AI slow motion

What it means

Real slow motion comes from shooting at a high frame rate, say 120 fps, and playing it back at 30. AI slow motion fakes those extra frames. A model studies two consecutive frames, estimates how every pixel moves between them and generates new frames in between, a technique known as frame interpolation.

Older methods based on optical flow already did this, but they smeared and warped edges. Neural interpolation predicts motion and occlusion better, so a 30 fps clip can often be slowed four times and still look convincing on calm, simple scenes.

How it works in practice

AI slow motion is a rescue tool. The best moment of the day happened while the camera was set to 30 fps, and you want it slowed down: interpolation makes that possible, especially for smooth, predictable motion such as a surfer carving or a drone gliding.

Results are best with sharp source frames. A short shutter time, plenty of light and little motion blur give the model clean edges to work with.

Where you know in advance that a moment deserves slow motion, still shoot it at 120 fps or more. Real frames beat generated ones every time, and the tool can then stretch true high-frame-rate footage even further if needed.

What to watch out for

Fast, chaotic motion breaks interpolation. Spray, snow dust, spinning wheels, propellers and limbs crossing each other often come out as wobbling blobs. Check every slowed shot frame by frame, and keep the slowdown modest when the scene is busy.

Rolling camera motion plus a moving subject is the hardest case. A shaky helmet camera combined with a rider crossing the frame produces warping around the edges of the subject.

Do not stack effects carelessly. A clip slowed by AI and then put through a speed ramp or heavy stabilization accumulates artifacts at every step. Apply the slowdown last, on a clip that is already trimmed and stable.