Favicon of HunyuanDiT

HunyuanDiT

A local text-to-image model generates images from Chinese and English prompts, with conversational refinement, ComfyUI support and NVIDIA GPU inference.

Screenshot of HunyuanDiT website

HunyuanDiT is a local text-to-image model for people generating images from Chinese or English prompts. It handles detailed descriptions and Chinese cultural references, including poetry, idioms, food and architecture.

  • Detailed prompts: Describe subjects, clothing, lighting and artistic style in long text prompts to guide image generation.
  • Conversational refinement: The companion DialogGen model uses follow-up requests and conversation context to produce revised image prompts.
  • Visual guidance: ControlNet supports pose, edge and depth guidance, while IP-adapter accepts image references.
  • Model adaptation: Training code and LoRA support let developers adapt the model to their own material.

Local use supports ComfyUI, Hugging Face Diffusers and Gradio interfaces. NVIDIA hardware with CUDA is required, and Linux is the tested operating system. An optimized Diffusers implementation can run below 6 GB of VRAM on supported NVIDIA cards, including RTX 3070, 3080, 4080, 4090 and A100. Standard inference needs more memory, and running DialogGen alongside image generation increases the requirement further.

Downloadable weights and inference code support generation on your own hardware. Tencent Hunyuan Bot is a separate hosted service for trying image generation through the web.

Similar to HunyuanDiT