Your product photo is ready to be a video. You just haven't asked it to move yet.

Most brand teams have the same folder: hundreds of good product photos, used once or twice and then left alone. Meanwhile social, display, and paid all want motion, usually in a few different formats.

A reshoot isn't always realistic. The good part is that it isn't necessary.

If you have a product photo, you can have a product video in about a minute. Here's how to do it so the result actually looks like your brand.

AI video models are eager. Ask for a little movement and they'll happily redesign your label, drift the camera, and add a lens flare nobody wanted.

So every good prompt does three things, in this order:

1. Name the one thing that moves. Be specific. Not "add motion"—"the dropper lifts slowly and settles back."

2. Lock everything else. Say it out loud. Camera, background, lighting, label.

3. Forbid what tends to creep in. Lens flares, particles, smoke, camera push-in, redrawn text.

That's the pattern that applies to most effective prompts of image and video models.

Start here. It's the highest hit rate of anything in this article.

Copy this:

Animate only [element]—it [specific motion]. Everything else in the frame stays completely still.

The camera does not move: no zoom, pan, tilt, parallax, or reframing. Keep the background, surface, and surrounding objects fixed and unchanged. Keep lighting, color, shadows, and brightness constant in every frame.

Do not add spotlights, light sweeps, lens flares, particles, smoke, or floating shapes. Do not redraw or restyle the image.

Filled in:

Animate only [the leaf beside the bottle]—it [sways gently once, as if someone just walked past]. Everything else in the frame stays completely still. The camera does not move: no zoom, pan, tilt, parallax, or reframing. Keep [the bottle, the surface, and the shadows] fixed and unchanged. Do not add lens flares, particles, or floating shapes.

Pro tip: pick the smallest thing in frame that could plausibly move. A sprig, a dropper, steam, a hand. Small motion reads as premium. Big motion reads as a screensaver.

Looping clips are the workhorses—they run on a product page, a story, a display unit, a screen in a store, and nobody notices the restart.

Copy this:

Create a seamless looping animation from this image. The first and final frames must match exactly—no jump, fade, or object displacement.

Animate only [element], completing one full cycle that returns precisely to its starting pose and position.

Camera completely locked. Maximum 5 seconds. No audio.

For a rotating product, this exact script is tested and reliable. Paste it as-is:

Locked static camera, no zoom, no pan, no cuts. Starting exactly from the reference image, the product rotates smoothly around its own vertical axis at a constant speed, completing exactly one full 360° turn, so the final frame matches the first frame exactly for a seamless loop.

Background, lighting, shadows, and scale stay completely still. No new elements or graphics anywhere in the frame. Duration: 5 seconds (0:05).

Pro tip: name your label elements individually—[the wordmark, the product name, the finish]. Models protect what you name and improvise on what you don't.

Pro tip: add more reference images. One photo is enough to get motion, but a second or third—a different angle, a detail shot, the label straight on—gives the model more to work from and less to invent. It's the fastest fix for a label that keeps drifting.

Don't rewrite from scratch. Reference the clip you got and subtract.

Copy this:

Same framing and composition as [@Video1]. Keep [what you liked] exactly as it is.

Remove all [wiggling / bouncing / drifting / camera movement]. Once elements have appeared and settled, they hold completely steady for the rest of the clip.

No camera movement, no zoom, no lighting changes.

This is the prompt most people skip, and it's the one that saves the most time. A second pass that subtracts almost always beats a first pass that starts over.

Shorter is usually better. Five seconds of controlled motion beats ten seconds of drift. Longer clips give the model more room to invent.

Specific beats poetic. "Cinematic" and "dynamic" mean nothing to the model. "The dropper lifts two centimeters and settles" means something.

Your first prompt won't be your last, and that's fine. Two or three passes is normal. Eight means the prompt is too vague—go back and get more specific.

Start with your best photo. AI video amplifies whatever it's given. Soft focus in, mushy video out.

Try a different model when you're stuck. They behave differently, and switching is often faster than rewriting. See the FAQ below for which to reach for.

Can I turn a product photo into a video without a shoot?

Yes. Upload the photo to Canvas, select it, and describe the motion you want. No studio, crew, or reshoot required.

Why does the AI keep changing my product?

Because you didn't tell it not to. Name the elements you need preserved—label text, logo, color, proportions—and state explicitly that everything except the moving element stays identical. If you're still having trouble, try a new model.

How long should an AI product video be?

Three to five seconds for social and display. Short clips hold detail better and loop more cleanly.

Can I make an AI video loop seamlessly?

Yes. Ask for matching first and final frames, and have the motion complete one full cycle back to its starting position.

Which model should I use?

Try more than one—they behave differently, and switching is often quicker than rewriting a prompt. MiniMax H3 Max returns outputs fastest, so it's the one to use while you're still dialing in the wording. Seedance 2.5 supports longer videos, more reference images, and can honor more prompt guidance, so try this for more specific outputs, but just remember it also takes longer.

Credit costs per model change as providers reprice them, so check the model picker in Canvas for current rates.

What's the most common mistake?

Vague prompts. "Make this move" gives the model permission to move everything. Name the one thing that changes and lock the rest.

Open any product shot in Canvas, paste the first prompt, and change one word.