Inpaint4DNeRF: Promptable Spatio-Temporal NeRF Inpainting with
Generative Diffusion Models

Videos of Inpainting Examples

3D Examples


Text prompt: an Argentina soccer player

Text prompt: an old professor, standing

Text prompt: telegraph pole

RGB, Final Result

Depth map, Final Result

RGB, Warmup Result

Depth map, Warmup Result

RGB, Initial NeRF

Depth map, Initial NeRF


4D Examples

Below we show 3 dynamic NeRF inpainting examples, including the example in our paper. For each example, we show a multiview video of the final result, as well as the "seed video" generated from the first seed images along with the original video.

For the third example "the girl", instead of generating new objects, we try to recover parts of the girl and the background that is occluded by the red dinosaur balloon in the original NeRF. While our final result is not perfectly clear, we can recover the static background and approximately the girl's motion in the NeRF. This example demonstrates that our method can not only generate brand new objects, but also generate missing content that is partially occluded.


Text prompt: a golden sword, side view

Text prompt: two storey red bus, driving towards left

Text prompt: the girl

Final Render

Seed Video

Original Video