Questions about WanVAE_tiny class

[nanwanj]

Question about encode_video

Regarding the implementation in:

LightX2V/lightx2v/models/video_encoders/hf/wan/vae_tiny.py

Line 12 in ca39ec4

class WanVAE_tiny(nn.Module):

This module provides the functions:

• decode
• encode_video
• decode_video

However, it does not define an encode function.

Would it be safe to call encode_video instead of encode in this case?

---

Question about the need_scaled flag

According to the README, when using the LightTAE series, the configuration should be:

{
    "use_tae": true,
    "need_scaled": true,
    "tae_path": "./models/vae/lighttaew2_1.pth"
}

However, in the following file:

https://github.com/ModelTC/LightX2V/blob/ca39ec400116cf57a4ee47e3f8d7f20bcd28f77e/lightx2v/models/video_encoders/hf/vid_recon.py

the need_scaled flag does not appear to be used.

Could you clarify:

1. When should need_scaled actually be set to True?
2. Is it required only for the LightTAE series?
3. If so, where in the pipeline is this flag expected to take effect?

[gushiqiao]

For TAE-series models, encode cannot be replaced with encode_video.The need_scaled parameter is only set to True for LightTAE, which is directly related to the video data preprocessing during our training phase.