Hello xfusion!

xfusion is a Python library specializing in assembling open-source AI models, with a particular focus on models like Stable Diffusion, Flux, and others.

Features

• Enhanced support for models from sources outside Hugging Face.
• Simple and efficient model loading using just a model URL.
• Extension for models.
• Multiple GPUs support.

Supported Models

• Stable Diffusion 1.5/2/3/3.5/XL
• Flux

Installation

pip install -q git+https://github.com/CyberVy/xfusion.git

or

curl -fsSL https://github.com/CyberVy/xfusion/raw/refs/heads/main/install.sh | bash

Code Example

Qucik start with bash

curl -Lso start.py "https://github.com/CyberVy/xfusion/raw/refs/heads/main/start.py"
export CIVITAI_TOKEN=""
export HF_HUB_TOKEN=""
export TG_TOKEN=""
export TG_ID=""
export MODEL_VERSION=""
python start.py

---

Jupyter Notebook

import os
os.environ["CIVITAI_TOKEN"] = ""
os.environ["HF_HUB_TOKEN"] = ""
os.environ["TG_TOKEN"] = ""
os.environ["TG_ID"] = ""
os.environ["DOWNLOAD_PATH"] = ""
os.environ["MODEL_VERSION"] = ""

import os
from xfusion import load_enhancer
from xfusion import load_stable_diffusion_ui,load_flux_ui
from xfusion.const import GPU_COUNT

telegram_kwargs = {"token":os.environ.get("TG_TOKEN",""),"chat_id":os.environ.get("TG_ID","")}
download_kwargs = {"directory":os.environ.get("DOWNLOAD_PATH","/xfusion_models")}
pipelines = [load_enhancer(None,model_version=os.environ.get("MODEL_VERSION",""),download_kwargs=download_kwargs,telegram_kwargs=telegram_kwargs) for i in range(GPU_COUNT)]

if os.environ.get("MODEL_VERSION","") in ["", "1.5", "2", "3", "3.5", "xl", "sdxl", "pony"]:
    server = load_stable_diffusion_ui(pipelines,_globals=globals(),pwa=True)
elif os.environ.get("MODEL_VERSION","") in ["flux"]:
    server = load_flux_ui(pipelines,_globals=globals(),pwa=True)

---

Directly use a pipeline in the backend

Code

from xfusion.enhancement import load_enhancer
import torch

telegram_kwargs = {"token":"","chat_id":""}
download_kwargs = {"directory":"./xfusion_models"}

model = "https://civitai.com/api/download/models/646523?type=Model&format=SafeTensor&size=pruned&fp=fp16"
pipeline = load_enhancer(model,model_version="xl", download_kwargs=download_kwargs, telegram_kwargs=telegram_kwargs).to("cuda")

prompt = """
young white woman with dramatic makeup resembling a melted clown, deep black smokey eyes, smeared red lipstick, and white face paint streaks, wet hair falling over shoulders, dark and intense aesthetic, fashion editorial style, aged around 20 years, inspired by rick genest's zombie boy look, best quality
"""
negative_prompt = """
bad hands, malformed limbs, malformed fingers, bad anatomy, fat fingers, ugly, unreal, cgi, airbrushed, watermark, low resolution
"""

num_inference_steps = 30
guidance_scale = 2
clip_skip = 0

seed = 13743883683399229202

width = None
height = None

images = pipeline(prompt=prompt,negative_prompt=negative_prompt,generator=torch.Generator(pipeline.device).manual_seed(seed),width=width,height=height,num_inference_steps=num_inference_steps,guidance_scale=guidance_scale,clip_skip=clip_skip).images

---

Acknowledgments

Xfusion leverages the Diffusers library and is inspired by the incredible work of the open-source community.