AI Portrait Generator with ComfyUI

TLDR: 1.Loads photo descriptions 2.Uses Ollama to enhance the descriptions 3.Sends them to ComfyUI 4.Saves the generated images

A stunning portrait showcasing intricate tattoo details and mesmerizing blue eyes

Overview

This project uses ComfyUI to generate hyperrealistic portraits using a custom workflow with the FLUX model. The system consists of two main scripts that work together to create beautiful, AI-generated portraits:

1. fetch_prompt.py - Fetches and processes portrait prompts
2. run-comfy.py - Generates images using ComfyUI with enhanced prompts

Features

• High-quality portrait generation using FLUX model
• Prompt enhancement using Ollama
• Custom workflow optimized for photorealistic results
• Batch processing capabilities
• Advanced noise seed management

Creative "bug's-eye view" demonstrating the system's capability for unique perspectives

Prerequisites

• Python 3.8+
• ComfyUI installation
• Ollama with qwen2.5:14b model
• Required models:
  • FLUX model
  • FluxRaslismLora.safetensors
  • Eye_Detail_Flux_Lora.safetensors
  • skin texture style v5.safetensors

Installation

1. Clone this repository:

git clone <repository-url>
cd ai-portrait-generator

2. Install dependencies:

pip install datasets requests

3. Ensure ComfyUI is properly set up with required models

Usage

Step 1: Generate Prompts

Run the prompt fetcher to create the initial prompt file:

python fetch_prompt.py

This will create simple_prompts.json with curated portrait prompts.

Step 2: Generate Images

Start the image generation process:

python run-comfy.py

Dynamic lighting demonstration with neon accents and moody atmosphere

How It Works

1. Prompt Preparation

   • Fetches carefully curated prompts from a dataset
   • Processes and saves them in a structured JSON format

2. Image Generation

   • Loads prompts from JSON
   • Enhances descriptions using Ollama
   • Processes through ComfyUI workflow
   • Generates high-quality portraits

3. Workflow Details

   • Uses FLUX model with custom LoRA adjustments
   • Implements advanced sampling techniques
   • Manages batch processing and seed variation

Configuration

The workflow can be customized through workflow.json, which includes:

• Image dimensions (1344x756)
• Sampling steps (25)
• Guidance scale (5.5)
• Various LoRA weights and strengths

Tips for Best Results

1. Prompt Engineering

   • Include specific details about facial features
   • Describe lighting conditions
   • Mention camera angle and perspective
   • Add emotional elements

2. Model Settings

   • Adjust guidance scale for different styles
   • Experiment with sampling steps
   • Fine-tune LoRA strengths

Contributing

Contributions are welcome! Please feel free to submit pull requests or create issues for bugs and feature requests.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• ComfyUI team
• FLUX model creators
• Dataset contributors