What I mean is that once you already have a textured mesh model, you can capture video through a flythrough/screen-recording walkthrough and then train it into a Gaussian Splatting model.

but this is not a real scene captured by a camera.. it is generated it seems! What do you mean by "500 photos extracted from a recorded video"?

500 photos extracted from a recorded video

Hi! Impressive How many photos did you train this scene on?