WAN2.1 FLF2V: Incorrect MASK Creation????

[KyujinHan]

Hello! I think that it is maybe error. (Or not, please explain it for me!!)

In WanImageToVideoPipeline class in pipline_wan_i2v.py,

(the code is the part of prepare_latents function)

For I2V, masking shape like as below:

[[1, 0, 0, ... , 0]
[1, 0, 0, ... , 0]
[1, 0, 0, ... , 0]
[1, 0, 0, ... , 0]]

I understood: when the mask is 1, input video frame does not change.
(*Mask shape: [1, 4, 21, 60, 104] = [B, C, F, H, W])

But in the FLF2V case, masking shape like as below:

[[1, 0, 0, ... , 0]
[1, 0, 0, ... , 0]
[1, 0, 0, ... , 0]
**[1, 0, 0, ... , 1]]**

Here, why the last frame mask has 1 only in last channel??
Is there anyone who can explain this part?

[KyujinHan]

@sayakpaul