By THE UNKNOWN BRAND · Editorial standards

Published 2026-08-01 · Updated 2026-09-06

If the image makes the product more beautiful by quietly changing the product, it failed.

01Kuwait Army02Gulf Bank03Midana04Warba Bank05Dabdoob06Boursa Kuwait07Palmy08World Defense Show09Hyundai10Kellyville

In brief

• The checklist proves a file matches its approved reference. It does not prove the impression the image creates can be substantiated.
• Google's Merchant Center image requirements set an accepted minimum of 500 x 500 pixels, and recommend 1500 x 1500 or above.
• The Saudi Food and Drug Authority defines a claim as any representation, including pictorial, that states, suggests or implies a product has particular characteristics.
• Meta says it adds an "AI info" label when it detects industry standard AI image indicators, or when the uploader discloses.
• Arabic is cursive: each letter takes an isolated, initial, medial or final form depending on position. A model drawing letter shapes produces disconnected text.

What this checklist is for

This is the review standard we ask clients to apply to AI product imagery, including our own. It exists because the most common failure in generated product work is not ugliness; it is a beautiful image of a slightly different product. If the image makes the product more beautiful by quietly changing the product, it failed. A checklist does not, by itself, guarantee a usable image. But it replaces “looks fine” with a sequence of specific questions that anyone on the team can ask and answer.

Use it at four moments: when references are collected, when individual frames are shortlisted, when the approved set is assembled, and when files are delivered. Each stage catches errors the previous one cannot see.

Reference coverage

Most quality problems are reference problems discovered late. Before any generation begins, collect:

• Front, back, sides and top of the product, where relevant.
• Packaging artwork as a separate high-resolution file.
• Dimensions or a scale reference.
• Material and finish notes.
• A colour target photographed or specified under known lighting.
• A written list of immutable details and allowed simplifications.

The last item is the one teams skip. Without an agreed list of what may never change and what may be simplified, every review becomes a negotiation.

Immutable details are the parts of a product that may never change in a generated image. The guide asks for them as a written list, agreed alongside the allowed simplifications before any generation begins.

Frame review

When shortlisting individual frames, confirm that:

• Silhouette and proportion match the reference.
• Cap, seams, buttons, ports and closures remain in the correct place.
• Logo and label are composited accurately, usually from supplied artwork rather than trusted to a generator.
• Material response makes physical sense.
• Reflections, shadow and contact points agree with the scene.
• Product scale is plausible relative to surrounding objects.
• Food quantity, cosmetic effect or technical function is not exaggerated.

The checklist

Work through each row against the approved reference, not from memory. A row passes only when the reviewer can say why it passes.

The order is deliberate. Geometry and packaging text fail most often and are the hardest to repair after the fact, so they are checked first.

Contact shadow is the shadow where a product meets the surface it rests on. It grounds the product where it touches that surface; the shadows row names a floating product as the common failure.

Set review

Compare every image to the approved master, then compare the set to itself. The model may produce ten individually plausible versions that do not describe one product. Catalogue crops, background values, shadow density and spacing should also be consistent across the set. A practical method is to tile the shortlisted frames beside the master at the same zoom; an inconsistency that survives a tiled comparison will survive into the catalogue.

Approved master is the single product image signed off before a large batch is produced. Every frame in the set is compared to it, then the set is compared to itself.

Delivery review

At handover, check dimensions, colour profile, compression, alpha, filenames, alt-text source, crop safety and whether the supplied image can legally be used in the planned channel. Keep the approved reference and the final file list together, so the next person can repeat the comparison without reconstructing it.

The destination sets part of the spec. Google's Merchant Center image requirements tell you to "accurately display the entire product" with "minimal or no product staging". They also tell you not to use an image containing "promotional elements or content that covers the product". Watermarks and added logos are refused unless they belong to the product itself. The accepted minimum is 500 x 500 pixels; 1500 x 1500 or above is recommended.

Alt text belongs to handover as well. W3C accessibility guidance is that where an image carries text not present elsewhere, the alt attribute should carry that text. Someone has to write it from the artwork, not from the render. Read the channel's rules before the render queue, not after the rejection notice.

Clearing a platform's file rules proves the image is accepted for upload. It does not prove the image is accurate.

Alt text is the text carried in an image's alt attribute. W3C accessibility guidance is that where an image carries text not present elsewhere, the alt attribute should carry that text. Someone has to write it from the artwork, not from the render.

How to run the review

Give the checklist to one named reviewer per set rather than a group thread; groups notice mood, individuals notice caps and seams. Approve a single master product image before a large batch is produced, so that volume inherits an approved standard instead of multiplying an unapproved one. And where a platform, regulator or client policy requires AI disclosure, decide the disclosure before the images exist, not after.

The underlying rule does not change with the tools. An image may idealise lighting, setting and mood; it may not change what the buyer will receive. Food quantity, cosmetic effect and technical function are the three places that rule is most often bent. That is why they appear by name in the frame review above.

Regulators treat a picture as a claim. The Saudi Food and Drug Authority's general rules for product claims define a claim as any representation, "including pictorial, graphic or symbolic representation". They count a representation as a claim when it states, suggests or implies that a product has particular characteristics. Its first principle is blunt: no "overstatement, ambiguity, or deception". Nothing in that sentence depends on how the image was made.

Platform standards land in the same place. Meta's advertising standards forbid ads that use "deceptive or exaggerated claims about the success of a product or service". TikTok's advertising policy states that ad content "must not promise or exaggerate results concerning a product's effect". It names before-and-after product comparisons as a case it refuses.

Disclosure is a separate rule again. Meta says it adds an "AI info" label when it detects industry standard AI image indicators, or when the uploader discloses. That is the platform's decision, not yours. Yours is whether the client and the category regulator want a disclosure of their own, and in what words.

So the craft gate and the claims gate are two different gates. Everything above proves a file matches its approved reference. It does not prove the impression the image creates can be substantiated. Nobody at the retouching stage is in a position to decide that.

Claims gate is the separate review that asks whether the impression an image creates can be substantiated. The craft gate above it proves only that a file matches its approved reference. Nobody at the retouching stage is in a position to decide the rest.

Sources

Every source below was opened and its quoted wording checked against the live page. A source we could not re-fetch was dropped, not softened. Where a rule could not be confirmed from the body that issues it, this guide says so rather than describe it from memory. Editorial responsibility sits with the studio, not an individual author. Found a moved link or a wrong citation? Email hello@theunknownbrand.com with the URL. We will correct the page and its modification date.

1. Google Merchant Center Help — The delivery review and the new channel-rules paragraph: a destination platform sets file and staging rules that a generated image must clear before it can be used.
2. Saudi Food and Drug Authority (SFDA), The SFDA General Rules for Products Claim, published 2021/09/21 — The post’s underlying rule that an image may idealise mood but may not change what the buyer receives — a regulator counts a picture itself as a claim, and requires it to be truthful.
3. Meta Transparency Center — Advertising Standards — The frame-review line that food quantity, cosmetic effect or technical function must not be exaggerated; platform ad standards refuse exaggerated product claims.
4. TikTok Advertising Policies — Misleading and false content — Same exaggeration rule, stated as a visual rule: product-effect comparisons and exaggerated results are refused, which is why the checklist tests cosmetic effect and technical function by name.
5. W3C — Arabic & Persian Layout Requirements (W3C Group Draft Note, 02 October 2025) — The packaging-text row: why generated Arabic comes out as disconnected letters, and why artwork must be composited rather than drawn by a model.
6. W3C Web Accessibility Initiative (WAI) — An alt Decision Tree — The 'alt-text source' item in the delivery review: where an image carries text, that text has to be carried into the alt attribute.
7. Meta Newsroom — Our Approach to Labeling AI-Generated Content and Manipulated Media — The instruction to decide AI disclosure before the images exist: the platform may label the content whether or not you planned for it.

Frequently asked questions