404-GEN

　　On April 14. 2025. @404gen_ tweeted on the X platform announcing the launch of the world’s largest open source 3D model dataset by 404-GEN, containing over 21.5 million AI-generated models. It is larger than all existing 3D datasets combined. The dataset is targeted at Gaussian splatting research for scenarios such as movie and TV special effects, virtual reality, and digital twins. The image accompanying the tweet shows a diverse range of 3D models, including gears, construction equipment, sculptures, and teddy bears, highlighting the diversity of the dataset. 404-GEN emphasizes that this is the largest Gaussian splatting dataset ever created.

　　The release of this dataset addresses the problem of data scarcity in AI training. According to Epoch AI Research (2024), human-generated data will run out by 2030. and synthetic data becomes a key solution. 404-GEN’s 21.5 million models provide researchers with a rich training resource for model development in areas such as autonomous driving and virtual reality. The dataset totals 40TB, and the full version is accessible via a request, but the sample set has been made publicly available via Hugging Face (Hugging Face, 2025), ensuring open source accessibility.

　　404-GEN’s use of decentralized miners from the Bittensor network to generate this dataset highlights the productive potential of decentralized AI, which can rapidly generate large-scale content by rewarding high-quality output in a way that is difficult for centralized systems to achieve. The community has responded enthusiastically, with @Airik saying “that’s crazy” and @ImBananas4 suggesting the potential of not needing human feedback. 404-GEN’s release not only pushes the boundaries of 3D rendering technology, it also sets a new benchmark for decentralized AI research.

Similar Posts