Photo to video AI: animate a still image you already have

Photo to video takes an image you already have and turns it into a short moving shot. The image is not a thumbnail or a background layer — it is the visual anchor. The model reads the subject, framing, colours and light from your file, and uses the prompt only to decide what should move and how the camera should behave.

That split is why the technique works well for real material. A photo of a product, a person, a room or a landscape already contains the details that are hardest to describe in words, and a sentence is very good at describing motion — a slow push in, a drifting cloud, a turning head. Put the picture in charge of appearance and the prompt in charge of movement.

The flow is short, and doing the cheap draft first saves both credits and time.

Pick a photo that has room for movement. A tight headshot gives the model almost nothing to animate beyond the face; a wider frame with sky, water, hair, fabric or foliage gives it something to move. If your image is a complex scene, say which part should move so the motion does not spread everywhere at once.

Sharpness and even lighting matter more than resolution, because a soft or noisy file tends to produce a soft or shimmering clip. Watch out for baked-in text and logos: the model re-renders the frame, and lettering that was crisp in your still can warp into something unreadable. For anything that has to stay pixel-exact, keep it out of the generated frame and composite it afterwards.

Because the source is images rather than a video, this is billed like a normal generation: per clip, by length and resolution. 720p and 1080p cost the same, and 4K costs more per clip. A new account receives 300 credits with no card, which covers 3 four-second tests or 2 clips at the default six-second length.

It is the cheapest way to add movement to material you already own. A catalogue photo becomes a product clip, a landscape shot becomes b-roll, a portrait becomes a short social post, and a concept image becomes something you can show someone rather than describe. Whenever the alternative is booking a shoot or a stock licence, animating the still is worth trying first.

It is the wrong tool when the shot must stay literally unchanged — a screenshot, a document, a diagram, or a packaging render that has to match a file exactly. Generative models reshape what they see, and that is the point, but it is also why they cannot be trusted with pixel-perfect requirements.

The first is describing the picture instead of the motion. The model can already see the photo; repeating its contents in words wastes the prompt and can drag the result away from your file. Spend the sentence on what changes between the first and last second of the clip.

The second is testing at full quality. A 4-second 1080p draft is the cheapest useful output; only move to longer or 4K once the motion reads correctly. The third is mismatching the aspect ratio: a photo composed for landscape will crop badly in 9:16, so choose the destination first and pick a source image that fits it, rather than generating a clip you cannot publish.