README.md

metadata

title: Avatar Generator
emoji: 👔
colorFrom: blue
colorTo: green
sdk: docker
sdk_version: latest
app_file: app.py
pinned: false

Avatar Generator API

A streamlined, Diffusers-based implementation for generating avatar images using SDXL, ControlNet, and face conditioning.

Overview

This application provides a direct alternative to ComfyUI for avatar generation, focusing on:

1. Reliability: Direct control over file paths and permissions
2. Performance: Optimized pipeline for faster generation
3. Simplicity: Clean FastAPI-based architecture
4. Extensibility: Easy to modify and enhance

Features

• Pose-controlled generation using ControlNet
• FastAPI endpoints for image generation
• Face conditioning for consistent identity
• Docker-based deployment for Hugging Face Spaces
• Proper path handling to avoid permission issues

API Usage

The application exposes a /generate endpoint that accepts:

• prompt: Text description of the desired image
• name: Person name (used for filename)
• role: Role/job (used for styling and filename)
• pose_image: Reference image for pose
• shirt_image: Reference image for clothing
• face_image: (Optional) Reference image for face identity
• steps: (Optional) Number of inference steps
• guidance_scale: (Optional) How strongly to weight the prompt
• width & height: (Optional) Output image dimensions
• seed: (Optional) For reproducible generations

Models

The application uses:

• SDXL Base: stabilityai/stable-diffusion-xl-base-1.0
• ControlNet: thibaud/controlnet-openpose-sdxl-1.0
• PuLID Models: Defter77/pulid-models

Environment Variables

The following environment variables can be set:

• HF_TOKEN: Hugging Face API token for private models
• HF_HOME, HF_HUB_CACHE, TRANSFORMERS_CACHE: Set to /app/.cache by default

Development

To run locally:

# Build the Docker image
docker build -t avatar-generator .

# Run the container
docker run -p 7860:7860 avatar-generator

Advantages Over ComfyUI

This implementation offers several benefits:

1. Direct File Control: Explicit handling of file paths and permissions
2. Simplified Architecture: No complex node graph to maintain
3. Reliable Error Handling: Proper error reporting and cleanup
4. Flexible API: Easy to integrate with other applications
5. Optimized Resources: More efficient memory usage

Credits

Created by Defter77, based on the Hugging Face Diffusers library.