Does model trained with ar or withour ar share the same inference pipeline?

[KaijingOfficial]

Thanks for the great work, I'm quite new in this.

[feifeiobama]

They do not. The model trained with ar does next frame prediction at every step, and the model trained without ar inference like DiT.

[KaijingOfficial]

Dear author, sorry to disturb you. Could you please share the generation code for the non-autoregressive version of the model? My implementation has encountered some issues. The first temporal unit always produces frames with a different appearance compared to other frames.