Decart has unveiled Lucy 2.0, a real-time video transformation model built to change live video while it is still running. The AI startup says the system can modify video at 30 frames per second in 1080p resolution with near-zero latency.
The core idea is direct control through text commands and reference images. Instead of waiting for a clip to be rendered and then edited, users can describe the change they want and see the live feed transform in response.
What Lucy 2.0 can change in live video
Lucy 2.0 is designed for real-time visual transformation, not just static editing. According to the source, users can swap characters, place products, change clothing, and completely transform environments while the video continues to run.
That makes the model different from a tool that only edits a finished video after the fact. The important point is timing: the system is presented as operating on live video, with changes controlled as the scene is already in motion.
The input methods are also notable. Lucy 2.0 can be guided by text prompts and reference images, giving users both language-based control and visual examples for the desired result.
- Character swaps: users can change who or what appears in the video.
- Product placement: products can be added into the running video.
- Clothing changes: outfits can be altered through the model.
- Environment transformation: entire surroundings can be changed.
Why real-time performance matters
The source states that Lucy 2.0 can work at 30 frames per second in 1080p resolution with near-zero latency. Those details define the claim Decart is making: this is not only a video generation model, but one meant to keep pace with a live stream of images.
In practical terms, low latency is central to the experience. If a transformation arrives too late, the video no longer feels live. Decart’s stated target is a system that keeps the transformed output aligned with the ongoing input.
The 1080p resolution claim also matters because it sets expectations for visual detail. The system is not described merely as a low-resolution preview or a rough test feed. It is described as modifying live video at a widely used high-definition resolution.
At the same time, the source does not provide independent benchmarks, comparison tests, or broader availability details beyond a demo. The key facts available are Decart’s stated performance level, the model’s real-time transformation abilities, and the demo address.
How Decart says the model understands motion
According to Decart, Lucy 2.0 does not rely on depth maps or 3D models. The company says the system’s understanding of physics comes entirely from patterns learned during video training.
That detail is important because it describes how the model approaches the problem of changing a moving scene. Rather than using explicit depth maps or a separate 3D representation, the system depends on learned video patterns to maintain the behavior of objects, people, and environments over time.
The source does not describe the training data, the architecture, or the full technical process behind the model. What it does state is Decart’s explanation that Lucy 2.0 learns physical behavior from video patterns and applies that learned understanding during live transformation.
Keeping quality stable over time
One challenge for any live video transformation system is consistency. A model may produce convincing frames at first, but quality can degrade if errors build up as the stream continues.
Decart says Lucy 2.0 uses a new technique called Smart History Augmentation to prevent image quality from degrading over time. According to the startup, this lets the model run stably for hours.
The source does not give a step-by-step description of Smart History Augmentation. The stated purpose, however, is clear: it is meant to help the system preserve image quality during extended real-time operation.
That matters because Lucy 2.0 is being presented as a system for live video, where long-running stability is part of the promise. A short demo and a system that can run for hours are different claims, and Decart is placing emphasis on the latter.
Where the system runs and how to try it
The technology runs on AWS Trainium3 chips, according to the source. That is the only hardware detail provided, but it gives one concrete clue about the compute platform behind Lucy 2.0.
A demo is available at lucy.decart.ai. The source does not include further details about pricing, access limits, supported workflows, or whether the model is available beyond the demo.
For now, the main takeaway is straightforward: Decart is presenting Lucy 2.0 as a real-time AI video transformation model that can reshape live 1080p video at 30 frames per second using text prompts and reference images. Its claimed advantages are near-zero latency, live control, no reliance on depth maps or 3D models, and stable operation over hours through Smart History Augmentation.