Hunyuan Image

Hunyuan Image 3.0 is a large-scale multimodal autoregressive image generation model from Tencent with 80 billion total parameters across 64 mixture-of-experts experts, trained on 5 billion image-text pairs. Unlike diffusion-based pipelines, it models text and image tokens in a unified framework — enabling superior world-knowledge reasoning, prompt adherence on complex thousand-word descriptions, and cinematic-quality output that prioritizes aesthetic polish and art-directed compositions.