Decart, the Israeli real time generative AI startup backed by Sequoia Capital and Benchmark, is advancing a new class of generative video technology built for continuous use. This week the company unveiled Lucy 2, a real time world model that runs live at 1080p and 30 frames per second with no buffering, resets, or predefined time limits, a technical threshold that has so far remained out of reach for most video generation systems.
Lucy 2 is designed to operate as a single continuous system rather than producing discrete clips. Motion, identity, lighting, and physical presence are generated frame by frame while a camera remains live, preserving full body movement and timing at near zero latency. The system also sharply reduces the cost of real time generation, bringing sustained operation down from hundreds of dollars per hour to roughly three dollars an hour, a shift that makes always-on use economically viable for the first time.
“This is the GPT 3 moment for world models,” said Dean Leitersdorf, Decart’s co- founder and CEO. “For the first time, a world model runs live, in real time, with no quality compromises. That shift doesn’t just improve video, it creates entirely new markets.”
Dean Leitersdorf, Decart’s co-founder and CEO.
Decart
Those markets span live entertainment, streaming, virtual try on, gaming, and robotics. Decart has already been testing Lucy based systems with Twitch creators, where latency tolerance is effectively zero. At TwitchCon last year, the company demonstrated real time character swaps, wardrobe changes, and environmental transformations that responded instantly to a streamer’s movement and audience prompts, all running inside standard live production workflows.
“They don’t storyboard,” Leitersdorf said of Twitch creators during that demonstration. “They improvise. Our AI needs to be just as fast.”
What distinguishes Lucy 2 from earlier experiments in generative video is persistence. Traditional video models generate short segments and rely on post production to correct errors. Lucy 2 remains live indefinitely, maintaining consistent anatomy, clothing behavior, lighting, and object interaction in environments where mistakes cannot be edited out later. That requirement for continuity is expanding interest beyond media and entertainment.
During a recent interview, Leitersdorf demonstrated Lucy 2 integrated into Nvidia’s Omniverse simulation tools. In those environments, the model dynamically altered lighting, textures, and visibility inside robotic training simulations. By degrading simulations in real time, Decart is working to narrow the long standing gap between clean virtual training data and the unpredictable conditions robots face in physical environments.
“Robots today are trained in simulations that are too perfect,” Leitersdorf said. “We can mess it up, add smoke, change materials, turn off the lights, and suddenly the robot learns to operate in conditions that look much closer to reality.”
That system-level emphasis has been central to Decart’s appeal to investors. Founded in 2023, the company has raised more than $150 million and is valued at over $3 billion, according to people familiar with the financing. Its investors include Sequoia Capital, Benchmark, and Aleph, and the company runs across Nvidia GPUs and Amazon’s Trainium infrastructure, making it one of the first generative video platforms optimized across multiple compute architectures.
Investors view efficiency as decisive in this category. Real time video generation has historically been constrained by cost and compute intensity, limiting deployments to short demonstrations. By reducing operating costs by two orders of magnitude, Decart is positioning Lucy 2 as infrastructure rather than a novelty. One investor involved in the company described the system as “one of the first generative video platforms that can scale without collapsing under its own compute bill.”
Leitersdorf remains cautious about predicting which applications will mature first. “I’m confident there are at least ten markets that this will change,” he said. “What I don’t know is whether next year it’s robotics, ecommerce, or Twitch that takes off first.”
Decart enters a competitive generative video landscape that includes startups and large technology companies pursuing different approaches. Platforms such as Runway focus on cinematic quality and creator workflows for offline production, while Google DeepMind’s Veo models emphasize longer, high fidelity video clips with synchronized audio. Enterprise focused tools like Synthesia target scripted avatar video, and Chinese players such as ByteDance’s Seedance are advancing low cost, high realism models at scale. Analysts typically distinguish these systems by latency, persistence, and cost, dimensions that place Decart’s real time world models in a distinct segment of the market.
With Lucy 2, Decart is moving generative video out of offline experimentation and into systems that respond continuously to people and machines. Whether the largest impact emerges in entertainment, commerce, or embodied AI, the company is betting that real time operation, rather than cinematic polish alone, will shape the next phase of generative video.