by Lightricks
Open-weights Lightricks video model with native audio, multi-shot generation, auto duration and up to 4K output.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
LTX 2.5 is the latest version of Lightricks' open-weights LTX video generation model. It builds on LTX 2.3 with a new diffusion video decoder ("Diffusion Fidelity Rendering"), a Gemma 4 12B text encoder with a prompt enhancer, automatic clip-duration prediction, and an improved distilled model trained on a larger dataset with reinforcement learning.
It supports text-to-video, image-to-video, video-to-video and audio-linked generation (including joint audio-video output), native multi-shot scenes that keep character, environment and voice consistent, and post-production tools such as refine, restore, SDR-to-HDR, 16-bit EXR input/output, layout-to-render and alpha matte generation. Creators report generation in tens of seconds locally in ComfyUI, with GGUF and int8 variants (about 22 GB for int8) and compatibility with LTX 2 LoRAs. Lightricks lists a 16 GB minimum VRAM. One reviewer found its quality generally behind MiniMax H3, with camera orbit as an exception.
Weights can be downloaded from Hugging Face and fine-tuned on custom data, with community IC-LoRAs for tasks like outpainting, face swap and lip-synced avatars. The model can also be used through LTX Studio, which has Free, Lite, Standard, Pro and Enterprise plans on a credit system; the free plan is for personal, non-commercial use. The Hugging Face license is listed as "other", so the terms should be checked before commercial use.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Open-source video generator that is an upgrade over LTX 2.3, with diffusion fidelity rendering, multi-shot generation, up to 4K and 50 fps, and up to 20 seconds. The creator installs it in ComfyUI and demos text-to-video, image-to-video, first/last frame, LoRAs and GGUF; generations took about 20-30 seconds, 2-3x faster than MiniMax H3.
Open-source video model with native audio and multi-shot, the fastest open video model. Tested against MiniMax H3 on several prompts and mostly lost on quality, winning only on camera orbit. Int8 version is 22 GB and works with LTX 2 LoRAs.
Open-weight video model that makes a 10-second clip from an image in about 6.8 seconds on NVIDIA superchips. It runs locally on RTX GPUs.
Closest neighbours by meaning, not just by category.
Library · Lightricks
Official Lightricks package for training and fine-tuning LTX-2 video models, including LoRAs for characters, effects and styles.
Model
Community GGUF-quantized builds of the LTX 2.5 video model for running on low-VRAM GPUs, e.g. in ComfyUI.
Model · MiniMax
Open-weight video generation model from MiniMax, part of its H3 family alongside H3 Image.
Model · Sand AI
Open-weights 114B MoE model from Sand AI that generates synchronized video and audio from text or image prompts.
Model · SenseTime
Open-weights 27B model from SenseTime that creates short videos by generating and rendering executable code.
Model · Nvidia
NVIDIA's 5B and 14B hybrid linear-attention video diffusion transformers generating up to 720p on a single GPU.
