- Lux3D generates individual 3D assets from text or a reference image.
- Manycore claims generation as fast as 20 seconds in Turbo mode; detailed benchmarks were not in the launch release.
- Interactive environments combining Lux3D and SpatialGen were still described as testing.
A released asset tool and a separate research roadmap
Manycore announced Lux3D on August 27 as a service for generating three-dimensional assets from text or images. At the same time, it described testing a broader system that combines Lux3D with SpatialGen to create interactive environments. The asset product and the world-model work are at different stages.
The company offers Standard mode for greater emphasis on geometry and materials, and Turbo mode for faster generation. Its claim of results in as little as 20 seconds refers to Turbo mode. The announcement says detailed benchmark results would be published separately, so that figure is a company performance claim rather than a general measured average.
Materials and usable output
Manycore emphasizes the way glass, metal, ceramic and plastic respond to lighting. It says the model was trained using both generated data and reconstructions of the physical world. API access and a batch-oriented Harness mode are intended to connect asset generation with existing production tools.
The public Aholo Lux3D repository describes browser previews, GLB export, physically based rendering materials, and integration through an API or ComfyUI. It also distributes a Codex plugin. This is evidence of an integration package, not a release of the model's training weights. Paid generation uses separate credentials for the Chinese and international services.
From an object to an environment
A downloadable 3D object can be repositioned, viewed from another angle and placed within an existing scene. Those operations distinguish an explicit geometric asset from a video that merely depicts an object. Materials then determine how its surfaces appear under the scene's lighting.
Manycore's longer-term plan adds component relationships, motion logic and interactive properties to generated assets. It describes the combined SpatialGen system as work in testing, rather than a finished general simulator. The current launch therefore supplies tools for producing objects while outlining a more ambitious route toward environments in which those objects can move and interact.
Sources & context
Go to the original material. Company claims remain attributed to their sources.
01Updates & corrections
— Expanded with additional reporting and source context.
