How long can an AI video made from a photo be?

It depends on the model. Veo 3.1 makes clips of 4, 6 or 8 seconds, while Kling and Grok Imagine Video go up to 15 seconds. For a longer video, generate several clips and edit them together.

Can AI add sound to a video made from a photo?

Yes, some models generate audio along with the picture. Veo 3.1 produces dialogue, sound effects and ambience in sync with the action, and Kling and Grok Imagine Video can add sound as well.

Why does the person in my AI video stop looking like themselves?

Every frame after the first is generated, so identity can drift, especially with big movements, long clips or a small, blurry face. A sharp, close photo, a simpler action and a shorter clip keep the likeness much better.

Guide · 4 min read

How to turn a photo into a video with AI

By the Luren AI team · Updated October 16, 2026

To turn a photo into a video with AI, upload the photo to an image-to-video model, describe the motion you want (what moves, how the camera moves, the mood and any sound), choose the length and aspect ratio, and generate. The model uses your photo as the opening frame and invents what happens next, so a sharp photo and a short, specific motion prompt matter more than anything else. Current models such as Veo 3.1 and Kling produce clips of several seconds, and some generate matching sound at the same time.

How image-to-video AI works

An image-to-video model starts from your picture and predicts the frames that follow, guided by your prompt. Because the first frame is fixed, the people, clothes and place in the clip come from the photo, while the movement, the camera and the timing come from the model and your words.

Some models also accept an end frame, a second image the clip has to arrive at. That turns the job into an animated transition between two pictures, useful for before-and-after reveals or moving between two poses. Text-to-video is the related task with no photo at all, where the model invents the whole scene from words; see text-to-video AI for that.

Step by step

The process is the same whichever model you use:

Pick a photo that is sharp and has room for movement around the subject.
Open an image-to-video tool and add the photo as the first frame, plus an end frame if you want one.
Write a motion prompt: one main action, the camera move, the atmosphere and any sound.
Choose the model, length, resolution and aspect ratio: 9:16 for Reels and TikTok, 16:9 for YouTube.
Generate, watch the result and change one thing in the prompt if something is off.
Download the clip and add captions or music in your editor if needed.

Choose a photo that animates well

The model can only move what it can see, and it has to invent anything that enters the frame. A few choices make a big difference:

Sharp and well lit. Blur and noise turn into flicker and smearing once things move.
The whole subject in frame. If the person will walk or dance, use a full-body photo, because cut-off feet or hands have to be made up.
Room around the subject, so a camera move or a gesture does not run off the edge.
A clear, simple background. Busy patterns tend to wobble.
A face turned roughly toward the camera if you want the person to stay recognisable.

Write a motion prompt

Describe what should happen, not what the picture already shows. Google's prompting guide for Veo 3.1 builds a prompt from five parts: cinematography (the camera), subject, action, context, and style and ambiance. In an image-to-video clip the subject and context are already in the photo, so most of your words go to the action, the camera and the sound.

For example: 'She turns toward the camera and smiles as the wind lifts her hair. Slow push-in, shallow depth of field. Soft city ambience and distant traffic.' One clear action per clip works better than a list of events; for a longer story, make several clips and join them. If the model generates audio and you want spoken words, put the line in quotation marks.

More ready-made examples for portraits, products and travel photos are in the image-to-video prompts collection.

Pick the model and settings

Models differ in length, resolution and whether they make sound. The Video tab in Luren AI lets you choose between these:

Veo 3.1 (Google): clips of 4, 6 or 8 seconds at up to 4K, with sound generated together with the picture, plus a first-and-last-frame mode. Strong at cinematic camera moves.
Kling (Kuaishou): photo animation up to 15 seconds, with an optional end frame and audio.
Seedance (ByteDance): quick, efficient animations from 720p up to 1080p.
Grok Imagine Video (xAI): clips from 1 to 15 seconds with native audio.

Fix common problems

Longer clips give the model more time to drift away from your photo, so start short and extend once a prompt works. For a closer look at the two most popular models, see Veo 3.1 vs Kling. The usual fixes:

The face changes during the clip: use a closer, sharper photo, ask for less movement and keep the clip short.
Hands or limbs warp: simplify the action and avoid fast gestures close to the camera.
Nothing much happens: use specific verbs ('spins around', 'raises a glass') and name the camera move.
The camera wanders: write 'static camera' or 'locked-off shot' if only the subject should move.
The motion feels rushed: ask for slow motion or a calmer action.

Ideas to try

Bring a portrait to life with a smile and a turn of the head, make a product spin slowly on a pedestal, add drifting clouds and moving water to a travel photo, or turn a still into a playful effect clip. Ready-made AI video effects do the prompting for you. If you animate a photo of someone else, ask them first: a realistic video of a person is still a video of them.