# @karenxcheng on Instagram

- **Type:** Video
- **Original URL:** https://www.instagram.com/p/Cf_06UgjbzW
- **Gondola URL:** https://gondola.cc/posts/16472567-karenxcheng-instagram
- **Thumbnail:** https://img.gondola.cc/tr:w-,h-,fo-auto/postThumbnails/91fc4c62b8.jpg
- **Posted:** 2022-07-14T15:17:01.000+00:00
- **Account Owner:** Karen X (@karenxcheng) — https://gondola.cc/karenxcheng

## Caption

Bullettime using phone footage, made possible with AI 🤳 collab with jperl #ad made for nvidiacreators

What's interesting about this tech (NeRF) is you can make a video from any angle -
Any camera path you want. You're not limited to the camera paths you shot from 🤯

Our process:
1) Have everyone stay PERFECTLY still - any movements will cause blurry scans or artifacts
2) Take a video from as many angles as possible. You can use your phone for this
3) jperl used COLMAP to calculate camera positions
4) Then he fed that data into Instant-NGP to create a NeRF (it's like a 3D model, but volumetric)
5) He wrote custom code to bring it into Blender to get the camera movements
6) I made a few final finishing touches (speed ramps, vignetting and color grading)

While this process is still technically challenging today, I do think within a few years it will be as easy as pressing a button on an app on your phone. Already, the latest version we used (NVIDIA's Instant-NGP) is much faster than the original NeRF. This tech will only keep getting better.

Challenges
▸ We tried about a dozen variations of this scene before getting it to work
▸ Humans are especially challenging bc we have a hard time staying perfectly still. That's why most NeRFs you see are of inanimate objects
▸ Even harder is when you need to scan multiple people. Because not only do they need to stay perfectly still - they need to do so for a LONG time. The bigger the scene, the longer it takes to scan
everyone
▸ After a few failed attempts, we decided to re-shoot this scene with professional dancers. Literally people
who are professionals at body control. This time, it worked!

Director karenxcheng
VFX jperl
Software: Instant-NGP
Hardware: We recommend an RTX 3090 or newer card

https://nvlabs.github.io/instant-ngp/
Researchers: Thomas Müller, Alex Evans, Christoph Schied, Alexander Keller

Dancers austingumban anthony_daste dielondielondielon erikuzin tofumeow karacharli
_justinvale
Restaurant doughzoneusa
BTS jennaaa7
Music: "Drift" by philipemorris

#photogrammetry #NeRF #neuralrendering #instantNeRF #neuralradiancefield #aiart #artificialintelligence #InTheNVIDIAStudio

## Stats

- **Views:** 0
- **Likes:** 20,408
- **Shares:** 0
- **Comments:** 182

## Tags

neuralradiancefield, photogrammetry, instantnerf, nerf, aiart, ad, artificialintelligence, neuralrendering, inthenvidiastudio

---
Copyright (c) Gondola