In This Article
Follow BestForAndroid
30 frames per second, under 40 milliseconds of delay. That’s how fast this new AI model can rewrite and edit a live video feed in real-time while it’s still streaming.
Decart AI released Lucy 2.5, a real-time video editing model that changes what’s happening in a live video feed as it streams, running at 30 frames per second with latency under 40 milliseconds.
It works on webcams, live streams, and pre-recorded clips, letting people swap outfits, add effects like fire or slime, restyle entire backgrounds, or remove objects using text prompts and reference images. It’s live now through the Decart API, on fal.ai, and at lucy.decart.ai.
Think about the last time you used a filter on a video call. It probably glitched a little when you turned your head too fast or moved out of frame. Lucy 2.5 is built not to do that. The edit holds even after you leave the shot and come back, the same way your reflection doesn’t reset just because you looked away from the mirror.
You can see Kfir Aberman, the founding member of Decart AI, interacting with the Lucy 2.5 model.
What’s actually new since Lucy 2.0
Lucy 2.0 could already swap clothes and backgrounds in real time when it launched back in January. Lucy 2.5 is about making those edits look and hold up better.
Decart trained a separate model just to carry image edits into video, then added a coarse-to-fine prompting system so people can start with a simple instruction and drill down into position, color, or size.
Effects like fire, sand, and slime now interact with a scene the way real materials do, and entire environments can shift style, a sunny street turning into a neon city, while the people and objects in frame keep their actual shape.
How Self-Anchoring keeps the edit from drifting
Live AI video usually drifts. A face swap starts sharp, then slowly warps the longer a stream runs.
Decart says its new Self-Anchoring method fixes this by having the model treat its own recent output as the reference point instead of the original frame. A costume change from the first few seconds of a stream still looks the same minutes later, even if the person walks off-screen and comes back.
Who’s using it and who’s funding it
Decart is pitching Lucy 2.5 at commerce, advertising, streaming, and gaming: live virtual try-ons, ads that adjust for different viewers, streamers letting their chat reshape the scene in real time.
The Tel Aviv-based startup, founded in 2023, has raised more than 450 million dollars from backers including Sequoia Capital, Nvidia, and Amazon, and it has already signed strategic partnerships with both. That kind of backing suggests this isn’t a demo reel headed nowhere.
The tech to make anyone look like anything on a live video call, and the person can even interact with it, this tech already exists. The only real question left is who decides when that gets used against you. Yes, the privacy concerns remain.
