What Is Wan 2.2? Where to Use It, Is It Free, 14B vs 5B, GPU Needs and Commercial Use

You give Wan 2.2 a text, an image or audio, and it generates a video. It is Alibaba's (Wan-AI) open-weight model family; it downloads for free under the Apache 2.0 license, commercial use is allowed, and no rights are claimed over the generated content. Running it on your own computer takes a powerful graphics card: the 14B models need at least 80 GB of memory on a single card, the 5B model at least 24 GB. To try it without installing there is wan.video; its pricing is not clear in the official sources. Turkish prompt support is not covered in the official sources.

What it does: Open model that generates video from text, images and audio
Where: wan.video, Hugging Face, ComfyUI, your own computer
Model cost: Free (Apache 2.0)
Commercial use: Allowed; must comply with the law
Graphics card: 5B: at least 24 GB; 14B: 80 GB
Turkish prompts: Not stated in the official sources

What is Wan 2.2 for?

You give Wan 2.2 a text or a photo, and it produces a short video. If you give it a voice recording or a reference video, it can also make a talking or moving character video. The videos are short: the model card for the 14B text-to-video model says 5-second videos.

Text to videoWhat you get: VideoNeeds a separate setup Photo to videoWhat you get: Video in the aspect ratio of the input imageNeeds a separate setup 720P video on a single GPUWhat you get: 720P, 24 fps (1280x704)Needs a separate setup

Who it's for: technical teams and content creators with a powerful graphics card who want to generate video on their own computer without uploading it to the cloud. For those who want a ready-made app there is wan.video; its pricing is not clear in the official sources.

The official README says Wan2.2 is good at cinematic looks and complex motion and that it leads both open and closed models. These are the maker's own claims.

Where can you use Wan 2.2? wan.video, ComfyUI and your own computer

Wan is not an app but a model: the files in which the AI has learned to generate video. The places where you can use it are below. Only the routes named in the official sources are listed.

Where	What	Watch out
wan.video	Wan's official online site (linked from the README)	Check plans, credits and pricing on the site with your account.
Hugging Face Spaces	The README shows working demos for 5B, S2V and Animate	Limits and sign-in requirements are not stated.
ComfyUI	Ready-made workflows for Wan2.2; on your own computer or on Comfy Cloud	The ComfyUI documentation says 5B should run on 8 GB of memory with memory offloading.
Diffusers (Python)	Ready-made Wan2.2 classes in Hugging Face's library	Requires coding knowledge.
Official command (generate.py)	Running it directly with the command in the repository	Installation problems are reported on Windows (below).

Pricing, limits and privacy details for wan.video and the Spaces are not covered here; check the relevant pages before using them.

How do you install Wan 2.2? ComfyUI and Python

Update ComfyUI. In the Template Library, search for “Wan2.2 5B” (or the 14B templates) and open it.
The models download from the Comfy-Org/Wan_2.2_ComfyUI_Repackaged repository: the diffusion model, wan2.2_vae.safetensors and the umt5_xxl_fp8_e4m3fn_scaled.safetensors text encoder. Folder locations are in the ComfyUI documentation.
If you like, enable the Load Image node and give it a photo (image to video). Run it with Run.

If your computer isn't enough, the ComfyUI documentation also links to opening the template on Comfy Cloud; check the cost of cloud usage separately.

Get the repository: git clone https://github.com/Wan-Video/Wan2.2.git
Install the packages: pip install -r requirements.txt (requires torch >= 2.4.0; if flash_attn throws an error, install the other packages first and flash_attn last)
Download the model: huggingface-cli download Wan-AI/Wan2.2-TI2V-5B --local-dir ./Wan2.2-TI2V-5B
Run it with the command below.

Text to video on a single card with 5B (README example)

› 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"

If you run out of memory (OOM), the README suggests these options: --offload_model True, --convert_model_dtype, --t5_cpu. The official README examples are for NVIDIA/CUDA; the official sources say nothing about AMD, Mac or Intel. The official pages also don't clearly state the disk size of the model files.

Reported by users It has been reported on GitHub that pip install -r requirements.txt fails on Windows 11 with a flash_attn and CUDA_HOME error (issue #111, open). Users suggest installing torch and the CUDA toolkit manually first. There is no official fix; the ComfyUI route looks easier.

How do you use Wan 2.2? A ready-made prompt example

The more detailed the prompt, the more controlled the result. For richer detail, the README suggests having a language model expand the prompt (--use_prompt_extend); this can optionally use the Alibaba Cloud Dashscope API, and your prompt is sent there. It can also be done with a local language model.

Example text-to-video prompt (English)"A red vintage car drives along a coastal road at golden hour, camera follows from the side, soft warm light, gentle waves in the background, cinematic, smooth motion."This is a sample prompt; try it with your own. The README examples are in English and Chinese; there is no official source on Turkish prompts.

Task	How
Animate a photo	Enable the `Load image` node in ComfyUI, or pass `--image` to generate.py. The size follows the aspect ratio of the input image.
Talking character	S2V-14B: image + audio file (`--audio`); the video length follows the audio.
Match a character to motion from another video	Animate-14B: first, a preprocessing step extracts the motion from the reference video.

Which version? 14B, 5B, S2V and Animate compared

Model	What for (README)	GPU memory
T2V-A14B	Text to video; 480P and 720P	At least 80 GB on a single card (README)
I2V-A14B	Image to video; 480P and 720P	At least 80 GB on a single card
TI2V-5B	Text and/or image to video; 720P, 24 fps (1280x704)	At least 24 GB, e.g. RTX 4090 (README)
S2V-14B	Video from a voice recording + image	At least 80 GB on a single card
Animate-14B	Matching a character to the motion in a reference video, or replacing the person in a video	The README gives no figure

Which one to choose?

If you have a single consumer card: TI2V-5B.
If you have a multi-GPU server or an 80 GB card: A14B (text or image).
If you want to make a person talk: S2V-14B.
If you want to animate a character with the motion from another video: Animate-14B.

Memory figures

The 24 GB and 80 GB figures come from the README. The ComfyUI documentation says 5B can fit in 8 GB; this relies on ComfyUI's own memory offloading. The documentation does not state 14B's actual requirement in ComfyUI.

Model list as of October 15, 2026. The official sources don't make clear whether newer Wan versions outside this repository are open-weight; this page covers Wan 2.2.

Does Wan 2.2 understand Turkish prompts?

Turkish promptsUnclear	The official sources don't mention Turkish prompt support. The Hugging Face model cards mark the languages as “en, zh” (English, Chinese).
InterfaceUnclear	As a model, it has no interface of its own. The ComfyUI interface is a separate product; the language of wan.video is not clear in the official sources.

What you can do: try your prompt in Turkish first; if it isn't understood, write it in English (the README examples are in English and Chinese). There is no benchmark for Turkish, so judge the result yourself.

Is Wan 2.2 free? License and commercial use

Model cost

The files download for free (Apache 2.0)
The graphics card, electricity and disk are on you

Allowed

Using, modifying and distributing the model for free (under the Apache 2.0 terms)
Using the generated videos: the README says “we claim no rights over the generated content”
Commercial use (the ComfyUI documentation also states that Apache 2.0 covers this)

Not allowed

Sharing content that violates the law
Harming people or groups, spreading harmful personal data
Spreading misinformation, targeting vulnerable groups

This is not legal advice. According to the README, responsibility for use is entirely yours; for the full list of prohibitions, read the license text and the license section of the README. The terms of use of the hosted service on wan.video are separate; read them on the site.

wan.video pricing

The plans, credits and prices of the hosted service are not on this page. Check the plan shown in your account yourself; see the price before paying.

Data

If you run it locally, your inputs don't leave your computer; if you enable Dashscope for prompt extension, your prompt goes to Alibaba Cloud. If you plan to use wan.video, read its privacy policy.

Frequently asked questions

What is Wan 2.2?

It is Alibaba's family of open-weight video generation models. It has separate models for video from text, images and audio, and for character animation.

Where can you use Wan 2.2?

On wan.video, in Hugging Face Spaces, in ComfyUI (local or Comfy Cloud), in the Diffusers library, and on your own computer with the generate.py command from the repository.

Is Wan 2.2 free?

The model files download for free under Apache 2.0; the cost is your graphics card and electricity. The pricing of the hosted service on wan.video is not clear in the official sources.

What is the difference between Wan 2.2 14B and 5B?

The 14B models generate 480P/720P video from text or images and need at least 80 GB of memory on a single card. The 5B model generates 720P at 24 fps and, according to the README, runs on a single 24 GB card (RTX 4090).

Can Wan 2.2 be used commercially?

Apache 2.0 allows commercial use, and the README states that no rights are claimed over the generated content; illegal and harmful content is prohibited. This is not legal advice.

Does Wan 2.2 understand Turkish prompts?

The official sources don't say; the model cards mark English and Chinese. Try Turkish, and if it doesn't work, write in English.

Tasks and setup kits that use Wan 2.2

Making a short product promo video with AI

Related guides

CapCut DaVinci Resolve

Sources

Not clear in the official sources: wan.video plans, credits and pricing · Hugging Face Space limits · Turkish prompt support · 14B's actual memory requirement in ComfyUI · disk size of the model files · whether Wan 2.5 and later versions are open. Each is covered in the relevant section along with what you can do.