Step 1: Generate the base shot with Seedance 2.5 text-to-video. A parkour kid leaps off a rooftop edge across the gap between two buildings. At the peak, time almost stops and the camera orbits halfway around him (bullet time). Then time resumes, he lands on the opposite roof and rolls to his feet.

Step 2: Pull frames from the base, restyle them frame by frame with an image model so he flips through different looks mid-action, Everything Everywhere All at Once style, then automatically cut it back into one full video.

When to start, how often to switch, which looks: your call. At least 6 looks. Action, pose, framing and background must match the base. Only the character's look changes.