Reproduction of results

[enesertas]

Hello, I tried to check the results in this repo using the huggingface demo https://huggingface.co/spaces/KAIST-Geometric-AI-Lab/syncdiffusion-demo and the notebook provided in the notebooks directory in this repo, however the structural alignment and the overall coherency were not really great. Is there any way to improve the outputs of the notebook? Haven't changed any parameters while using it.

Here are the results I obtained using "Skyline of New York City":

Image from huggingface demo

Image from the notebook