AI product video prompts: 8 shots for product photos

For a multi-shot ad, prompts are only part of the solution. Use the product-consistency workflow to preserve SKU shape, labels, color and materials across separate generations.

Open Magic Hour Image-to-Video, upload your own product photo, and paste one prompt below. Select duration, aspect ratio, and an available model in the tool. The prompts are starting points to adapt; the cover is an editorial illustration, not a set of measured model outputs.

Want to inspect recorded attempts before choosing a model? The commercial image-to-video benchmark includes the source images, fixed prompts and full output gallery for skincare, footwear, coffee and interior scenes. Its completion and credit figures are historical measurements, not quality scores or a promise that these recipes will pass your brief.

Choose a shot for the photo you have

1. Locked camera and moving light

Use for: packaging where the front label must remain recognizable. Start with an evenly lit photo in which the full label is legible.

Check: compare the label at the beginning, middle, and end. If lettering changes, keep the real product photo as a fixed layer in your editor and animate the background separately.

2. Gentle product push-in

Use for: a simple hero shot of a bottle, jar, or box. Leave room around the product so the closer framing does not cut it off.

Check: watch for a cap changing size, a label bending, or the object growing independently of the camera move. Reduce movement if these appear.

3. Sideways slide for shoes and accessories

Use for: a side-view product image with space on both sides. This prompt requests a small change in viewpoint, not a full rotation.

Check: inspect laces, straps, handles, and the sole or base. A single photo cannot establish what the hidden side of a real product looks like. Use additional real views or filmed footage for a full rotation.

4. Material and texture detail

Use for: an existing close-up of fabric, leather, wood, or another material. The source needs enough detail for the area you want to show.

Check: reject invented seams, changing grain, or texture that looks like a different material. Generation does not verify the product's actual construction.

5. Steam above a still drink or dish

Use for: a photograph of a product that is genuinely served hot. Avoid this treatment if it would misrepresent the item.

Check: the steam should be the change. Reject extra ingredients, a moving logo, a melting container, or a different portion. Use the clip to illustrate presentation, not to prove a product claim.

Recorded coffee example: source, prompt and output

This AI-generated, unbranded scene is a retained study input, not a customer product or a test of the eight suggested prompts in this guide. Download the original source image and download the LTX 2.5 clip (MP4), then play the complete clip to inspect the motion.

Recorded September 10, 2026 UTC through Magic Hour’s API: LTX 2.5, requested eight seconds, 720p, audio off. The file returned 1278 × 722 pixels at 24 fps; the final charge was 384 account credits. These are historical API-run settings and charges, not the guest offer or a current quote. Inspect all fifteen coffee attempts. Watch the bag, cup and beans throughout; completion alone does not establish an acceptable ad.

6. Small furniture pull-back

Use for: furniture already photographed in a room with enough surrounding space. A tight crop gives the model little evidence about the wider scene.

Check: compare leg count, straight edges, scale, and nearby objects. If the move reveals invented room details, use a wider source image or a fixed-camera shot.

7. Move the background, hold the product

Use for: a product with a simple background and a clean outline. Name the exact background element that should move.

Check: inspect the outline and translucent surfaces for background changes bleeding into the product. For strict packaging fidelity, compositing the original product over a separately animated background gives you direct control over that layer.

8. End-card hold with space for copy

Use for: a photo already composed with empty space beside or above the product. Add your final price, offer, and call to action in an editor after generation.

Check: make sure the generated motion leaves enough stable space for your message. Add readable text yourself rather than depending on generated lettering for an offer or legal disclosure.

Turn three shots into a simple product video

Use one establishing shot, one detail shot, and one end-card hold. For example: show a shoe with the sideways slide, show its material using a real close-up, then finish with the original product composition and a clear next step. Assemble the accepted clips in your video editor. Reuse the same product references so the shots describe the same item.

For the spoken message or on-screen sequence, choose a product-video script template. For choosing the source image, use the reference-image workflow. These solve different parts of the job: the script determines what you communicate; the shot prompt requests how an image moves.

The five checks before export

• Identity: shape, label, color, material, and visible features still match the real product.
• Continuity: inspect the whole clip for flicker, morphing, duplicate parts, and sudden camera changes.
• Truthfulness: the video does not invent a product function, ingredient, included accessory, or observed result.
• Delivery: the crop, readable copy, audio, and export match the destination's requirements.
• Cost: include rejected completed generations and editing time when comparing usable-output cost. Use the video-cost worksheet.

Do these prompts work with every model?

They describe a shot rather than an API request, so you can adapt them to an image-to-video interface that accepts an image and a text prompt. Results, controls, and supported inputs differ by model and provider. Follow the selected tool's documented settings; writing a duration or resolution in the prompt does not change the actual output setting.

Choose the camera movement you actually mean

A gentle push-in is not automatically a zoom. Runway’s camera-direction guide distinguishes a dolly, where the viewpoint moves and perspective changes, from lens magnification. For recipe 3, a sideways slide means the camera trucks parallel to the product; a pan swivels from a fixed position. An arc is a partial curved move, while an orbit travels around the subject. If one approved product photo shows only the front, keep the requested move inside the visible evidence rather than asking a full orbit to reveal an unverified back.

Keep camera controls and prompt direction consistent. When the selected tool offers a path control or motion preset, check it before copying a recipe. A locked-camera prompt paired with an orbit preset gives conflicting direction. Runway’s guide calls this double-directing: use one source of truth for the path, and use the prompt for the action, purpose or visual treatment. Where the interface provides only text, describe the move directly. This is a preparation check, not a promise of an exact camera trajectory.

Magic Hour's model reference documents supported output options. Runway's camera-direction guide explains common camera terms. Product access and source links were checked September 9, 2026.

What if the product must match perfectly?

Keep the real product photograph or filmed product as a fixed source layer and use AI for surrounding motion where appropriate. Inspect the composite too. A prompt asking for an unchanged label is a request to the model, not a guarantee of exact preservation.

To try one of these shots, upload a product photo to Magic Hour Image-to-Video and start with the simplest movement that serves your message.

For a current platform shortlist, compare the best AI video generators by inputs, output limits and production cost.