- CUDA (tested on cu121 only for now)
- pytorch 2.0+
Install additional dependencies:
pip install -r requirements.txt
from cg2real import CG2Real
from PIL import Image
cg = CG2Real()
image = Image.open("your image path here")
realistic = cg("prompt to describe your image", image)
If using less than 16GB-ish VRAM, you can call cg = CG2Real(low_memory=True) to offload pipeline components as they are called ; this will slow down inference.
To increase or decrease fidelity to the original image, use: realistic = cg("prompt to describe your image", image, fidelity=[value between 0 and 1])
To run more denoising iterations, use the iterations parameter.