Basic Information
| Attribute | Value |
|---|---|
| Organization | Luma AI |
| Type | Commercial SaaS |
| Launch Date | Genie 2023, Dream Machine 2024 |
| Current Version | Genie, Dream Machine 1.6 |
| URL | lumalabs.ai |
| Pricing | Free tier with attribution, paid tiers |
Capability Profile
| Capability | Support | Notes |
|---|---|---|
| Text-to-3D | Full | Via Genie |
| Video-to-3D | Full | NeRF-based reconstruction |
| Gaussian Splat Output | Full | Native representation |
| Mesh Export | Partial | Quality loss in conversion |
| Video Generation | Full | Dream Machine for video |
Standards Compliance
| Standard | Compliance | Notes |
|---|---|---|
| glTF 2.0 | Partial | Mesh export available |
| PLY | Full | Gaussian splat export |
| Proprietary | Full | Native Luma format |
Key characteristic: Gaussian splat native - mesh is secondary output.
Maturity Assessment
SRL Score: SRL-5 (Early Adoption)
Evidence:
- Pioneered consumer-accessible neural 3D capture
- Genie supports 10+ languages
- Free tier with attribution requirement drives widespread adoption
Interoperability
| Aspect | Details |
|---|---|
| Export Formats | GLB, PLY (splats), OBJ, FBX |
| Import Formats | Text, images, video |
| API Availability | Yes (commercial tiers) |
| Known Integrations | Unity (via plugins), Unreal (via plugins), web viewers |
Strengths
-
Neural representation pioneer - High visual quality via Gaussian splats
-
Accessible free tier - Enables widespread experimentation with attribution requirement
-
Multi-language support - 10+ languages broadens global accessibility
Limitations
-
Quality loss in conversion - ~15% visual quality loss when converting splats to mesh
-
Custom renderers required - Neural formats need specialized rendering in game engines
-
No physics or interaction - Pure visual representation; no collision or navmesh
Neural vs. Traditional Output
Gaussian Splat (Native)
- Millions of Gaussian particles
- Photorealistic rendering
- ~15% quality loss when converted to mesh
- Requires custom renderers
Mesh Export (Converted)
- Traditional triangulated geometry
- Compatible with standard engines
- Lower visual fidelity than splats
- Physics-capable
Product Suite
Genie (Text/Image to 3D)
- Text-to-3D generation
- Image-to-3D reconstruction
- Multi-language prompts
- Free with attribution
Dream Machine (Video Generation)
- Text-to-video
- Image-to-video
- 1.6 version current
- Separate from 3D pipeline
Output Gallery
MSF Relevance
Luma AI is representative of the “neural-native” path producing Gaussian splats rather than traditional meshes. This highlights:
-
Critical need for KHR_gaussian_splatting - Standard format for neural representations
-
Neural-to-mesh conversion standards - Quality preservation during conversion
-
Rendering pipeline considerations - How to integrate neural assets in game engines
-
Quality validation challenges - How to measure quality for non-traditional geometry
Key insight: The future likely involves both neural and mesh representations coexisting, requiring interoperability standards between them.