An Ozil goal AI prompt that keeps the action readable

Give every input one job

A useful prompt tells the model what each reference controls. The photo establishes the intended scorer’s face, hair, proportions and clothing. A video reference, when present, establishes the action order and camera timing. Describe these roles directly. Do not ask the photo to be an exact opening frame while also treating it as a flexible identity reference; those are separate input modes in the model adapter. The Ozil Goal creator uses reference-image mode so the person can enter a new football setting.

Use an ordered action prompt

Try this direction: “Use the person in the reference photo as the only scorer, keeping their face, hair and outfit recognizable. In one wide continuous tracking shot, receive a through ball, chip the approaching goalkeeper, collect the ball, pause and step around two sliding defenders, then roll the ball into the empty net. Keep the scorer, ball and goal visible. Use natural anatomy and consistent ball contact.” This describes a requested result, not a guarantee. The prepared complete-goal scene already supplies this structure; extra direction should not repeat it.

Choose a camera that can show the play

A close-up can show a face but cannot show the keeper, defenders, ball and net at once. A wide sideline or three-quarter tracking shot is more useful for understanding the whole action. Avoid combining a drone shot, a face close-up, a reverse angle and a slow-motion replay in a short single-take request. Each cut asks the model to re-establish positions and identity. If you need a separate celebration shot, use the celebration scene rather than adding it to an already crowded action sequence.

Keep the football physically present

Ball continuity is a useful review criterion, so ask for it plainly. Specify one football, visible foot-to-ball contact and a net that remains in the scene. Adding technical jargon does not guarantee physical correctness. Avoid asking the scorer to juggle, turn, chip, slide, celebrate and switch outfits during the same short play. The chip and finish are the important contacts. A simpler action with a readable trajectory can be more convincing than a complex scene where the ball disappears halfway through.

Use sound direction carefully

Ask for original stadium ambience and synchronized contact sound if you want audio. Do not promise a named song, a broadcaster’s commentary or a famous voice. A reference clip can guide sound where the selected model supports it, but the result may still differ. If silent output is selected, make that intention consistent across the form and prompt. After generation, listen to the whole clip: a plausible image does not rule out an unexpected voice, a sharp audio cut or music that is inappropriate for the intended post.

Revise a prompt after reviewing one failure

Write down the exact failure before changing the direction. “The keeper appears after the chip” calls for a clearer action order. “The player’s face changes at the turn” calls for better identity evidence and fewer camera changes. “The shot ends before the finish” calls for a sequence that fits the selected duration. Change one main variable so you can understand the next result. Do not describe a completed but imperfect render as a technical task failure; the refund rules distinguish those cases.

Common questions

Take the next shot

Choose a prepared scene, use a clear authorized photo, and review the quote before generating. Inspect the complete result before sharing.