ComfyUI Pinecone Prompt Upserter

ComfyUI Pinecone Prompt Upserter

A custom ComfyUI node for managing image generation prompts using Pinecone vector database with OpenAI embeddings. Features interactive approval with image preview for quality control.

Features

• Interactive Approval: Visual preview with buttons before saving prompts
• Automatic Trigger Word Sanitization: Replaces character names (e.g., Sarah01) with <trigger_word> placeholder
• Deduplication: Prevents duplicate prompts using SHA256 hashing
• Smart Memory: Remembers decisions - only prompts first-time prompts
• OpenAI Embeddings: Uses text-embedding-3-large model (3072 dimensions)
• Image Preview: See generated image before deciding to save prompt
• Beautiful UI: Gradient buttons with hover animations in centered popup

Prerequisites

• Python 3.8+
• ComfyUI installation
• OpenAI API key
• Pinecone account with an index created (3072 dimensions)

Installation

1. Navigate to your ComfyUI custom nodes directory:

cd /path/to/ComfyUI/custom_nodes/

2. Clone this repository:

git clone https://github.com/Hearmeman24/ComfyUI-Pinecone.git

3. Install dependencies:

cd ComfyUI-Pinecone
pip install -r requirements.txt

4. Set environment variables:

export OPENAI_API_KEY=your_openai_api_key_here
export PINECONE_API_KEY=your_pinecone_api_key_here

5. Restart ComfyUI

Usage

1. In ComfyUI, add the Pinecone Prompt Upserter node to your workflow (found under prompt_management category)

2. Connect inputs:

   • image (optional): Connect the generated image output to see it during approval
   • prompt_text: The prompt used (connect from prompt node)
   • model_name: Name of the model (e.g., "flux", "qwen", "wan", "sdxl")
   • index_name: Your Pinecone index name (default: "prompts")
   • pinecone_api_key (optional): Your Pinecone API key (uses env var if empty)
   • openai_api_key (optional): Your OpenAI API key (uses env var if empty)

3. How it works:

   First Time Seeing a Prompt:

   • Workflow pauses when it reaches the node
   • Popup appears automatically showing the generated image
   • Two buttons below image: "Insert to Index" (green) and "Cancel" (red)
   • Make your decision based on the image quality
   • Popup closes and workflow continues

   Subsequent Runs with Same Prompt:

   • Node recognizes the prompt instantly
   • Silently skips without interrupting
   • No popup, no pause - smooth workflow execution

Configuration

Environment Variables

• OPENAI_API_KEY: Required for generating embeddings
• PINECONE_API_KEY: Required for accessing Pinecone index
• INDEX_NAME: Optional, defaults to "prompts"

Pinecone Index Setup

Make sure your Pinecone index is configured with:

• Dimensions: 3072 (for text-embedding-3-large)
• Metric: cosine (recommended for text embeddings)

Example index creation:

from pinecone import Pinecone, ServerlessSpec

pc = Pinecone(api_key="your_api_key")
pc.create_index(
    name="prompts",
    dimension=3072,
    metric="cosine",
    spec=ServerlessSpec(cloud="aws", region="us-east-1")
)

Metadata Structure

Each prompt stored in Pinecone includes:

• text: The full prompt text (with trigger words sanitized)
• model: The model name you specify (e.g., "flux", "qwen", "wan", "sdxl")
• prompt_hash: SHA256 hash for deduplication

Trigger Word Sanitization

The node automatically replaces character-specific trigger words with a generic placeholder:

• Pattern: Name01 (capital letter + letters + "01")
• Examples: Sarah01, Onyx01, MaryJane01
• Result: All replaced with <trigger_word>
• Purpose: Maintains consistency and allows flexible character substitution

Deduplication

Prompts are deduplicated using SHA256 hashing of normalized text (lowercase, stripped whitespace):

• First time: Popup appears for approval
• Already seen: Silently skips (no interruption)
• Already in Pinecone: Won't duplicate even if approved again

Preview Features

• Auto-Appear: Popup automatically displays when a new prompt is detected
• Image Display: Large centered preview (up to 90% of viewport)
• Styled Buttons: Gradient green (Insert) and red (Cancel) with hover animations
• Manual Re-display: Click "📷 Show Preview" on node if you close the popup
• One-Click Decision: See image and click choice in single popup window

Troubleshooting

"API key not found" error

• Ensure environment variables are set before starting ComfyUI
• Alternatively, provide API keys directly in the node inputs

"Index not found" error

• Verify your Pinecone index name matches your configuration
• Ensure the index exists and has 3072 dimensions

Node not appearing in ComfyUI

• Restart ComfyUI completely after installation
• Check console for errors
• Verify dependencies installed in correct Python environment

Preview popup not appearing

• Check browser console for JavaScript errors
• Ensure you're connecting an image to the node's image input
• Try clicking "📷 Show Preview" button manually

Dimension mismatch error

• Pinecone index must be created with 3072 dimensions
• This matches OpenAI's text-embedding-3-large model

License

MIT License