[kungfupawnda]

It's not remixing training data. It is actually generating video using a diffusion transformer architecture.

More info here: https://www.reddit.com/r/LocalLLaMA/comments/1aspxox/explana...

[bamboozled]

What would the module produce without the training data?