LTX-2.5 AI Video Model Generates 10-Second Clip in 6.8 Seconds
LTX has released LTX-2.5, an AI model capable of generating a 10-second video from an image in just 6.8 seconds on Nvidia superchips. The model is available as open weights.
LTX, an open world model company spun out of Lightricks, has launched LTX-2.5, the latest version of its open-weights video and "world" model. The model can generate a 10-second, 720p AI video from a static image in 6.8 seconds when utilizing Nvidia's high-end superchips. This new version is natively integrated into ComfyUI, a popular node-based workflow tool for generative media prototyping.
LTX-2.5 is available as open weights on Hugging Face and within ComfyUI. For teams requiring managed generation, the LTX API is also offered. The model is free for organizations with annual recurring revenue under $10 million, with larger companies negotiating licensing terms. LTX reports that its models have accumulated over 33 million downloads, positioning the LTX family as the most widely used "open world" model line on the market.
Significant upgrades have been made to the model, including a new diffusion video decoder designed to reduce visual artifacts and enhance the reproduction of details like text and faces. LTX-2.5 also supports native multishot generation, ensuring consistency in character and scene across multiple clips. Additionally, the model features an improved Gemma 4 language backbone and a dedicated prompt enhancer for more accurate handling of complex instructions.
LTX emphasizes the model's efficiency and quality. According to the company, LTX-2.5 produces video at approximately one-seventh the cost and time compared to competing models. The model is capable of running on various hardware configurations, from data center GPUs to standard Mac computers. LTX is focusing on open models as an alternative to closed API-based solutions dominating the market.