Open video model licenses: Wan 2.2, Mochi, Hunyuan and LTX-2.5

Wan 2.2 and Mochi 1 are listed as Apache 2.0 on their own repositories. HunyuanVideo-1.5 uses the Tencent Hunyuan Community License, which excludes the EU, UK and South Korea, and LTX-2.5 is free below an annual revenue threshold of $10 million and needs a paid agreement above it. "Open weights" therefore does not mean one thing, and the license decides whether you can ship.

What does each project say?

The table reflects what I read on each page today. I am reporting the headline terms only; the full text controls.

Which limits are easy to miss?

Two stand out. The Hunyuan license is territorial: use outside its territory is not licensed, which matters for teams and customers in the EU, UK or South Korea. It also bars using outputs to improve competing AI models and requires disclosing the actual provider to end users.

LTX's threshold is about the licensee's revenue, so a startup can start free and cross the line later. The existing LTX-2 license revenue threshold post goes deeper on that.

Does a license apply to hosted output?

When you call a hosted route you are using the provider's service under its terms, not downloading the weights, so the weight license is not the document that governs your request. I did not find a Sume doc that restates each upstream license, so read the terms of service for the route you use. Sume's Video generation docs describe the API, not upstream licensing.

How should I pick?

If you need the fewest strings: Wan 2.2 or Mochi 1 under Apache 2.0. If you need the newest quality and you are a small company: LTX-2.5 under its revenue cap. If you serve users in Europe, the UK or South Korea, do not build on Hunyuan weights without legal advice. If none of that is acceptable, pick a hosted id from GET /v1/videos/models and keep your own infrastructure out of the license question.

Also note that Apache 2.0 repositories can still link to components with other terms, such as a text encoder. Check every file you download, not just the headline repo.

What should be in my release checklist?

Before shipping anything built on downloaded weights, write down four things: the exact repository and commit you downloaded from, the license file at that commit, the territories your users are in, and your company's revenue band if the license has a threshold. Re-check the license when you upgrade, because a new version can come with new terms.

Keep copies of the license and any notice files alongside your deployment, since several of these licenses require you to pass the text on to recipients. Where a license restricts using outputs to train other models, make sure your data pipeline does not feed generated clips into a training set.