Tag: #3d-generation40 results
The Trellis.cpp project has resolved critical bugs, bringing its image-to-3D generation quality on par with the original reference implementation. It now supports efficient 3D asset production without requiring CUDA.
This article demonstrates how an autonomous agent can orchestrate multiple Hugging Face Spaces to generate and display 3D content. It highlights the power of chaining specialized AI tools to create complex, multi-modal applications.
AutoNavi has launched the internal testing phase for ABot-Earth0.5. This model moves beyond traditional 2D distillation methods to achieve high-consistency scene generation using a 3D native approach.
BrickAnything is an autoregressive framework that generates physically buildable brick structures from 3D shapes. It utilizes structure-aware tree tokenization to ensure structural stability and geometric fidelity during the assembly process.
3DS has launched Meshy.ai in Japan, an AI platform that generates 3D models from text or image inputs. It aims to streamline design and prototyping workflows by removing the need for specialized 3D modeling skills.
A Shenzhen company mastered advanced 3D technology two years before Fei-Fei Li's recent hype. The article emphasizes that the real competition lies beyond the generation stage.
Benchmark on MineBench shows Kimi K2.6 builds superior 3D Minecraft structures vs K2.5, with higher detail despite some inconsistency. Total cost $2.35, deemed most cost-effective for performance. Links to benchmark repo and prior comparisons provided.
Roblox upgrades its AI assistant with agentic features: planning mode analyzes game code and proposes actions, generates procedural 3D models and meshes, plus self-correcting test loops. It integrates MCP client with tools like Claude and Cursor, with more on roadmap.
Gemma 4 31B impresses with high-quality 3D models from complex images like an F1 car, outperforming Claude Sonnet 4.6, Gemini 3.1 Pro, ChatGPT, and Qwen3.5 27B. It generates superior results in just 3600 tokens versus competitors' 6800+ and flawed outputs. Also excels in coding, math, reasoning, and conversation.
HY-World 2.0 introduces one-click generation of interactive 3D worlds from text or images. It provides editable 3D outputs compatible with Unity and Unreal Engine, including meshes, 3DGS, and point clouds. The tool features a unified model system for both synthetic and real-world scenes with real-time interactive exploration.