Skip to content
EN
English 简体中文 soon 日本語 soon

LTX-2

Keyframe-controlled video generation for high-fidelity work

Visit official site

What LTX-2 is

LTX-2 is the video generation model from the LTX team at Lightricks, and it is the engine that runs inside LTX Studio. It is a model rather than an application, so most people meet it through that studio or through someone else's deployment.

The official page at ltx.video returned a forbidden response when fetched, so what follows comes from the surrounding ecosystem and from directory listings rather than from the vendor directly.

The model and the studio built around it are separate things, and knowing which one you are dealing with matters once questions of terms arise.

What you can do with it

Picture and sound are produced together by one model instead of being generated apart and joined later, which is the technical claim that separates it from most of the field. Resolution reaches four times high definition, and the pitch is coherence over length rather than isolated shots.

Direction is the emphasised part: shots are meant to be directed rather than merely described, which is why keyframes and reference guidance matter more here than raw prompting. The weights and training framework have been published, so the model can be run on your own hardware, and several third parties now host it as a service. Prices, audience figures and certification could not be read from the blocked page, and the hosted route carries different terms from self-hosting.

Who it is for

Creators and teams who want a controllable, openly published video model are the audience, either by running it themselves or by working through the studio built on top of it.

Developers who need the weights locally have a clear reason to be here. Anyone who simply wants a web interface is better served by a hosted product.

Researchers and engineers who want to inspect the training approach have a further reason to be here. Teams needing support commitments should use a hosted product instead.

What to watch out for

An open model shifts responsibility onto whoever deploys it: consent, content rules and output rights sit with the operator, or with the terms of whichever host you use. The official page being unreachable means the licence, any hosted pricing and usage restrictions all need confirming from the live site before you rely on them.

No SOC 2, ISO or GDPR certification could be read from the blocked page. Teams taking the hosted path should read that studio's trust centre, since the compliance story belongs to the host rather than to the model.

Pros & cons

✓ What we like

  • Picture and sound come from one model rather than being joined later
  • Weights and training framework are published, so it can be self-hosted
  • Built for direction and long-form coherence rather than single shots

! What to watch out for

  • The official site returned a forbidden response, so claims are unverified
  • Responsibility for output sits with whoever deploys the model
  • No certification is readable for the model itself

FAQ

Is this an app I can log into?

No, it is a model, usually reached through LTX Studio or by self-hosting.

Can I run it myself?

Yes, the weights are published, which is why third-party hosts exist.

Are prices published?

Not on the blocked page, so check the hosted service you intend to use.

Last reviewed: 2026-09-16

More AI video generation tools

View all →

How we review