Could diffusion model work for n channels (n>3)?

[random-var-x]

Hello,
My question is as the title. Why I have n > 3 is because I use DCT signals to represent an image.
I tried to modify your Dataset class to incorporate my data and set calculate_fid = False. The loss decreased from ~0.9 to ~0.26 and didn't go down anymore, which showed poor results.
So it can't work for n channels where n > 3? Or I need to dig deeper into the codes?