Artificial intelligence startup Decart continues to stretch the boundaries of what’s possible with the release of its latest live video model, Lucy 2.5, introducing real-time streaming visual effects capabilities for the first time.
The startup claims that Lucy 2.5 is unique in its ability to generate AI VFX continuously, on the fly, and embed them into live videos as they’re being created. For the first time, media production firms and independent creators alike will be able to generate realistic fireballs, have characters breathe out smoke, burst into flames or even inflate and explode in a gory mess at the flick of a prompt.
What it’s promising is Hollywood movie-style effects, limited only by the imagination, and without the need for post-production processing.
Lucy 2.5 has the potential to unlock a real jump in creativity, but its usefulness doesn’t stop there. The model’s mastery of complex physics and intricate visual detail also provides a stronger footing for Decart to fulfil its broader ambitions in the physical AI arena. Lucy 2.5 doesn’t just deliver cinematic flair, but serves as a dual-purpose engine that can enable greater accuracy in its world model simulations used for robotics training.
What’s new in Lucy 2.5?
Decart has enhanced the Lucy model significantly since the release of version 2 just five months ago. The biggest change is the addition of realistic visual effects, with users able to insert explosions, fire, tsunamis and completely alter the appearance of on-screen characters, such as by giving them the appearance of a melting chocolate bar.
Creatives will be able to prompt special effects to their heart’s content, and they’ll look far more realistic than the effects created by earlier models, thanks to its improved physics engine and prompt adherence.
There’s also more precision. Decart says users are being given finer control over exactly how AI-generated elements are placed and behave when integrated into a live video. They will also be able to erase unwanted objects and characters in an instant, without any noticeable artefacts that give away the fact it has been edited. These capabilities all add up to a far more realistic-looking end result, and they’re already available through the Decart API.
Advertising and marketing is an obvious target for Decart, with Lucy 2.5 enabling these types of users to create personalized versions of promotional clips in real time.
Shopping experiences should be much wilder, too. Gone are the days of trying to imagine how you’ll look wearing a new overcoat or skimpy dress. Simply connect Lucy to your camera, then select an image of the product you want to see, and Lucy’s ultra-realistic physics engine will merge them into a new, high-definition video that shows exactly how it would look in the real world.
The implications for immersive ecommerce virtual try-on and even in-store magic mirrors are promising.
A leap forward for world models
The launch of Lucy 2.5 is just the latest milestone in the meteoric rise of Decart, which has emerged over the past two years as one of AI video generation‘s leading lights.
Over the last few months, all of the pieces have fallen into place for Decart to establish itself as a serious rival to the likes of Runway, Luma AI, Pika Labs and even Google. In December, Decart struck a deal with Amazon Web Services to leverage its Amazon Bedrock service and Trainium processors to offer Lucy as a plug-and-play product. In March, Comcast and Nvidia announced a partnership with Decart to build out a distributed AI grid based on Comcast’s nationwide edge architecture.
That infrastructure boost helped pave the way for the launch of Lucy 2 earlier this year, which was followed by a massive $300 million funding round that closed in May. Backed by Amazon’s VC arm, Nvidia, Adobe Ventures, Atreides Management, Sequoia Capital and others, Decart established itself not only as an AI video heavyweight, but also one of the hottest innovators in the related field of world models and video compute infrastructure.
World models are closely related to AI video, as they’re used to generate environments that can simulate physical, real-world environments, and Decart has made significant strides in this area too. Thanks to its powerful simulation engine Oasis 3, the company has started making real traction in robotics training, enabling “physical AI” developers to fine-tune the systems that can safely pilot autonomous robots and self-driving cars in the real world.
The launch of Lucy 2.5 promises to accelerate Decart’s world model efforts in the same way it’s going to expand creative video making. Lucy 2.5’s improved precision and visual effects have the potential to enhance physical AI simulations in some clever ways. For instance, developers might be able to test how an autonomous car performs in dense fog, when it can no longer rely on its camera views.
Alternatively, they can create even more extreme scenarios, teaching drones how to avoid tornados, or robots how to respond if there’s an explosion in the industrial facility they’re working in. By marrying Lucy 2.5’s capabilities with Oasis 3, Decart should be able to generate far more unpredictable, and therefore realistic, training simulations.
Get the TNW newsletter
Get the most important tech news in your inbox each week.