A suite of powerful and versatile utility nodes for ComfyUI, designed to streamline complex workflows involving Large Language Models and text manipulation. This pack is created by FeiMao-326.

✨ Nodes Included

This pack contains the following nodes, all available under the FeiMao-326 category:

General API Node: A robust Vision LLM API node.
- Supports multi-image inputs.
- Features automatic GPU cleanup for local Ollama.
- [NEW] Native Google Gemini API support (Chat & Image Generation).
- Includes advanced seed control (fixed, increment, randomize). =======

🇬🇧 English

A suite of powerful and versatile utility nodes for ComfyUI, designed to streamline complex workflows involving Large Language Models and text manipulation. This pack is created by FeiMao-326 and is modernized for 2026 models.

✨ Nodes Included

This pack contains the following nodes available under the FeiMao-326 category:

General API Node: [Core] The Ultimate Multimodal LLM Engine.
- 2026 SOTA Model Support: Native adaptation for Gemini 3.1 Pro/Flash, GPT-5.4, Llama 4, and Claude 3.5.
- Smart Task Routing: Automated redirection for Google Gemini. Use standard names like gemini-3.1-pro or gemini-3.1-pro-image-preview for chat/vision; it automatically switches to Image Gen mode when the model name includes the image- keyword.
- Double VRAM Cleanup: Specialized for local Ollama. When cleanup_local_gpu is enabled, the node triggers a REST API keep_alive: 0 and a CLI fallback to physically purge model data from VRAM immediately after execution.
- Advanced Multimodal Vision: Supports concurrent parsing and comparison of 2 or more images.
- Seed Logic: Includes industrial-grade seed control (fixed, increment, randomize).

e9ec5ee (feat: Release v1.0.6 with 6 new advanced text nodes and Nodes 2.0 (Vue UI) optimization)

Text Batch Replace: A powerful text utility for performing up to 8 find-and-replace operations in a single node.
JSON Parser: Deconstructs complex, nested JSON payloads into multiple separate text outputs with a built-in labeled preview. It deeply searches for user-defined keys, perfect for handling structured prompts.
Show Text: Displays input text directly on the node UI.
Simple Text: A simple text input node for passing strings to other nodes.
Text Split By Delimiter: Splits a text string into a list based on a delimiter.
Get List Element: Retrieves a specific element from a list by index.
Text Iterator: Splits text by delimiter and outputs it as a batch, triggering batch execution for downstream nodes. <<<<<<< HEAD =======
Markdown Extractor: Intelligent regex-based extraction of ```json or code blocks from LLM responses.
Text Template (Dynamic): [Recommended] Dynamic slot template. {var} automatically generates input ports. Supports all languages.
Prompt Cleaner: Normalizes prompts by removing duplicates, extra commas, and empty brackets.
Text Logic Switch: Flow control (If-Else) based on content matching (Contains, Regex, Exact, etc.).
Regex Extractor Pro: Advanced pattern capture with capture group support.
Dictionary Translator: Bulk mapping/replacement via JSON or Key=Value dictionaries.

e9ec5ee (feat: Release v1.0.6 with 6 new advanced text nodes and Nodes 2.0 (Vue UI) optimization)

🔧 Installation

Clone the Repository

Open your terminal.
Navigate to your ComfyUI custom_nodes directory:
```
cd path/to/your/ComfyUI/custom_nodes/
```

Clone this repository:

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:
```
pip install -r requirements.txt
```
Restart ComfyUI
- After completing the steps above, please restart ComfyUI completely.

💡 How to Use

General API Node

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 the seed output of the node back to its own seed input. This creates a loop that updates the seed after each run.
Usage Scenarios:
- 📝 Text-Only Generation: Leave both image_1 and image_2 disconnected.
- 🖼️ Single Image Description: Connect an image to image_1.
- 🎬 Dual Image for Video Transitions: Connect a start frame to image_1 and an end frame to image_2.
- 📸 Multi-Image Analysis: You can connect multiple images (image_1, image_2, image_3...etc) for complex analysis tasks. <<<<<<< HEAD
API Connection Examples:
- Local Ollama:
  - api_baseurl: http://127.0.0.1:11434/v1
  - api_key: ollama
  - model: gemma3:4b (or another vision-capable model)
  - cleanup_local_gpu: Keep it checked (True).
- External API (e.g., OpenAI):
  - api_key: Your OpenAI API key (sk-xxxxxxxx)
  - model: gpt-4o
- Google Gemini (Native):
  - api_baseurl: https://generativelanguage.googleapis.com/v1beta/
  - api_key: Your Google AI Studio Key
  - model: gemini-1.5-pro (Chat) or imagen-3.0-generate-001 (Image) Below is a complete example workflow for the dual image transition task:

Text Batch Replace

Input any text into the text field.
Fill in the find_x and replace_x fields to perform sequential replacements.

JSON Parser

Paste your complex JSON into the json_payload field.
In the keys_to_extract field, enter the keys you want to extract, separated by commas (e.g., shot, subject, audio).
The node will find these keys anywhere in the JSON, output their combined values to the corresponding output_x ports, and show a preview inside the node.

Show Text

Connect any string output to the text input.
The text will be displayed on the node itself.

Simple Text

Enter your text in the text box.
Connect the text output to any node requiring string input.

Text Split By Delimiter

Input text and a delimiter (default is comma).
Outputs a LIST type that can be used with Get List Element.

Get List Element

Connect a LIST input.
Specify the index (0-based) to retrieve a specific string.

Text Iterator

Input text and a delimiter.
Outputs each split item as a separate batch item.
Useful for iterating over a list of prompts or parameters.

📜 License

This project is licensed under the Apache 2.0 License. See the LICENSE and NOTICE files for details.

🇨🇳 中文

一套为 ComfyUI 设计的、功能强大且用途广泛的实用节点包，旨在简化涉及大语言模型和文本处理的复杂工作流。本节点包由 FeiMao-326 创作。

✨ 包含的节点

本节点包包含以下节点，您都可以在 FeiMao-326 分类下找到它们：

General API Node: 一个强大的视觉LLM API节点。
- 支持多图像输入。
- 为本地 Ollama 提供自动 GPU 清理功能。
- [新增] 原生 Google Gemini API 支持 (对话 & 绘图)。
- 包含高级种子控制（固定, 递增, 随机 等）。
Text Batch Replace: 一个强大的文本工具，可在单个节点中执行多达8次的查找与替换操作。
JSON Parser: 可将复杂的、深度嵌套的JSON结构，解析为多个独立的文本输出，并自带带标签的预览功能。它会深度搜索用户定义的关键字，非常适合处理结构化提示词。
Show Text: 直接在节点界面上显示输入的文本。
Simple Text: 一个简单的文本输入节点，用于将字符串传递给其他节点。
Text Split By Delimiter: 根据分隔符将文本字符串分割为列表。
Get List Element: 通过索引从列表中检索特定元素。
Text Iterator: 根据分隔符分割文本并作为批次输出，触发下游节点的批次执行。

🔧 安装方法

克隆仓库

打开您的终端。
导航到您的 ComfyUI custom_nodes 文件夹：
```
cd path/to/your/ComfyUI/custom_nodes/
```

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
```
- 安装所需的依赖项：
```
pip install -r requirements.txt
```
重启 ComfyUI
- 完成以上步骤后，请完全重启 ComfyUI。

💡 如何使用

General API Node

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。
- 📸 多图分析: 您最多可以连接多张图片 (image_1, image_2, image_3...等) 进行复杂的分析任务。
API 连接示例:
- 本地 Ollama:
  - api_baseurl: http://127.0.0.1:11434/v1
  - api_key: ollama
  - model: gemma3:4b (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/v1
  - api_key: 填入您的 OpenAI 密钥 (sk-xxxxxxxx)
  - model: gpt-4o
- Google Gemini (原生):
  - api_baseurl: https://generativelanguage.googleapis.com/v1beta/
  - api_key: 填入您的 Google AI Studio 密钥
  - model: gemini-1.5-pro (对话) 或 imagen-3.0-generate-001 (绘图) 下面是一个完整的双图转场任务的示例工作流：

Text Batch Replace

在 text 字段中输入任意文本。
填写 find_x 和 replace_x 字段以执行顺序替换。

JSON Parser

将您的复杂JSON粘贴到 json_payload 字段中。
在 keys_to_extract 字段中，输入您想提取的关键字，用逗号分隔（例如 shot, subject, audio）。
节点会在JSON的任何位置找到这些关键字，将其合并后的值输出到对应的 output_x 端口，并在节点内显示预览。

Show Text

将任何字符串输出连接到 text 输入。
文本将直接显示在节点上。

Simple Text

在文本框中输入您的文本。
将 text 输出连接到任何需要字符串输入的节点。

Text Split By Delimiter

输入文本和分隔符（默认为逗号）。
输出一个 LIST 类型，可与 Get List Element 配合使用。

Get List Element

连接 LIST 输入。
指定索引（从0开始）以检索特定字符串。

Text Iterator

输入文本和分隔符。
将每个分割项作为单独的批次项输出。
用于迭代提示词列表或参数列表。

📜 许可证

本项目采用 Apache 2.0 许可证。详情请参阅 LICENSE 和 NOTICE 文件。

API Connection Examples:
- NVIDIA NIM (Official/Local):
  - api_baseurl: https://integrate.api.nvidia.com/v1
  - api_key: Your NVIDIA API Key (nvapi-xxxxxxxx)
  - model: meta/llama-4-maverick, nvidia/nemotron-3-ultra
- OpenAI GPT-5 (Latest):
  - api_baseurl: https://api.openai.com/v1
  - api_key: Your OpenAI API key
  - model: gpt-5.4-thinking or o3-pro (Reasoning model)
- Local Ollama:
  - api_baseurl: http://127.0.0.1:11434/v1
  - api_key: ollama
  - model: gemma4:e4b (Current default, supports vision analysis)
  - cleanup_local_gpu: Keep it checked (True).
- Google Gemini (Native):
  - api_baseurl: https://generativelanguage.googleapis.com/v1beta/
  - api_key: Your Google AI Studio Key
  - model: gemini-3.1-pro (Chat/Vision) or gemini-3.1-pro-image-preview (Image Gen)
  - Tip
  - Important
    
    : For Image Generation, the model name MUST contain the keyword image- (e.g., gemini-3.1-pro-image-preview). For Chat/Reasoning, use standard names like gemini-3.1-pro.

Text Batch Replace

Perform multiple find-and-replace operations sequentially.

JSON Parser

Automatically parses LLM JSON outputs into independent string slots based on defined keys.

Simple Text / Show Text

Basic input and visual display modules.

Text Split / Get Element / Iterator

Essential tools for handling lists and batch execution flows.

Markdown Extractor

Problem: LLMs often reply with chatter like "Sure! Here is your JSON: json ... ".
Solution: Automatically identifies and extracts clean content from Markdown code blocks.
Scenario: When LLM generates a JSON structure, passing it directly to JsonParser would crash due to Markdown syntax. This node ensures a "smooth" connection.

Text Template (Dynamic)

Concept: Write a template with variables: A {subject} in {style} lighting, {mood} background.
Magic: The node instantly creates input slots for subject, style, and mood.
Note: In Nodes 2.0 (Vue UI), right-click and "Refresh" the node after modifying variables.

Prompt Cleaner

Function: Removes duplicate keywords, merges extra commas, and cleans inconsistent spaces.
Scenario: Multiple prompt concatenations often result in messy punctuation (e.g., , ,). Use this as a "final touch" before KSampler.

Text Logic Switch

Function: Flow control (If-Else). If input text contains a keyword (or matches regex), output travels to the "True" port, otherwise "False".
Scenario: Check if LLM output contains "Error" to stop the workflow or switch to a fallback model.

Regex Extractor Pro

Function: Precise data extraction using Capture Groups.
Scenario: Extract specific info like HEX color codes (#FFFFFF) or numeric coordinates from a long LLM description.

Dictionary Translator

Function: Bulk tag replacement using a mapping table (JSON or Simple Text Pairs).
Scenario: Map common Chinese terms to professional SD tags (e.g., replace "Cinematic" with "cinematic lighting, 8k, highly detailed").

📜 License

This project is licensed under the Apache 2.0 License.

e9ec5ee (feat: Release v1.0.6 with 6 new advanced text nodes and Nodes 2.0 (Vue UI) optimization)