HY-WorldMirror 2.0 · Tencent's feed-forward multi-view → 3DGS model.
32 input images → ~1.5s GPU inference → millions of gaussians.
No COLMAP. No per-scene optimization. Same model, three different scenes —
pick from the dropdown.
drag orbit · scroll zoom · auto-spins when idle ·
lyra chapel ·
plant (splatfacto) ·
room