Generated videos are pure noise without parameter changes; possible model class mismatch?

[Anastasiawd]

Hi, thank you for releasing this project.

I am encountering an issue where all generated videos are pure noise, even though I did not modify any parameters in the provided scripts. The problem occurs consistently across runs.

During inference, I also see the following warnings/errors:
`Expected types for vae:
(<class 'diffusers.models.autoencoders.autoencoder_kl_cogvideox.AutoencoderKLCogVideoX'>,),
got <class 'robomaster.models.autoencoder_magvit.AutoencoderKLCogVideoX'>.

Expected types for transformer:
(<class 'diffusers.models.transformers.cogvideox_transformer_3d.CogVideoXTransformer3DModel'>,),
got <class 'robomaster.models.transformer3d.CogVideoXTransformer3DModel'>.`

I am not sure whether this type mismatch between diffusers and robomaster implementations is the root cause of the noisy outputs, or if the warnings can be safely ignored. Below is one of the output video:

close.mp4