Tag
Direct3D-V3 proposes a latent-free diffusion transformer framework that discards the VAE and directly models raw SDF voxel grids in the native voxel domain, using implicit manifold regularization and camera-prediction auxiliary tasks to produce richer 3D mesh detail.
A set of comparison images shared by developer Scott shows that in just one year, AI-generated 3D scenes using Three.js have evolved from crude low-poly graphics to visuals rivaling AAA games, featuring detailed Chinese architecture and cinematic lighting.
World Labs launches Atlas, a next-generation world model that turns images into explorable 3D worlds, with tools like Viewfinder and Chisel plus developer APIs for converting splats to meshes for games, film, architecture, and robotics workflows.
Hugging Face publishes a definitive guide with ready-to-use LoRA fine-tuning scripts and containers for TRELLIS.2, enabling adaptation of the 3D generation model to specific object categories on local machines or Hugging Face Jobs.
A 4B open-weight coding model (Qwen3.5-4B-GLB-3D-Coder) that generates Three.js scenes with GLB export for creating 3D game assets usable in engines like Godot is being released as an initial checkpoint with full weights and training data to follow.
HairDeck, a Tsinghua/CMU/CUHK project, generates animation-ready stylized 3D hair cards with spring-bone chains from a single character image, exporting to GLB or VRM with code and weights released under Apache 2.0.
HairDeck 提出使用集合扩散 transformer 从单张图像生成风格化发卡与弹簧骨动画,将 CHARM 的 Chamfer 距离降低 44%,速度提升 33 倍,可直接导出为可用的下游资产。
作者发布了其4B 3D编码模型的新版本,一次性生成结果和视觉质量显著提升,并在Hugging Face上更新了Q8和Q4量化版本,同时承诺公开数据集和完整FP16权重。
Hi3D released Hi3D V1.0, an open-source 3D model built on Sparc3D that adds detail to rough 3D meshes from a reference image, with inference code and geometry refinement weights available on GitHub, while texture generation remains in the commercial V3.
CCORE 发布了基于 Qwen3.5-4B 的独立微调模型 Qwen3.5-4B-3D-Coder,专注于通用代码编写与高质量 Three.js 交互式 3D 场景生成,目前已提供 Q4_K_M 和 Q8_0 量化版本,完整 FP16 权重和训练数据集将在训练完成后公开。
Google DeepMind 展示了 Gemini 4 Argon 通过 Blender MCP 工具,仅凭一张 Nano Banana 图片在 Antigravity 中自动构建出哥特式大教堂的 3D 场景,并迭代优化几何、纹理与光照。
Qwen4 and Kimi K3.1 appear to have surfaced on LMArena, with both models generating complex interactive 3D helicopter scenes (takeoff/land, assembly, exploded views) in 15-20 minutes, signaling major advances from Chinese AI labs.
Viggle 推出 JST-2:一个物理原生的世界基础模型,可从单一模型生成 3D 场景、角色和全身动作,支持文本/图像/视频输入,并提供 PINOC 浏览器应用、API 及 MCP 接入。
Viggle has released the JST-2 model, which uses a "physics-native" Joint Space-and-Time architecture to represent the world as 3D atoms with spatial positions and physical properties. It can generate interactive 3D scenes, skeletal characters, and full-body motions in about one second for less than a cent. The model is available via API but is not yet open-sourced.
Viggle released JST-2, described as the world's first full-stack physics-native world foundation model built entirely in 3D rather than pixel/video space, along with a PINOC MCP server that lets agents like Claude Code generate characters, animate them from text, and assemble scenes in one flow with a shared skeleton eliminating retargeting.
提出在由2D基础模型导出的高维隐式3D潜在空间中直接进行扩散建模,实现3D-grounded的场景级3D生成,相比2D多视角方法在空间一致性和效率上更具优势,并验证了其在表示与模型容量上的可扩展性。
An AI blogger tested Gemini 4 Argon's 3D content creation capabilities and showcased a comparison of the results after tuning across four versions.
Alibaba's Qwen team released BlenderWorld, a new benchmark repository that measures how well visual agents can build 3D scenes in Blender from real-world references. The release is seen as a signal that Qwen may focus on agent-driven 3D design and tool use (e.g., Blender via MCP) in future models.
Google recently released Gemini 4 Argon, and users are already using it to create stunning 3D worlds and games, with the post showcasing 10 impressive examples.
SUPRA-Tex 提出通过解耦可见材质估计、表面对齐与观测引导的3D补全来生成可重光照的3D材质,在单图3D生成中实现可见区域的高保真还原与未观测表面的连贯补全,在纹理化和端到端资产生成任务上达到SOTA。