Practical checklist

Kling Motion Control Reference Video Checklist

Use a short, readable reference with one clear performer, then make the target image agree with its direction, crop, and visible body parts. If those inputs disagree, change the inputs before adding more prompt detail.

Run the four input gates

Gate one: one performer. Decide who the motion belongs to before uploading. A crowded clip can contain several people, but the useful test is whether one person stays visually dominant from start to finish. If the performer disappears behind another person, use a cleaner clip.

Gate two: compatible framing. Compare the first readable pose in the reference with the target image. A full-body turn needs a target image with room for the shoulders, arms, hips, and legs that will be seen. A close portrait is a poor starting point for a wide gesture, even when its face looks excellent.

Gate three: matching direction. Check which way the head, torso, and camera-facing shoulder point. When the source motion starts to the left and the image is built to face right, you have introduced a problem that no background prompt can solve. Fix the pairing first.

Gate four: readable face. If expression transfer matters, choose a reference where the face is visible through the transitions, not only in the opening frame. Use a target image with clear facial detail; otherwise judge the result as a body-motion test, not an identity test.

Choose the test that reveals the failure

Do not make the first run a showcase. Make it diagnostic. Keep the scene plain, use the shortest representative motion, and include only one action. A clean test makes it possible to separate an input mismatch from a hand, face, or contact error.

Review the render in this order: direction change, face transition, hands, then the moment the performer touches an object or another surface. Stop at the first break. That point tells you which input gate to revisit instead of giving you a vague reason to regenerate.

Use prompts for what the reference does not carry

Let the reference establish timing and movement. Let the prompt establish the setting, light, atmosphere, and style. Asking both inputs to define the same gesture produces a result that is harder to inspect because it is unclear which instruction won.

For a prop scene, test the physical interaction before styling it. A simple object, ordinary lighting, and one contact point create a better baseline than an elaborate scene. Once the contact is stable, add the visual treatment in a later pass.

Before you commit to a longer sequence

Keep a small input note beside each successful render: which reference clip was used, the target crop, the first motion beat, and the first point where a failure appeared. This turns a good result into a repeatable setup instead of a lucky file you cannot reconstruct.

The goal is not to make every reference work. It is to reject the mismatched ones early, while a clean replacement is still cheap to test.