Big Tech giants like OpenAI and Google dominate the artificial intelligence industry, but Decart, a Silicon Valley-based unicorn startup, aims to establish itself as a player in real-time video generation models.
The company has announced a partnership with Amazon Web Services to offer scalable generative AI video solutions that outperform larger competitors with lower latency and superior performance. Decart’s novel model architecture, paired with AWS’s cutting-edge computing infrastructure, enables the delivery of unrivaled low-latency and high-throughput video generation.
Decart has agreed to run its most powerful video generation model, Lucy, on Amazon’s most advanced AI processors, specifically the custom-designed AWS Trainium3 processors. The company has been optimizing Lucy to run on AWS’s Trainium processor chips for months, starting with Trainium2 before moving to Trainium3.
Decart’s leadership chose Amazon’s infrastructure due to its belief that AWS can achieve better performance for video generation than Nvidia’s general-purpose GPUs. Trainium is a specialized processor customized for AI training and inference workloads, thanks to its systolic array design that enhances matrix operations.
Decart CEO Dean Leitersdorf told the Wall Street Journal that before the breakthrough, the company’s models could only generate video for 1.5 to 2 seconds before degrading. “The models would get too confused and everything would blur into oblivion…. Trainium really allowed us to run a bigger model which is smarter and doesn’t crash quickly.”
Decart’s partnership with Amazon is an example of how startups can compete with established players by targeting specific verticals. This partnership demonstrates that developing bespoke models for highly specific problems can disrupt the industry. Other AI startups, such as AiHello and SoundHound AI, have also found success by focusing on niche challenges.
Generating AI videos in real-time is a challenging task that requires a performance trade-off to achieve extremely low latencies. However, Decart claims that its streamlined model architecture and Amazon’s Trainium infrastructure allow it to avoid making this sacrifice. A recent Decart demo at Twitchcon showcased the potential of real-time AI video for everyday content creators.
The performance boost from Amazon enables Decart to establish real-time generative AI video as a distinct market segment, enabling new applications in industries such as gaming and social media. Lucy has already demonstrated impressive capabilities on AWS Trainium2, generating video outputs at up to 40 frames per second within milliseconds.
Decart is now working to optimize its models for Trainium3 processors, aiming to deliver real-time AI video at up to 100 FPS. Customers can access Lucy on AWS Trainium as of this week, with the model being distributed through the Amazon Bedrock platform.
Leitersdorf stated that generative AI video is one of the toughest challenges in AI due to its high computation demands and latency requirements. He added, “We’re making real-time video generation practical and cost-effective at scale. Together with AWS, we’re enabling customers to create and stream new forms of interactive content, from personalized live experiences to large-scale simulations, faster and more efficiently than ever before.”
