HunyuanVideo 1.5

What it is

HunyuanVideo 1.5 is Tencent’s open-weights video model, and its significance is as much about who is publishing it as what it does. It is released under the Apache 2.0 licence, which makes HunyuanVideo 1.5 one of only three genuinely clean open-source options in this section, alongside LTX-2.3 from Lightricks and Wan 2.7 from Alibaba. That a company the size of Tencent is giving away a capable video model, rather than locking it behind a paid API, is the detail worth pausing on.

Open weights versus open source

Open weights and open source are worth separating. Many models described as open are, in practice, only free to look at, with licences that restrict commercial use or reserve rights that make a lawyer nervous. Apache 2.0 is the real thing: a permissive licence that lets a business use, adapt and ship the model commercially without the obligation to open-source its own work in return. That is why HunyuanVideo 1.5 belongs in the small group of clean options rather than the larger, murkier category of models that are open in name only. For a company deciding what to build on, the difference is the one between a model it can depend on and a model it merely borrows.

The three clean options

The comparison with LTX-2.3 and Wan 2.7 places it precisely. All three carry the Apache 2.0 licence, so all three offer the same basic freedom to use and modify without a commercial trap. Each comes from a different kind of company: Lightricks, a consumer-software specialist; Alibaba, a platform giant with a deep open-model programme; and Tencent, a hyperscaler with interests spanning games, messaging and cloud. That three organisations of such different shapes have all settled on the same permissive licence tells you something about where the open end of the video market is heading.

Breadth across media

What sets Tencent apart is breadth. HunyuanVideo 1.5 is one product in the wider Hunyuan family, which covers text, image, video, 3D and world models. Very few organisations publish open weights across all of those categories at once, and doing so makes Tencent arguably the most complete open-weights publisher of any large company anywhere. The strategic logic is that a lab which can generate text, pictures, moving footage, three-dimensional assets and interactive world models from a single family is building an entire open toolkit rather than a single open product, and each piece reinforces the others. A studio that adopts one Hunyuan model has an obvious reason to try the next.

That completeness is the thing to watch. Owning an open stack that spans several media is a different proposition from shipping one strong model, because it lets Tencent set a baseline of free capability across the whole creative pipeline rather than in a single corner of it. For the open video market specifically, the model adds a third serious, cleanly licensed choice to sit beside LTX-2.3 and Wan 2.7, which deepens the pool a business can adopt without a contract and without a provenance worry hanging over the licence. For a business, the appeal of adopting from a single family is practical: shared tooling, consistent licensing and one relationship to understand rather than a patchwork of vendors, each with its own terms.

Where it sits

Where this leaves the wider AI video field is a market splitting cleanly in two. At one end sit the paid, closed APIs that compete on peak quality and integration; at the other, a growing set of Apache-licensed models, three of them now serious, that compete on price and freedom. HunyuanVideo 1.5 is a marker of how far the open side has come, and of how willing the largest technology companies now are to compete by giving capability away. For the broader picture of what these labs are releasing, the AI hub tracks the same pattern across text and images.