How to turn a photo into an AI video

A photograph gives you a scene. A video needs something to happen inside it. The useful starting question is not “How do I make this cinematic?” It is “What should move, and what should stay still?” A person can turn toward the camera. A curtain can lift in the breeze. A camera can move slowly toward a product. Each is a complete idea for a short shot.

This guide gives you a practical planning workflow, three original prompts, and a review sheet. It is not a claim that every model supports the same controls. Choose an image-to-video option that accepts your reference, and check the available settings before you generate.

1. Choose a photo that can become a first frame

Look at the image as a director would look at the beginning of a shot. Is the person already framed at the right distance? Is the product fully visible? Is there space for the movement you want? If someone needs to raise a hand, a tight face crop leaves little room to show the gesture. Solve the framing before asking the model to invent missing parts of the scene.

• Portrait: choose a sharp face, clean separation from the background, and enough room above the head.
• Product: keep the important shape, label, and contact with the surface visible.
• Landscape: decide whether water, foliage, clouds, or the camera should provide the movement.
• For any subject: remove accidental crops and obvious defects before generating. Use a photo you have permission to use.

Runway’s image-to-video documentation describes the input image as the source of the composition and visual appearance, while the text directs motion. That is a useful division of work for planning a shot: establish the look in the photo and explain the change in the prompt.

2. Describe a small, observable action

Write a sentence that someone could act out. “Looks confident” gives a mood; “lifts their chin slightly and looks toward the lens” gives a visible action. For a first attempt, select one action that fits naturally with the starting pose. You can build a longer story from several successful shots later.

This example asks for a modest change, so you can judge the face without a complicated camera move competing for attention. If the source photo shows a different pose, change the action to fit it instead of asking for an abrupt transition.

3. Decide whether the camera moves

Camera movement changes how viewers read the action. A still camera makes it easier to compare a face or product shape across the shot. A slow move forward creates a reveal. A sideways move needs a scene that makes sense as more of it becomes visible. Choose one direction for the first version and write it in ordinary language.

4. Set the output before spending credits

Check the actual options shown by the tool you are using: image support, duration, aspect ratio, resolution, audio, and cost. These can differ by model or plan. A prompt alone does not set a control that the tool requires you to choose elsewhere. For a vertical project, inspect the crop before generating so the subject is not lost at the edges.

Keep your first attempt focused. Write down the exact input photo, prompt, and settings. If you change the image, motion, camera, and duration together, the next result tells you very little about which change helped. A small revision gives you a more useful comparison.

5. Review the beginning, middle, and end

Play the full clip, then pause at several moments. A strong opening can hide a face that changes later or a cup that slowly bends. Choose a result for the complete movement, not only for its first frame.

A worked example: revise one thing

Imagine that you used the ceramic-cup prompt above and the handle changed shape halfway through the shot. That is a hypothetical review finding, not a result from a test. Before rewriting the whole scene, make a small comparison plan. Keep the same source photo, model, output settings, and steam description; change only the camera instruction.

Save both clips and record your observations at the beginning, middle, and end. A single pair does not prove what caused a difference: generation can vary between attempts. It does give you a concrete basis for choosing the next version. If both fail at the same detail, inspect the source image and simplify that detail in a separate experiment. Keep the successful shot and its settings together so you can reproduce the brief later.

Build the next shot from what worked

Save the version you would actually use, along with its settings and the reason you chose it. For a small campaign, that note becomes more valuable than a folder of unlabeled generations. You might discover that the portrait works best with a steady camera, while the product benefits from a slow approach. Carry that decision into the next shot.

When you are ready to explore Yanora, check the current creation options and available templates. Use the same shot-planning process with the controls that are available to you. You do not need a complicated scene to make a useful first video; you need a clear beginning and a change worth watching.