Generative video becomes a different creative medium when an artist can see an idea before the clip would have finished playing. Lightricks says its new LTX-2 foundation model can create a stylised six-second high-definition video in five seconds, giving creators an almost immediate preview to refine.
Speed is only part of the proposition. LTX-2 generates synchronised sound alongside the image, can produce native 4K video at up to 48 frames per second, and is designed to run on consumer-grade GPUs. Lightricks plans to release the model weights, datasets, and tooling under an open-source licence.
Together, those features aim to reduce three forms of friction: waiting for an iteration, assembling audio separately, and paying for remote computing every time an idea changes. Faster feedback can change creative behaviour, encouraging more alternatives before a director commits to one visual direction.
Picture and Sound Emerge From One Model
LTX-2 is a diffusion model, gradually removing noise until an output resembles the patterns learned during training. Lightricks has accelerated that process enough to provide rapid previews while generating dialogue, ambience, effects, or a soundtrack in sync with the visuals.
Previously, creators often produced sound separately and then spent time aligning it with the generated clip. Google’s Veo models have made integrated audio a prominent capability; LTX-2 brings a similar workflow into an openly released alternative.
Professional users can trade some speed for quality in LTX Studio, reaching 4K and 48 frames per second. Planned controls include pose and depth guidance, video-to-video generation, and alternative rendering options, giving creators more influence than a single text prompt provides. Those controls are essential when generated footage must match an existing production rather than stand alone as a short experiment.
Faster Iteration Changes the Economics of Direction
Earlier LTXV models laid the groundwork. LTXV-13B supported multi-scale rendering, allowing detail and colour to be added progressively, and the family later expanded to videos of up to 60 seconds. Users could introduce prompts while output streamed, making generation feel more like direction than a one-time request.
A distilled LTXV-13B reduced generation to four to eight diffusion steps and supported LoRA fine-tuning for a project’s visual style. Lightricks says the model was trained using licensed material from Getty and Shutterstock, addressing some of the copyright uncertainty surrounding generative media datasets.
LTX-2 is available through LTX Studio and an API. Pricing begins at $0.04 per second for fast HD output, $0.07 for a Pro tier, and $0.12 for Ultra generation at 4K, 48 frames per second, and full-fidelity audio. The company claims efficiency makes it up to 50 percent cheaper than competing models.
Open Source Moves Control Toward Creators
The open release is scheduled for November. Lightricks distinguishes “truly open-source” from open access by promising pretrained weights, datasets, tooling, and the model itself on GitHub. Local operation can reduce ongoing costs, support experimentation, and give studios more control over how the system fits into production. Developers will also be able to inspect the available components rather than relying solely on a hosted claim of openness.
Consumer-GPU compatibility is particularly important for independent creators and small teams that cannot reserve large cloud budgets for iteration. Local execution also allows developers to inspect, adapt, and fine-tune the technology rather than depend entirely on a hosted interface.
Claims about quality, speed, and hardware requirements will need independent testing across different scenes and devices. Six-second stylised clips are not equivalent to maintaining characters, physics, and narrative consistency across a finished film.
Still, LTX-2 points toward a more responsive production loop. When video and sound arrive together, previews appear faster than playback, and the underlying system can run locally, generative video begins to behave less like a remote novelty and more like a creative instrument an artist can repeatedly play.
