Tencent open-sources HunyuanWorld 1.0 for 3D world generation

Background

Most image generation models produce flat pictures. HunyuanWorld 1.0 was aimed at a different output: a 3D environment that a user can move through. Tencent's Hunyuan team built the model on 360° panoramas, which give a full surround view of a scene.^[1] Generating a 3D environment is harder than generating a flat picture, because a usable world needs a structure that holds together from every viewpoint a user might take.

What happened

On 26 July 2025 Tencent released HunyuanWorld 1.0 and its technical report.^[1] The model creates explorable 3D worlds from text descriptions or images.^[1] Tencent described it as the first open-source, simulation-capable, immersive 3D world generation model.^[1]

Key details

Generated worlds can be exported as meshes, so they can be used in standard 3D software.^[1] A mesh is the polygon surface that describes a 3D object's shape. Interactivity was one of the stated goals: the disentangled object representation is meant to let a user act on individual objects in a generated scene.^[1]

A lite quantized version for consumer GPUs followed on 15 August 2025.^[1] Quantization lowers the memory a model needs, which is why a lite version can run on a single consumer card.

Why it matters

The release marked the start of a line of Tencent world models. Later releases included HunyuanWorld-Voyager, WorldMirror (version 1.1), WorldPlay (version 1.5) and, in April 2026, HY-World-2.0.^[1] Each of these followed the first release in the same product line, all published under the HunyuanWorld name.^[1] The model's code and checkpoints were published in a GitHub repository, so developers who want to try the model can download it and run the released versions on their own hardware.^[1] Open weights and a technical report also let other researchers check the method and build on it. The technical report sets out the method in more detail than this summary can.^[1]