TRELLIS is a state-of-the-art 3D asset generation model that produces high-quality 3D outputs from text or image inputs. It leverages a unified Structured Latent (SLAT) representation, enabling versatile outputs such as Radiance Fields, 3D Gaussians, and meshes. With up to 2 billion parameters trained on a diverse dataset of 500K 3D objects, TRELLIS outperforms existing methods in quality and flexibility. Key features include intricate shape and texture details, support for multiple 3D formats, and capabilities for local editing and generating variants of objects. It also offers pre-trained models, a large-scale dataset, and tools for training and fine-tuning, making it a powerful solution for scalable and adaptable 3D generation.