This is cool. Can see it's still hard to get nice camera behavior (motion, angle, composition), because image inputs are not exact cinema shots in angle and composition, and i2v models will tend to keep the camera still. Call it "viewport inertia" t2v is better at the camera