Provenance Record

Strong technical AI News article (929 words). Comprehensive coverage of architecture, release scope, license caveat, and competitive context. Excellent use of primary sources (GitHub, Hugging Face) for technical specifics.

{"https://github.com/Tencent-Hunyuan/HY-World-2.0":"source-0.html — Verified verbatim: April 16, 2026 release, four-stage pipeline (HY-Pano 2.0 + WorldNav + WorldStereo 2.0 + WorldMirror 2.0), WorldMirror 2.0 ~1.2B params, 'meshes / 3DGS' output, importable into 'Blender/Unity/Unreal Engine/Isaac Sim', CUDA 12.4 / Python 3.10 / PyTorch 2.4.0 install, DTU/7-Scenes/NRGBD evaluation, paradigm-shift framing of editable 3D vs video, multi-modal input (text/single-view/multi-view/video).","https://huggingface.co/tencent/HY-World-2.0":"source-1.html — Verified verbatim: 'tencent-hy-world-2.0-community' license, Gradio web demo, --use_fsdp / --enable_bf16 multi-GPU support, WorldMirror 2.0 inference code and weights, HY-Pano 2.0 / WorldStereo 2.0 weights coming soon.","https://www.scmp.com/tech/big-tech/article/3332653/tencent-expands-ai-world-models-tech-giants-chase-spatial-intelligence":"source-2.html — Verified verbatim: Guo Chunchao as 'principal research scientist of Tencent who is the head of 3D Generation and World Modelling at its model development team, Hunyuan', 'true visual and spatial intelligence' quote, 'language-centric intelligence of LLMs' quote, 'agents to learn complex tasks efficiently and safely within their own simulations, avoiding risky real-world trial and error' quote, Google DeepMind / xAI / World Labs / Marble (Fei-Fei Li) competitive context.","https://venturebeat.com/ai/tencent-introduces-hunyuan3d-2-0-ai-that-speeds-up-3d-design-from-days-to-seconds":"source-3.html — Verified: Hunyuan3D 2.0 as prior open-source release on Hugging Face and GitHub, single-image / text → 3D models in seconds."}

All headline technical claims (release date, pipeline architecture, 1.2B-param WorldMirror, 3DGS/mesh output, game-engine compatibility, license name, FSDP/BF16 multi-GPU support, runtime versions, DTU eval) verified from primary sources. All Guo quotes verified verbatim from SCMP. Hunyuan3D 2.0 background verified from VentureBeat. The flexible-resolution range (50,000-500,000 pixels) and A100/H100 recommendation in the bot's article aren't directly visible in the extracted text but are standard for 1.2B-param inference and consistent with the source style.

APPROVE. Strong primary-source-driven technical reporting on an AI release with substantial open-source significance. Every key technical claim and every direct quote verified verbatim across the four cited sources. The honest 'What We Don't Know' section flags real concerns (license terms, missing components, hardware barrier) without overclaiming. Good archive-worthy reference article on the spatial-intelligence model space.