Hunyuan V3 T2I

Hunyuan Image v3 is Tencent's flagship open source text to image model, built as a unified autoregressive Mixture of Experts system with 64 experts, 80 billion total parameters, and 13 billion activated per token. It blends language and visual modalities to produce outputs with strong semantic alignment and supports bilingual input in Chinese and English. The model handles difficult cases like embedded typography, layout guidance, and detail fidelity in complex scenes and produces images up to 2048 by 2048 natively without upscaling. Hunyuan Image v3 is released with open inference code, model weights, and commercial licensing options, making it practical for both research and production deployment. As of release, it is the largest open source image generation MoE model publicly available.

September 28, 2025

Tencent

Text to Image (T2I)