How to make an AI product demo video: 8-shot workflow

A moving packshot can introduce a product. A demo also needs to show what someone does with it. If the video skips that action, the viewer may understand the appearance without understanding the use. Start by deciding which action you can show accurately, then choose where AI-generated motion adds something useful.

Choose the format with the product-video examples, and use the product video script templates for narration structures. The steps below assemble a demonstration around one real action.

1. Choose one action the viewer should understand

Write a sentence with a person, task and observable endpoint. For example: “A desk worker opens, fills and closes this bottle, then puts it within reach.” For software: “A new user creates a project, uploads an image and finds the completed download.” These are useful demo briefs because each contains actions you can capture.

Separate visible behavior from claims that require other evidence. Showing a bottle being filled does not establish its capacity, durability or resistance to leaks. Showing a software result does not establish typical processing speed or a success rate. Include those claims only when you have the relevant product facts and evidence; do not ask a generated shot to supply them.

Decide the next step at the same time. A product-page demo might point to the item shown. A software tutorial might point to the exact workflow. That choice determines what the viewer must see before the final card: the product variant, the relevant starting screen, or the result they will receive.

2. Collect the proof before generating anything

For a physical product, photograph the exact item and record its real operation. Keep the package, label, attachments and color consistent with the item offered. Use an approved product photograph for a hero shot; film hands performing the key action rather than generating a fictional demonstration of the mechanism.

For software, record the current interface in a demonstration account with appropriate sample data. Capture the pointer or touch action and the resulting state. If an upload, permission or plan restriction matters to the workflow, show or explain it. Do not replace a missing feature with a generated button or an imagined dashboard.

Make a small source-to-message map. It prevents an attractive shot from acquiring a claim it cannot support. The following entries are a planning example, not evidence about a particular product.

Do not record every possible angle. Capture the action wide enough to understand, then add a detail only if the wide view hides something essential. For an interface, increase readability by recording the relevant panel at a useful scale, rather than shrinking a full desktop beneath decorative graphics.

3. Build an eight-shot plan around that evidence

This illustrative plan uses a hypothetical refillable desk bottle and a 30-second edit. It is a proposed storyboard, not a filmed or generated result. The timings add to 30 seconds but are adjustable; they are not a recommended universal runtime. Follow the actual product’s operation and care instructions when adapting the action shots.

The middle shots carry the demonstration. The opening and reveal earn attention; the action sequence explains use; the recap and end card orient the next step. If the edit needs to be shorter, trim repeated context before removing an action that makes the sequence understandable.

For a SaaS demo, keep the same logic and substitute actual screen capture for shots three through six: enter the workflow, perform the key action, inspect the resulting state, and show where to continue or download. Do not show four unrelated features simply to fill four slots. An app demo should retain readable controls and the result after the tap, even when cropped for a vertical layout.

4. Assemble a rough edit using the real assets

Place the proof shots on a timeline before ordering generated clips. Play the sequence without music or narration. Can someone see what changes from one step to the next? If the cap leaves the frame during opening, or a pointer clicks outside the crop, repair the capture instead of explaining around the missing evidence.

Keep an action and its consequence together. A cut can remove dead time, but it should not imply a different process. For software with a wait, use an explicit time jump or a brief waiting-state shot when needed. Speeding up the recording is an editing choice, not evidence that the product responds that quickly.

Add temporary words only where they help locate the step: “Open,” “Fill,” “Close,” or a real interface label. Reserve space for captions and the final CTA. This rough edit tells you whether you need a supporting visual at all, and how long that visual can be without crowding the explanation.

5. Generate one supporting shot from the approved photo

For an optional reveal clip, open Magic Hour Image-to-Video, upload the approved product photo and describe the desired movement. The current tool takes a starting image and an optional motion prompt, then lets you preview and download the video. Available duration, aspect ratio, resolution and audio options depend on the selected model.

Use generation where the shot does not have to prove a mechanism. A gentle camera move around a closed bottle can be a reveal; a generated cap opening can misrepresent how the cap works. Keep the real opening footage in the demonstration sequence.

Illustrative motion prompt:

This is a request to the model, not a promise of fidelity. Review the entire downloaded clip against the original photo. Inspect the cap seam, silhouette, label and color at the beginning, during the movement and at the end. If the image changes in a way that could misidentify the item, replace the clip with the approved still and a conventional editor move.

One photo-to-video generation supplies a clip, not a complete eight-shot demo. Assemble it with the real footage, fixed text and end card in your editor. If you need additional motion ideas, use the dedicated product-photo prompt guide instead of adding more generated actions to the proof sequence.

For supporting shot choices, see eight product-photo motion prompts. For a product that changes across generated clips, use the product-consistency workflow.

6. Put narration beside the action it explains

Write narration after the rough sequence is understandable. Each line should describe the visible step or give necessary context. Avoid introducing a new benefit over footage that shows something else. Record or import the voiceover, then adjust the shot holds so an instruction finishes while its action remains visible.

For the illustrative bottle plan, a restrained draft is: “Keep water close while you work. Open the cap. Fill according to the care instructions. Close the cap. Place the bottle within reach. See the product and care information.” Adapt these lines to the actual item; they do not assert capacity, safety, insulation or performance.

If the voice is racing through a necessary step, remove a redundant line or extend that shot. Do not make the action unreadably fast to meet the proposed 30-second allocation. Use music underneath the explanation, and review the video with captions when speech is essential. Keep any generated clip audio out of the mix if it conflicts with the narration or suggests an action that did not occur.

7. Keep exact text outside generated motion

Add prices, specifications, product names and offer details as controlled text layers in the editor. Use the current approved wording. A generated label or dashboard may look plausible while saying something different, so do not rely on it to communicate an exact fact.

The final card should identify the same product or workflow shown and give one clear action. For the bottle example, that could be “View this bottle and care instructions.” For software, it might be “Create an image-to-video project.” Check the destination before export, including the item variant and any conditions your narration mentions.

When adapting the edit for another aspect ratio, inspect the proof shots again. A vertical crop that cuts off the opening hand, interface result or full product label changes what viewers can understand. Reframe or rebuild the card rather than assuming a resized export communicates the same demonstration.

8. Review the exported demo with three concrete checks

First, watch with the sound off. Verify that the action sequence is visible, the product stays identifiable and the final destination is readable. For software, pause after the action and confirm the resulting state is legible. If the viewer needs the narration to guess what happened, change that shot.

Second, listen while watching the proof shots. Match each factual sentence to the source footage or approved specification. Remove an unsupported claim; do not leave it in because the surrounding visuals look convincing. Confirm captions and narration name the same steps.

Third, inspect the final exported file, including the joins between generated and captured material. Reject product drift, invented interface text, incomplete actions, cropped instructions or an end card that points somewhere different. A pleasant transition does not repair a misleading shot. Keep the simplest accepted version: real proof, enough context, and a precise next step.

Official sources and product references

https://docs.magichour.ai/get-started/starter-recipes

https://docs.magichour.ai/tools/video/image-to-video

For the distinction between a demonstration and a product introduction, see Shopify’s product-demo guide. The storyboard, timing allocations and bottle prompt above are illustrative editorial examples.