On April 16, Tencent officially released and open-sourced the Hunyuan 3D World Model 2.0 (HY-World 2.0). HY-World 2.0 is a multimodal world model capable of understanding different types of inputs such as text, images, and videos, automatically generating, reconstructing, and simulating 3D worlds. It also supports the export of various 3D asset formats (Mesh/3DGS/point clouds, etc.) and can seamlessly integrate with existing game workflows for rapid generation of game maps and level prototypes.
<img src="https://imageproxy.pbkrs.com/https://cdn.pingwest.com/portal/2026/04/16/J7TnraBJtYjpE7p60012Gb71w87yrRF3.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://cdn.pingwest.com/portal/2026/04/16/J7TnraBJtYjpE7p60012Gb71w87yrRF3.png"/>
The Hunyuan World Model 2.0 allows AI to evolve from &#34;dialogue,&#34; &#34;drawing,&#34; and &#34;shaping objects&#34; to &#34;creating worlds&#34;—a complete world that includes people, objects, and scenery. Unlike Google's Genie 3 and Hunyuan World Model 1.5, which can only generate video files, the World Model 2.0 offers further practicality: it can directly generate editable 3D asset files that can be imported into game production or embodied simulation engines, achieving true playability and usability.
Currently, the World Model 2.0 is available on the Tencent Hunyuan 3D Creation Engine official website, and users can apply for access. (Application address: https://3d.hunyuan.tencent.com/sceneTo3D)
<img src="https://imageproxy.pbkrs.com/https://cdn.pingwest.com/portal/2026/04/16/A65G6WpMPx0K68572QHtdw82FeGhzzJS.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://cdn.pingwest.com/portal/2026/04/16/A65G6WpMPx0K68572QHtdw82FeGhzzJS.png"/> Live demonstration: Input &#34;generate a Japanese RPG style medieval dungeon,&#34; and a 3D spatial asset will be generated! <img src="https://imageproxy.pbkrs.com/https://cdn.pingwest.com/portal/2026/04/16/rh88TFY3H4NewANy60PE3Tr0083CXG36.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://cdn.pingwest.com/portal/2026/04/16/rh88TFY3H4NewANy60PE3Tr0083CXG36.png"/> Live demonstration: Input &#34;generate a cozy picture book style cabin,&#34; and game characters can freely navigate through the generated 3D scene.
All content is in 3D files, all generated by the HY-World 2.0 large model with one click.
The Hunyuan World Model 2.0 supports multiple inputs including text, images, and videos. By inputting a piece of text or an image, the model can accurately parse complex semantics and generate a roamable world in various styles with one click. Compared to Hunyuan World Model 1.0, the architecture of 2.0 has been comprehensively upgraded, significantly enhancing image detail and realism. Users can also customize styles to generate 3D spaces in realistic, comic, or game styles as needed
After generation, users can import assets into engines like Unity and UE for secondary editing, significantly lowering the creative threshold for game maps.
<img src="https://imageproxy.pbkrs.com/https://cdn.pingwest.com/portal/2026/04/16/2hBcFZGsHEAw7ijSH27Ks4Yd6ZrAGD57.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://cdn.pingwest.com/portal/2026/04/16/2hBcFZGsHEAw7ijSH27Ks4Yd6ZrAGD57.png"/>
The model also supports character mode, allowing users to control characters to freely explore streets, buildings, and scenes without time limits, featuring physical collisions, just like in a game.
<img src="https://imageproxy.pbkrs.com/https://cdn.pingwest.com/portal/2026/04/16/153iFSX0bzjCCCdTNbj_sx7187p92b8s.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://cdn.pingwest.com/portal/2026/04/16/153iFSX0bzjCCCdTNbj_sx7187p92b8s.png"/>
At the same time, the World Model 2.0 supports the replication of real 3D scenes. Users can input a video of a real space or multi-angle images, and the model can construct a high-precision digital twin space. Based on the upgraded WorldMirror 2.0 architecture, the model supports arbitrary size image and video inputs, predicting dense point clouds, multi-angle depth maps, surface normals, and camera parameters in one go, generating permanently reusable outputs. In the future, scenarios such as interior decoration previews, urban planning, and cultural heritage protection can achieve rapid construction and restoration based on this capability.
<img src="https://imageproxy.pbkrs.com/https://cdn.pingwest.com/portal/2026/04/16/PKhN85H5HP6Ne0zzwT9N2WSxM9kZW6sJ.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://cdn.pingwest.com/portal/2026/04/16/PKhN85H5HP6Ne0zzwT9N2WSxM9kZW6sJ.png"/>
The Hunyuan World Model 2.0 is centered around 3D, achieving SOTA-level generation effects through a unified architecture for spatial understanding, generation, and reconstruction. Compared to other world models, the Hunyuan World Model 2.0 performs better in scene completeness (object sides and backs) and adherence to input images. The mixed representation of 3DGS and Mesh generated by Hunyuan also allows users to enter character mode for interactions with real object collisions.
Since its release, the Hunyuan World Model (HY-World series) has been continuously evolving: from the first open-source 3D world model HY-World 1.0, to the real-time online interactive HY-World 1.5, and then to the one-click generation of 3D spatial assets with HY-World 2.0, Tencent's Hunyuan 3D world model is gradually turning the concept of &#34;AI creating worlds&#34; into reality.
Since the second half of 2025, Tencent has introduced several senior talents, accelerated the upgrade of its R&amp;D architecture, and continuously increased the efficiency of Hunyuan large model R&amp;D, achieving significant results. In December 2025, Hunyuan released the 2.0 version of its large language model, with reasoning capabilities and efficiency ranking among the top in the country;
The new model HY3, which has a higher level of intelligence, is also undergoing internal business testing and has achieved significant positive returns in the Yuanbao App testing. The release of the Hunyuan World Model 2.0 marks another breakthrough for Hunyuan in the multimodal field

HERO

GAMR

ESPO

On April 16, Tencent released and open-sourced the HyunYuan 3D World Model 2.0 (HY-World 2.0), which supports various input forms (text, images, videos) and can automatically generate and reconstruct 3D worlds, seamlessly integrating with existing game workflows. Compared to the previous version, version 2.0 has significantly improved visual detail and realism, allowing users to customize styles and generate editable 3D assets, lowering the barrier for creating game maps. The model also supports the replication of real 3D scenes, enabling users to build high-precision digital twin spaces through videos or multi-angle images

PingWest

Global X Video Games & Esports ETF

Wedbush ETFMG Video Game Tech ETF

VanEck Video Gaming and eSport ETF

Tencent

TENCENT

Tencent Holdings Limited

Tencent releases and open-sources the Hunyuan World Model 2.0, creating a 3D world with one sentence, compatible with game engines!