Comfyui-General-API-Node

<div align="center">

ComfyUI General API Node Pack

Created by FeiMao-326

English | 中文说明

</div>

---

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:

1. 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.
   • Built-in Proxy Support: Native system_proxy parameter (e.g. http://127.0.0.1:7890) safely bridges network gaps without polluting global OS logic.
   • Force JSON mode: Enable force_json_format to guarantee 100% structured API responses ready for the downstream JSON Parser.
   • Seed Logic: Includes industrial-grade seed control (fixed, increment, randomize).
2. Text Batch Replace: A powerful text utility for performing up to 8 find-and-replace operations in a single node.
3. 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.
4. Show Text: Displays input text directly on the node UI.
5. Simple Text: A simple text input node for passing strings to other nodes.
6. Text Split By Delimiter: Splits a text string into a list based on a delimiter.
7. Get List Element: Retrieves a specific element from a list by index.
8. Text Iterator: Splits text by delimiter and outputs it as a batch, triggering batch execution for downstream nodes.
9. Markdown Extractor: Intelligent regex-based extraction of ```json or code blocks from LLM responses.
10. Text Template (Dynamic): [Recommended] Dynamic slot template. {var} automatically generates input ports. Supports all languages.
11. Prompt Cleaner: Normalizes prompts by removing duplicates, extra commas, and empty brackets.
12. Text Logic Switch: Flow control (If-Else) based on content matching (Contains, Regex, Exact, etc.).
13. Regex Extractor Pro: Advanced pattern capture with capture group support.
14. Dictionary Translator: Bulk mapping/replacement via JSON or Key=Value dictionaries.
15. Save Text: Persist generated prompts or long texts to local TXT/JSON files, supporting append mode.
16. Load Text: Read TXT/JSON files from your input folder directly into the workflow for templating or context injection.

🔧 Installation

1. 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
     

2. Install Dependencies

   • Navigate into the newly cloned folder:

     cd Comfyui-General-API-Node
     

   • Install the required dependencies:

     pip install -r requirements.txt
     

3. Restart ComfyUI

   • After completing the steps above, please restart ComfyUI completely.

💡 How to Use

General API Node

1. Finding the Node: In ComfyUI, you can find the node by right-clicking -> Add Node -> FeiMao-326 -> FeiMao-326 General API Node.

   

2. 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.

3. 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.

4. Universal API Connection Guide (Configuration Table): Built on the OpenAI-compatible protocol, this node seamlessly integrates with 99% of cloud LLMs and local frameworks.

| Vendor / Framework | api_baseurl | model (Example) | Notes | | :--- | :--- | :--- | :--- | | Local Ollama | http://127.0.0.1:11434/v1 | gemma4:e4b / llama3 | Native VRAM cleanup support | | Local LM Studio | http://127.0.0.1:1234/v1 | (Loaded model name) | Popular cross-platform tool | | Local WebUI (Oobabooga)| http://127.0.0.1:5000/v1 | (Loaded model name) | Classic open-source WebUI | | Local vLLM / TGI | http://127.0.0.1:8000/v1 | (Deployed model name) | Enterprise-grade engines | | OpenAI GPT-5 | https://api.openai.com/v1 | gpt-5.4-thinking / gpt-4o | Vision (VQA) supported | | DeepSeek | https://api.deepseek.com/v1 | deepseek-chat | High performance, cost-effective | | Alibaba Qwen | https://dashscope.aliyuncs.com/compatible-mode/v1 | qwen-vl-plus / qwen-max | vl models support Vision | | Zhipu AI (GLM) | https://open.bigmodel.cn/api/paas/v4 | glm-4v / glm-4 | 4v models support Vision | | Moonshot (Kimi) | https://api.moonshot.cn/v1 | moonshot-v1-8k | Great for long-context tasks | | Groq (Ultra-Fast) | https://api.groq.com/openai/v1 | llama-3.1-70b-versatile | Ultra-fast LPU inference | | NVIDIA NIM | https://integrate.api.nvidia.com/v1 | meta/llama-4-maverick | Official or Self-Hosted | | Google Gemini (Native)| https://generativelanguage.googleapis.com/v1beta/ | gemini-3.1-pro / ...-image-preview| Node auto-converts to REST |

> [!TIP]
> **Google Gemini Image Generation**: The API supports native Gemini Image Gen. To enable **Image Generation modes**, the model name MUST contain the keyword `image-` (e.g., `gemini-3.1-pro-image-preview`). For standard chat or vision (VQA), use regular names like `gemini-3.1-pro`.

5. 🌐 Built-in System Proxy (system_proxy) Guide:

This node has built-in proxy support. **No need to modify system environment variables.** Just enter your local proxy address.

**How it works**: Your proxy software (e.g., Clash, V2rayN) opens a local HTTP listening port when running. Simply enter that port address into the `system_proxy` field.

**Common proxy software defaults**:

| Proxy Software | Default Value |
| :--- | :--- |
| **Clash / Clash Verge** | `http://127.0.0.1:7890` |
| **V2rayN** | `http://127.0.0.1:10809` |
| **Shadowsocks (SS/SSR)** | `http://127.0.0.1:1080` |
| **Qv2ray** | `http://127.0.0.1:8889` |

> [!IMPORTANT]
> **When do you need to fill this in?**
> - Using **local Ollama / LM Studio** or **domestic APIs (DeepSeek, Alibaba, Zhipu)** → **Leave blank**.
> - Proxy software is running in **Global/System Proxy** mode → **Leave blank** (system already routes traffic).
> - Proxy software is running in **Rule-based mode** (browser works, but Python apps can't access foreign APIs) → **Fill this in**. This is the exact pain point this feature solves.

6. 🗂️ Force JSON Format (force_json_format) Guide:

When enabled, the node forces the LLM to return a **100% clean JSON string** without any surrounding chatter like "Sure! Here is your JSON:".

> [!TIP]
> **Recommended Pipeline**: `General API Node (force_json_format ✅)` → `Markdown Extractor` → `JSON Parser`. This pipeline achieves fully automated, zero-intervention structured data extraction.

![General API Node Workflow Example](./assets/workflow_example.png)

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").

---

Save Text

• Function: Saves any text output from your workflow to a local file in ComfyUI's output directory.
• filename: Enter a filename like result.txt. Subdirectories are supported (e.g., prompts/scene_01.txt — folders are created automatically).
• append toggle: When OFF, the file is overwritten each run. When ON, new content is appended to the end (ideal for batch runs that accumulate results).
• Scenario: Save 1000 generated prompts to a txt file during a batch production session.

Load Text

• Function: Reads a local file from ComfyUI's input directory and outputs it as a string.
• Supported formats: .txt, .json, .md, .csv.
• How to use: The node automatically scans the input folder on startup and lists available files in a dropdown. Just select a file.
• Scenario: Place a pre-written character bio, world-building doc, or long prompt template into the input folder, then inject it into General API Node with one click — no manual copy-paste needed.

📜 License

This project is licensed under the Apache 2.0 License.