supercool! i love the bike. i havent had much success on really shiny things. i think different training solutions (specific gaussian splat computing engine like original 3DGS or newer ones excel at specular) might help but i have not experimented with that yet. i use postshot.
the only useful tip i might have is training with close up shots of the details you want to be able to zoom in on, but i think that makes the files a lot bigger.