HunyuanImage

HunyuanImage is Tencent's text-to-image model for people who want to generate images on their own hardware. It handles detailed scene descriptions, including requests for lettering within an image.

• 2K images and varied layouts: Generate images at 2048 × 2048 resolution, with support for square, landscape and portrait aspect ratios.
• Chinese and English prompts: Describe subjects, scenes and compositions in either language, including complex instructions and longer prompts.
• Text within images: Character-aware processing based on ByT5 helps the model render lettering requested in a prompt.
• Prompt rewriting and refinement: An automatic rewriting model adds descriptive detail to prompts, while a separate image refiner improves clarity and reduces artifacts.

The downloadable model runs locally on Linux, and the project provides pretrained weights and PyTorch generation code. Tencent also offers a hosted website for trying the model; that service is separate from local generation.

Local use requires an NVIDIA GPU with CUDA support. The stated minimum for 2K generation is 24 GB of GPU memory with FP8 quantization and CPU offloading enabled. A GPU with sufficient memory can run without offloading for faster generation.