How to generate “photoreal synthetic data” from Omniverse environments with Cosmos?

[azazdeaz]

Hi There,

I can't figure out where to find this Cosmos feature

From the press release:

Physics-based photoreal synthetic data generation, using Cosmos models to generate photoreal videos from controlled 3D scenarios developed in the NVIDIA Omniverse™ platform.

NVIDIA Newsroom

In the video demo: https://youtu.be/9Uch931cDx8?t=58

Which model should be used for this? Probably i’m missing something, but the models i currently see in the catalog are all seem to be for generating new frames, and none of them are for enhancing existing ones. Also, i can't find anything about it in the paper or on the paper website

Thank you!

[anm-ol]

Same questions! Did you figure out how to do it?
I'm assuming there is some kind of video-to-video model but I have only seen I2V so far

[sophiahhuang]

Thanks for your interest in Cosmos and request for enhancing existing videos. Currently we support this model family. We have exciting plans to continuously expand our capabilities to better support our community and showcase the full potential of Cosmos. Stay tuned.

[azazdeaz]

Thank you for your answer. Since this feature has already been announced, do you have an estimate of if and when it will be available?