For a two-person contact action, first make the starting positions physically compatible. The people must be close enough to reach each other, facing usable directions and standing at plausible heights. Then plan one contact and one consequence. Asking for a handhold, turn, embrace and lift in the same short clip increases the number of anatomical transitions that need to succeed.
Describe the starting geometry
“They hold hands naturally” describes the desired result but leaves the approach unresolved. Identify which person's hand moves, which hand receives it and where each body is supported. If one partner stands on a step, their hands meet at a different height than on level ground. Do not ask the shorter person to reach a raised shoulder without changing distance or posture.
Look at the reference image as an actual starting state. A tightly crossed arm cannot instantly become an extended hand without an intermediate movement. A hidden hand behind a torso gives the model less visible evidence about fingers and wrist position. A clearer preparatory pose can be more valuable than a more elaborate prompt.
Try a single assisted step
Fictional planning example: Ari waits beside a low ceremony platform and offers one open hand. Sam stands one step below, places a hand into Ari's and steps up. Ari remains on the platform; neither person lifts the other. The scene ends with both standing together. This is a staging exercise, not a safety instruction or a filmed customer outcome.
Check that Sam has a visible foot placement and enough space to move upward. The offered hand should not begin above Sam's comfortable reach. Keep the camera far enough back to show why the body rises. A close-up of joined hands alone would hide the step and make a change in shoulder height look like unexplained floating.
Use a contact preflight before video generation
| Question | Approve when | Simplify when |
|---|---|---|
| Can the arms meet? | Distance and orientation are plausible | Partners begin on opposite sides of the aisle |
| Is support clear? | Feet or a stable seat explain posture | Both bodies appear suspended |
| Is one hand pair identifiable? | Wrists and contact are readable | Several overlapping hands compete |
| Does the end follow the start? | One step changes height | A seated person abruptly appears standing |
In Imagild, review the keyframe or source image before generating the motion. If it already has merged wrists or incompatible arm positions, do not expect movement to fix it. Rewrite that scene's staging or use a simpler reference. Preserve the rest of the approved story instead of replacing the entire project because one contact is difficult.
Review the middle, not only the endpoints
A clean beginning and ending can conceal a malformed transfer between them. Watch at normal speed, then inspect the contact interval carefully. Look for a wrist passing through a sleeve, hands changing ownership or fingers multiplying when bodies overlap. Reject a visually confusing action even if the final pose seems attractive.
Do not assume a model can execute dangerous lifts safely because the generated picture looks plausible. For fictional acrobatics, keep the description clearly fictional and separate it from instructions for a real couple. Production planning here concerns visual readability, not physical training or stunt advice.
Decide what a retake should change
Change one geometry problem at a time: reduce separation, lower the offered hand, expose the receiving wrist or remove a second movement. That makes the retake interpretable. If you change wardrobe, venue, camera, action and contact together, you cannot tell which correction helped and may create new continuity problems.
Would a wider camera solve every hand problem?
No. A wider view can clarify bodies and feet but may make fingers less readable. Choose a framing that explains the action without making detailed hand anatomy the only focus.
Should I use two endpoint images for every contact?
Only when their body positions describe a compatible transition. Two beautiful but geometrically incompatible poses can ask for impossible motion. A simpler single-reference action may be a better pilot.