Wan AI Video

Wan AI video is the Wan family of open video models, used either by downloading the weights and running them yourself or through a hosted workspace that runs them for you. This site is the second kind: you write a prompt or supply a frame, pick a route, and get a clip back, with no install and no GPU of your own.

The routes, in plain terms

Different jobs map to different routes, and mixing them up is the most common reason a first attempt disappoints:

• Text to video — you describe a shot and the model builds it. Best when you already know what the shot is.
• Image to video — you supply a first frame and describe only the movement. Usually the fastest route to a usable result, because the composition is already decided.
• Reference to video — you hand it one or two example clips and it carries their look onto new content.
• Extend a video — continues a clip you already have instead of rebuilding it.
• Edit an image and translate an image — change part of a still, or read and re-render the text inside one.

Model generations sit alongside these: Wan 2.5 exposes the widest set, with Wan 2.2 and Wan 2.1 available too. The live form is the source of truth for which routes are up right now, not this page.

What separates a good result from a wasted run

Three things, in order of impact:

1. Naming the camera move. Subject, action, and how the shot moves. The third is the one people omit, and it changes the output most.
2. Iterating at a low output band, then re-running only the winners at the size you will publish.
3. Fixing the seed when comparing two prompts, so you learn which change did what.

Prompt guidance in depth is on Wan 2.2 Prompts, and why the band matters is on Wan 2.2 Resolution.

What it costs, and when it is free

Hosted generation here is credit-based: the credit cost of your chosen settings is displayed beside the form before you submit, so a test is a decision rather than a guess. Credit packs and prices are on pricing.

Nothing here is unlimited, and this site does not hand out large free balances — if a page promises unlimited free generation of a hosted video model, that is worth treating with suspicion. "Free" for Wan usually means the weights are free to download and the cost is your own hardware; Wan 2.2 GGUF covers what that involves.

What to expect from the output

Every generation is a test, not a final asset. Quality varies by route and by prompt, and the same route can produce a strong clip from one prompt and a weak one from its neighbour. Check motion, artefacts and whether the shot matches your intent before building on it — and check the licence that applies to the model route you used before publishing.

Questions people ask

Do I need to install anything? No. The routes run in the browser.

Which route should I try first? If your idea starts from a still image, image to video. If it starts from words, text to video.

Is it free? There is a credit cost per generation, shown before you submit; see pricing. The models themselves are open, which is a different kind of free — Wan 2.2 Hugging Face explains that trade.

Where do I start? Open the Wan workspace and choose a route.