While some AI-generated videos can be impressive, most people will agree that they’re still very far from being the finished article. As realistic as some of the visuals can be, most models still lack significant functionality in terms of being able to generate special effects, make edits while livestreaming, and swap elements in and out at the speed of thought. With the recent release of Decart’s newest video model, Lucy 2.5, there’s a new baseline for what AI video creators can come up with. By making Lucy 2.5 available through its API, Decart has just delivered the first easily accessible model of its kind that can implement hyper-realistic VFX visuals in real-time, even as the video is being livestreamed, with minimal latency.
In addition, the company has introduced a number of new editing tricks for creators to play around with and adapt their videos on the fly, simply by entering text-based prompts.
Immersive Discovery That Curbs Buyer’s Remorse
Decart says Lucy 2.5 will enable dozens of previously impossible new applications, but few are as compelling as the virtual dressing room. The company says it’s the first in the world to close the “imagination gap,” which has persisted since the days of the first mail order catalogs.
The imagination gap is where people struggle to properly visualize how a new garment might look when they slip themselves into it, and it’s something most have experienced at some point. You see a new pair of sneakers or a shirt that looks really cool, you go ahead and order it, but when it arrives on your doorstep a few days later, and you try it on, it’s a massive disappointment.
Lucy 2.5’s major promise is to prevent this from happening again. It’s the first model of its kind to be able to dynamically insert products such as a new pair of jeans or accessories such as shades or a new bag directly onto a character within a video. And of course, that character can be the actual viewer. By enabling their webcam to sync with a shopping app, they’ll be able to visualize exactly how a particular dress or scarf looks when it’s draped around their body.
The Secret Sauce
The secret sauce here is the way Lucy 2.5 precisely maps each item onto the body. The model’s immense processing power enables it to get everything from the physics to the drape of the fabric just right, so someone can virtually try on almost any kind of garment and see how it looks in different scenarios.
While the model doesn’t yet have the capability to map a garment’s size and cut to each individual user’s dimensions, it does empower people to visualize how things might look, creating a more immersive shopping experience.
One of the most astonishing aspects of Lucy 2.5 is its real-time speed, which is possible thanks to Decart’s compute infrastructure expertise, offered to other enterprises as DOS (Decart Optimization Stack), an inference service that reduces the hardware constraints on high-resolution, low-latency AI workloads.
With Lucy, a shopper could potentially try out multiple luxury handbags, one after another, to see which one looks best. The physics engine will ensure that each bag sways perfectly in tandem with their natural stride – so it looks as if the person is genuinely holding a real bag.
Social media influencers could make use of this too. One possibility is the idea of dynamic advertising campaigns, where they can instantly swap in and out different products during a livestream broadcast.
Hollywood-Grade Aesthetics
Lucy 2.5 isn’t just for shopping apps and influencers, though. With the new live AI VFX feature, it’s now capable of generating high-quality video clips with Hollywood-esque special effects, such as fireballs, smoke, explosions, perhaps even a tsunami, without having to spend a single dime on post-production.
In traditional video creation, this kind of computer-generated imagery can only be done by professional editors who charge hundreds or even thousands of dollars for their services. But Lucy 2.5 can conjure up the same ultra-convincing visuals from a single prompt.
A TikTok influencer can whip up a cloud of smoke before teleporting themselves to an entirely different location, live. Decart says it has significantly enhanced Lucy 2.5’s precision and prompt adherence to ensure that these effects don’t just sit on the video frame with the usual awkwardness found in earlier AI generators, but instead interact with the environment and its characters in a much more realistic way.
Decart says it’s trying to usher in a revolution of visual effects in AI video, and it’s backing up this evolution with support for new kinds of prompts and dynamic settings. For instance, Lucy 2.5 gains the ability to seamlessly adapt attributes such as the texture and color of characters and objects, swap people in and out of videos at will, and apply sweeping changes to the background with minimal ghosting problems.
