CutDiffusion

Code release for "CutDiffusion: A Simple, Fast, Cheap, and Strong Diffusion Extrapolation Method" Paper Link.

Experiments environment

Set up the dependencies as:

conda create -n CutDiffusion python=3.9
conda activate CutDiffusion
pip install -r requirements.txt

Higher-image generation

python cutdiffusion.py --experiment_name="cutdiffusion" \
    --model_ckpt="stabilityai/stable-diffusion-xl-base-1.0" \   # your model ckpt path
    --num_inference_steps=50 \
    --view_batch_size=1 \
    --seed=6 \
    --resolution="2048,2048" \
    --prompt="A corgi sits on a beach chair on a beautiful beach, with palm trees behind, high details." \
    --stride=64 \
    --shuffle 

Acknowledgement

We would like to express our profound appreciation and respect towards MultiDiffusion, which has been a significant source of inspiration for our concept. Furthermore, our code is heavily influenced by the principles and methodologies established in DemoFusion.