LTX-Video: what Lightricks has published

Lightricks publishes LTX-Video under Apache 2.0, including a 13B model with distilled and fp8 variants for lower VRAM. The repository states support for up to 60 seconds of video. LTX-2.5, sold through the vendor's API, has no published weights. As of 2026-09-12.

The LTX-Video repository page as it appeared on 16 September 2026. — Fig. 1 Published weights from the same company that sells hosted seconds, which is why this family appears twice on this site under different headings.Screenshot of github.com/Lightricks/LTX-Video, captured 2026-09-16. Reproduced to show where this entry was read from.

Fig. 2 Filled where the repository states something, hollow where it does not. A blank is not an inference.

Fig. 3 How many releases state something in each column, counted off the rows below.

LTX-Video releases against whether their weights are published. Recorded 2026-09-12.
Release	Weights	Licence	Note
LTXV-13B	Published	Apache 2.0	Up to 60 s stated
Distilled variants	Published	Apache 2.0	Faster inference
fp8 variants	Published	Apache 2.0	Lower VRAM
LTX-2.5	No weights published	-	Sold through the vendor's API

Inclusion rule. Releases named in the vendor's own repository, plus any version that appears only on the vendor's priced pages. The latter is recorded as having no published weights. Order. Published releases first, then versions sold only as a service.

1Variants are the useful part of this release

Publishing one set of weights is a gesture. Publishing distilled and fp8 variants alongside the full model is a decision about who can run it: the same architecture at several memory footprints means the hardware question has more than one answer, and the vendor rather than the community has answered it.

The 60 second figure matters for the same reason. Most published video weights are trained and documented around clips of a few seconds, and a stated 60 seconds changes what the model is for. It is the vendor's own claim in its own release notes and is recorded as that.

2The same split as everywhere else

What is open is LTX-Video; what is sold by the second is LTX-2.5. The pattern matches the other family in this register exactly, and it is worth stating plainly because the shared brand makes the two easy to conflate. A production that evaluates the open weights is not evaluating the hosted product.

No hardware figures appear in the repository in the form of a stated minimum, so nothing is recorded in that column here. The variant names imply the intent without stating a number, and an implication is not a requirement.

Licence
Apache 2.0stated in the repositoryLightricks, LTX-Video repository / recorded 2026-09-12
Published models
A 13B model, with distilled and fp8 variants published for lower VRAM and faster inferenceseveral variantsLightricks, LTX-Video repository / recorded 2026-09-12
Length
The repository states support for up to 60 seconds of video on LTXV-13Bfrom the release notesLightricks, LTX-Video repository / recorded 2026-09-12

3Open weights, hosted

Instead of local hardware for “LTX-Video”, SceneMixer runs hosted models; its guides describe the result:

Video is 16:9 landscape by default, with 9:16 vertical available; you set it once per project. Segments are typically 4 to 15 seconds, and some model tiers go to 30.
SceneMixer, AI Short Drama Generator — “What length and aspect ratio does it output?” (read 2026-10-13)

SceneMixer writes dialogue into the script analysis, assigns each line to a shot with a speaker and a short delivery note, and the video model performs it — picture and audio generated together, so the mouth forms that sentence in the language you chose.
SceneMixer, AI Dialogue Generator (read 2026-10-13)

4Sources

Read from the official LTX-Video repository, checked 2026-09-12. What the hosted service sells is on its hosted page.