# Wan 2.2 — card for AI agents

Open MoE video model family for text-to-video, image-to-video, speech-to-video and character animation at 480p/720p.

Category: AI video generators and platforms · maker Wan team (Alibaba) · checked 2026-10-25

## Price
- Price: Free
- Free tier: Open weights, free to download; self-hosted (the 5B model runs on a single RTX 4090).

## License
Apache-2.0 — https://github.com/Wan-Video/Wan2.2/blob/main/LICENSE.txt

- commercial: **yes** (Commercial use OK)
  > The models in this repository are licensed under the Apache 2.0 License.
  (https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B)
- attribution: **no** (No credit needed)
  > We claim no rights over the your generated contents, granting you the freedom to use them while ensuring that your usage complies with the provisions of this license.
  (https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B)
- in app: **yes** (Can ship inside your app)
  > The models in this repository are licensed under the Apache 2.0 License.
  (https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B)
- youtube: **yes** (OK in monetized YouTube videos)
  > We claim no rights over the your generated contents, granting you the freedom to use them while ensuring that your usage complies with the provisions of this license.
  (https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B)

## Rules to follow
- Keep the Apache-2.0 license and notices when you redistribute the weights.
- The model card bars sharing illegal or harmful content, misinformation and personal data meant to harm.
- You are fully accountable for what you generate.

## How to connect
- Download only
- Repository: https://github.com/Wan-Video/Wan2.2
- Website: https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B

## Notes
Hugging Face API, Oct 25, 2026: Wan-AI/Wan2.2-T2V-A14B 3939 downloads, 591 likes; Wan-AI/Wan2.2-I2V-A14B 8212 downloads, 926 likes; Wan-AI/Wan2.2-TI2V-5B 18641 downloads, 874 likes; GitHub Wan-Video/Wan2.2: 18165 stars. Variants on the Wan-AI account: T2V-A14B, I2V-A14B, TI2V-5B, S2V-14B, Animate-14B (all Apache-2.0).

## Numbers (AgentGid data)
- Gid Score: 40
- GitHub stars: 18,766

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

```
- Wan 2.2 (AI video generators and platforms). Open MoE video model family for text-to-video, image-to-video, speech-to-video and character animation at 480p/720p. Price: free; commercial use OK; no credit needed.
  Connect: download from https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B (no API).
  Rule: Keep the Apache-2.0 license and notices when you redistribute the weights.
  Rule: The model card bars sharing illegal or harmful content, misinformation and personal data meant to harm.
  Rule: You are fully accountable for what you generate.
  License: https://github.com/Wan-Video/Wan2.2/blob/main/LICENSE.txt (checked 2026-10-25; details: https://agentgid.com/assets/wan/)
```

## Links
- AgentGid page: https://agentgid.com/assets/wan/
- Is it free for commercial use? https://agentgid.com/assets/wan/commercial-use/
- Source: https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B
- Source: https://huggingface.co/Wan-AI/Wan2.2-I2V-A14B
- Source: https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B
- Source: https://github.com/Wan-Video/Wan2.2
- Source: https://github.com/Wan-Video/Wan2.2/blob/main/LICENSE.txt


## Wan 2.2: open AI models that make short videos

Wan 2.2 is a free, open family of AI video models from Alibaba's Wan team. It makes short video clips from a written description and brings a still picture to life. It can also make a person in a photo talk or dance along with sound or another video. You run it on your own powerful graphics card, and you may use the clips in paid work.

### Where to use it
- A clip from a written idea: Describe a scene in a sentence, such as two cats boxing on a stage, and get a short 720p video. Useful for social posts, mood videos or story drafts.
- Bring a product photo to life: Start from your own picture, such as a product shot or an illustration, and add movement described in words. The video keeps the shape of the picture.
- A talking or singing person from audio: Give one photo of a person and a voice or song recording. The model makes a video of that person speaking or singing along, as long as the audio.
- Make a character copy a dance: Take a video of someone moving and a picture of a character. The character repeats the moves, or replaces the person in the video.
- Make videos from Python code: Call the model from a Python script, for example to make a batch of clips for a product catalog.
- Moving backgrounds for a landing page: Make a calm looping clip, like waves or clouds, from a short description and use it as the hero video of a site. (idea, not from the official pages)

### How to start
- Get the code and install the packages. PyTorch 2.4 or newer is needed; install flash_attn last if it fails.

```
git clone https://github.com/Wan-Video/Wan2.2.git
cd Wan2.2
# Ensure torch >= 2.4.0
pip install -r requirements.txt
```

- Download a model. The 5B model is the one that fits a single 24 GB card.

```
huggingface-cli download Wan-AI/Wan2.2-TI2V-5B --local-dir ./Wan2.2-TI2V-5B
```

- Run generate.py with the task, size, model folder and your prompt.

```
python generate.py --task ti2v-5B --size 1280*704 --ckpt_dir ./Wan2.2-TI2V-5B --offload_model True --convert_model_dtype --t5_cpu --prompt "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage"
```


### License
The Wan 2.2 models are under the Apache 2.0 license, and the makers claim no rights over the videos you make. You may use them in paid work, but you are responsible for the content and must not make harmful or misleading videos.

### Good to know
- The 5B model needs a graphics card with at least 24 GB of memory, like an RTX 4090. The larger 14B models need 80 GB cards.
- The files are big. The 5B model folder alone holds about 34 GB, with single files near 10 GB.
- An optional step that rewrites your prompt in more detail can call Alibaba Cloud with a paid key. A local Qwen model can do the same on your machine.
- Free online demos exist on Hugging Face and on the wan.video site if you want to try before setting it up.
- It also works inside ComfyUI, a visual tool where you connect blocks instead of typing commands.

Checked 2026-11-10 against: https://github.com/Wan-Video/Wan2.2, https://github.com/Wan-Video/Wan2.2/blob/main/README.md, https://github.com/Wan-Video/Wan2.2/blob/main/LICENSE.txt, https://github.com/Wan-Video/Wan2.2/blob/main/requirements.txt, https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B, https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B/raw/main/README.md, https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B/raw/main/README.md, https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B-Diffusers/raw/main/README.md, https://huggingface.co/api/models/Wan-AI/Wan2.2-TI2V-5B/tree/main

### Mistakes to avoid
- Don't: --size 1280*720 with --task ti2v-5B → Do: --size 1280*704 (or 704*1280) for the 5B model (The 5B model uses its own 720P sizes.)
- Don't: running the A14B or S2V-14B single-GPU commands on a 24 GB card → Do: use the TI2V-5B model on 24 GB cards; the 14B commands need at least 80 GB or several GPUs (The README states the memory each command needs.)
- Don't: stopping at an out-of-memory error → Do: add --offload_model True, --convert_model_dtype and --t5_cpu (These options lower the graphics memory needed.)
- Don't: --use_prompt_extend --prompt_extend_method 'dashscope' without asking → Do: use --prompt_extend_method 'local_qwen', or ask the person for a DASH_API_KEY first (The DashScope method sends the prompt to Alibaba Cloud and needs a key the person must create.)
- Don't: pip install -r requirements.txt fails on flash_attn and the setup is abandoned → Do: install the other packages first and flash_attn last (The README gives this order as the fix.)

### Questions
- Can I use Wan videos in ads or on YouTube? Yes. The makers claim no rights over generated videos, and the models are under Apache 2.0; the content must stay lawful and not harmful.
- Which model runs on one home graphics card? The TI2V-5B model, on a card with 24 GB of memory such as an RTX 4090.
- Can it make a video without a written prompt? Yes, from the image alone; a prompt can be written for you from the picture with the optional prompt rewriting step.
- Can I set the video shape when I start from an image? The size you give sets the area; the shape follows your input image.

---
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.
