Source-led prompt workflow

Image-to-video prompts for controlled motion

Build a motion-first image-to-video prompt, then add camera direction and a repair pass only when the first render needs it.

Reviewed2026-11-08
Evidence statusEditorially reviewed · Not hands-on tested
Model tagsMulti-model · Runway · Kling
Short answer

What makes this prompt more controllable?

For image-to-video, let the image carry appearance and composition. Use the text prompt to describe one subject action, one camera behavior and the timing of the motion.

Interactive prompt builder

Build your image-to-video prompts for controlled motion

A short first pass that isolates the main motion.

the person in the foreground turns toward the window and takes one slow step. holds a steady medium shot. slow, continuous motion over five seconds.

How to use this workflow

  1. Upload a clean source image without obvious face, hand or compression artifacts.
  2. Generate the Quick variant first and change only one variable between attempts.
  3. Move to Controlled only when the basic motion works; use Repair for a specific observed failure.
  4. Record the model, version, settings and number of attempts before treating a pattern as reliable.

Failure diagnosis and next prompt

Observed failureLikely causeNext change
The face or outfit changesThe requested action or camera move is too large for the reference.Shorten the action, reduce speed and keep the camera static for the next attempt.
The shot invents objectsThe prompt introduces scene details that are absent from the frame.Remove appearance descriptions and prompt only motion already supported by the image.
Motion looks chaoticToo many actions or camera instructions compete in a short duration.Keep one subject action and one camera behavior; split a sequence into separate shots.

Filled editorial example

Illustrative input: a woman beside a cafe window, curtains visible behind her.

The camera holds a steady medium shot as the woman turns toward the window and takes one slow step. Curtains move gently in the breeze. Slow, continuous motion over five seconds. The shot remains continuous and visually consistent with the input frame.

Editorial example only — this is not a generated output or a claimed test result.

Compatibility and limits

Model-agnostic starting point. Exact camera controls, negative prompts and reference features vary by model and version.

Generation quality and credit use can change by model, version, input asset and settings. Check the product interface and current official documentation before paying for a plan or running a large batch.

Primary sources

Continue the workflow

Use rights-cleared inputs. Review identity, factual claims, disclosures and commercial-use terms before publishing. AI RankLab does not promise a model result or income outcome.