LTX-Video
Real-time video generation model by Lightricks using video DiT.
About
LTX-Video is Lightricks' open video generation line built on a Diffusion Transformer, notable for pushing generation speed: the distilled variants need as few as 8 diffusion steps and can run faster than real time on an H100 GPU. The repository covers text-to-video, image-to-video, multi-keyframe conditioning, and forward or backward video extension, and the LTX-2 release adds synchronized audio generation with output up to roughly 10 seconds, 50 frames per second, and native 4K resolution. Model options span a 13B full model, 13B and 2B distilled versions, and FP8 quantized variants for lower VRAM, with macOS supported through MPS. Weights ship under the OpenRail-M license, permitting commercial use, and the ecosystem includes official ComfyUI workflows, Hugging Face Diffusers support, and community tools like LTX-VideoQ8. It attracts creators and developers who want controllable local video generation without datacenter-scale hardware.
Should you use LTX-Video?
Pick it when
Pick LTX-Video when iteration speed matters: distilled checkpoints need as few as 8 steps, the 2B and FP8 variants suit mid-range GPUs or macOS via MPS, and multi-keyframe conditioning plus forward or backward extension give you control.
Look elsewhere when
Skip it if you need one clear license: early 2B checkpoints use OpenRail-M while the 13B and later releases use Lightricks' own open weights license, so check the file you deploy. For new audio-video work, start with LTX-2.5.
Alternatives to LTX-Video
- LTX-2.5
The current 22B generation with synchronized audio and multishot consistency, heavier to run and revenue-gated above 10 million dollars.
- Wan 2.1
Apache-2.0 across sizes with image-to-video, first-last-frame, and VACE editing, simpler to license but without LTX's few-step distilled speed.
- FastVideo
Brings few-step distilled speed to Wan models through FastWan checkpoints under Apache-2.0, if you want LTX-like speed on a different base.
- WanGP
Runs LTX-Video alongside Wan and Hunyuan on 6 GB cards through a GUI with queues, if you want the model without scripting it yourself.
Reviews (0)
Leave a Review
No reviews yet. Be the first to review!
Details
- Category
- Video Generation
- Price
- Free
- Platform
- Local/Desktop
- Difficulty
- Intermediate (3/5)
- License
- LTXV Community
- Minimum VRAM
- 12 GB
- Added
- Apr 3, 2026