# HunyuanDiT — card for AI agents

Tencent's diffusion-transformer text-to-image model with fine-grained Chinese and English prompt understanding.

Category: AI image generators · maker Tencent Hunyuan · checked 2026-10-25

## Price
- Price: Free
- Free tier: Open weights, free to download.

## License
Tencent Hunyuan Community License — https://huggingface.co/Tencent-Hunyuan/HunyuanDiT/blob/main/LICENSE.txt

- commercial: **conditions** (Commercial use with conditions)
  > is greater than 100 million monthly active users in the preceding calendar month, You must request a license from Tencent
  (https://huggingface.co/Tencent-Hunyuan/HunyuanDiT/blob/main/LICENSE.txt)
- attribution: **no** (No credit needed)
  > All distributions to Third Parties (other than through a Hosted Service) must be accompanied by a “Notice” text file
  (https://huggingface.co/Tencent-Hunyuan/HunyuanDiT/blob/main/LICENSE.txt)
- in app: **conditions** (Ship inside your app with conditions)
  > You must provide all such Third Party recipients of the Tencent Hunyuan Works or products or services using them a copy of this Agreement
  (https://huggingface.co/Tencent-Hunyuan/HunyuanDiT/blob/main/LICENSE.txt)
- youtube: **conditions** (Monetized YouTube with conditions)
  > Tencent claims no rights in Outputs You generate. You and Your users are solely responsible for Outputs and their subsequent uses.
  (https://huggingface.co/Tencent-Hunyuan/HunyuanDiT/blob/main/LICENSE.txt)

## Rules to follow
- Products with over 100 million monthly active users need a separate license from Tencent.
- Follow the Acceptable Use Policy; do not use outputs to improve other large language models.
- Give recipients a copy of the license and ship the Tencent Hunyuan Notice file with redistributed weights.

## How to connect
- Download only
- Repository: https://github.com/Tencent-Hunyuan/HunyuanDiT
- Website: https://huggingface.co/Tencent-Hunyuan/HunyuanDiT

## Notes
Hugging Face API, Oct 25, 2026: 0 downloads, 508 likes; GitHub Tencent-Hunyuan/HunyuanDiT: 4,290 stars.

## Numbers (AgentGid data)
- Gid Score: 24
- GitHub stars: 4,292

## Lines to paste into your prompt or AGENTS.md

```
- HunyuanDiT (AI image generators). Tencent's diffusion-transformer text-to-image model with fine-grained Chinese and English prompt understanding. Price: free; commercial use with conditions; no credit needed; ship inside your app with conditions.
  Connect: download from https://huggingface.co/Tencent-Hunyuan/HunyuanDiT (no API).
  Rule: Products with over 100 million monthly active users need a separate license from Tencent.
  Rule: Follow the Acceptable Use Policy; do not use outputs to improve other large language models.
  Rule: Give recipients a copy of the license and ship the Tencent Hunyuan Notice file with redistributed weights.
  License: https://huggingface.co/Tencent-Hunyuan/HunyuanDiT/blob/main/LICENSE.txt (checked 2026-10-25; details: https://agentgid.com/assets/hunyuandit/)
```

Not verified: The Hugging Face API reports 0 downloads for this repo (its weights sit in a t2i/ subfolder, which the counter may not track). This repo holds the original v1.0 release; later versions are published in separate repos.

## Links
- AgentGid page: https://agentgid.com/assets/hunyuandit/
- Is it free for commercial use? https://agentgid.com/assets/hunyuandit/commercial-use/
- Source: https://huggingface.co/Tencent-Hunyuan/HunyuanDiT
- Source: https://huggingface.co/Tencent-Hunyuan/HunyuanDiT/blob/main/LICENSE.txt
- Source: https://huggingface.co/api/models/Tencent-Hunyuan/HunyuanDiT
- Source: https://github.com/Tencent-Hunyuan/HunyuanDiT


## Hunyuan-DiT: open image model for Chinese and English prompts

Hunyuan-DiT is an open AI model from Tencent that draws pictures from a written description in Chinese or English. It is known for understanding Chinese culture and long, detailed prompts, and a helper model can refine a picture over several chat turns. You run it on your own NVIDIA graphics card; business use is free under Tencent's own license with some rules.

### Where to use it
- Pictures from Chinese prompts: Make illustrations from prompts written in Chinese, including Chinese art styles and cultural details that many other models miss.
- Quicker drafts with the fast version: Use the distilled model with fewer steps when you need many draft pictures, for example to pick a direction for a poster.
- Refine a picture by chatting: Describe a picture, then ask for changes in follow-up messages; a helper model rewrites the prompt each turn.
- Follow a pose or outline: Give it an edge sketch, a pose or a depth picture, and it draws a new image that keeps that layout.
- Teach it your own style: Train a small add-on on your own pictures, such as a product line or a house style, and reuse it in every prompt.
- Visuals for Chinese-speaking customers: Make banners and social images for a shop or app aimed at Chinese-speaking customers, written straight from Chinese briefs. (idea, not from the official pages)

### How to start
- On Linux with an NVIDIA card of at least 11 GB, install PyTorch 2.0 or newer and the diffusers library.

```
pip install diffusers
```

- Load the v1.2 model in Diffusers format and draw a picture from a Chinese or English prompt.

```
pipe = HunyuanDiTPipeline.from_pretrained("Tencent-Hunyuan/HunyuanDiT-v1.2-Diffusers", torch_dtype=torch.float16)
pipe.to("cuda")
```

- For the chat-style refining and the full scripts, clone the code and download all model files into ckpts.

```
huggingface-cli download Tencent-Hunyuan/HunyuanDiT-v1.2 --local-dir ./ckpts
```


### License
Free to use, also in paid products, under the Tencent Hunyuan Community License. Very large services need Tencent's permission, and the license file published with the code does not cover the European Union. Pictures you make are yours, but posted AI pictures must be marked as machine-made.

### Good to know
- It needs an NVIDIA graphics card with at least 11 GB of memory; the chat helper model raises that to 32 GB. Tencent tested it only on Linux.
- The license file published with the code on GitHub excludes the European Union, while the files on the model pages do not. Check with a lawyer before any use in the EU.
- The usage rules forbid posting its pictures publicly without clearly saying they are machine-made, and forbid copying real people without consent.
- This page's model is the first release from May 2024. Newer versions 1.1 and 1.2 sit in separate model pages.
- Pictures or results from it may not be used to improve another large language model.

Checked 2026-11-12 against: https://huggingface.co/Tencent-Hunyuan/HunyuanDiT, https://huggingface.co/Tencent-Hunyuan/HunyuanDiT/raw/main/README.md, https://huggingface.co/Tencent-Hunyuan/HunyuanDiT/raw/main/LICENSE.txt, https://huggingface.co/Tencent-Hunyuan/HunyuanDiT/raw/main/Notice, https://huggingface.co/Tencent-Hunyuan/HunyuanDiT-v1.2/raw/main/LICENSE.txt, https://huggingface.co/api/models/Tencent-Hunyuan/HunyuanDiT, https://github.com/Tencent-Hunyuan/HunyuanDiT, https://github.com/Tencent-Hunyuan/HunyuanDiT/blob/main/README.md, https://github.com/Tencent-Hunyuan/HunyuanDiT/blob/main/LICENSE.txt, https://pypi.org/pypi/diffusers/json, https://raw.githubusercontent.com/huggingface/diffusers/v0.42.1/src/diffusers/pipelines/hunyuandit/pipeline_hunyuandit.py

### Mistakes to avoid
- Don't: HunyuanDiTPipeline.from_pretrained("Tencent-Hunyuan/HunyuanDiT") → Do: HunyuanDiTPipeline.from_pretrained("Tencent-Hunyuan/HunyuanDiT-v1.2-Diffusers", torch_dtype=torch.float16) (The official Diffusers example loads the -Diffusers model page; the original page holds files for the project's own scripts in t2i/ and dialoggen/ folders.)
- Don't: Loading the prompt helper and the image model together on a 16 GB card → Do: Start with --no-enhance, or load the helper in 4-bit with --load-4bit (DialogGen plus Hunyuan-DiT needs about 32 GB; the image model alone needs 11 GB.)
- Don't: Publishing generated pictures on a site or social account with no AI label → Do: Mark them clearly as machine-generated, for example in the caption or alt text (The Acceptable Use Policy in the license requires it for public content.)
- Don't: Running the GitHub code for users in the European Union without checking the license → Do: Check where your users are; the license in the code repository is limited to a territory that excludes the EU (Use outside the territory is unlicensed.)
- Don't: Shipping the model files in your product without the license and Notice file → Do: Give recipients the license and include the Notice text file (The license makes both a condition of distribution.)

### Questions
- Can I use the pictures in a paid product? Yes. Tencent claims no rights in the pictures you make, and commercial use is allowed below 100 million monthly users, if you follow the usage rules.
- Does it understand English prompts? Yes. It reads both Chinese and English; its strength is fine detail in Chinese prompts.
- Will it run on a smaller graphics card? The makers released scripts for cards with 6 GB of memory; the standard setup needs 11 GB.
- Can I use it in the European Union? Unclear. The license in the GitHub code repository excludes the EU, but the license files on the v1.0 and v1.2 model pages have no such clause.

---
AgentGid's catalog is curated: every entry was checked against the maker's official pages (registry, repository, license and pricing pages) on the date shown; a license value shows the quote it stands on wherever AgentGid has one. Usage numbers come from AgentGid's own scan of public repositories set up for AI coding agents. Claims by makers are reported, not tested by AgentGid.
