Timelapse Revolution with Artificial Intelligence

Timelapse creators, it's time to embrace AI for that wow factor! Explore the wonders of Relighting.

We're somewhat proud to say that this is the only AI network designed exclusively for timelapse-related issues, without being affected by video resolution. Think of the potential!

As shown in earlier posts, blending images in succession can remove unwanted movement in your timelapses. This works great for natural landscapes. However, it falls short when it can't tell the difference between repetitive changes (like shadows or day-to-night transitions) and random variations. This could be an issue for city settings or construction sites where you aim to keep the dynamic activity around your focal point.

This is where AI comes in again. Our trained network, known as “Relighting,” learns how a scene changes with time (weather, day-to-night cycles, etc.) and separates it from objects that appear or disappear.

The second phase of this innovative algorithm uses this analysis to consistently regulate the brightness of each pixel across time, adapting to abrupt changes in appearance. The outcome? Your extended timelapse's central theme is clear and distinct (the “information”), unaffected by the scene's brightness alterations (shadows, weather, referred to as “noise”), while ensuring a lifelike and lively portrayal of the environment (vehicles, people, etc.).

Be sure to watch the video below. Remember, in the video's initial segment we used the identical frames for the top and bottom sections! For comparison, the latter half showcases improved smoothing techniques.

Special thanks to our collaborator Frédéric LARDIN for permitting us to demonstrate the relighting with his project.

Ready to enhance your timelapse game with Relighting? Share your insights and any timelapse challenges you're eager to have us address next! Use hashtags #Timelapse #Enlaps

Delve into the realm of AI-guided video smoothing and its capacity to elevate your timelapse projects. Learn more on our website. 

Article authored by Adrien Fontvielle, R&D Director at Enlaps