AI video prompts for text and image to video

Write an AI video prompt as a short shot: what is in frame, what happens and where the shot ends. Add a camera move only if you want one. On the video page, start with text-to-video when you need to create the scene, or image-to-video when you already have a starting picture.

For image-to-video, focus on motion rather than redescribing everything in the still. Google's video prompt guide lists subject, action, scene, camera and lighting as useful components; you don't need them all in every prompt.

Choose a camera move you can describe

• Fixed: the view stays in place, without a pan, orbit or zoom. Useful for a small action you want to see clearly.
• Pan: the camera turns left or right from a fixed position. Name what it turns from and what it finishes on.
• Push-in: the viewpoint moves closer to a subject. Say which detail should fill the ending frame.

"Slow push-in toward the bicycle bell, then stop" is easier to evaluate than "cinematic, dynamic camera." For an early test, choose one camera move rather than combining several.

Three scenes, six prompts

These original examples pair a text-to-video prompt with an image-to-video version of the same scene. They haven't been rendered for this guide. For the image version, prepare the starting still described with each example.

After generating, check whether the intended action finishes and the ending is recognizable. That tells you what to change next.

The cup on the shelf

One gesture, fixed camera. For image-to-video, the still shows a hand already on the cup, and the handle is not yet facing forward. If the handle already faces the camera, the action is over before the clip starts.

Text-to-video:

A pottery shelf in even daylight, one unglazed cup in front. A hand enters from the left, turns the cup until the handle faces the camera, and leaves. Camera stays fixed at eye level, no zoom. End with the handle facing forward and the hand gone. No other cups move, no text.

Image-to-video:

The hand already on the cup turns it until the handle faces the camera, then leaves. Camera stays fixed. End on the handle facing forward and the hand gone.

The toast that misses

For a funny video prompt, try a simple setup and an unexpected finish. Here the toast rises and pauses as though it will stay upright, then flops onto the counter. The brief pause before the fall is the timing to experiment with.

Keep the camera fixed so the fall stays in frame. Do not add a hand, a second slice, or a camera move in the same clip. For image-to-video, the still should show the slice just coming out of the toaster and an empty counter. The still is the setup. The prompt only plays the payoff. If the toast is already flat in the still, there is no fall left.

Text-to-video:

A kitchen counter and one toaster in morning side light. A single slice of toast rises, pauses at the top, then flops forward onto the counter. Camera stays fixed, medium distance, no zoom. End on the toast lying flat. No hands, no faces, no text.

Image-to-video:

The toast pauses at the top, then flops forward onto the counter. Camera stays fixed. End with the toast lying flat.

The bell on the bike

The bicycle stays still. The camera pushes in. For image-to-video, start from a wide still. A still that is already a close view of the bell gives the push-in no distance to travel.

Text-to-video:

A bicycle leaned against a brick wall in overcast daylight. Camera starts wide on the whole bike and pushes in slowly toward the handlebar bell, then stops. The bike does not move. End closer on the bell. No people, no other vehicles, no text.

Image-to-video:

Slow push-in toward the handlebar bell, then stop. The bike stays still. End closer on the bell.

To make this a pan, replace the camera sentence. Keep one move.

Camera pans slowly from the brick wall to the bicycle, then stops. The bike stays still. End with the whole bike in frame.

For that version, start with more wall visible on one side of the bicycle, leaving space for the pan.

For model-specific examples, continue with the Kling guide or Grok Imagine guide.

Change one thing on a short clip

On video creation, try the cup shot at a short supported duration. Check the current model, settings and credit quote before generating.

Keep that clip as the control: same prompt, same still if you are using one, same duration. The next run changes a single variable. If the push-in is too fast, change only the pace. If a hand appears beside the toast, leave rain and a new lens out of that retry. Leave the other settings alone. Another take can still vary at random, so the pair shows whether you aimed at the failure you saw. It does not prove the new word was the only cause.

Watch once and name the failure.

The subject changes

The cup becomes a bowl, or the bike changes color before the end. Remove simultaneous events first. If the object still will not hold, move the shot to image-to-video and let the still define it. Switch models after the shot is simpler. When you do, keep the same short prompt, so you can tell what changed.

Too much is happening

The toast flops, a hand catches it, the camera creeps, and steam fills the room. Return to one action and one end frame before you tune adjectives. If the control was already the short toast prompt and the rise, the pause, and the flop will not all finish, shorten the action to a tip forward and a flat landing.

The camera wanders

You asked for a fixed frame and the view drifts or zooms. State one move and its stop, and remove cinematic and dynamic. If it still wanders, generate from a still whose framing you already accept, and describe no new angle in that retry.

Check sound before planning the edit

Audio support depends on the selected model and what the integration offers. Listen to the actual result before relying on dialogue or room tone; words in the prompt may be missing or mistimed.

Once a short shot has the action, subject and camera move you wanted, save it before trying a longer version. If you need to build a video around an existing song, the song-to-short-film workflow is a separate starting point.