KINETIC controls how much physical motion a scene uses to explain the prompt. Whatever "action" is happening, KINETIC increases its intensity.

At negative weights, actions and scenes become more static, posed, restrained, or frozen. At positive weights, the checkpoint increasingly expresses the same prompt through active body movement, gesture, interaction, displaced clothing/hair, dynamic posing, and other evidence of motion.

This isn't simply a camera-motion or action-effect LoRA. It can change how physically active the subjects need to be for the prompted event to make sense. That makes it useful on prompts that already imply action or interaction: dancing, kissing, embracing, boxing, grappling, cooking, moving water, group portraits, and even sculpture.

A particularly interesting behavior is that KINETIC can distinguish a state from an event. A family can go from a relatively conventional posed portrait toward an unruly moment that appears to be actively happening, while the basic scene remains recognizable.

Recommended range: approximately −3 to +3.

Around ±4 and beyond, the direction becomes increasingly aggressive and may produce exaggerated poses, stronger composition changes, or visual shorthand such as motion-line-like artifacts. Those weights can still be fun for deliberate overdrive, but ±3 is the more reliable working range.

Negative: quieter, more stationary, posed, restrained, frozen.
Positive: active, gestural, physically involved, dynamic, in motion.

No special trigger phrase is required; use the LoRA weight as the control.

That wording also avoids promising literal velocity. The thing we actually observed is closer to amount of motion required to explain the action, which I think is the genuinely distinctive bit.