AI video generation
Write a prompt someone could shoot
Describe a coherent shot, then refine the result one decision at a time.
Request consultingDescribe what happens
Name the subject, the visible action and the setting. Add the framing, camera behavior and light when those choices matter. Keep the instructions consistent: a camera cannot remain fixed and circle the subject in the same uninterrupted shot.
Use visual descriptions that explain the result you want. Instead of relying only on “premium,” describe the surface, the quality of the light and the pace of movement. Lens names and style references are not guarantees that a model will reproduce physical optics.
An illustrative starting point
Example: “A ceramic bowl sits on a dark stone worktop beside a window. A hand sets a pear inside it and leaves the frame. Close view from table height, fixed camera. Soft light from the left reveals the uneven glaze. The action finishes with the bowl still.”
This example is newly written and has not been tested against a model. Set duration, aspect ratio and other structured options in the tool where supported. Consult that model’s guidance before adding negative prompts or control syntax.
Revise the instruction that caused the problem
If the camera moves unexpectedly, simplify the camera instruction and check the tool’s controls. If several actions are incomplete, reduce the shot to the most important one. If a reference image is supplied, focus the text on the change you want rather than redescribing every visible detail.
Keep a record of the prompt, model, settings, references and result. Compare variations with one deliberate change at a time. Save the successful configuration with the clip so the decision can be revisited later.