Advertising & measurement

How to make AI ad images and video without misleading customers

Plan a product shoot and AI edits, review the final assets against approved references, and check Meta, YouTube and Merchant Center disclosures.

Prepared with AI assistance. Source links appear in the relevant sections.

Content creator preparing video with smartphones and a camera

Use generative AI to explore concepts, storyboard a video or adapt a background around an approved product image. Use verified references for the item you sell, the people you show and any claim a customer might rely on. Before publishing, compare the final files with those references, record permission and check each channel’s disclosure rules.

Start with the claim the viewer will take away

Ask someone who has never seen the brief to describe what the ad promises. A scene intended to make a bag look good may also make it appear waterproof. A desk lamp without a visible cable may look rechargeable. A realistic clip of a spokesperson may imply that the person actually endorsed the product. Those impressions affect the decision to buy, even when the copy makes no explicit claim about water resistance, charging or endorsement.

That distinction gives you three practical production options. Use real footage where the ad needs to demonstrate something, such as a latch opening, a product fitting into a space or a feature working in real time. Use controlled AI editing where a verified subject can be isolated and the background can change independently. Consider a fully generated scene for a clearly imagined setting whose details do not carry the evidence behind the offer. A storyboard can include invented sets, temporary packaging and rough scale markers. Keep those files out of the handover folder used for paid placements.

The right method can change within a single spot. A generated background may establish the mood in the opening second. A close-up recorded with the real product can answer how it works. A manually composed end card can show the offer and destination without distorted type. Deciding shot by shot lets the team spend its time where the viewer actually needs proof.

Evidence to record before a decision
ObservationEvidenceWhat it does not establish
Product appearanceApproved product referenceGenerated detail is real
Person or voiceAuthorised usage scopePermission for every future use
Exported creativeFinal crop, disclosures and metadataPlatform acceptance or performance

Build a reference pack before anyone generates a frame

For a physical product, collect the approved front and side images, product variant identifier, visible label and a brief description of dimensions and materials. Add real footage of any action the ad will demonstrate. Ask the owner which features are fixed and which parts of the setting may change. In a studio scene for a matte product with two controls, for example, the surface finish and controls stay fixed while the desk, background colour and lighting direction may vary.

Do the equivalent for brand assets. Keep the approved logo, typography and end-card wording as editable layers in the design file. Reuse a verified caption or write a new one against the product record. A generated substitute for a logo can look plausible at full width and still become unreadable on a phone. Clear references give the reviewer a specific correction to make. “Restore the original two-line product label” is more actionable than “make it look more premium.” If the project is still at the planning stage, use a design brief that specifies what the team needs to deliver.

Check who can authorise each input and each use of the output. A recorded interview, product photograph, music track and voiceover can each have different usage terms. Record the allowed territories, period, placements and relevant editing permissions. If a new job involves synthetic voice or a recognisable likeness, check that the approval explicitly covers that use. A person who agreed to appear in one company video may never have agreed to have a synthetic version of their voice used in other campaigns.

Also check whether the chosen service may receive the underlying files. Review its terms and data handling settings against your team’s approval for that job. An unreleased package design or a recognisable customer clip deserves a different decision from a neutral mock-up. Use a placeholder for the initial experiment where it can answer the question just as well. Keep the final file’s origin, usage records and source reference together so another team member can establish what was approved later.

Walk through a desk lamp ad

Suppose an online shop wants a short ad for a wired lamp that comes in two colours. This is an illustrative brief. The reference pack contains both approved colour variants and footage showing the switch and cable. The opening shot shows the actual lamp. For the next shot, the team tests a light wood surface and a darker background using the same approved product image. It keeps the cable visible and uses real footage when the switch is pressed. The model name and call to action are added during editing, outside the generated scene.

View the vertical preview without sound. The product and the reason to pay attention should be apparent early enough to survive a quick scroll. Read the model name and captions at the size they will actually appear. If a model-generated reflection suggests a second light source in the lamp or the clip makes the cable vanish, fix that shot before comparing performance. Where the background keeps affecting the product, use the original shot with a simpler crop or reshoot against a clean surface.

Put the advertisement next to its product page. Do both show a wired lamp and the same colour option? Is the furniture in the background clearly separate from the offer? Can the shopper find the same model at the destination? When a test is intended to compare two background treatments, keep the product, offer and destination fixed. Agree on the outcome before launch and consider what visitors do on the site as well as clicks. The guide to interpreting ROAS for ecommerce advertising explains why one reported metric may leave important questions unanswered.

Check exports and placements, not only the edit

Inspect the final image alongside the reference product at a useful zoom level. Check edges, fastenings, label text, colour, texture and the number of parts. Where scale is important, compare against a real image that shows the object in context. Check the landscape, square and vertical versions independently. A disclosure or subtitle may be legible in a master file and hidden by interface controls in a placement preview.

Watch video at normal speed and step through any uncertain movement. An object should retain the same shape while the camera moves. Text and logos should remain stable between shots. Listen for added audio that changes the meaning of a claim, and read subtitles alongside the approved script. Use a muted preview to catch any sales claim communicated by the picture alone. Invite the product owner to check factual details and have the editor verify placement crops and caption timing. One review can catch both only if the person knows which source to use for each decision.

Keep a short release record for each version. Include the source image, edits made, rights evidence, placements approved, date and decision. Give the live files a version name that matches the record. If a problem is discovered after launch, you should be able to identify the affected variant and put a verified version in its place. Check other placements that reuse the same asset before treating the issue as resolved.

Review sequence for this task
  1. Select a real product reference
  2. Confirm rights and claims
  3. Review the final export
  4. Check the actual ad preview

Treat disclosures as a channel-by-channel check

Meta ads. Meta describes its “AI info” labels for ads created or significantly edited with certain Meta generative features. It also says it has started detecting some edits made with third-party tools using industry-standard signals. Review the finished ad and the available information in “About this ad.” The absence of a visible label tells you little about whether a file is cleared for use or whether the offer is accurate. Meta’s advertising transparency update

YouTube uploads. YouTube asks creators to disclose photorealistic content meaningfully altered or generated with AI. Its examples include depicting a real person saying something they did not say, altering a real place or generating a realistic scene that never happened. Check the relevant upload setting and review the advertising policy separately if the same video will run as an ad. YouTube’s AI disclosure guidance

Merchant Center product images. Google requires applicable IPTC source metadata on product images created with generative AI and tells merchants to preserve existing tags. Inspect the copy sent to the feed after resizing and compression, where that data can be lost. Separately, read the Google Ads misrepresentation policy when evaluating what the ad and the destination tell a prospective buyer. A platform disclosure and an accurate offer are two independent checks. Merchant Center AI-generated content rules · Google Ads misrepresentation policy

Scope the work around approved deliverables

Ask for estimates against a shot list and an export list. A studio may generate numerous drafts while only two still images and one short vertical edit enter approval. Define which items need filming, which existing images can be reused, and how many finished crops, subtitle languages and revision rounds the team needs. Agree on whether the client receives raw footage, a source project or only finished files. Those terms affect the cost of revising the campaign later.

The schedule depends on when products arrive, whether references are complete and who signs off on factual claims and asset rights. Generating the first backdrop may be quick. Correcting product edges across several cuts, keeping captions legible and checking whether every variant matches the store listing can take longer. A short shoot is worth pricing if it supplies reliable shots for several future ads. Ask the producer to identify what can be approved before shooting and what needs a physical sample in hand. That often tells you more about the turnaround than a list of tools.

A small team can start with one approved product, one verified claim and two deliverables. Send the reviewer the exact copy, the product reference and the placement previews together. Ask them to mark “approved,” “revise” or “hold” for each format and state the reason. If the copy is still waiting for sign-off, use the session to assess the visual only and keep the file out of the campaign handover until both are approved.

Talk through your shot list with WeAreMedia

Bring the product you plan to promote, the channels you want to use and the evidence the ad needs to show. We can discuss which shots would benefit from a shoot, where existing photos are enough and which formats need individual review. See photo and video production services for the production scope.

Call to discuss your product shoot, describe the footage and formats you need on WhatsApp, or open the short production enquiry form.

Related guides

Apply this review to your website

Share your website and priority business goal. We can identify the missing information, technical checks and next step.