Wan 2.2 is a family of open video generation models, and "Wan 2.2 AI" is what most people type when they want to try it without touching a terminal. That is the whole point of this workspace: the model runs on our machines, you work from a browser tab, and nothing gets installed on your laptop. If you have been reading about the 2.2 release and want to see what it actually does with your own prompt, you can start here rather than shopping for a GPU first.
The 2.2 generation is not one model but a set of related checkpoints, split by task. Some are tuned for turning a written description into motion, some for animating a still image, and some for extending footage that already exists. The workspace exposes those as separate entry points instead of hiding them behind a single button, because the choice changes what kind of prompt works. You can browse the full set of routes on the /wan page, or jump straight into /text-to-video if you already have a scene in mind.
The practical difference from running it locally is not quality, it is friction. No CUDA setup, no quantisation decisions, no waiting on a multi-gigabyte download before your first attempt. You type, you submit, you watch the result.
The single biggest factor is the prompt, and the second biggest is your patience with the first attempt. Treat the initial render as a test rather than a finished asset. You are checking whether the model understood your subject, your camera movement, and your pacing. Almost nobody gets a usable clip on the first try, and that is normal rather than a sign you did something wrong.
Write prompts as if you were describing a shot to a camera operator: who or what is in frame, what they are doing, where the camera sits, and how the light behaves. Vague mood words alone tend to produce vague motion. If you are animating a still, the source image carries a lot of the composition, so your text should describe movement and change rather than re-describing what is already visible.
Keep clips short while you iterate. A short render tells you whether the direction is right at a fraction of the wait, and you can lengthen once the motion reads correctly.
Model variant matters more than most settings. A text-to-video checkpoint and an image-to-video checkpoint will respond differently to the same sentence, and picking the wrong one is a common reason a prompt appears to be ignored. If you are starting from a photograph, use the image route. If you are starting from nothing, use the text route.
Resolution and duration trade against each other. Higher resolution and longer clips both cost more time and more credits, so it rarely makes sense to max both while you are still exploring. The credit cost of a run is displayed beside the form before you submit, so you can see the trade before committing rather than after.
If you want to understand how the 2.2 line differs from its predecessor, the /wan-2-2 overview and the /wan-2-1 page cover the generational differences without hype.
This is not a local installation guide, and this site does not distribute model weights or offer any downloads. If your goal is to run Wan 2.2 on your own hardware, you need a weights host and a setup guide, not this workspace. Pages about file formats and local runtimes exist for reference, but nothing here hands you a file.
It is also not the right tool for precise editing work. If you need to change one object in an existing shot while everything else stays pixel-identical, generative video is a clumsy instrument. And it is not a replacement for a real camera when you need a specific person, a specific location, or a guaranteed outcome on a deadline. Generative clips are best treated as exploration and previsualisation, not as a locked final deliverable.
Do I need a GPU or any installation?
No. The models run on our hardware and you interact through the browser. Nothing is downloaded to your machine.
Can I download the Wan 2.2 weights here?
No. This site does not host or distribute model weights and provides no downloads. We only run the models for you as a hosted service.
Why did my first clip come out wrong?
Because first renders usually do. Adjust the prompt, confirm you picked a variant that matches your input type, and try again with a shorter clip before scaling up.
How much does a run cost?
The credit cost is shown next to the form before you submit, so you always see the price of a specific run before spending anything.