[

LTX open source

]

Open source, by necessity

We’re building world models that understand and simulate physical reality. Technology this foundational has to be built in the open where it can be inspected, improved, and trusted.

GitHub

HuggingFace

Documentation

~28M

Downloads on HuggingFace

22B

Parameters, open weights

1 GPU

Runs locally on a single RTX 5090

Our vision

Why LTX is open

Open source isn’t a feature. It’s how world models scale. The next generation of AI will run on your hardware, your data, and your world. No centralized model can understand every environment it needs to serve. 

Build in the open

Everything you need to start building with LTX.

Model weights

Inference & training code

Reference pipelines

Benchmarks & evals

How LTX works for you

Run on your terms

Your infrastructure. Your hardware. Complete control over deployment.

Adapt to your environment

Fine-tune for your environment. Models that understand your world.

Customize freely

Modify the full stack. Extend, adapt, and evolve.

Powered by community

Open architecture means faster iteration and shared improvements.

Academic Programs

Research

Join our programs

Whether you're just getting started or building the next generation of Physical AI, we're here to help.

HuggingFace

GitHub

Join the Discussion

Developer Program

Giving builders early access, technical resources, and collaboration opportunities.

Apply Now

Physical AI Program

Helping developers build for robotics, automation, manufacturing, and beyond.

Apply Now

Academic Collaboration

Supporting universities, research labs, and students advancing world models.

Learn More

[

Community Library

]

Made with LTX

Discover products, ComfyUI workflows, and LoRAs created by builders across the community. Built something with LTX? Submit it to the Community Library. Apply Now →

Relight

This IC-LoRA relights existing footage according to an input 3d sphere's lighting direction

huggingface

LTX Team

License:

LTX License

Compatibilty:

LTX-2.3

Foley

Generates synchronized Foley sound for silent video—footsteps, impacts, environments, machinery—while suppressing music and speech. Works across a wide range of footage, handles both subtle sounds (rain, typing) and intense effects (hammers, explosions),

huggingface

LTX Team

License:

LTX License

Compatibilty:

LTX-2.3

Clean Plate

This IC-LoRA cleanly removes people, along with the moving shadows they and their vehicles cast. It was trained across a range of weather conditions, camera angles, and lenses, so it performs reliably on a wide variety of real-world footage. In post-production with After Effects, you get before/after for masking - only specific objects disappear. Perfect for VFX artists who need selective background removal.

huggingface

LTX Team

License:

LTX License

Compatibilty:

LTX-2.3

Cinemagraph

Create animated cinemagraph-style videos with subtle, looping motion. Great for social media content and eye-catching marketing materials.

huggingface

LTX Team

License:

LTX License

Compatibilty:

LTX-2.3

LTX Desktop

LTX Desktop is an open-source desktop app for generating videos with LTX model: locally on supported Windows/Linux NVIDIA GPUs, with an API mode for unsupported hardware and macOS.

github

LTX Team

License:

Apache-2.0

Compatibilty:

LTX-2.3

WaterSimulation

Add water to any shot including rivers, surf, rain, floods and splashes while keeping the subject, framing, and camera move exactly as filmed.

huggingface

LTX Team

License:

LTX License

Compatibilty:

LTX-2.3

In-Outpainting

Extend the canvas or modify any region of a video with generated content that matches the original scene

huggingface

LTX Team

License:

LTX License

Compatibilty:

LTX-2.3

InstantShave

Remove beards, mustaches, and stubble from video while keeping the subject's identity, motion, and expression intact.

huggingface

LTX Team

License:

LTX License

Compatibilty:

LTX-2.3

Ingredients

Provide characters, props, locations, and visual references, and turn them into complete video scenes while preserving their visual identity.

huggingface

LTX Team

License:

LTX License

Compatibilty:

LTX-2.3

Decomposition

Remove compression artifacts including macroblocking, chroma bleed, ringing, and banding from low-bitrate video while preserving identity and framing.

huggingface

LTX Team

License:

LTX License

Compatibilty:

LTX-2.3

Deblurring

Restore sharpness to out-of-focus footage, feed the blurry clip as reference and get a sharp re-render with identity and framing preserved.

huggingface

LTX Team

License:

LTX License

Compatibilty:

LTX-2.3

DayToNight

Re-render any daytime video as a nighttime shot with matching composition, camera movement, and motion intact.

huggingface

LTX Team

License:

LTX License

Compatibilty:

LTX-2.3

CrossEyed

Transforms straight eyes in portrait video into crossed eyes — preserving expression, head movement, and framing.

huggingface

LTX Team

License:

LTX License

Compatibilty:

LTX-2.3

Colorization

Restore natural color to grayscale or monochrome footage

huggingface

LTX Team

License:

LTX License

Compatibilty:

LTX-2.3

KJ Node Pack

Utility node pack for ComfyUI — Set/Get nodes, model optimization, and workflow QoL tools.

github

Kijai

License:

GPL-3.0

Compatibilty:

No model dependency needed

Cockpit Cam LoRA

Generates open-wheel motorsport video from a T-cam/cockpit perspective.

huggingface

mxturbo

License:

Apache 2.0

Compatibilty:

LTX-2.3

ID-LoRA

Generate a person's visual appearance and voice together in a single pass from a reference image and audio clip.

huggingface

AviadDahan

License:

Compatibilty:

LTX-2.3

HDR AnyClip workflow

Batch-process any number of clips through LTX IC-LoRA HDR automatically — one click, walk away.

huggingface

d3nnyvg3org3

License:

Compatibilty:

LTX-2.3

JoyAI-Echo

Generate up to 5-minute multi-shot stories with consistent character identity and voice.

huggingface

jdopensource

License:

LTX Community License

Compatibilty:

LTX-2.3

Seed Hunter Multiroll Workflow

Generate and preview multiple seeds in one run, then pick the best one to finalize.

civitai

foxydits

License:

Compatibilty:

LTX-2.3

CRT LoRA

Renders video in a late-80s CRT monitor look — scanlines, phosphor glow, choppy motion.

huggingface

lovis93

License:

Apache 2.0

Compatibilty:

LTX-2.3

DramaBox

Prompt-driven TTS with voice cloning — control emotion, delivery, and sound effects through text.

github

manmay

License:

LTX Community License

Compatibilty:

LTX-2.3

Vertical Greenscreen IC LoRA

Generate talking avatar videos against a flat green chroma-key background with natural lip sync and subtle head movement.

huggingface

OmerHagage

License:

LTX Community License

Compatibilty:

LTX-2.3

FPV Motion LoRA

Adds smooth, low-speed FPV drone-style camera motion to LTX video generation.

huggingface

chsengni

License:

Apache 2.0

Compatibilty:

LTX-2.3

AV LoRA

Generate talking head videos with synchronized lip sync from a reference image, with internalized voice characteristics.

huggingface

elix3r

License:

OpenRAIL

Compatibilty:

LTX-2.3

V2V Inpainting

Mask and regenerate specific regions of an existing video.

huggingface

RuneXX

License:

Compatibilty:

LTX-2.3

Music Video Creator

Sequence multiple scenes with custom audio sync for full music video generation.

huggingface

RuneXX

License:

Compatibilty:

LTX-2.3

FFLF LoRAs

Generate video that starts and ends on exact frames — control both the opening and closing shot of your LTX-2.3 generation with custom audio support.

huggingface

RuneXX

License:

Compatibilty:

LTX-2.3

Inflated Balloon LoRA

Adds inflated balloon physics and look to objects in generated video.

huggingface

NebSH

License:

Compatibilty:

LTX-2.3

LTX-2 workflows

High-rank LoRA adapter that improves image-to-video quality and motion coherence without complex preprocessing workflows.

huggingface

MachineDelusions

License:

LTX Community License

Compatibilty:

LTX-2.3

HuMo integrated workflows

LTX-2.3 + HuMo audio-visual model for reference-audio-driven talking head generation.

github

ckinpdx

License:

Compatibilty:

LTX-2.3

IC-LoRA: Audio-Only

Feed only an audio reference and let it steer both the generated video and audio output.

huggingface

FredBliss

License:

LTX Community License

Compatibilty:

LTX-2.3

Obscura Remova LoRA

Remove foreground occlusions from video — curtains, plants, windows, haze — using a prompt-driven video-to-video LoRA.

huggingface

WepeNerd

License:

LTX Community License

Compatibilty:

LTX-2.3

Colorizer IC LoRA

Add color to black-and-white footage using a prompt-guided IC-LoRA workflow.

huggingface

Doctor Diffusion

License:

Apache 2.0

Compatibilty:

LTX-2.3

IC-LoRA-Cameraman

Transfer camera motion from a reference video into a new generated scene — pan, tilt, zoom, orbit, and compound moves.

huggingface

cseti

License:

Apache 2.0

Compatibilty:

LTX-2.3

Googly Eyes LoRA

Slaps oversized googly-eye stickers onto any subject in an LTX-2.3 generation. Trigger with googlyeyes.

huggingface

TheBurgstall

License:

Apache 2.0

Compatibilty:

LTX-2.3

BodyPositivity LoRA

Transforms subjects into a fuller body type — same scene, same motion. One trigger word.

huggingface

TheBurgstall

License:

Apache 2.0

Compatibilty:

LTX-2.3

Multi LoRA Loader

Stack multiple LoRAs in one node with per-layer LTX2 strength controls and drag-and-drop reordering.

github

phazei

License:

Apache 2.0

Compatibilty:

ST-DRC (Spacial-Temporal Decoupled Reference Conditioning) Workflow

Identity-preserving image-to-video using Spatial-Temporal Decoupled Reference Conditioning.

github

PineAmbassador

License:

MIT

Compatibilty:

LTX-2.3

Best Face Swap Video

Swap a face onto any guide video while preserving motion, expressions, and temporal consistency.

huggingface

alissonerdx

License:

LTX Community License

Compatibilty:

LTX-2.3

Fantasy Painterly LoRA

Renders generated video in a rich fantasy oil-painting style with magical lighting and brush textures.

huggingface

VRGameDevGirl

License:

Compatibilty:

LTX-2.3

LTX Director

Drag-and-drop timeline node for multi-segment LTX videos with image guides and audio.

github

WhatDreamsCost

License:

GPL-3.0

Compatibilty:

LTX-2.3

Deforum LoRA

Evolving, morphing animations in the style of early AI animation tools.

huggingface

S4f3ty_Marc

License:

Apache 2.0

Compatibilty:

LTX-2.3

TTM IC LoRA

Styled 3D transfer from flat 2D cutouts — brings still images into motion with a distinctive look.

huggingface

SIRAX

License:

Apache 2.0

Compatibilty:

LTX-2.3

OmniNFT

Improves joint audio-video generation with separate reward signals for each modality.

huggingface

zghhui

License:

Research only

Compatibilty:

LTX-2.3

ICEdit-Insight

IC-LoRA adapter family for restoration, HD enhancement, watermark removal, and subtitle removal.

huggingface

JoyFox

License:

Apache 2.0

Compatibilty:

LTX-2.3

Outpainting LoRA

Extend any video canvas — sides, top, bottom — with content consistent with the original scene.

huggingface

oumoumad

License:

LTX Community License

Compatibilty:

LTX-2.3

Edit Anything LoRA

Mask any region of a video and regenerate it from a prompt.

huggingface

civitai

alissonerdx

License:

Apache 2.0

Compatibilty:

LTX-2.3

FLW IC-LoRA

Insert gray frames between clips and let LTX generate a seamless transition between them.

huggingface

SYSTMS

License:

Apache 2.0

Compatibilty:

LTX-2.3

360° Outpainting IC-LoRA

Turns widescreen footage into a full 360° equirectangular projection for VR viewing.

huggingface

TheBurgstall

License:

Apache 2.0

Compatibilty:

LTX-2.3

LTX ATI (Any Trajectory Instruction)

Instruction IC-LoRADraw motion paths on an image and animate it along those paths. Expect trippy results.

huggingface

Yuvraj Seegolam

License:

Apache 2.0

Compatibilty:

LTX-2.3

VBVR LoRA

Fine-tuned on 1M reasoning videos to improve multi-object motion, physical interactions, and prompt accuracy.

huggingface

LiconStudio

License:

LTX Community License

Compatibilty:

LTX-2.3

Upscale IC-LoRA

Bicubic upscale followed by an LTX second pass for sharper, more detailed output.

huggingface

Zlikwid

License:

Compatibilty:

LTX-2.3

FAQs

What is LTX-2.3 Open Source?

LTX-2.3 is a 22B-parameter video generation model with fully open weights, built for commercial teams that need complete control over their infrastructure. Fine-tune on proprietary IP, run fully on-premise, and integrate directly into existing pipelines.

How does LTX-2.3 generate videos?

LTX-2.3 uses a diffusion-based transformer architecture to generate video content. This approach allows the model to learn temporal structure across frames while producing coherent, high-quality video outputs suitable for research and experimentation.

Can I use LTX-2.3 for text-to-video generation?

Yes. LTX-2.3 supports text-to-video generation workflows, allowing enterprise teams to explore how textual prompts can be translated into generated video sequences using an open source model.

Does LTX-2.3 support image-to-video generation?

Yes. LTX-2.3 can be used for image-to-video generation research, enabling experimentation with transforming still images into dynamic video outputs using open source AI techniques.

Where can I access the LTX-2.3 open source model?

LTX-2.3 weights and code are available on GitHub and Hugging Face. There is also licensing opportunities for those that require. Ready to deploy, fine-tune, or integrate into your existing infrastructure today.

Is LTX-2.3 suitable for universities and students?

Yes. LTX-2.3 is designed with students, universities, and researchers in mind. Its open source release makes it suitable for coursework, academic research, benchmarking, and hands-on experimentation with AI transformer models for video generation.

Is LTX-2.3 an open source video generator?

LTX-2.3 is free and open source to use under its license. Users can access the model, code, and resources without cost, making it suitable for both commercial and non-commercial use.

Why is LTX-2.3 considered the best open source AI video model?

LTX-2.3 gives commercial teams something closed models can't: full access to weights, training code, and inference pipelines so you can fine-tune on proprietary IP, deploy on your own infrastructure, and build production workflows without cloud dependency or data exposure. Text-to-video and image-to-video out of the box, with a diffusion-based architecture built to be extended.