Fill the template with action, camera, and ending
The start image already supplies appearance, so spend prompt space on change. Reusable manual template:
Begin from the start frame. [Subject] keeps [starting body, hands, and held object]. After [trigger], [complete one continuous main action, including contact and object-state change]. [Relevant secondary environmental response]. The camera [position or movement], keeping [visual priority] readable. End with [visible final position, hands, object state, and gaze]. Preserve [subject, clothing, location, and composition relationships required from the start frame]; add no [likely task-breaking object or text].
Make the verbs one continuous line: who moves, what is contacted, and how the object changes, followed by camera behavior. Avoid combining fast walking, looking back, opening an umbrella, waving, and changing location. Give the camera one purpose. A slow pullback, for example, can keep the opened umbrella and exit readable together. The ending should be drawable as a still rather than merely “ending with emotion.”
Inspect the real start image before writing motion
Image-to-video starts from the actual uploaded picture. Zoom in and inspect subject identity, body facing, hand-to-object contact, visible text, and whether the framing has room for the next action. A motion prompt should not silently repair a wrong stance or prop state already present in the image; replace or edit an unsuitable start first.
Kling’s image-to-video guide treats the input image as the scene basis, focuses the prompt on subject movement, optionally adds background movement, and recommends simple motion compatible with the image. Asking for a different place or radically different pose can create a visible cut or transition.
This article uses an original paper plan with no actual generation or test. The imagined start image shows an adult woman in a mustard raincoat on a riverside walkway. Her right hand holds a closed red umbrella pointed down, her body is side-on to the river, and her left hand rests on the railing. The shot will have her open the umbrella and turn toward the walkway exit.
Use the real start frame, then bind an Element if needed
The VIDEO 3.0 guide describes uploading a real start frame, then optionally binding a separately created Element through “Bind Subject to Enhance Consistency.” This supplements the frame; ordinary reference images are not automatically created Elements. Start with the frame-only route here. Choose the frame-plus-subject route only with a compatible prepared subject and an account interface that exposes binding.
Original case: repair a compressed prompt
Weak version: “A woman walks by the river, opens a red umbrella, looks back and smiles as the camera circles her, finally reaching the street.” It stacks actions and never explains how the umbrella opens. Keep one action: open it in place, then turn toward the exit.
Begin from the real start frame. The adult woman in a mustard raincoat keeps her body side-on to the river, left hand on the metal railing, and right hand around the straight handle of the closed red umbrella, its tip pointing down. She releases the railing and uses her right hand to turn the closed umbrella tip upward; her left hand slides the runner along the shaft toward the tip while the right hand steadies the handle, and opens the red canopy fully at her right side. A few droplets shake from the canopy and thin riverside branches move lightly in the breeze. She raises the open umbrella overhead and turns her body one quarter toward the walkway exit by pivoting her feet in place, without stepping toward the exit. At eye level, the camera slowly pulls back only far enough to show her full body, the complete canopy, and the exit direction together; it does not travel behind her. End with her facing the exit, right hand holding the umbrella overhead, left arm resting at her side, and gaze following the walkway. Preserve the mustard raincoat, red umbrella, riverside railing, wet stone path, and overcast light. Add no pedestrian, vehicle, subtitle, sign, or second umbrella.
This text is a copyable paper plan, not a claim of a successful clip. If the actual frame shows the umbrella in her left hand or the exit on the opposite side, revise hand use and turning direction to match the image.
Check each input in Kling before submitting
Choose the required VIDEO 3 image-to-video input route in the account interface. Upload the inspected real frame, confirm that its preview shows this run’s image, and paste the motion prompt. If you chose the supplemental-subject route above, finish its preparation and binding before checking the current inputs.
Pre-submit check:
| Check | Answer for this case |
|---|---|
| Start image | Woman, closed red umbrella, railing, and exit direction are visible and correct |
| Main action | Turn the umbrella tip upward, open it, and pivot toward the exit |
| Camera | Eye-level slow pullback to hold person, canopy, and exit together |
| Ending | Right hand holds umbrella overhead; body faces exit without leaving the starting area |
| Exclusions | No pedestrian, vehicle, text, or second umbrella |
Complete the manual submission using the actual inputs and settings shown in the interface.
Inspect Motion and Continuity in the Actual Clip
Inspect the returned video against the shot job. This plan has not been executed:
- The first frame matches the upload instead of jumping to an already-open umbrella.
- The left hand leaves the railing; the right hand turns the tip upward before the left pushes the runner while the right keeps the handle; hand roles do not swap.
- The umbrella changes continuously from closed to open without a second umbrella or canopy mutation.
- The subject turns in place toward the exit; her feet pivot in place without stepping toward it.
- The camera performs only the slow pullback, keeping person, canopy, and exit readable.
- Riverbank, railing, wet path, raincoat, and overcast light remain continuous.
- The ending can freeze as right hand overhead, body toward exit, left arm down, gaze forward.
For an opening jump, compare upload and written start. For hand or umbrella failure, shorten the chain and clarify contact. If camera movement overwhelms the action, remove it. Change one major issue per revision.
FAQ
How many actions should one prompt contain?
Start with one continuous chain from start to ending. Opening the umbrella needs release, reorientation, runner contact, and lifting; those steps belong together. Walking to the street, looking back, or waving belongs in another shot unless this shot requires it.
Should the prompt repeat every visible detail from the start frame?
Repeat the identity, clothing, held objects, geography, and light needed for continuity, then spend most words on contact, action, camera, and ending. Inspect the real image before upload.
Do it with the skill
Use $short-drama-video-prompts to write one copyable Kling VIDEO 3 image-to-video prompt for this accepted shot. The actual start frame is /ABS/start-frame.png; first inspect body facing, hands, held object, location, and composition. The main action is [one continuous chain], the camera is [one primary move], and the ending is [a freezeable state]. Follow the selected manual input plan and write the prompt only; generate no media. I will check inputs and submit manually in Kling, then inspect the actual result.
Read the method: Real-image, motion, and boundary rules for short-drama video prompts · Kling AI image-to-video guide · Kling VIDEO 3.0 model user guide
About Drama SkillsThe short-drama-video-prompts skill on GitHub