> For the complete documentation index, see [llms.txt](https://guide.404.xyz/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://guide.404.xyz/404-gen-user-guide/extended-resources/readme.md). # 404—GEN Synthetic 3D Dataset ### Overview The **404—GEN** Synthetic 3D Dataset is the world's largest open-source 3D asset repository. Comprising over 21.5 million high-fidelity, AI-generated 3D models, this dataset surpasses the volume of all other existing 3D datasets combined. ### Key Highlights & Specifications * **Dataset Size:** 21.5M+ High-fidelity 3D assets * **Total Storage Volume:** 40 Terabytes (TB) * **Licensing:** Open-source and attribution-ready * **Metadata:** Each asset includes detailed metadata, usage rights, and ownership attribution. * **Target Applications:** 3D generative AI, Gaussian Splatting research, neural radiance fields (NeRFs), AR/VR development, and gaming studios. ### Why Synthetic Data Matters #### Overcoming Data Scarcity AI models require massive amounts of high-quality data to continue scaling and learning. However, research from organisations like Epoch AI indicates that the tech industry faces a looming "data wall," predicting that we may run out of high-quality human-generated data for training by 2030. #### The Synthetic Solution "Synthetic" datasets—data generated programmatically by other AI models—provide a scalable alternative to human data. When produced at an elite standard and a massive scale, synthetic data enables next-generation 3D models to train efficiently without hitting the constraints of human data availability. ### Powered by Decentralised Intelligence Achieving this milestone through traditional, centralised infrastructure would have been nearly impossible. **404—GEN**’s rapid scaling and production speed are a direct result of operating within the Bittensor network. * **Incentivised Architecture:** Independent network miners are directly rewarded based on the quality and volume of their outputs. * **Unprecedented Throughput:** This decentralised design allowed **404—GEN** to generate, score, filter, and aggregate 40TB of synthetic 3D content in record time, demonstrating the raw power of decentralised compute. ### Dataset Access & Downloads Because hosting a 40TB dataset for public download is an immense infrastructure challenge, we offer flexible access tiers depending on your project needs. #### 1. Light/Sample Tier (404—Mini) For researchers and developers looking to evaluate the dataset or run smaller-scale experiments, a curated sample set [**404—Mini**](https://huggingface.co/datasets/404-Gen/404mini) is publicly hosted on Hugging Face. #### 2. Full Dataset Access The complete 40TB dataset is available for large-scale institutional research, commercial training, and studio pipelines. Priority access is given to projects actively advancing the boundaries of 3D research and development. **Request Portal:**[ dataset.404.xyz](https://dataset.404.xyz/) #### 3. Immediate API Access If you are a researcher, developer, or studio requiring direct, high-speed API integration into your training pipeline, please reach out directly via email: ****