Hub sample - same image always generated

I am trying to run Inference on server to generate some images from text.
Here is the sample I am using: Run Inference on servers

(I added my personal token to bypass the limit).

Everytime I run this sample with the same prompt, the same image is generated. I would like to have something new generated always, like ChatGPT is doing.

I experimented with guidance_scale but the image turns to complete garbage.

I used only default model and stabilityai/stable-diffusion-2-1