Code release for "AccDiffusion: An Accurate Method for Higher-Resolution Image Generation"
- 2024.11.19: 🔥 AccDiffusion v2 is available!
- 2024.07.18: 🔥 AccDiffusion has been accepted to ECCV'24!
conda create -n AccDiffusion python=3.9
conda activate AccDiffusion
pip install -r requirements.txt
python accdiffusion_sdxl.py --experiment_name="AccDiffusion" \
--model_ckpt="stabilityai/stable-diffusion-xl-base-1.0" \ # your sdxl model ckpt path
--prompt="A cute corgi on the lawn." \
--view_batch_size=16 \
--num_inference_steps=50 \
--seed=88 \
--resolution="4096,4096" \
--upscale_mode="bicubic_latent" \
--stride=64 \
--c=0.3 \ # c can be adjusted based on the degree of repetition and quality of the generated image
--use_progressive_upscaling --use_skip_residual --use_multidiffusion --use_dilated_sampling --use_guassian --use_md_prompt --shuffle