LTX-Video
Legacy v0.9.x LTX video models: text-to-video, image-to-video, keyframes and extension. Superseded by LTX-2.5.
The v0.9.x models. The README points new work to LTX-2.5 and says this repo is no longer active.
LTX-Video is the original LTX video generation model line (v0.9.x). The README now says LTX-2.5 supersedes it and this repository is no longer under active development, so it suits people maintaining existing workflows, not new projects.
What it is
The repository provides inference code and configs for the v0.9.x models, including 13B and 2B variants with distilled and FP8 versions, for text-to-video, image-to-video, multi-keyframe conditioning, video extension and video-to-video. ComfyUI and Diffusers integrations, plus IC-LoRA control models for depth, pose and canny, are listed.
Who it's for
It suits technical video creators and researchers who need to reproduce published results or keep an existing LTX-Video pipeline running. If you are starting fresh, the README recommends LTX-2.5 in a separate repository.
Key features
Text and image to video
Generate clips from prompts, images or multiple keyframes.
Video extension
Extend a clip forward or backward from existing segments.
Distilled and FP8 models
Smaller and quantized variants trade some quality for speed and lower VRAM.
IC-LoRA control models
Depth, pose and canny control models guide generation.
ComfyUI and Diffusers
Use the models through ComfyUI workflows or the Diffusers library.
Up to 60 seconds
v0.9.8 added longer shot generation up to 60 seconds in the 13B model.
Up and running in 3 steps
Our honest verdict
What's great
- Apache-2.0 code with a range of model sizes, from 2B distilled to 13B.
- A ComfyUI integration offers a graphical workflow for people who avoid scripts.
- Hosted options named in the README (LTX-Studio, Fal.ai, Replicate) let you try it before installing.
- Quantized FP8 variants and CPU offloading help on smaller GPUs.
Where it falls short
- The README says this repository is no longer under active development and recommends LTX-2.5 for all new projects.
- Local use needs Python, CUDA-capable PyTorch and a capable GPU; the README notes the best results come through ComfyUI, not inference.py.
- Model weights are licensed separately (the v0.9.5 note mentions OpenRail-M), so check commercial terms.
- The repo has no tagged releases on GitHub; versions are distributed as Hugging Face checkpoints.
Pricing & license
The code is free under Apache-2.0 and the older v0.9.x models can be run yourself; the README links hosted options (LTX-Studio, Fal.ai, Replicate) without prices. Model weights have separate licences, and commercial licensing details are at https://ltx.io/model/license.
Open remote AI-training roles
870 open remote AI-training roles, updated November 14, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Should I start a new project on LTX-Video?
No. The README says LTX-2.5 supersedes it and recommends LTX-2.5 for all new projects, with code at github.com/Lightricks/LTX-2.
Can I use the output commercially?
The code is Apache-2.0, but the README says model weights are licensed separately and points to ltx.io/model/license for commercial licensing.
Is there a hosted way to try it?
The README links LTX-Studio, Fal.ai and Replicate.
