Product explainer from a spec sheet and your own photos: a script and claims-table method

This page is about one product. If you have hundreds of items and need to decide what to film, generate or skip, read the catalogue guide instead. If your worry is that a label or logo warps when a photo becomes video, read turning a product photo into video. Here the question is different: how do you get a clear, short explainer out of a spec sheet without the video saying something false? I use Higgsfield, a web service that offers many image and video models in one account, as the AI example, and I say where a phone and a tripod do better.

This page may contain affiliate links from which Blur Studio earns commissions.

Short answer: the spec sheet is the script, the photos are the truth

Most failures in AI product videos are not bad pictures. They are quiet errors: a wrong number, a feature the product does not have, a button in the wrong place, a claim nobody can back up. A video tool that writes the words for you can invent benefits. So flip the process. Start from the facts, write only what the facts allow, and prove each line. The video is an advertisement, even if a tool wrote the words, so the person who publishes it is responsible for it.

Freeze a source pack

Before you write a word, put these in one folder and stop changing them:

• The current spec sheet, as PDF and as plain text.
• Your own product photos: front, back, sides, label close-up, in-use shots. Photos where the label and shape are clean.
• The product name, model code and price exactly as they appear on the product page.
• Any test report or certificate behind a claim, such as a water rating or a battery figure.
• A list of phrases that must not be rewritten (a model name, a legal line).

Note the date. If something changes later, you will know which version the video was built from.

Pick 3 to 5 specs and turn each into a benefit

A spec sheet has dozens of lines. A viewer remembers three or four. Pick the ones that matter to the buyer, then ask "so what?" until you reach an outcome. Vendor guides on explainers give the same advice: choose a few specs and state what each means for the customer. Worked example, with made-up numbers:

Spec (from the sheet)	So what?	Safe sentence
Capacity 20 L, internal measurement	Fits a week of lunches	Only if the sheet says 20 L; do not claim lunches you have not measured
Weight 1.2 kg	Easy to carry daily	"Weighs 1.2 kg" is literal; "easy to carry" is an opinion, keep it light
Rated IPX4 (splash resistant)	Survives a rainy commute	Say "rated IPX4" and what that rating means; never "waterproof" unless the test supports it

Keep the sentence tied to the spec. Do not borrow a benefit that sounds right but is not in the sheet. A claim like "charges a phone three times" is only safe if the maker's own test supports it.

Write the script and a claims table

Use a simple structure: the problem, the product, three benefits, one call to action. Short lines. For a 30 to 60 second video, that is roughly 75 to 150 spoken words. Pacing of about 150 words a minute is a common convention, not a rule, and the claim that explainers of 60 to 90 seconds perform best is repeated by agencies without a source. Your own audience decides. Shorter is easier to check.

Now build the claims table. One row per claim, whether it is spoken or shown on screen:

#	Claim as it appears in the video	Spec cell or evidence	Visual on screen	Checked by
1	"Weighs 1.2 kg"	Spec sheet, row Weight	Photo of the product on a scale or in hand	Name and date
2	"Rated IPX4"	Spec sheet plus test certificate	Photo of the rating mark on the unit	Name and date
3	"From $49" (made-up price)	Product page, today's date	Price card, not spoken	Name and date

If a row has no spec cell or evidence, the claim goes. If a claim says "best", "fastest", "waterproof" or "lasts all day", it needs evidence you can show to someone else. The Federal Trade Commission (FTC, the US consumer protection agency) says in its small business advertising guidance that ads must be truthful and not deceptive, that you need a reasonable basis for each claim, and that advertising agencies can also be liable and cannot simply rely on the advertiser's word. That is information, not legal advice, and other countries have their own rules, so check the rules where you sell.

Tools help but do not verify. An explainer tool such as invideo (an online video maker) says it reads pasted text, not Excel or PDF files, and that you should check the benefit claims it generates. Another vendor guide puts it well: narration is not verified by the tool, and fixing frames does not fix narration. The claims table is how you check it yourself.

Plan scenes from your own photos

Make a scene list with one scene per claim, and put each spec in one scene only. That way, when one spec changes, you redo one scene and not the video.

Scene	Seconds	Line	Visual	Source photo
1 Problem	4	One-sentence problem	Real photo, slow push-in	In-use photo
2 Product	4	Name and what it is	Clean front photo	Front
3 Benefit one	6	Weight claim	Hand-held photo	Side
4 Benefit two	6	Rating claim	Label close-up	Label
5 Benefit three	6	Capacity claim	Open view	Open
6 Call to action	4	One action, one address	Product plus address as text	Front

For more on movement from a single photo, see multiple angles from one photo.

What AI may add, and what it must not

AI may add	AI must not add
Camera moves: slow push-in, pan, orbit	A different product, shape or colour
Light and shadow changes	Buttons, ports, a screen or an app interface
A background or mood scene around the product	Text on the product or label that was not there
Abstract opening shots and B-roll	A feature, accessory or size that is not in the sheet

Kling's own explainer page tells users to verify diagrams, charts and technical labels against the source because generated text may be wrong, and to use dedicated graphics tools when exact detail matters. A generated interface is not proof that the product works that way. Show the real thing. Ask the model for motion, not for a different object. If a result changes the product, throw it away; do not hope nobody notices. If your product is software, the truth layer is the real screen, and the software explainer guide covers it.

Voice, captions and one call to action

Record the narration yourself or generate it from the approved script. Use the script as the caption source, then check numbers, units and the model code letter by letter. Order matters: voice and music first, clips after, captions last. I explain that order in captions, voice and music for AI clips. End on one action: one address, one button, one phone number. If the video will go on a marketplace or TikTok Shop, check what each requires; see TikTok Shop and Reels product clips and marketplace product images.

The three-layer accuracy check

A buyer's tool review splits accuracy into three layers, and the split is useful. Check each layer separately, and mark it pass or fail.

Layer	Question	How to check	Fail means
1. Asset fidelity	Is the product on screen the real one?	Put frames next to your photos: shape, colour, label, ports, proportions	Redraw detected; regenerate or use the plain photo
2. Information fidelity	Are names, units, prices and codes literal?	Read every on-screen word and every spoken number against the claims table	Fix the text; regenerate nothing
3. Correspondence	Is the right visual up when a feature is named?	Watch with the script in hand: at each claim, does the picture show it?	Re-cut the scene

A final pass: ask someone who does not know the product to watch it and write down what they think each feature is. If their notes disagree with the sheet, a layer failed.

Test with hard cases before the real video

Before you spend credits on the real video, run a test script through your chosen tool. Put in four things that tools often mangle: a model code (for example AB-1234-X, made up), a number with a unit (1.2 kg), a price ($49) and a phrase that must not be rewritten. Check what comes out. If the tool changes any of them, you have learned its failure rate on your kind of data for the cost of a short test. Use the same four items every time you try a new tool.

Updating when a spec changes

Specs change: a new battery, a new price, a new rating. Because each spec lives in one scene, you change the claims table row, regenerate or re-edit that scene and re-run the three checks on that scene only. Keep a version note: date, what changed, who approved it. Take down an old version if its claim is no longer true.

Where Higgsfield fits

Higgsfield is a reasonable place to turn your real product photos into short moving shots: a slow camera move, a lighting change, a mood background. Its Marketing Studio is built for product ads, described in the Marketing Studio guide. Several of its video tools do not publish a credit price and show the cost on the button, so work out the cost per approved scene: clips times retries times credits per clip, with the blanks filled from your own account. See how to spend credits wisely.

It is a weak fit for a spec-heavy product where exact detail is the whole point. It does not read your spec sheet and verify the words, and the model can redraw small features. The terms also say you must hold the rights to what you upload, so use your own photos or ones you have permission for; manufacturer photos need the manufacturer's agreement.

Who should skip Higgsfield and film instead

• Your product has to be shown working: a hinge, a pour, a fit. A 60 second phone demo or a screen recording beats AI for how an item works.
• Label or print accuracy is critical, as with food, medicine, chemicals or regulated goods. Use a product photographer or plain photos.
• You want a quick slide-style explainer from text. A script-to-video tool such as Pictory (which turns scripts into videos) or invideo is quicker.
• You cannot show evidence for the main claims. Fix that first.
• The product is a software screen. Capture the real screen.

I would use AI for the opening shot and abstract B-roll and film everything that proves a claim.

How I'd approach this

I would start with one product and one spreadsheet: the source pack, the three to five specs, the claims table and the scene list. Write the script from the table, not the other way round. Run the hard-case test on the tool. Make the scenes from real photos, add the voice, then run the three checks and have a second person sign the claims table. That table is also your record if anyone asks where a number came from.

If you have more than one product, Blur Studio can build the shared source pack, claims table and approval step so each new video starts from the same checked facts. And if you want to try the AI motion part, you can test Higgsfield on one product before you commit to anything.