ChatGPT cannot produce a black image

It's probably because the underlying diffusion model is started with grayish noise and can't easily turn that into complete black. At least Stable Diffusion had a similar problem. https://www.crosslabs.org/blog/diffusion-with-offset-noise

I asked it to generate one using matplotlib and it worked.

Worked the first time here: "please give me a 640x480 image where ALL pixels are black"