Using non square images for training

#35
by demo001s - opened

Can I modify the code for resizing the image and use images with different scales for training, such as 3:2. I don't know if it's feasible?

🧨Diffusers org

I think that depends bit if the underlying base model would support good generations at the scale (SDXL UNet in this case). It might require fine-tuning the base model first and then doing the ControlNet training. I am not very sure, though.

Sign up or log in to comment