Really cool. I wonder if it would work to use the DUSt3R project[1] to get the camera intrinsic and extrinsic parameters so you could create a scene that can be turned into a nerf from any pictures without the need for special apps to use for capture.

[1]: https://dust3r.europe.naverlabs.com/

I'm not sure if Dust3r can produce better poses than colmap given the number of photos you need to properly train GS.