World Labs Marble

Basic Information

AttributeValue
OrganizationWorld Labs (Fei-Fei Li)
TypeCommercial (SaaS)
Launch DateBeta 2024, Production 2025
Current VersionMarble 2.x
URLworldlabs.ai
Pricing$35-95/month
SRL-5 Early

Capability Profile

CapabilitySupportNotes
Text-to-WorldFullPrimary input modality
Image-to-WorldFullReference image drives generation
3D Export (Mesh)FullGLB, PLY export
3D Export (Gaussian Splat)FullNative output format
Spatial Editing (Chisel)FullBlock out before style generation
Physics CollisionPartialDual-mesh for collider + visual; no navmesh
Real-time GenerationPartialSimple prompts in seconds; complex in minutes

Standards Compliance

StandardComplianceNotes
glTF 2.0PartialGLB export available
OpenUSDNoneProprietary format primary
KHR_gaussian_splattingNonePre-dates draft; proprietary splat format

Maturity Assessment

SRL Score: SRL-5 (Early Adoption)

Evidence:

  • $230M+ funding achieved unicorn status
  • Production SaaS with paying customers across VFX, games, and enterprise
  • First commercial “Large World Model” product

Interoperability

AspectDetails
Export FormatsGLB, PLY (Gaussian splats), native Marble format, video
Import FormatsText, images, video
API AvailabilityYes (commercial tiers)
Known IntegrationsRosebud AI (gameplay), VFX studios (undisclosed)

Strengths

  1. First commercial world model with exportable geometry - Proves the feasibility of combining world model approaches with practical asset pipelines

  2. Spatial editing via Chisel - Provides designer control through block-out editing before AI style generation

  3. Multi-platform output - Outputs compatible with VR headsets (Vision Pro, Quest 3)


Limitations

  1. No native navmesh generation - Must be generated in-engine post-export

  2. Quality loss in conversion - ~15% visual quality loss when converting splats to mesh

  3. Commercial licensing required - $35-95/month for production use



MSF Relevance

World Labs Marble is the first production system demonstrating that world models can produce engine-compatible output. This proves the feasibility of standards for:

  • AI generation metadata
  • Quality validation profiles
  • Neural-to-mesh conversion specifications

Marble represents the target use case for MSF_ai_generation_metadata extension work.