Demo about counterfactual object insertion

[choyingw]

Hi,

Thanks for the cool work. I wonder how to make the counterfactual object insertion, like an elephant walking on the street. Do you have an example using the inference code? Thank you!

[YuxueYang1204]

Hi, thanks for your interest!

The generation model takes rendered images, masks, depth, and Plucker embeddings as conditions. You can mask out the region you want to edit and use a prompt to control the generation.

That said, NeoVerse is not a dedicated video editing tool, so the success rate for complex edition may be limited. We'd recommend using a more specialized video editing model like VACE to perform the editing on a single-view video first, and then feed the edited video into NeoVerse to generate multi-view data or other novel views you need.

We are also preparing a tutorial for producing these demos — stay tuned!

[Jumponthemoon]

Hi, great work! I'm wondering any planned updates on the tutorial?