# Run Wan 2.1

The best-looking open video model. Slow and worth it for finished work. Ready in about 90 seconds on an H100 80GB, from $2.25/hr. No install, no code.

## What it does

Wan 2.1 generates video from a description or from a starting image, at noticeably higher quality than the faster models.

Motion is more coherent, subjects hold their shape across the clip, and it can render readable text in the frame, which almost no other video model manages.

It is slow and it needs a large card, so a clip costs real money. Treat it the way you would treat a render: get the shot right in a cheap model first, then run the final here.

## Your setup

We run Wan 2.1 on an H100 80GB with 80GB of memory, at $2.25 per hour, billed by the minute. It takes about 4 minutes per clip.

The 14 billion parameter version of Wan 2.1 needs roughly 30GB for the model plus another 25GB or more of working space for the frames of the clip. That does not fit on a 48GB card at 720p without cutting the clip short. The H100 also has substantially faster memory than the L40S, which matters here because this model is genuinely compute-bound rather than waiting on data.

## What it costs

20 clips takes about 80 minutes of GPU time, which is about $3.00.

## Two ways to run it

- **Open the app.** Generate and preview one clip at a time, with the cost meter running.
- **Run a batch job.** Queue a shot list overnight. The GPU stops itself when the last clip finishes.

## Model facts

| Field | Value |
|---|---|
| Parameters | 14B |
| License | Apache 2.0 |
| Memory required | 60GB minimum, 80GB recommended |
| Base model | Trained from scratch by Alibaba |
| Publisher | Alibaba (Wan-AI) |
| Recommended hardware | H100 80GB, 80GB |
| Category | [Video and film](https://gpuvault.io/working-on/video-and-film/) |

## Questions people ask

### How much does one finished video clip cost?

About $0.15. A five second clip takes roughly four minutes on an H100 at $2.25 per hour. The real cost is iteration, not the final render. Most people generate ten or fifteen versions before they are happy, which puts a usable clip closer to $2. Use LTX-Video to find the shot first and you will spend a fraction of that.

### Do I need to install anything to use Wan 2.1?

No. We start a machine with Wan 2.1 already loaded and hand you a link. Everything runs in your browser, there is nothing to download, and nothing is left on your computer afterwards. It works the same on a Mac, a Windows laptop, or a Chromebook. A workspace is usually ready in about 90 seconds.

### How much does it cost to run 20 clips?

About $3.00. Wan 2.1 takes about 4 minutes per clip on the H100 80GB we recommend, so 20 clips is roughly 80 minutes of GPU time at $2.25 per hour. Billing is by the minute, and a batch job shuts the GPU off the moment the last item finishes. The calculator above works this out for your own numbers.

### Can I run Wan 2.1 on a cheaper card?

Sometimes, and the calculator will not always make it look worth it. The 14 billion parameter version of Wan 2.1 needs roughly 30GB for the model plus another 25GB or more of working space for the frames of the clip. That does not fit on a 48GB card at 720p without cutting the clip short. The H100 also has substantially faster memory than the L40S, which matters here because this model is genuinely compute-bound rather than waiting on data. If you want to try a different card anyway, the advanced catalog lets you pick one and shows the estimated time before you commit.

### Can I use Wan 2.1 commercially?

Yes. Wan 2.1 is released under the Apache 2.0 license, which permits commercial use with no conditions worth worrying about. You can use the output in paid client work and in products you sell.

### What happens to my files after I stop?

Anything you save into your workspace folder stays in your account and is there when you come back. Everything else is destroyed: the machine is terminated and its disk is wiped. Batch job inputs and outputs are deleted from our storage 24 hours after you download them. We do not read your files or use them to train anything.

---

*Source: https://gpuvault.io/model/wan-2-1/ — GPUVault, GPU rental by the minute.*