text / image -> 3D in 2026
prompt or image
text, 1 photo, or 3-16 photos
multi-view diffusion
SV3D, CAT3D, MVDream, Zero123
8 consistent views
3D fit
Gaussian splat or mesh extract
3D Gaussian Splatting (Kerbl 2023)
~1M Gaussians per scene, differentiable render, 100 fps
direct text/image to mesh
Meshy 4, Rodin Gen-1.5, Hunyuan3D 2.0
output: PBR mesh (albedo, roughness,
metallic, normal)
→ direct import into Unity, Unreal, Blender
30s - 60s per asset on hosted endpoints
2022 → 2026 time per asset
DreamFusion 2022: 1 hour; LRM 2023: 5s; TripoSR 2024: 1s; Meshy 4 2025: 30s with PBR
NeRF → 3D-GS → direct generative mesh: quality up, time down 100x