Why perform the following operations on the latent condition?

[trouble-maker007]

in the code :https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/wan/pipeline_wan_i2v.py
line 395-404:

latents_mean = (
    torch.tensor(self.vae.config.latents_mean)
    .view(1, self.vae.config.z_dim, 1, 1, 1)
    .to(latents.device, latents.dtype)
)
latents_std = 1.0 / torch.tensor(self.vae.config.latents_std).view(1, self.vae.config.z_dim, 1, 1, 1).to(
    latents.device, latents.dtype
)

latent_condition = (latent_condition - latents_mean) * latents_std

The official inference code of Wan2.1 does not perform similar operations：
https://github.com/Wan-Video/Wan2.1/blob/main/wan/image2video.py#L237

[hlky]

The original code does perform the same operations inside the VAE, we have moved it to be uniform with the rest of our codebase.

https://github.com/Wan-Video/Wan2.1/blob/b58b7c573776b76b6fe8d36086590e033173f9b1/wan/modules/vae.py#L536-L540

https://github.com/Wan-Video/Wan2.1/blob/b58b7c573776b76b6fe8d36086590e033173f9b1/wan/modules/vae.py#L629-L663

[a-r-r-o-w]

I believe we can mark this issue as closed since OP reacted to above hlky's message, which resolves the issue