Decart launches real-time AI tool for live video transformation
The startup says its Mirage system can change video style instantly while preserving motion.
Israeli AI startup Decart has announced what it describes as the world’s first real-time, autoregressive video-to-video model. The new system, named Mirage, is designed to convert live video streams into different visual styles while maintaining motion, structure and frame rates, an advance that could open new possibilities for how video is produced and adapted in real time.
Unlike previous AI image or video generators that have struggled with visual glitches, Mirage works by generating each video frame in sequence, conditioning it on the one before and a simple text prompt describing the desired style. This autoregressive approach is intended to keep motion consistent and visuals stable, an area where other diffusion-based models have often fallen short.
Decart says Mirage can run at 20 frames per second at a standard web resolution, with an end-to-end processing delay of around 100 milliseconds per frame. That performance could make it feasible for live streaming, interactive video calls, and other real-time applications if hardware and cloud infrastructure keep pace. The company plans to expand to full HD and 4K resolution as performance improves.
Related articles:
Mirage is being launched alongside an early version of a user platform where people can upload or record videos, type in a style prompt, and watch as the model applies the changes immediately. Unlike conventional video editing, which requires separate production stages, Mirage aims to make live video a more flexible format that can be adapted on the fly.
Dean Leitersdorf, Decart’s founder and CEO, said the company wants to make video a more responsive medium. “Content is no longer fixed but adaptive,” he said in a statement, adding that Mirage could give viewers more ways to influence how video looks while it plays.
Founded in 2023 by Dr. Dean Leitersdorf and Moshe Shalev, Decart has raised $53 million to date from the likes of Benchmark, Sequoia Capital, and Zeev Ventures.
