Hunyuan Image 3.0

image

by Tencent · Updated December 1, 2025

Hunyuan Image 3.0 is Tencent's massive open-source text-to-image model with 80 billion total parameters (13B activated). It uses a Mixture-of-Experts architecture with Transfusion, trained on 5 billion image-text pairs. It excels at world knowledge reasoning, multilingual text rendering, and diverse artistic styles. Apache 2.0 licensed for free commercial use.

Best For

Open-sourceMultilingual textWorld knowledgeCommercial useDiverse styles

Prompting Tips

1Fully open-source with Apache 2.0 license — free for commercial use
2Renders both Chinese and English text accurately in images
3Leverages world knowledge for culturally accurate imagery
4The MoE architecture means high quality with efficient inference
5Supports thousand-character semantic understanding for complex prompts
6Use for culturally diverse and globally representative imagery

Syntax & Constraints

Natural language prompts in Chinese or English. Open-source (Apache 2.0). Configurable inference parameters. Supports various aspect ratios.

Example Prompts

👤 Portrait Photography

Generational cooking portrait

A warm, candid portrait of a grandmother teaching her granddaughter to make traditional dumplings in a cozy kitchen. Flour on their hands, genuine laughter, steam rising from a bamboo steamer. Natural warm interior light, storytelling documentary style.

candidwarmstorytelling

🏔 Landscape & Scenery

Guilin Li River dawn

A photorealistic image of the Li River in Guilin, China at dawn. Towering karst peaks emerge from morning mist, their reflections perfect in the still river. A lone bamboo raft with a cormorant fisherman and his lantern glowing warmly. National Geographic quality.

guilinriverdawn