Timelapse Revolution with Artificial Intelligence

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

We're a bit proud to say that this is the only AI network of its kind solely trained for timelapse-specific issues, independent from video resolution. Imagine the possibilities!

As we've seen in previous posts, blending successive images can eliminate unwanted motion in your timelapses. It’s very effective for landscapes, for instance. But there's a major drawback: it can't distinguish between cyclic changes (shadows, day-night cycles) and sporadic shifts. This can be a hard stop for urban scenes or construction projects where you still want to capture the buzzing activity around your subject.

That's where AI steps in, once again. We trained a network, dubbed “Relighting”, that learns a scene's evolution over time (weather, day-night cycles, etc.) and disentangles it from appearing/disappearing objects.

The second part of this unique algorithm uses this analysis to smoothly adjust the brightness of each pixel over time while adapting to sudden appearance changes. The result? Your long-term timelapse's main subject stands out prominently (the “information”), without disruptions from scene brightness changes (shadows, weather - aka “noise”). All while maintaining a realistic and vibrant appearance of surrounding activity (vehicles, people, and more).

Be sure to check the video below. Keep in mind that in the first part of the video we used exactly the same frames in top and bottom parts! In the second part, we included advanced smoothing for comparison.

Thanks to our partner Frédéric LARDIN for letting us use its project to illustrate the relighting.

Ready to level up your timelapse game with Relighting? Share your thoughts and any timelapse challenges you'd love to see us tackle next! Use hashtags #Timelapse #Enlaps

Dive deeper into the power of AI-driven video smoothing and its ability to transform your timelapse projects. Discover more on our website. 

 Article written by Adrien Fontvielle, R&D Manager at Enlaps