Video Extension Prompts: 24 Examples and a Formula

Nov 18, 2026

When you extend a video with AI, the clip shows the model what the scene looks like. The prompt tells it what happens next. A vague prompt like "continue the video" leaves the model guessing, which is the most common reason extensions stall, drift, or change the subject. This guide gives you a simple formula, 24 copy-and-adapt examples, and the mistakes to avoid. Use them with any model in Video Extend.

The formula

A good continuation prompt answers four questions:

  1. Continuity: Is this the same shot? ("Continue the same uninterrupted shot.")
  2. Action: What does the subject do next? Use verbs: walks, turns, lifts, pours.
  3. Camera: What does the camera do? Tracks, pans, pushes in, stays still.
  4. What must not change: Subject appearance, lighting, setting, and anything the model should not add.

Put together:

Continue the same uninterrupted shot. [Subject] [one or two actions]. The camera [movement], keeping the same [distance/height/angle]. Preserve [subject details, lighting, setting]. No cuts, no new people or objects, no text.

A real example

This is the prompt we gave nine models on our model comparison page:

Continue the input shot seamlessly in one continuous take. The cowboy slows to a stop, then looks back over his right shoulder so that his face becomes visible. He raises his right hand, touches the brim of his hat once, then lowers his hand naturally. During this action, the camera smoothly arcs about 45 degrees from directly behind him toward his right side, maintaining the same distance and height. Preserve his hat, dark clothes, body proportions, warm backlighting and the layout of the street. The buildings stay fixed in place with natural parallax as the camera moves. Keep the dust drifting naturally. Real-time movement, no cuts, no slow motion, no new people or objects, no text or logos.

It names the actions in order, says exactly how the camera moves and how far, lists what to preserve, and rules out common failures. For a simpler shot, a shorter prompt is fine:

Continue the same uninterrupted tracking shot. The lone cowboy keeps walking forward at a steady natural pace along the dusty street. The camera follows smoothly from behind at the same distance and height. Preserve his hat, dark clothing, the buildings and warm backlighting. No cuts, new characters, sudden camera moves, text or logos.

24 prompt examples by shot type

Replace the bracketed parts with details from your own clip.

People and characters

  1. Continue the same shot. [The woman] turns toward the camera and smiles, then looks back at [the window]. Slow push-in. Keep her outfit, hair, and soft window light unchanged.
  2. Continue the same tracking shot. [The runner] keeps running along [the river path] at the same pace. Camera tracks alongside at the same distance. Same morning light, no new people.
  3. Continue the shot. [The man] stops walking, takes [his phone] from his pocket, and looks at it. Camera stays still at eye level. Preserve his clothes and the street behind him.
  4. Continue the same shot. [The dancer] completes the spin and extends both arms, then holds the pose. Camera slowly orbits right. Same stage lighting and costume.
  5. Continue the shot. [The child] runs toward the camera and jumps into [the leaves]. Camera stays low and still. Same autumn colors.

Animals

  1. Continue the same shot. [The dog] keeps running across [the beach] toward the camera. Camera tracks backward at the same speed. Same afternoon light.
  2. Continue the shot. [The lion] lifts its head, looks left, then slowly blinks. Locked-off camera, shallow depth of field, same golden light.
  3. Continue the same shot. [The bird] takes off from [the branch] and flies out of frame to the right. Camera stays still. Branch keeps swaying slightly.

Camera moves and landscapes

  1. Continue the same aerial shot. The drone keeps flying forward over [the coastline], then slowly tilts down toward the waves. Same sunset colors.
  2. Continue the shot. The camera keeps panning right across [the mountain range] at the same slow speed. Clouds drift gently. No people.
  3. Continue the same shot. The camera keeps pushing forward through [the misty forest] at walking pace. Light rays stay in the same direction.
  4. Continue the same timelapse. [The sun] keeps rising behind [the peaks], light spreads across the valley. Locked-off camera.

Cities and night scenes

  1. Continue the same shot. The camera keeps moving down [the rainy neon alley] at walking pace. Reflections on the wet ground, same pink and blue signs. No new people.
  2. Continue the same aerial shot. The camera keeps orbiting [the city skyline] at the same height as the sun sets. Lights in the buildings begin to turn on.
  3. Continue the shot. Traffic keeps flowing across [the intersection]. Locked-off camera, same light trails, same exposure.

Products and food

  1. Continue the same shot. The camera keeps orbiting [the sneaker] slowly to the right on the same turntable. Same studio lighting, clean background, no text.
  2. Continue the shot. [The chef] finishes slicing [the tomato] and slides the pieces into [the pan]. Close-up, camera stays still. Same kitchen light.
  3. Continue the same shot. Coffee keeps pouring into [the cup] until it is full, then the stream stops. Macro, locked-off camera.

Action and nature

  1. Continue the same shot. [The surfer] keeps riding through the wave barrel and exits into open water. Camera follows from the side.
  2. Continue the same shot. The waves keep rolling onto [the rocks] and spray into the air. Locked-off camera, same overcast light.
  3. Continue the shot. [The car] keeps driving along [the coastal road] and disappears around the bend. Camera stays at the same position.

Sci-fi and stylized

  1. Continue the same shot. The camera keeps floating down [the sci-fi corridor] at the same speed. Lights pulse in the same rhythm. No characters appear.
  2. Continue the same underwater shot. The camera keeps gliding over [the coral reef] as [a school of fish] passes in front. Same blue light.
  3. Continue the shot in the same [anime/watercolor] style. [The character] looks up at the sky as [petals] drift past. Slow tilt up.

For more ideas, browse the video prompt library, which covers scenes like coral reefs, neon alleys, product orbits, and character walks.

Common mistakes

Mistake What happens Better
"Continue the video" Motion stalls or the model invents something random Name the next action with a verb
A whole story in one prompt Rushed or skipped actions One or two actions per extension
No camera direction Unexpected zooms or angle changes "Camera stays still" or name the move
Describing a different scene Visible jump at the seam Start from what is on screen at the end of the clip
Not saying what to keep Outfit, lighting, or background changes List what must stay the same
Asking for text or logos Garbled lettering Add text in your editor afterwards

Tips for longer sequences

  • Plan beats first. Split a long sequence into short extensions, one action each. ChatGPT can help you break a script into beats; see Can ChatGPT Extend Videos? for a ready-made prompt.
  • Repeat the "preserve" line in every extension so details don't drift over several passes.
  • Stay with one model for a continuous shot. Our model comparison shows how differently each model behaves.
  • Check every seam before extending again. If something looks off, see how to fix seams in extended videos.

Ready to try one? Open the AI video generator, upload your clip, and paste a prompt from this list.

FAQ

How long should a video extension prompt be?

One to four sentences is usually enough. Longer prompts are fine for complex shots, as long as each sentence adds a specific action, camera instruction, or constraint.

Do I need a prompt if I already uploaded the video?

Yes. The video gives the model visual context, but without a prompt it has no direction for what happens next. The AI video extender page explains how the clip and prompt work together.

Should I describe the whole scene again?

Only the parts that must stay the same. The model can already see the scene, so focus on the action, the camera, and the details you want preserved.

Can I use negative prompts?

Some models have a separate negative prompt field. With others, add constraints at the end of the main prompt, such as "no cuts, no new people, no text."

Video Extend Team

Video Extend Team