One Portrait to a Vertical AI Video: Framing, Motion, Review, and Export

A vertical video needs more than a tall export. The source photo, the action, and the final crop need to work together. A face can look sharp in a wide image and still become too small or leave the frame in a phone-sized clip. Decide where the person should sit in the frame before asking for motion.

This guide follows one portrait through preparation, generation, review, and finishing. You can use the general image-to-video tool for an open motion brief or choose a template for a specific effect. The examples are suggested workflows, not a claim that a measured test produced a particular success rate or number of retries.

Start with the purpose of the clip

Choose one job: a reaction, a fantasy reveal, a style change, or a short greeting. Write the main action in a sentence. That sentence should explain what happens without relying on a list of effects.

For a reaction, the face needs to be large enough to read on a phone. For an outfit reveal, the shoulders and garment need room. For a two-person greeting, both faces and the space between them matter. These are different framing problems even if all three clips will be vertical.

Decide whether you need a caption, a product name, or a final call to action. Keep a quiet area for it. Add exact words in a normal editor after generation rather than asking the model to draw tiny letters in the scene. The video’s first job is to give you a coherent visual sequence.

Prepare the photo for a tall frame

Use a source with a clear face and enough body context for the chosen movement. A head-and-shoulders portrait can support a small smile. It cannot reliably explain a full-body takeoff if the legs and lower body are missing.

Make a copy of the photo and test the intended crop before uploading. A 9:16 frame is tall and narrow. Keep the important feature away from the side edges and leave room in the direction of motion. If a hand will rise, include the hand’s starting position and the area above it. If hair will grow, leave space below the current ends.

Do not stretch a wide photograph into a narrow one. Crop or choose a different source. Stretching changes face proportions before the model begins. A plain background can be easier to extend visually than a wall of shelves or signs, but a generated extension is still invented content.

Match the template to what the source shows

Choose the effect after checking the pose, not just the template name. A calm face can suit a small smile or a sleepy reaction. A torso-visible image has more room for an armor reveal. A sky journey such as Cloud Trip needs space around the person so the environment can explain the movement.

Two-person templates need additional planning. Follow the actual image-slot labels and use references with compatible scale and light. Do not expect two tight face crops to supply a clear handshake, carry, or kneeling scene. The source images should help establish the bodies and the shared action.

Browse the public preview to understand the direction. It does not guarantee that your photo will follow the exact same path. The selected model and current settings also affect the result.

Keep the original preset as the main action

A template’s optional detail field can refine the existing action. A short note such as “Keep the camera still” or “Keep my face visible” has a clear role. It protects a part of the sequence without asking for a different story.

Read the interface note about long custom prompts. In this workflow, sufficiently long text can replace the preset instructions. If you want a completely different scene, the general image-to-video tool is a clearer place to write that full brief.

The following are short detail ideas to adapt, rather than changes to the site’s stored template prompts:

Keep my face visible. Use a gentle motion and a still camera.

Leave room above my head. Keep the original outfit and expression.

Hold the final pose briefly so the ending is easy to trim.

Choose the detail that addresses the actual risk in your source. Adding all three to every template can create unnecessary demands, especially if the preset already includes a moving camera or a larger expression change.

Choose the available vertical settings

If the current model offers 9:16, select it in the generator. Some workflows derive output shape from the uploaded image or use an automatic aspect option. In those cases, a correctly prepared tall source matters. Use the choices actually shown rather than assuming every model has the same menu.

Select a duration that leaves room for the action and a short ending. A longer clip can introduce extra gestures after the main idea has finished. Resolution affects output detail, but it cannot recover a face that was unclear in the source or repair an impossible body move.

Check the quoted credits after changing model, duration, resolution, or aspect settings. The quote is for the current configuration. Do not infer the cost of your task from the example video or from another model’s rate. Review pricing for purchase options and use the generator to confirm the task itself.

Review the whole frame on a phone-sized view

Watch the result once without pausing. Can you understand the action immediately? Is the face large enough? Does an important hand leave the frame? A technically tall file can still have a weak vertical composition.

Then pause at the opening, the largest movement, and the ending. Check eyes, mouth, hands, neck, and the line where the subject meets the background. For a costume reveal, inspect seams and coverage. For a sky scene, inspect the layers around the body. For a greeting, inspect the point where hands or props meet.

Look at the final crop as well as the full output. If the post will show interface overlays, keep important detail out of the area likely to be covered. Different apps change their presentation over time, so use the current in-app preview for the destination instead of relying on a universal safe-area number.

Diagnose a failed vertical attempt

| Problem | What to change first | | --- | --- | | The face becomes too small | Use a closer source or a smaller camera pullback; keep enough room for the action. | | Hair or a raised hand leaves the top | Add margin in the source or reduce the movement. | | The camera reveals an invented body side | Reduce the orbit and keep closer to the source angle. | | The face changes during the effect | Use a sharper front view and one simpler action. | | A wide output is hard to crop | Prepare a tall source and choose a supported vertical or image-derived option. | | The ending adds an unwanted gesture | Ask for a held final pose or use a supported shorter sequence. |

Change one condition per test. If you switch the source, model, prompt, and crop together, a better result will not tell you which change helped. Keep notes even if they are only a few words beside the downloaded file.

Finish the edit after the visual sequence works

Trim the beginning so the viewer has a moment to understand the subject. Keep the main action and a short settled ending. Add captions, music, and any required source credit afterward. This gives you control over spelling and timing without asking the image model to solve a layout task inside a moving scene.

Listen to the actual output before assuming it includes sound. Audio depends on the selected model and task preset. If you add music or a voice later, check that you have the rights to use it in the intended context.

Export using the editor’s real output settings and inspect the saved file, not only the timeline preview. Check that the crop, motion, and text remain readable. Download the generation you want to keep promptly; a result link is not a promise of permanent storage.

Publish with clear context

Use a portrait you own or have permission to animate. If a realistic clip could be mistaken for a real event, describe it as AI-generated. A synthetic smile, costume, or greeting should not be used to imply a person’s endorsement or relationship without their agreement.

For a second clip, reuse the parts that worked: the source crop, one clear action, and suitable model settings. Change only the part that needs improvement. That produces a useful series of decisions rather than a growing folder of unrelated experiments.

Read the reference-image guide for source preparation or the pet workflow when paws, fur, and animal identity are the main challenge.