// AI NEWS · MODEL RELEASE

LTX-2.5 Launches as an Open-Weight Video Model That Makes a 10-Second Clip in Seconds — On a GPU You Already Own

LTX, the open-model company spun out of Lightricks, released LTX-2.5 with open weights, synchronized native audio, and speeds fast enough to render a clip quicker than the clip runs — free for organizations under $10M in annual revenue.

2026-08-12 · by Moe Ameen

What happened

On August 11, 2026, LTX — the open world-model company spun out of Lightricks — released LTX-2.5, an open-weight video and world model. Its headline claim is speed paired with openness: on two NVIDIA GB200 chips it generates a 10-second, 720p image-to-video clip in roughly 6.8 seconds, faster than the clip itself plays back, and through LTX's managed API it returns a 1080p clip in about 23.7 seconds. The model does both text-to-video and image-to-video and keeps the synchronized native audio LTX introduced in its 2.3 generation.

The part that changes the math is the license. LTX-2.5's weights are open and free to use for organizations under $10 million in annual recurring revenue, with larger companies negotiating a license. It ships day-one on Hugging Face, inside ComfyUI, and through fal.ai and the LTX API. LTX also tuned it to run on hardware creators already own — a minimum of roughly 16GB of VRAM, optimized for local NVIDIA RTX GPUs and the NVIDIA DGX Spark desktop, with a distilled variant for faster local inference. LTX's own benchmarks claim the on-prem model generates several times faster than the nearest closed alternative.

On capability, LTX-2.5 adds native multishot — one generation produces several connected shots that hold character, environment, lighting, and voice across cuts — plus a new diffusion video decoder for cleaner high-motion footage, a custom Gemma-based text encoder with a prompt enhancer, automatic clip duration, native 4K HDR with a RAW finishing workflow, and a beta precise-editing mode. Treat specific speeds, resolutions, and license terms as launch-day figures and confirm them against ltx.io before relying on them.

What LTX-2.5 does not do is anything after the render. It outputs a clip, with sound — not a captioned, reframed, on-brand post. There is no editor, no persona system, and no scheduler or publishing, which is exactly the layer a creator needs to turn a fast, cheap, local render into something a feed rewards.

Why it matters for creators

  • Fast, free, local video generation moves the scarce skill. When a frontier-class clip renders in seconds on a GPU you own, the edge stops being "can you make a clip" and becomes "can you turn a firehose of clips into a channel."
  • Open weights under a $10M revenue threshold means most creators and small teams can run it at no per-render cost — the generation bill goes to near-zero, and the remaining cost is captioning, formatting, and distribution.
  • Native audio still is not muted-feed audio. Most short-form is watched on mute, so even a clip that ships with perfect sound needs burned-in captions to earn a view — the finishing step did not disappear.
  • Local batch generation creates a consistency problem. A folder of clips rendered overnight has no shared look or voice; without a brand layer, a week of posts assembled from them reads as a grab bag, not a body of work.
  • It is a model, not a studio. There is no scheduler and no publishing, so the labor of posting across platforms is precisely what near-free local generation does not buy you.

How to act on this with Kompozy

The instinct on a launch like this is to fixate on the render time. But when generation is this fast, free, and local, speed stops being the bottleneck — it moves the bottleneck downstream, to everything that happens after the MP4 lands. That is the specific gap [Kompozy](/) fills. Kompozy is a full AI generation and multi-platform publishing engine, not another model. Point it at a folder of LTX-2.5 clips and it burns in branded, on-style captions so the sound-off feed still lands, reframes each clip to 9:16, 1:1, and 16:9 per destination, and stacks a hook overlay through [HyperFrames](/glossary/hyperframes) — then schedules and publishes across the eight social platforms plus blog and email from one queue behind a per-post review gate with [Autopilot](/glossary/autopilot). The [Persona Brief](/glossary/persona-brief) applies one voice and HyperFrames one pixel-exact look to every clip, so a week of overnight local renders reads as a single channel rather than a stock-footage mixtape.

It also makes what a video model can't. Feed the concept behind an LTX-2.5 clip into Kompozy and one render fans into a [carousel](/glossary/output-buckets) breaking down each beat, a quote card pulled from the script, a face-locked persona photo, a captioned [Persona Short](/glossary/persona-shorts), a blog draft, and platform-native captions in your voice. LTX-2.5 made raw video nearly free; Kompozy is where that near-free clip becomes a business.

Quick takeaways

  • LTX-2.5 is open-weight and free for organizations under $10M ARR; larger companies negotiate a license.
  • It renders a 10-second, 720p image-to-video clip in about 6.8 seconds on GB200 hardware, ~23.7s at 1080p via API.
  • It runs locally on NVIDIA RTX GPUs and DGX Spark from roughly 16GB of VRAM, with synchronized native audio and native multishot.
  • It is a model, not a content tool — Kompozy adds captions, brand styling, formats, and multi-platform publishing on top.

Frequently asked questions

What is LTX-2.5?

LTX-2.5 is an open-weight AI video and world model released August 11, 2026 by LTX, the open-model company spun out of Lightricks. It generates a 10-second, 720p image-to-video clip in about 6.8 seconds on NVIDIA GB200 hardware, supports text-to-video and synchronized native audio, and runs locally on RTX GPUs.

Is LTX-2.5 really open source and free?

The weights are open and free to use for organizations under $10 million in annual recurring revenue; larger companies negotiate a license. It is available on Hugging Face, inside ComfyUI, and through fal.ai and the LTX API. Confirm the exact license terms on ltx.io.

What makes LTX-2.5 different from closed models like Veo or Kling?

Open weights and local execution. LTX-2.5 runs on NVIDIA RTX GPUs and DGX Spark from around 16GB of VRAM, so a frontier-class video model runs on hardware you already own rather than behind a metered cloud API, with LTX claiming it generates faster than the clip runs.

Is an LTX-2.5 clip ready to post?

Not quite. It arrives as a rendered file, often with native audio, but no captions for muted feeds, no per-platform aspect ratios, no brand styling, and no way to publish. A content engine like Kompozy handles captioning, reframing, brand voice, and publishing across platforms.

Related news

← All AI news · Get started →