How Brands Are Using AI Video for Marketing

Most coverage of this is either a vendor case study or a warning. The useful version is narrower: which categories of marketing video generation genuinely displaces, which it cannot, and where the legal exposure sits.

Where it is actually displacing spend

The pattern is consistent, and it is not "replacing film crews".

Stock footage. The largest and least discussed shift. A great deal of marketing video was already generic B-roll bought from a library. Generated clips do that job at lower cost, and nobody was emotionally attached to the stock clip.

Cutaways and transitions. Short pieces of motion between talking-head segments.

Concept and pitch material. Showing a client an idea before committing to production. Fast, cheap, and nobody expects it to be final.

Animating existing stills. Turning a product photograph you already own into a few seconds of motion for social. This is the lowest-risk category and often the highest return, because the product is untouched and the asset already existed.

Localised variants. Producing several versions of a background or setting without reshooting.

Where it does not help

The parts that were never tedious:

• Your premises, your team, your product actually being used. The value was the specificity, which is the one thing generation cannot supply.
• Anything requiring hands doing something. Still the most reliable failure.
• Anything with legible text. Smears in motion.
• Long-form. Clips remain short enough that a generated piece is a component, not a film.
• Lifestyle where the setting is the point.

The line that matters legally

One rule covers most of the exposure:

Never let a generative tool touch the product itself.

Background, lighting, motion, setting — all fine. Colour, texture, finish, size, or anything a customer will compare against what arrives — not fine.

Consumer protection rules in most markets prohibit misleading representations of goods, and those rules attach to the claim the image makes rather than to how it was produced. A photograph retouched to change a product's colour has always been a misrepresentation; nothing about generative tools changes that, and the ease of doing it accidentally is the new part.

The commercial cost arrives before the legal one: a customer who receives something that does not match the video returns it and often says so publicly.

Disclosure

Moving from courtesy toward requirement, at different speeds in different places.

What is already true: several jurisdictions, including Norway and France, require disclosure where a person's body has been digitally altered in advertising. Platforms including TikTok and Meta apply AI labels based on signals such as C2PA Content Credentials, which attach signed metadata at creation.

A workable internal policy, more conservative than current requirements and cheaper than being caught behind them:

• Generated background or setting behind real product: no disclosure needed
• Generated motion applied to a real product still: no disclosure needed
• A person who does not exist: disclose, prominently
• Any alteration to a real person's appearance: disclose
• A product that does not exist as shown: do not publish

The test is whether a viewer would be surprised to learn it was generated. Surprise is the thing disclosure exists to prevent.

Synthetic presenters

Worth its own note, because it is where brands most often misjudge the reaction.

Audiences respond badly to discovering a presenter is synthetic — considerably worse than to being told upfront. The format implies a person vouching for something, and finding out nobody did feels like a broken promise rather than a clever production choice.

If you use one, make it evident. A stylised, obviously non-real presenter carries almost none of this risk. A photorealistic one that viewers assume is a person carries all of it.

A worked example: one product, three assets

A small candle brand is launching a new scent. It has one good product photograph and no video budget. Applying the rules above:

Asset 1: the listing image. The photo was taken on a cluttered workbench. Replace the background with a plain warm surface and leave the candle untouched. Check the label colour and the wax tone against the real candle in daylight. No disclosure needed: the setting is generated, the product is real.

Asset 2: a short social clip. Animate the same still: a slow push in, the flame flickering. Reject any attempt where the label text has smeared or the jar has changed shape, which will be most of them. Keep the one where the candle looks exactly like the one in the box. No disclosure needed under the policy above.

Asset 3: a lifestyle image. The temptation is to generate a person relaxing by the candle in a beautiful living room. Under the policy, that person does not exist, so it needs a prominent label, and it is also the asset most likely to look generic. The better choice is a real customer photo, used with permission, or skipping the asset.

The two assets that used generation both left the product exactly as it is, and the one that would have needed a fictional person was dropped. That is the whole policy working on a real launch.

For assets 1 and 2, background replacement runs in the browser, and the photo to video tool animates the still.

Starting sensibly

1. Animate product stills you already own. Lowest risk, immediate use, product untouched.
2. Replace cluttered backgrounds on existing photography. See background replacement.
3. Check every output against the real product at full size, specifically colour and texture.
4. Write the disclosure policy before you need it, not during an incident.

The practical production side — shooting stills that animate well, and expecting a low hit rate — is in preparing for AI video. The equivalent guidance for small businesses without a marketing team is in AI photo tools for small business owners.

Photo to Video runs in the Kitana app and in the browser studio, next to several photo tools.