Your Face as a LoRA: A Stable Diffusion Plugin

What it does

EasyPhoto is a plugin for the Stable Diffusion WebUI that automates the creation of personalized AI portraits. You upload 5–20 photos of a person, and it trains a dedicated LoRA model to generate that face into preset or user-uploaded templates. The project also supports multi-person scenes, video inference without retraining, and SDXL template generation.

The interesting bit

Instead of leaving you to wrestle with prompts and inpainting, the plugin orchestrates ControlNet, background fine-tuning, and similarity scoring to keep the likeness intact. It also bundles recent speed tricks like LCM-LoRA to cut sampling from 50 steps down to 12.

Key highlights

• Trains a personal LoRA from 5–20 portrait photos inside the WebUI.
• Supports multi-person generation, video inference without retraining, and SDXL template generation.
• Integrates ControlNet (requires at least three models) to preserve pose and composition.
• Bundles attribute editing and virtual try-on via Concepts-Sliders.
• Demands serious hardware: verified on GPUs from 12 GB to 40 GB VRAM and roughly 60 GB of free disk space.

Caveats

• Docker builds trail the GitHub repository, so the latest features may not be present out of the box.
• Windows users may encounter OOM kills and need to zero out certain num_threads settings.
• The UI is English-only for now; Chinese localization is on the TODO list.

Verdict

Good for Stable Diffusion users who want a guided, UI-driven way to generate personalized portraits or group scenes without building a LoRA pipeline by hand. Skip it if you are running on modest hardware or prefer lightweight, API-first workflows; this is a heavy plugin that expects ControlNet and plenty of disk space.