Hysocs/Aozora_SDXL_Training
A highly memory-efficient layer-selective fine-tuning script for Stable Diffusion XL (SDXL) that trains only your chosen UNet layers while freezing the rest, allowing full-quality fine-tuning on just 12 GB VRAM.
This tool helps artists and designers fine-tune Stable Diffusion XL models with their own images, even if they only have a single graphics card with 12-16GB of VRAM. You provide your images and captions, and the tool produces a personalized SDXL model capable of generating images in your unique style or incorporating specific elements. It's designed for individual creators who want to customize AI art models without needing expensive, high-end hardware or complex coding.
Use this if you have a consumer-grade NVIDIA GPU (like an RTX 3060/4060/4070) and want to train high-quality Stable Diffusion XL models efficiently using a user-friendly graphical interface.
Not ideal if you have a multi-GPU setup, as its memory-saving features designed for single GPUs might actually make training slower for you.
Stars
10
Forks
1
Language
Python
License
—
Category
Last pushed
Mar 15, 2026
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/diffusion/Hysocs/Aozora_SDXL_Training"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
TheLastBen/fast-stable-diffusion
fast-stable-diffusion + DreamBooth
XavierXiao/Dreambooth-Stable-Diffusion
Implementation of Dreambooth (https://arxiv.org/abs/2208.12242) with Stable Diffusion
VinAIResearch/Anti-DreamBooth
Anti-DreamBooth: Protecting users from personalized text-to-image synthesis (ICCV 2023)
bits-bytes-nn/py-dreambooth
Easily create your own AI avatar images!
HemantKArya/HqStableDiffusionColab
A High Quality (HD / 2K / 4K) Image Generation Using Stable Diffusion and Real-ESR / SwinIR /GFPGAN