AI Product Video Generator: One Product Photo to a Finished Clip

A product video that sells is three shots, and most of them are five seconds long. The hero move shows the whole object doing one thing: turning, opening, pouring, lighting up. The detail shot goes close on the part that justifies the price: the stitching, the crystal, the texture of the cream. The context shot puts the product in a life: on a wrist, on a desk, in a hand on a sunlit kitchen counter. One product photo can seed all three, because each is a different prompt on the same start frame.

• Hero move: one motion, a slow camera, the product filling the frame, 5 to 8 seconds.
• Detail: a macro framing on one feature, light that rakes across it, 3 to 5 seconds.
• Context: a hand, a surface, a room, with the product doing what it is for, 5 to 10 seconds.

Render them separately rather than asking one clip to do all three. A model asked for an orbit, a macro and a lifestyle scene in one prompt gives you a muddle; asked for one at a time, it gives you three usable clips.

The clip inherits everything in the first frame: the light, the dust on the lens, the crooked label, the cluttered background. Fix the still before you animate it, because a bad frame one costs you every second after it. If you have a decent product photo, clean it. If you have none, make one.

Nano Banana Pro does both. It edits a photo you upload, so you can ask it to put the bottle on white marble by a window, remove the price sticker, or straighten the label, and it renders stills at 1K, 2K or 4K. For a new product shot from a description, start on our image generator page or the Nano Banana page, which have prompts to begin from. For a photo you want to keep but change, image to image walks through the edit.

The Editor's image modes handle the mechanical fixes. Background removes or replaces what is behind the product, Retouch paints out a scratch or a reflection, Resize extends the canvas when a square photo needs to be 9:16, and Upscale takes a small web image to a size that survives a full-screen phone.

• One product, centred, with room around it for the move you plan.
• Soft, directional light; the model will keep it, so choose it now.
• A plain or purposeful background. Clutter in the still becomes clutter that moves.
• The shape the clip will be: a 9:16 still for a vertical ad, 1:1 for a grid post.

Four video models cover every product job in this guide. Lengths, sizes and sound in this table come from our catalog, so they match what the studio lets you pick today.

• KlingKling 3.03 to 15 s · 720p, 1080p
• KlingKling V3 Turbo3 to 15 s · 720p, 1080p
• GoogleVeo 3.1 (with audio)4, 6 or 8 s · 720p, 1080p
• ByteDanceSeedance 2.54 to 30 s · 480p, 720p, 1080p

Shapes: Kling 3.0 and Kling V3 Turbo render 16:9, 9:16 and 1:1. Veo 3.1 renders 16:9 and 9:16. Seedance 2.5 renders those plus 4:3, 3:4 and 21:9, and it is the only one of the four that takes an end frame.

Open Kling 3.0 with a start frame, upload the still, and describe the motion rather than the product: the model can already see the product. Say what moves, how the camera moves, and what the sound is. These five moves cover most product categories, and you can run the prompts as written.

The macro row is the prompt family behind the Kling 3.0 4K preview, a slow push in on a mechanical watch movement lit by a jeweller's lamp; the 4K card renders the same lengths and shapes as Kling 3.0 and is the one to use for the final delivery when a client will view it large. For the unboxing, the template gallery has a ready unboxing prompt to start from. Hold the product still for the first second, give each clip one move, and render in the shape of the still. Our image to video page covers the general method, and the Kling page has more product prompts.

Kling 3.0 renders sound with the picture, on by default, and for products that means the action's own sound: the pour, the click of a clasp, the scrape of a lid. Leave it on for the hero and detail shots; it makes the object feel real. Switch it off, which lowers the cost per second, when the clip will sit under a voiceover or music.

For a spoken line, Veo 3.1 renders the voice, the lips and the room in one pass: put the words in quotes and name the space. For a voiceover over any clip, ElevenLabs v3 reads your script and takes direction tags for pace and emotion; our text to speech page has ad scripts to start from. For one cue, such as a cap twisting off, Sound Effects renders it from a description. And when a real face should present the product, Talking Avatar animates one portrait to your recorded line.

Marketplaces and ad networks often want the product on a plain plate, and some want the background removed entirely. Two routes. Render the plate clean in the first place: a still on white or on one surface, animated on Kling 3.0 with sound off, or on Kling V3 Turbo, which is silent, at 720p or 1080p. Or render the shot you want and strip it afterwards: the Editor's Remove Background mode runs the clip through Bria and returns the product on a transparent background, ready for a composite.

Two things to know before you strip a clip. Fine edges such as fur, mesh and glass are where any background removal struggles, so a plate rendered clean beats a plate cut out. And an upscale comes last: the Editor's Upscale mode lifts a finished 1080p clip to 4K with Topaz, sharpening what is there without inventing detail, so a soft draft stays soft.

Render the shape you will post rather than cropping. Kling 3.0, Kling V3 Turbo, Veo 3.1 and Seedance 2.5 all render 9:16 directly for Reels, TikTok and vertical ad placements, and all but Veo 3.1 render 1:1 for grid posts. A 9:16 still is the easiest start: if the product photo is square or landscape, the Editor's Resize mode extends the canvas first so the model composes for the tall frame.

Amazon, Shopify stores and other ecommerce marketplaces publish their own video requirements for listings, and they change, so check the current spec for length, shape and whether a plain background is required before you upload. The clips here come out as ordinary video files and drop into any listing tool or editor. For the short-form side, our Reels and TikTok roundup covers hooks, sound and captions.

Here is the route applied to six product categories, each from one still, with the move, the model and the prompt to run. Change the nouns and keep the structure.

Four of the six are Kling, three on Kling 3.0 and the watch on Kling 3.0 4K, because four of the six are objects in motion. The coffee bag is Seedance 2.5 because the shot wants 15 seconds and an end frame on the cup, and the headphones are Veo 3.1 because someone speaks. That is the whole decision rule.

Plan the route around these limits before you pay for a long take.

• Kling 3.0 renders 1080p at most and cannot land on a second image: deliver 4K on Kling 3.0 4K, and use Seedance 2.5 when the clip must end on a packshot.
• Kling V3 Turbo is silent by design, which suits a marketplace plate and not a finished ad.
• Veo 3.1 stops at 8 seconds, renders no 1:1, and tops out at 1080p here although Google lists 4K for it; it is for the spoken line, not the orbit.
• Seedance 2.5 renders up to 1080p, and a long take wanders when the prompt asks for more than one move.
• Nano Banana Pro edits stills, not clips. Fix the frame first; it cannot fix the video after.
• Background removal struggles on fur, mesh and glass edges; render the plate clean where you can.
• No model reproduces your product's logo or text perfectly in motion every time. Check the label on every render before you post.

Video cards bill by the second, so draft at the short end and render the final length once; the price a model starts from is printed on its card.

1. Clean or make the still on Nano Banana Pro, in the shape the clip will be.
2. Open Kling 3.0 with a start frame, upload the still, and write the hero move: what moves, how the camera moves, what it sounds like.
3. Render it short first, 4 or 5 seconds at 720p. If it misses, change one word and run again; you are not billed for a product clip that fails.
4. Render the keeper at the real length and size, then the detail and context shots the same way, one move each.
5. Unsure which model suits a move? Put the prompt on up to four models in Compare. It renders from the text alone, without your still, in 16:9 at each model's shortest length and smallest size, so treat it as a cheap read on the model, then open that model in the Video studio with your still as the start frame.
6. Add the voiceover from the Voice studio or a spoken line on Veo 3.1, strip the background if the listing needs it, and export the 9:16 and 1:1 cuts.

The whole route runs in one account: the still, the clip, the voice and the clean plate.