close
close
Light tricks strive for his deepseek moment with open source AI video model

Light tricks strive for his deepseek moment with open source AI video model

by

Israeli unicorn -lighth tricks makes a new advance to bring Israel to the global generative AI card. The best known for its viral image editing apps and once with an annual turnover edge of over $ 200 million, the company is now converting to the race in order to create advanced AI video models, with which high-quality video content can be generated from simple text meetings.

The new model named LTX Video-13b contains 13 billion open source parameters and claims to generate high-quality videos at a fraction of the costs of competitors such as Openais Sora, Google and other ladders in the generative video area. According to Light tricks, its model can achieve results that are comparable to first -class systems and at the same time train for only tens of million dollars. The video rendering with the model costs only a few cents per clip -possibly a big breakthrough for affordability in the AI ​​video.

According to Light tricks, its model is also the only open source video model published by a western company, since other open models that are currently available come from China. The company hopes that this publication will mark its own “Deepseek moment” -a breakthrough that increases both its profile and Israel in the generative AI landscape.

Although the AI ​​video -video is still a smaller market than large language models (LLMS), it is subject to a quick disturbance. Lightricks estimates the commercial potential of this emerging sector at 600 billion US dollars, especially since the industries are increasingly generative tools from digital marketing to film production. A main shift is the ability to create video ads without physical sets, actors or expensive post -production.

“Our goal in this publication is to promote acceptance among academic researchers and startups,” said Shaul Meridor from Lightricks' CFO. “Yes, we compete with the Giants, but we have previously proven that we can create extraordinary products. Models are important, but the real influence is to integrate them into user-friendly tools-and that is exactly what we do.”

The LTX Video-13b model creates videos about a multi-stage process. It begins with a rough sketch to identify movement and composition with minimal arithmetic resources. This sketch then forms the basis for later layers that add sharper movement and lighting effects – similar to progressive levels of illustration. The rendering speed of the system is said to be 30 -faster than that of competing models.

To train his model, Light tricks worked with the media giant Shutterstock and Getty Images to gain access to their extensive content libraries. This publication follows the company's earlier entry into the video in the end of 2024, when a product performed that could create clips for five seconds.

Light tricks were founded in 2013 and was one of the first unicorns of Israel. Despite several rounds of discharge in recent years, it remains an important player in KE-controlled creative tools and currently employs around 500 people.

Leave a Comment