404-GEN, the decentralized platform for AI-generated 3D content, has announced the release of the world’s largest open-source 3D dataset. Comprising over 21.5 million AI-generated 3D models, with detailed metadata, prompts, and contributor attribution, this release significantly surpasses the combined scale of all existing 3D model datasets, including Objaverse.
The dataset is designed to support a wide range of research and development initiatives, particularly in emerging areas such as 3D Gaussian Splatting (3DGS), a rendering technique that uses millions of translucent ellipsoids ("splats") to enable photorealistic, real-time 3D visualization. 404-GEN offers both a Unity add-on and a Blender extension for creators, alongside integration support for platforms like Spline and SuperSplat, and emerging tools for Unreal Engine via third-party plug-ins.
Built on a decentralized architecture as a permissionless subnet of Bittensor, 404-GEN relies on an open network of compute contributors who are rewarded based on model quality. This decentralized structure enables large-scale, rapid generation, validation, and distribution of 3D content without centralized control or storage bottlenecks. Each model is tagged with its generation prompt and contributor ID, promoting transparency, reproducibility, and trust.
Current initiatives include:
- Curating and storing high-scoring 3D models with full metadata
- Decentralized model storage for long-term access and data integrity
- A technical front end for scalable dataset interaction
- Prompt-based generation tools optimized for clarity and specificity
A sample of the dataset is available on Hugging Face, while full access is offered by request, prioritizing academic and industrial projects advancing 3D modeling, simulation, gaming, and spatial computing. API access is also available for qualified researchers and developers.