EasyPhoto is an AI-powered tool for generating personalized digital doppelgangers and portraits. It targets users who want to create custom AI-generated images of themselves or others, offering a user-friendly interface for training and inference without requiring extensive knowledge of Stable Diffusion WebUI. The primary benefit is the ability to generate realistic and personalized AI portraits using a small set of user photos.

How It Works

EasyPhoto leverages Stable Diffusion's image-to-image capabilities, specifically using LoRA (Low-Rank Adaptation) for fine-tuning. It trains a personalized "digital doppelganger" model from 5-20 user portrait images. During inference, it combines this LoRA model with template images and ControlNets (canny with color for fusion images, OpenPose for replaced images) to ensure similarity and stability. A two-stage diffusion process is employed, with the second stage using higher resolution for enhanced detail.

Quick Start & Requirements

• Install: git clone https://github.com/aigc-apps/EasyPhoto.git, cd EasyPhoto, pip install -r requirements.txt, python app.py
• Prerequisites: Python 3.10+, PyTorch 2.0.1+, CUDA 11.7+, cuDNN 8+. Requires an NVIDIA GPU (e.g., 3060 12GB or higher).
• Disk Space: ~60GB for weights and data.
• Docs: https://github.com/aigc-apps/EasyPhoto

Highlighted Details

• Supports training and inference for single and multi-person generation.
• Offers a Stable Diffusion WebUI plugin.
• Allows fine-tuning of background and similarity calculation.
• Utilizes ControlNets for enhanced image fidelity.

Maintenance & Community

• Active development with recent updates in September 2023.
• Community support via DingTalk group (38250008552) and WeChat.

Licensing & Compatibility

• Licensed under the Apache License (Version 2.0).
• Permissive license suitable for commercial use and integration with closed-source projects.

Limitations & Caveats

The project notes that dataloader num workers is not effective on Windows due to potential errors. Training is recommended with images not featuring glasses, as mixed usage may affect results.