Baylands, take T4 (cycling, 4K100), 20 seconds from the drones' own pixels

How it's made. A 3D body fitted on every frame to all three drones gives the shape. Each pixel of it is coloured from the drones' 4K frames at that instant, from whichever drones see that point best. Nothing is generated. The body's outline matches the athlete in every drone view (silhouette overlap 0.94–0.97). The field and tree line are built from the same footage with the athlete and their shadow removed; the shadow you see follows the real sun.

Generated fill. Angles none of the three drones saw at that moment (often the front, when all three follow from behind) are filled from six views generated by 4DAnyone, a research video model, guided by our three-drone measurement, and placed as virtual cameras that move with the rider. The drones' own pixels win wherever a drone saw the point. Turn it off with Generated fill to see the drones alone. Research benchmark: the model's weights are not licensed for commercial use.

Where it breaks. The free camera is limited to the heights the drones covered; hands and other small, fast parts are the least accurate. Drone 3 has no pose for the first 4.6 s, so early frames use two drones. The field is flat by construction, so anything standing on it apart from the athlete is flattened.