cd path/to/your/ComfyUI/custom_nodes/ - Clone this repository using
git clone:git clone https://github.com/FeiMao-326/Comfyui-General-API-Node.git
-
Install Dependencies
- Navigate into the newly cloned folder:
cd Comfyui-General-API-Node - Install the required dependencies using
pip:pip install -r requirements.txt
- Navigate into the newly cloned folder:
-
Restart ComfyUI
- After completing the steps above, please restart ComfyUI completely.
💡 How to Use
-
Finding the Node: In ComfyUI, you can find the node by right-clicking ->
Add Node->FeiMao-326->FeiMao-326 General API Node. -
Seed Control Setup: To enable automatic seed changes (e.g.,
randomize), connect theseedoutput of the node back to its ownseedinput. This creates a loop that updates the seed after each run. -
Usage Scenarios:
- 📝 Text-Only Generation: Leave both
image_1andimage_2disconnected. - 🖼️ Single Image Description: Connect an image to
image_1. - 🎬 Dual Image for Video Transitions: Connect a start frame to
image_1and an end frame toimage_2.
Below is a complete example workflow for the dual image transition task:
- 📝 Text-Only Generation: Leave both
-
API Connection Examples:
- Local Ollama:
api_baseurl:http://127.0.0.1:11434/v1api_key:ollamamodel:llava(or another vision-capable model)cleanup_local_gpu: Keep it checked (True).
- External API (e.g., OpenAI):
api_baseurl:https://api.openai.com/v1api_key: Your OpenAI API key (sk-xxxxxxxx)model:gpt-4o
- Local Ollama:
📜 License
This project is licensed under the Apache 2.0 License. See the LICENSE file for details.
🇨🇳 中文
一款为 ComfyUI 设计的、功能强大且极为稳健的通用视觉大语言模型 (LLM) API 节点。它旨在为各种兼容 OpenAI 的 API 服务提供统一的接口,支持多图像输入,并为本地 Ollama 用户提供智能的 GPU 自动清理功能。
✨ 主要功能
- 灵活的图像输入: 支持零、一或两张图像输入,完美适用于从纯文本生成,到为视频转场创建复杂提示词的各种任务。
- 通用 API 兼容性: 支持任何兼容 OpenAI API V1
chat/completions接口的服务,例如 Ollama, DeepSeek, Groq, 以及 OpenAI 官方 API。 - 智能 GPU 清理: 自动检测本地 Ollama 实例,并在每次运行后执行
ollama stop <model_name>命令以释放您的宝贵显存。 - 高级种子控制: 在每次生成后,为种子值提供
fixed(固定),increment(递增),decrement(递减), 和randomize(随机) 选项,实现创造性的自动化工作流。 - 稳健且友好: 拥有安全的种子处理机制以防止错误、强大的图像转换以提高兼容性、可调节的
max_tokens,以及智能的错误提示,让调试更轻松。
🔧 安装方法
-
克隆仓库
- 打开您的终端或命令行工具。
- 导航到您的 ComfyUI
custom_nodes文件夹:cd path/to/your/ComfyUI/custom_nodes/ - 使用
git clone克隆此仓库:git clone https://github.com/FeiMao-326/Comfyui-General-API-Node.git
-
安装依赖
- 导航到刚刚克隆下来的节点文件夹:
cd Comfyui-General-API-Node - 使用
pip安装所有必需的依赖项:pip install -r requirements.txt
- 导航到刚刚克隆下来的节点文件夹:
-
重启 ComfyUI
- 完成以上步骤后,请完全重启 ComfyUI。
💡 如何使用
-
找到节点: 在 ComfyUI 中,您可以通过右键菜单 ->
Add Node->FeiMao-326->FeiMao-326 General API Node找到它。 -
设置种子控制: 若要启用自动种子变更(例如
randomize),请将节点的seed输出端口连接回它自身的seed输入端口。这个“循环”连接会在每次运行后自动更新种子值。 -
使用场景:
- 📝 纯文本生成: 将
image_1和image_2保持断开。 - 🖼️ 单图描述: 连接一张图片到
image_1接口。 - 🎬 双图视频转场: 连接起始帧到
image_1,连接结束帧到image_2。
下面是一个完整的双图转场任务的示例工作流:
- 📝 纯文本生成: 将
-
API 连接示例:
- 本地 Ollama:
api_baseurl:http://127.0.0.1:11434/v1api_key:ollamamodel:llava(或其他支持视觉的模型)cleanup_local_gpu: 保持勾选 (True)。
- 外部 API (以 OpenAI 为例):
api_baseurl:https://api.openai.com/v1api_key: 填入您的 OpenAI 密钥 (sk-xxxxxxxx)model:gpt-4o
- 本地 Ollama:
📜 许可证
本项目采用 Apache 2.0 许可证。详情请参阅 LICENSE 文件。

