AI Product Video Generator

Product video used to mean a studio day, a lighting kit and an editor. Now the shortest path is a product photo, a written shot description and a model that can hold the subject still while the camera moves. This guide covers the full path on VidVid: which model to pick, what a prompt needs to say, where generations break, and how to budget credits before you commit to a batch.

Who this is for — and who it is not for

Product video generation works best when the product already exists and you are adding motion, not inventing it.

• A good fit: e-commerce listing clips, paid social hooks, launch teasers, lookbook motion, app and SaaS promos, A/B creative variants, macro detail shots.
• A poor fit: anything that must show exact on-screen text (price tags, legal disclaimers, ingredient lists), regulated claims, or a packshot where the logo must stay pixel-perfect. Render those elements in post instead.
• Never do this: let a model invent product specs. Area, price, ingredients, certifications and addresses belong in your copy, checked by a human.

The workflow that survives contact with a real catalogue

1. Fix one measurable outcome. Decide whether the clip has to earn a click, show material detail, stay faithful to a reference photo, or fill a reusable social slot. This single choice determines the model and the duration.
2. Prepare the still. Centred subject, clean or plain background, no harsh reflections, edges sharp. A weak input photo is the most expensive problem you can have, because you pay for every retry.
3. Write the shot, not the mood. Subject + motion + camera + lighting + format. “Slow macro push-in on a ceramic bowl, studio black background” beats “beautiful premium product shot”.
4. Test short and cheap. Use the shortest supported duration for drafts at the lowest suitable supported resolution first. Lock subject stability, camera speed and lighting before you raise duration or resolution.
5. Generate per-shot, not per-video. Give each shot one purpose, choose a supported duration and assemble accepted shots in editing.
6. Finish outside the generator. Subtitles, logos, price stickers, music and end cards are editing work. Keep them out of the prompt.

Keep labels, logos and product geometry trustworthy

A product clip is usable only when its required facts survive the entire shot. Before generating, classify the shot by what a buyer must be able to verify. A prompt asking for unchanged text is a request, not proof that the result preserves it.

Required evidence	Source to prepare	Accept only when	If it fails
Brand or ingredient label	Sharp photograph of the actual package and approved artwork	Every visible letter and mark agrees with the source throughout the shot	Use a real packshot or tracked artwork in external editing; check occlusion and perspective
Shape and accessory count	Photo of the exact SKU, including cap, handle and included accessories	Silhouette and count stay correct, including during motion	Reduce the movement once; if still wrong, use real footage
Back, interior or opening mechanism	Real photograph or footage of that view	The depicted structure is supported by that source	Obtain the missing view; a front photo is insufficient
Texture or finish	Close-up under relevant lighting	Material detail remains recognizable without invented seams or patterns	Replace the reference or use a filmed close-up
Offer, dimensions or feature claim	Approved product information	Copy matches the approved information and remains readable in the final crop	Add verified copy in editing and review the final export

Do not repair a wrong bottle shape by adding a correct logo over it. A label overlay addresses text, not the underlying geometry. Likewise, a generated rotation cannot establish a view that was never photographed.

A review worksheet and a stopping rule

Copy this record for each shot. It is an editorial review template, not a new measured benchmark.

Field	Record before or after generation
Product identity	SKU, source-photo filename and the version of approved product copy
Purpose and invariant	One shot purpose; details that must not change, such as cap count or label spelling
Generation setup	Selected model, input mode, supported duration, resolution, aspect ratio and displayed credit quote
Attempt limit	Maximum attempts and total credit ceiling agreed before submitting
Review result	Accepted or rejected; timestamp of each defect and the source detail it contradicts
Next action	Change one variable, replace source material, use real footage or finish in editing
Handoff	Accepted file, actual charges, caption/artwork version and final reviewer

If the first result fails, name the defect before trying again. For motion-related shape drift, try one simpler camera move with the same reference and supported settings. If the same required detail still fails, or the agreed budget is exhausted, stop that configuration. This is a suggested budget rule, not a claim that two attempts are sufficient for every product. Missing evidence, such as an unseen back panel, calls for new source material rather than more retries.

Review the exported clip at its delivery aspect ratio, including frames where an overlay crosses the product. External editing and this worksheet are user-managed steps; they are not automated VidVid quality checks.

Sources and limits of this update

Reviewed October 8, 2026: Runway’s official Image-to-Video Prompting Guide explains that the input supplies visual composition and that motion prompts can be refined incrementally; it also notes that input defects can carry into video. That guidance is specific to Runway’s documented models and does not establish capabilities for every model in VidVid.

The decision table and worksheet above are our proposed product-review workflow. No new paid generation, label-fidelity measurement or cross-model benchmark was performed for this update. The historical H3 files below retain their original scope and do not validate label accuracy.

Picking a model for product video

Filter for image-to-video support and your required output settings, then compare a small number of runs using the same product photo and shot description. Prices at different settings are not a quality ranking.

Record silhouette, colour, label accuracy, background stability, camera motion and actual credits. One sample does not guarantee performance on your product. Check the model directory and current pricing, then confirm settings and credits before generation.

Prompts you can paste today

These are untested editorial prompt examples. Adapt the subject and motion; select duration and aspect ratio in the supported model controls. Written constraints do not guarantee fidelity.

PROMPT 1 — Clean product reveal (image → video)
Camera slowly pushes in on the ceramic bowl from the reference image. Mango cubes stay sharp, pulp texture unchanged, soft top-light, matte studio background, shallow depth of field, 5 seconds, 16:9, no camera shake, product does not deform.

PROMPT 2 — Vertical social hook
Energetic handheld move, product enters frame in the first second, bright accent lighting, clean background, 3 seconds, 9:16, subject stays centred and readable at thumbnail size.

PROMPT 3 — Material macro detail
Extreme macro of the fabric weave, very slow lateral drift, raking side light reveals texture, no change to colour, 4 seconds, 1:1, steady lateral movement.

PROMPT 4 — Packaging turntable
Product rotates gently on a clean white surface, soft even studio lighting, subtle contact shadow, packaging text remains legible and undistorted, 5 seconds, 1:1.

PROMPT 5 — Lifestyle cutaway
Product sits on a sunlit wooden table, steam rises, shallow depth of field, warm practical light, camera holds still, 4 seconds, 16:9.

Test cards: three shots to run before you scale

Run these three before committing a catalogue. Each card is a suggested shot brief — the same inputs, the same settings, the same checkpoints.

These three suggested test cards cover a product loop, a vertical opening and a material detail. Use your own authorized product assets, supported settings and explicit acceptance rules. They are planning suggestions, not the exact prompts used for the historical samples below.

Historical MiniMax H3 samples and their actual prompts

The three existing samples below use different prompts: a ceramic-object push-in, a vertical text-to-video hook and a brushed-metal text-to-video detail. The prompts and output settings are listed beside each file. They are single historical runs, not a controlled model comparison or a guarantee for your catalogue. The original image for sample A is not included here, so this page alone cannot reproduce that run exactly.

Open this clip in a new window →

Prompt we sent (as run): A slow macro push-in on a matte ceramic object, realistic reflections, black studio background, premium commercial lighting, the object keeps its exact shape, no deformation, 5 seconds.

Open this clip in a new window →

Prompt we sent (as run): Vertical 9:16 product hook, energetic but controlled camera move, subject fully visible in the first second, bright accent colour, shallow depth of field, object shape stays fixed.

Open this clip in a new window →

Prompt we sent (as run): Extreme macro of a brushed metal surface, slow lateral camera move, reflected light crawling across the material, no distortion, commercial product photography look.

Run	Mode	Settings	Output file	Submit → ready
A	image → video	5s · 768P	1344×768, 24 fps, 5.2 s, 726 KB	196 s
B	text → video	5s · 768P · 9:16	768×1344, 24 fps, 5.2 s, 1.0 MB	117 s
C	text → video	4s · 768P · 1:1	768×768, 24 fps, 4.5 s, 341 KB	83 s

Scope: three single runs on one model. We did not measure the credit deduction inside VidVid, we did not time any other model, and we did not score output quality — that stays a human call. Check the current credit quote before submitting; these recorded runs do not establish today’s price.

How to compare another model

Keep the input and shot goal unchanged. Match duration, resolution and aspect ratio only where both models support them; record any differences so the results are not presented as a controlled comparison. Check the current quote, set an attempt limit and review every result against the same criteria. One output does not establish a general quality ranking.

Choose supported settings in the VidVid studio. No other model was timed in the sample runs on this page.

Troubleshooting product distortion

Compare the output with the source image for altered handles, caps, labels and silhouettes. Input quality, model settings and requested motion can all affect the result.

1. Check that the reference clearly shows the required details and angle.
2. Test a shorter supported duration or simpler camera move, changing one variable at a time.
3. Use the actual product image when image input is supported; it is a reference, not a guarantee of accuracy.
4. Split detail and motion shots, then assemble them in editing.

If the same defect persists at your attempt limit, stop and use a new photograph, real footage or another supported configuration. Do not publish invented product details.

Budgeting: what a product video actually costs

Record the current quote for each shot and multiply it by your maximum paid attempts. Three shots, two attempts each and a hypothetical quote of 100 credits gives a 600-credit ceiling. This is an arithmetic example, not a model quote or a promise of accepted clips.

Count accepted clips and actual charges rather than applying a universal retry multiplier. Validate using shorter duration and lower resolution where supported, then recheck the final quote. See the pricing and accepted-clip cost guide.

Product accuracy checklist before publishing

• Check silhouette, colour, quantity, labels and accessories against the source photo throughout the clip. Reject altered details.
• A front-view photo does not establish the back or internal structure. Supply real images for those angles; do not present invented details as product facts.
• Add and verify captions, prices, trademarks and feature claims during editing.
• Play the full clip at the target aspect ratio to check cropping, jitter, transitions and caption coverage.
• Save the input, prompt, settings and review notes before scaling.

Mistakes that quietly burn credits

• Generating the final resolution first. Use a lower supported resolution for draft review.
• Asking for text in the frame. Add price, titles and captions in post.
• Retrying a multi-action shot without isolating the action that fails.
• Reusing one input photo for every angle — shoot or source a second and third angle.
• Publishing a model’s invented detail as a product fact.

FAQ

Can I turn one product photo into a video?

You can use image-to-video to add motion, but shape, colour and packaging may still drift. Review against the source image throughout the clip; add exact labels and prices in editing.

Which model should I start with?

Choose a model that supports your input and required output settings. Compare the same product image and shot brief at a supported draft setting; check the current credit quote and inspect product details before selecting a final render.

How long should a product clip be?

Use the shortest supported duration that communicates the shot. Test one action per clip and assemble the listing video in editing; inspect the full result before accepting it.

Why does my product change shape?

Shape drift can depend on the input, motion, model and settings. Compare a simpler shot with the same reference image, changing one variable at a time. Written constraints cannot guarantee product geometry.

Is the output commercial-use ready?

Check the current service terms, the selected model conditions and your rights to the source images, artwork and music before commercial use. A generated file alone does not establish those rights.

Can I see real MiniMax H3 output before I spend anything?

Yes. Three complete H3 runs are published on this page with their prompts, settings and the MP4 files themselves — 768P, 4–5 seconds, 83–196 seconds from submit to file ready. These samples are not a controlled comparison of models. Check the current quote before testing the same brief with your own product image.

Where to go next

• Open the VidVid studio and run Card A with your own product photo.
• Compare models side by side on the AI video model comparison page.
• Read the AI video pricing and credits guide before a batch run.
• Making vertical cuts? See the vertical video generator guide.

Troubleshoot the output

Use the AI video artifact checklist to isolate one defect before generating again.

Plan a three-shot product video before generating

Shot	Job	Reference and acceptance rule
Opening	Show what the product is	A real product photo; subject is recognizable in the intended crop
Detail	Show one visible feature	A close-up of that actual feature; texture and shape remain faithful
Closing	Leave room for the offer	A clean composition; add verified price and call to action in editing

These are shot-planning suggestions, not tested performance claims. Choose the shortest supported duration that communicates each shot, quote each separately, and set a retry limit. A single front view does not establish the back or inside of a product. Supply another real photo or filmed detail if those views matter.

Start in the image-compatible model directory, use the cost worksheet, and consult the artifact checklist when a named defect blocks acceptance.