Have you compared with open-sora's video wav encoder decoder?

[chenerg]

The open-sora-plan also provided a video encoder-decoder called WF-VAE, which decomposes video into sub-bands using wavelet transforms. The details can be found https://github.com/PKU-YuanGroup/Open-Sora-Plan/blob/main/docs/Report-v1.3.0.md .

[qqingzheng]

WF-VAE has only released the distilled weights, not the direct training weights. By the way, I'm really thrilled to see this outstanding work cosmos tokenizer being open-sourced!