LTX Video for Real Estate Agents: Open Video You Can Run Free

All Things AI · Season 3 · The New Models · Episode 8

LTX is an open-weights video model from Lightricks, and the reason it belongs in this season is that you can run it for free on a rented GPU with no per-clip charge. We tested the smaller version on a free-tier cloud GPU: about seven and a half seconds to generate each five-second clip, at no cost.

What it is

LTX-2.5 is Lightricks’ current flagship: a 22-billion-parameter open-weights model that generates synchronized video and audio together from text, an image, or an existing video. It generates connected multi-shot sequences in one pass — a wide shot, a medium shot, a close-up, cut together — while keeping the same character, environment, and voice consistent across the cuts. It runs at up to 4K resolution and, on the hardware Lightricks used for its own benchmark, produced a 10-second 720p clip in under seven seconds.

LTX-Video 0.9.8 13B distilled is the earlier, smaller model in the same family: 13 billion parameters, image-to-video only, no audio, built to be faster and lighter than the full-size version at a small cost in quality. This is the version we actually measured.

How you get to it

Because the weights are open and published, there is more than one door to the same model.

Run it yourself, free

The open weights download from Hugging Face and run through the Diffusers library, ComfyUI, or the LTX-Video GitHub repository, on hardware you pay for by the hour or already own. We ran LTX-Video 0.9.8 13B distilled on a free-tier cloud GPU included in an existing subscription — no separate charge for the render itself.

A hosted service

LTX Studio, the browser app from the same company, offers a free tier with 800 one-time credits, though that free tier is licensed for personal use only, not business or paid work. Paid tiers start at $15 a month (the Lite tier); the commercial-use license is confirmed starting at the $35-a-month Standard tier and above. The model is also available through an API at $0.09 to $0.37 per second depending on resolution, and through third-party hosts like fal.ai and Replicate.

Why the multishot feature is worth knowing about

Most video models generate one continuous shot per request, which means a wide-medium-close sequence has to be assembled from several separate generations, each one a fresh roll of the dice on whether the subject still looks the same. LTX-2.5’s native multishot generates the whole connected sequence in one pass instead, holding the same character, environment, and voice steady across every cut within that one generation. It uses a text encoder with a built-in prompt enhancer, so a shorter prompt gets filled out with more of the detail a longer, more careful prompt would have specified by hand.

For an agent, that’s the difference between one clean sequence and stitching together several separate clips that don’t quite match.

What to use it for

The profile is free b-roll and background motion, not talking heads.

  • Ambient or scenic clips — motion in a setting, not a person speaking — generated free on a rented GPU.
  • A multi-shot sequence from one prompt, where LTX-2.5’s native multishot handles the wide-medium-close cut for you.
  • Testing a video idea on the free tier before deciding whether a paid render or a different model is worth the cost.
  • Any pipeline where you’re generating enough clips that a per-second API charge would add up, and running your own GPU session removes that charge entirely.

Where it is weaker

Two honest limits worth knowing before you build around it.

The smaller 0.9.8 model has no audio track at all — LTX-2.5 is the version built for synchronized audio and video together.

For a talking head specifically, an earlier free demo of LTX’s audio-to-video model produced lip movement that did not hold up on close inspection. This model’s strength is motion and scene, not a face speaking in sync — for that job, a dedicated voice-and-avatar tool is the better fit.

What it costs

The open weights themselves are free to use under Lightricks’ Community License for any organization under $10 million in annual revenue, with no per-generation billing when you self-host. Renting the GPU you run it on may have its own cost depending on the provider, though a free-tier GPU session, as we used, brings that to zero as well. Hosted options range from a limited free tier on LTX Studio, up through $15-$125-a-month paid tiers, to a per-second API charge of $0.09 to $0.37 for anyone who’d rather not manage the hardware.

Running it on modest hardware

The full model has options built for machines that don’t have a data-center GPU sitting behind them: an fp8 quantized mode that trades a little precision for a much smaller memory footprint, and CPU offload for a GPU that doesn’t have enough memory on its own to hold the whole model. Between those two options and the smaller distilled model, there’s a version of LTX that fits on hardware well short of what the flagship benchmark used.

Where this goes wrong

One habit to watch.

Using LTX Studio’s free tier for anything you’ll actually publish or sell. That tier’s own license restricts it to personal use, not business or marketing work — the free option for commercial use is running the open weights yourself, not the hosted free tier.

What we actually measured

We ran the smaller LTX-Video 0.9.8 13B distilled model on a free-tier cloud GPU included in an existing subscription. Setup — installing the software and downloading the model weights, about 47.6 gigabytes — took roughly two and a half minutes total. After that, each 5-second vertical clip at 512×896 resolution rendered in about 7.6 seconds. The motion held up: real drone-glide movement, steam, and leaves moving naturally, though the image was noticeably softer at that resolution than a sharper still.

That combination — a two-and-a-half-minute one-time setup, then under eight seconds per clip, at zero cost — is the actual case for trying this before paying for video generation anywhere else.

What to try in the first week

One exercise, sized to the free tier.

Generate a single short scenic or ambient clip on a free cloud GPU session, and time it start to finish — the setup, the download, and the render itself. Once you’ve watched that whole process once, you’ll know exactly what ‘free video’ actually costs in your own time, which is the number that decides whether this becomes a habit.

How it sits next to the others

Against a paid per-second video API, LTX’s advantage is the same advantage every open-weights model offers: self-hosted, the marginal cost of the next clip is zero, not a per-second charge that adds up across a season of content.

Against a model built specifically for talking-head or avatar video, LTX’s strength is scene and motion — ambient footage, multi-shot sequences, background video — rather than a face speaking a script in sync.

Questions agents actually type

Is LTX video free to use?

The open weights are free under Lightricks’ Community License for any organization under $10 million in annual revenue, with no per-generation charge when self-hosted. LTX Studio’s hosted free tier exists too, but its license restricts that tier to personal use, not business work.

How fast is LTX at generating video?

We measured the smaller LTX-Video 0.9.8 13B distilled model at about seven and a half seconds to generate a five-second clip on a free-tier cloud GPU. Lightricks’ own benchmark for the newer LTX-2.5 model reports a 10-second 720p clip in under seven seconds on dedicated hardware.

Can LTX make a talking-head video?

Not reliably. An earlier free demo of LTX’s audio-to-video model produced lip movement that didn’t hold up on close inspection. LTX’s strength is scene and motion — ambient clips, multi-shot sequences — rather than a face speaking a script in sync.

Where can I run LTX video?

Self-hosted, through Hugging Face weights, Diffusers, or ComfyUI, on your own or a rented GPU with no per-clip charge. Or hosted, through LTX Studio, an API priced per second, or third-party services like fal.ai and Replicate.

Can LTX generate a multi-shot video sequence in one pass?

Yes. LTX-2.5’s native multishot feature generates a connected wide-medium-close sequence in a single generation, holding the same character, environment, and voice consistent across every cut, rather than assembling separate clips by hand.

Does LTX video work on a smaller GPU?

The full model supports fp8 quantization and CPU offload for lower-memory hardware, and the smaller 13-billion-parameter distilled model is built specifically to run faster and lighter than the flagship, at a small cost in quality.

Book a 30-minute strategy call →

Season 1 is Claude Code for Real Estate Agents, filmed and on its own hub. All three seasons: AI for real estate agents.