Repolode
Submit
Tencent-Hunyuan

Tencent-Hunyuan/HunyuanDiT

An open-source bilingual text-to-image diffusion transformer from Tencent that understands both Chinese and English prompts.

Open-source alternative to:MidjourneyDALL-E 3

View on GitHubWebsite
Stars
4,290
Forks
365
Open issues
124
Last commit
14 days ago

What is it?

Hunyuan-DiT is a text-to-image diffusion model built on a transformer architecture that works in latent space, using a pre-trained VAE to compress images and a bilingual CLIP plus multilingual T5 text encoder to process prompts. It is designed for fine-grained understanding of Chinese as well as English, and supports multiple output resolutions.

The repository provides PyTorch model definitions, pretrained weights (v1.0 through v1.2, plus a distilled version), and inference scripts for command line, Gradio web UI, and Hugging Face Diffusers. It also includes full-parameter and LoRA training code, ControlNet and IP-Adapter support, and integrations for ComfyUI and Kohya.

The project also includes DialogGen, a multimodal prompt-enhancement model that supports multi-turn text-to-image dialogue, and Hunyuan-Captioner for re-captioning training data. Inference requires an NVIDIA GPU with CUDA, with a minimum of about 11GB of GPU memory; a 6GB inference path based on Diffusers is also provided.

Key features

  • 01Bilingual Chinese and English prompt understanding for text-to-image generation.
  • 02Supports multiple image resolutions with multi-resolution training.
  • 03Provides LoRA and full-parameter training scripts.
  • 04Includes ControlNet (canny, pose, depth) and IP-Adapter support.
  • 05Offers multi-turn text-to-image dialogue via the DialogGen prompt-enhancement model.
  • 06Integrates with Hugging Face Diffusers, ComfyUI, and Kohya.

Use cases

  • Generating images from Chinese-language prompts for creative or design work.
  • Fine-tuning a text-to-image model on a custom style with LoRA.
  • Running controllable image generation with ControlNet conditions such as pose or depth.
  • Building a multi-turn image creation assistant on top of the provided Gradio demos.

Quick start

  1. 1

    Try the model in the browser via Tencent Hunyuan Bot

  2. 2

    Follow the Installation Guide for Linux in the README

  3. 3

    Download pretrained models as described in the README

Things to know

  • The README's installation guide is written for Linux.
  • Running inference locally needs a capable NVIDIA GPU, and the README lists a 6GB VRAM option.
  • The metadata gives no clear license (NOASSERTION), so check the license terms before commercial use.

Topics

#text-to-image#diffusion-model#transformer#chinese-nlp#pytorch#lora#controlnet#generative-ai

FAQ

What is HunyuanDiT?
An open-source bilingual text-to-image diffusion transformer from Tencent that understands both Chinese and English prompts. Hunyuan-DiT is a text-to-image diffusion model built on a transformer architecture that works in latent space, using a pre-trained VAE to compress images and a bilingual CLIP plus multilingual T5 text encoder to process prompts. It is designed for fine-grained understanding of Chinese as well as English, and supports multiple output resolutions.
Is HunyuanDiT free to use? What license does it have?
HunyuanDiT is open source under the Other license. Check the license terms before using it in commercial products.
Is HunyuanDiT still maintained?
Actively maintained. The last commit to HunyuanDiT was on Oct 8, 2026.
What can I use HunyuanDiT for?
Generating images from Chinese-language prompts for creative or design work. Fine-tuning a text-to-image model on a custom style with LoRA. Running controllable image generation with ControlNet conditions such as pose or depth. Building a multi-turn image creation assistant on top of the provided Gradio demos.
How do I install HunyuanDiT?
Try the model in the browser via Tencent Hunyuan Bot Follow the Installation Guide for Linux in the README Download pretrained models as described in the README
Where can I get HunyuanDiT?
The source code and installation instructions are on GitHub: https://github.com/Tencent-Hunyuan/HunyuanDiT

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects