Video Autoregression Dreamer

[DevinCheung]

Thanks for your awesome work! May I ask when the video-version world dreamer would be released? Thanks!

[DevinCheung]

And the pretrained model "SDv1.5_mv_single_ref_nus" is only for multi-view image generation, not multi-view video generation? Thanks!

[jokester-zzz]

Thank you for your recognition.

"SDv1.5_mv_single_ref_nus" is a single-frame implementation of the autoregressive generation algorithm, it does has limitations in preserving temporal consistency. We are actively working on a more continuous version and have already achieved some promising results. For more details, you can refer to our report. We may also consider open-sourcing this part of the code in the future.