About noise difference in video & image generation

[XuWuLingYu]

In image generation, different views use the same noise as validation initial. torch.stack([latents]*6,dim=0) But in the video generation, it seems not work to do the same way. The nosie between views must be different. Is there any research can tell why?

[flymin]

We did not strictly verify these options.

We did try training image generation by adding the same noise to different views and it does not work.

I think the expected operation may be use different noise to each single view, no matter multi-view or video generation. Thus, it may not be necessary to use the same initial noise for image generation.

[github-actions[bot]]

This issue is stale because it has been open for 7 days with no activity. If you do not have any follow-ups, the issue will be closed soon.