The author expresses strong satisfaction with Gemma 4's quality, highlighting its coding ability and adaptability in conversations and reasoning. A test involving 3D model generation from an F1 car image demonstrated that Gemma significantly outperformed models like Claude Sonnet, Gemini Pro, and ChatGPT, which exhibited notable flaws.
DreamGaussian4D introduces a novel method for generative 4D content creation, utilizing Gaussian Splatting to efficiently model dynamic scenes over time. This approach enables the synthesis of realistic, animated 3D environments.
HY-World 2.0 has been released, enabling one-click generation of interactive 3D worlds from text or images. It offers pipeline-ready outputs for Unity and Unreal Engine, a unified world model system, and an interactive character mode.
Este trabalho propõe um framework generativo para sintetizar veículos 3D Gaussianos animáveis a partir de uma única imagem ou entradas multi-view esparsas. Ele visa superar as limitações dos modelos de veículos rígidos atuais em simulações de direção autônoma, introduzindo um módulo de refinamento para articulação de partes.
A solo developer is building Vextrude, a free, open suite of browser-based 3D generation tools. The project aims to offer functionalities for 3D creation directly in the browser.