Draft page for internal preview — source-backed claims and links must be reviewed before publishing.

Open Source
Text to Video
Image to Video
Low VRAM

LTX-Video: Lightricks' Lightweight Open Video Model

Lightricks’ open video model line, useful when you want local T2V/I2V without a giant GPU budget.

Last updated 2025-12-12·6 min read·By Editorial Team

Quick answer

LTX-Video is Lightricks' open video model family, published on HuggingFace.

The distilled builds are the practical local pick when speed, low VRAM and cheap iteration matter more than peak cinematic quality.

Self-hosting has no per-video model fee; the real cost is local or rented compute.

Run locally

One of the cheapest open models to test locally.

Local costs

Download weights

Official Lightricks HuggingFace model cards.

HuggingFace

Hosted providers

Use hosted inference when you need an API around open weights.

Cost paths

01LTX-Video standard vs distilled builds

LTX-Video’s main advantage is practical local iteration. Smaller/distilled builds trade peak quality for lower hardware cost.

LTX-VideoDistilled builds
PositioningMain open model familyLower-cost local variants
WeightsOpen · HuggingFaceOpen · HuggingFace
Model fee$0$0
Best forQuality experimentsFast low-VRAM iteration

02Generation modes

LTX-Video is one of the cheapest open video models to run locally.

Text to Video

T2V

Generate video from prompt-only descriptions.

text → video

Image to Video

I2V

Animate still images with prompt steering.

image + prompt → video

Distilled local builds

Distilled

Distilled variants trade peak quality for speed and lower hardware requirements.

small model → faster iteration

03Run locally

LTX-Video is a good first local model to test because smaller builds reduce setup cost and wait time. HuggingFace GPU examples start at $0.40/hour.

VRAM by GPU

SetupConcrete costBest forNotes
Consumer GPU$0 model fee + your hardwareLow-cost local testingSmaller builds are the main appeal
Laptop/workstation$0 model fee + slower runsExperimentsDepends heavily on memory and acceleration
Cloud GPUHF examples: T4 $0.40/hr, L4 $0.80/hr, A10G $1.00/hrBatch generationOften cheaper than renting huge GPUs for open models

04Official pricing & registration

LTX-Video has no per-video model fee because the weights are open. To make the price concrete, use compute pricing: HuggingFace Spaces lists CPU Basic as free, Nvidia T4 small at $0.40/hour, Nvidia L4 at $0.80/hour and Nvidia A10G small at $1.00/hour.

Path / hardwareOfficial priceWhat you getSource
RegisterFree accountDownload/access model repositoriesOfficial source ↗
Model weights$0 model feeLightricks/LTX-VideoHuggingFace model card
HF Spaces CPU BasicFREE2 vCPU / 16 GB CPU SpaceOfficial source ↗
HF Spaces Nvidia T4 small$0.40/hour16 GB VRAM GPU SpaceOfficial source ↗
HF Spaces Nvidia L4$0.80/hour24 GB VRAM GPU SpaceOfficial source ↗
HF Spaces Nvidia A10G small$1.00/hour24 GB VRAM GPU SpaceOfficial source ↗

05How to prompt

Formula: Subject + Action + Short duration + Camera + Style
Example:
A paper boat floats through a tiny rain gutter, water ripples around it, close-up tracking shot, 5 seconds, charming miniature cinematic style
Open LTX-Video model card

06vs the competition

LTXLTX-Video
vs
WanWan 2.2

Short version: LTX-Video is better for low-VRAM local iteration. Wan 2.2 is better when you can afford more compute and want broader open video workflows.

Read full comparison →

07FAQ

Is LTX-Video open source?

Yes — Lightricks publishes the weights on HuggingFace. Check the model card for license terms.

Is LTX-Video good for low-VRAM GPUs?

Yes — the 2B distilled line is designed to run on consumer hardware; exact requirements vary by provider.

How much does LTX-Video cost?

The weights are free to self-host. The practical cost is your GPU, cloud compute, and any hosted provider you use.

Where do I register for LTX-Video?

For open-weight downloads, create a HuggingFace account at https://huggingface.co/join and use the Lightricks/LTX-Video model pages.

08Sources

Was this helpful? · Report an error · Suggest an updateUpdated 2025-12-12