question about the init latents and motion layers

[jianshu-qin]

The code inits latents by generated 16 different noise, and it works well . But when I try to generate 1 noise and repeat it to batchsize 16, the output will be 16 same images when I offload the motion module. It's strange that with the motion module, the output is 16 noise ! I wonder if someone can explain it. Thanks.