Lightricks’ LTXV: The first open source AI video generation model, powered by Google Cloud
50% faster time-to-market
50% savings in person hours during the development process
Billions of data parameters processed in seconds
Able to generate 5-second videos in 4 seconds or less
Lightricks’ LTXV, an AI video generation model, leverages Google Cloud to enable users to create compelling visuals in seconds. Google Cloud provides the infrastructural and computational support that powers LTXV’s first-to-market, open source model.
Lightricks’ LTXV, an AI video generation model, leverages Google Cloud to enable users to create compelling visuals in seconds. Google Cloud provides the infrastructural and computational support that powers LTXV’s first-to-market, open source model.
Democratizing AI powered text-to-video creation, using an open source model
Generative AI tools that allow users to create videos with text prompts have widely been embraced by content creators and marketers.
In fact, the global generative AI in video creation market is projected to reach USD 2,298.8 million by 2033, driven by increasing demand for personalized and engaging video content across various sectors.
But oftentimes, AI video generation solutions are prohibitively expensive, requiring costly subscriptions that render them out-of-reach for the average person.
Lightricks, the market leader in video generation and photo editing software tools, envisions a world in which AI innovation is accessible to a wider audience than ever before.
With more than 15 million monthly users and over 730 million app downloads, Lightricks is the leading provider of video and photo solutions on the market.
Creating LTXV, our AI-powered text-to-video model, required the world’s best data storage, processing, and infrastructure solutions. Google Cloud was the clear choice, thanks to their unmatched expertise in data engineering and efficient environment for building advanced models. The resources provided by Google Cloud are ideal for innovation and exploration.
Ofir Bibi
VP Research, Lightricks
The company began developing an open source model, LTXV, with the goal of empowering users, whether creators on social media or professional marketers, to create videos from text prompts in a matter of seconds.
Lightricks needed to secure extensive computational power, processing for more than one billion parameters, and intense infrastructural support in order to create the model.
Additionally, in order to reinforce the company’s position as an industry powerhouse, Lightrick needed LTXV to be the first solution of its kind to reach the market.
"Creating LTXV, our generative AI-powered text-to-video model, required the world’s best data storage, processing, and infrastructure solutions. Google Cloud was the clear choice, thanks to their unmatched expertise in data engineering and efficient environment for building advanced models. The resources provided by Google Cloud are ideal for innovation and exploration," says Ofir Bibi, VP Research at Lightricks.
"Google Cloud is the driving force behind LTXV, helping ensure that more users have access to the power of AI innovation."
Lightning-fast processing times, revolutionizing video creation
LTXV revolutionizes AI video generation, enabling users to write text prompts that create compelling visual imagery. The model slashes video creation times by over 90%, setting a new industry standard. Its innovative architecture rivals larger, resource-intensive models, achieving stunning visual output with a fraction of the computational load.
The model’s real-time processing speed is unparalleled in the market, generating a 5-second video in 4 seconds or less. This speed empowers creators to produce content faster than ever, while maintaining exceptional quality. LTXV also supports scalable long-form video production, consistently delivering extended, high-quality videos with enhanced flexibility and control.
The solution tackles common challenges in video generation, including motion and structural inconsistencies. With its unique frame-to-frame learning system, LTXV ensures seamless transitions between frames, eliminating flickering and maintaining coherence throughout scenes.
Google Cloud’s TPU was instrumental for the creation and success of LTXV, Bibi emphasizes. TPU’s immense memory capacity enabled Lightricks to feed multiple video frames into a single chip, allowing the model to be trained on longer videos, which significantly enhanced motion consistency.
"We achieved our goals of launching on time and being first to market, in large part thanks to TPU enabling us to innovate faster. The solution gave us room to grow and experiment within Google’s limits, turning a six-month project into four, and accelerating our time-to-market by 50%," Bibi says.
We achieved our goals of launching on time and being first to market, in large part thanks to TPU enabling us to innovate faster. The solution gave us room to grow and experiment within Google’s limits, turning a six-month project into four, and accelerating our time-to-market by 50%.
Ofir Bibi
VP Research, Lightricks
TPU's parallel processing capabilities ensured that LTXV unlocked incredibly fast generation speeds. Using Trillium, LTXV generated four 5-second videos in just 9 seconds. With TPU’s four cores working in parallel, LTXV achieved the same result, marking a major leap in efficiency and performance.
"To train the LTXV model, we used a large-scale setup involving 2 billion parameters and hundreds of millions of videos, requiring massive data collection, filtering, and processing. Trillium played a vital role here, providing a cost-effective platform for processing vast amounts of data," he adds.
Bibi highlights that TPU and Trillium are more cost-effective than other options. Trillium, used for inference, offered efficient video generation at scale, while TPU supported both training and inference, allowing us to leverage its computational power without compromising affordability.
"Additionally, Google Cloud Storage became a critical component of our infrastructure. Its high-speed connections to training machines allowed us to handle massive datasets more efficiently than using traditional hard drives. This setup not only accelerated training but also ensured smooth and reliable data handling at scale."
Stellar infrastructural and technical support
Lightricks relies on Google Cloud’s managed solutions and works hand in hand with the Google Cloud team, which allows it to relieve its developers from management burdens and focus on achieving their goals, rather than being bogged down by tedious, time-consuming tasks.
"Google Cloud handled DevOps for us, letting us focus on core engineering. Free from distractions like cluster management or admin tasks, we were able to operate with just 50% of the team we’d typically need at the start. This streamlined our process and allowed us to prioritize what truly mattered," says Yoav HaCohen, Generative AI Research Team Lead at Lightricks.
HaCohen stressed that hands-on, round-the-clock support from Google Cloud’s engineers are incredibly important for LTXV’s success and ongoing improvements.
"Our close relationship with Google Cloud’s engineers has proven to be an invaluable asset for us. They’ve invested time and resources in us, and are always available to provide the support we need to continue innovating and developing according to our scalable, data-first strategy."
Google Cloud handled DevOps for us, letting us focus on core engineering. Free from distractions like cluster management or admin tasks, we were able to operate with just 50% of the team we’d typically need at the start. This streamlined our process and allowed us to prioritize what truly mattered.
Yoav HaCohen
Generative AI Research Team Lead, Lightricks
Open source is the best approach for building foundational models, as it empowers innovation through community collaboration and learning, says HaCohen. This benefits Lightricks’ brand, legitimacy, and developers by fostering growth through shared insights and experiences.
By learning from the mistakes and successes of others, innovators can leverage collective knowledge to build better products, he says. When others interact with LTXV, they may provide valuable feedback or build upon it, creating an ongoing dialogue to improve the model.
This openness cultivates a thriving community, where contributions are shared, ideas are exchanged, and the model evolves through collective effort and innovation. HaCohen emphasized that Lightricks’ is proud of releasing LTXV as an open source model, and that Google Cloud’s support is a major force powering ongoing innovation in AI video generation.
Lightricks is the leading, AI-first company providing innovative photo editing and video generation software to consumers, content creators, and businesses. With more than 15 million monthly users and 730 million downloads, Lightricks offers many of the most popular photo and video apps on the market. LTXV by Lightricks is the first real-time AI video generation open source model.
Industry: Media & Entertainment, Technology
Location: Israel
Products: Google Cloud TPUs, TPUv5, Trillium
