Tag
3DCodeVerse introduces a research ecosystem with a 2M-scale dataset of executable 3D code, a multi-agent DAG harness for automated generation and verification of 3D content, and empirical guidance showing fine-tuned open models achieve a 90% win rate against Claude Design (3D Object).
Intern Lumina U2 is presented as the first diffusion language model using multi-codebook discrete visual tokens to unify image, video, and 3D generation with text in a single backbone, achieving leading performance and 4.45x faster image generation via stability-aware sampling.
UniForge3D 提出在共享的锚点引导空间中联合建模 3D 重建与生成,解决相机相对重建与规范空间生成之间的坐标不匹配问题,并在 Toys4K 和 GSO 数据集上取得了最先进的外观与几何结果。
A user open-sourced a prompt that can generate a rotatable, interactive 3D miniature island scene, claiming that any agent or model can produce it with one click. They also mentioned that the new Chinese model Step 5 Preview topped OpenRouter’s Trending chart on its second day on the platform, delivering results comparable to Opus 5.5 at a lower price.
阶跃星辰发布旗舰基座模型 Step 5 Preview,登陆 OpenRouter 热榜第一,主打 AI 编程与软件工程,支持 1M token 上下文。用户实测将其接入 Claude Code,用 Three.js 生成可交互 3D 小岛,效果媲美 Opus 5.5 且成本更低。
本文提出 Artisan 系统,通过多智能体流水线与物理仿真反馈扩展 Articraft,生成具有物理合理关节结构的 3D 资产(如碰撞自由运动、重力稳定性和准确的质量材料属性),并构建涵盖空间保真度、物理合理性、下游任务表现和整体质量的统一基准。
SpaceFlow 提出了一种免训练的局部可控 3D 生成流程,用户可通过几何基元为不同区域指定控制强度,并分别路由文本或图像提示以合成外观,从而在几何保真与生成自由度之间取得平衡。
A user tested image-to-3D model Seed3D 2.0 via Atlas Cloud on a single photo of a fire hydrant, finding that the reconstructed back and overall shape looked convincing, but fine details like raised cast lettering were completely missing from the mesh—highlighting how AI 3D demos can appear plausible while losing object-specific fidelity.
NVIDIA开源了名为Lyra 2.0的模型,可将任意单张图片直接转换为可自由探索的交互式3D世界,甚至可用于向生成的场景中放入机器人进行仿真训练。模型已发布到Hugging Face,UI代码也已在GitHub开源。
image-blaster is an open-source MIT skillset for Claude Code that converts a single photo into an explorable 3D environment in under 5 minutes, chaining World Labs Marble, Hunyuan 3D, nano-banana and ElevenLabs to produce meshes, gaussian splats, physics and ambient audio for Unity, Unreal, Godot, Blender or Three.js.
A viral post claims that a GitHub repository can turn any single photo into a fully explorable 3D world — complete with physics-enabled meshes, background splats, and ambient audio — within minutes, shaking up an entire industry of 3D content creation.
A tweet hypes a new open-source GitHub repository called image-blaster that turns a single photo into an explorable 3D world — complete with meshes, physics effects, background splats, and environmental audio — claiming the tool could disrupt traditional 3D modeling workflows.
该论文提出 LMBuild,一个用于评估 LLM 智能体生成可建造、功能性结构能力的基准,包含交互式装配环境、基于 CAD 数据集构建的评测集以及涵盖结构可靠性、功能可供性和物理可实现性的评估框架,在 30 个系统上的评测揭示了功能推理仍是前沿模型的主要瓶颈。
Mold 是一个 CLI 原生的本地 AI 媒体生成开源项目,支持在自有的 NVIDIA CUDA 或 Apple Silicon 上运行图像、视频与 3D 生成模型,统一了原本分散的本地 AI 媒体工具流程。
Google's new Gemini 4 Argon is reportedly performing on par with Claude's best models at 3D game generation, marking a notable advance in generative 3D content and game development capabilities.
MeshyAI teased a new "Meshy Edit" feature that adds 3D inpainting, letting users select a region of a 3D model and describe the change they want made.
作者用同一套带硬性预算上限(20分钟、$4、60次请求)的 agent 循环测试了 14 个模型的 photo-to-Blender 3D 场景生成能力,GPT-6 Astra 以持续优化到预算耗尽赢得全部三项,而重度推理模型常因耗尽时钟而交付为空。
The paper proposes Autoregressive Frontier Expansion, a generative framework that grows tree-like branching structures iteratively, using a flow-matching model parameterized by an SO(2)-equivariant GNN to decide bifurcation versus termination at each frontier step. It is evaluated on cortical neuron and botanical tree morphology generation in unconditional, class-conditioned, and morphology-guided settings, closely matching reference distributions.
MIT CSAIL, Google, and Northeastern researchers unveiled InstructMesh, an AI-driven design tool that pairs Microsoft's TRELLIS 3D generator with GPT-4 so users can generate, prompt-edit, and repair functional 3D-printable models before fabrication.
用户分享使用6.1 Sol在Codex中生成3D内容的体验,称其3D建模与纹理质量可与Astra媲美,并与Blender结合还原了法拉利F1赛车的爆炸装配视频,且仅消耗约10%周额度,性价比极高。