AI-Generated Image Review: A Checklist for Your Business

An AI-generated image review should begin with accuracy, not realism. A convincing picture can still show the wrong product or suggest a service your business does not offer.

The most important question is: “What would a customer reasonably believe after seeing this?” A beautiful workshop image might imply a real location. A small change to a product’s shape might imply a feature. Those errors can matter more than an obvious visual glitch.

Give the review a named owner and an agreed source of truth. The process below is a practical editorial check, not legal clearance or a guarantee that every risk has been removed.

First, decide what kind of image it is

An image can play several roles:

• Product evidence: showing the actual item someone may buy.
• Process evidence: documenting how work is done.
• Illustration: explaining an idea or setting a mood.
• Concept: exploring a direction that is not yet a finished offering.

These roles have different standards.

A conceptual café interior should not be presented as a photograph of your real location. An illustrative package should not appear to show the packaging a customer will receive.

Write the intended role in the brief. If the finished image could reasonably be mistaken for evidence, clarify it in the presentation or choose a more accurate asset.

Start the AI-generated image review with product accuracy

Compare the image with approved source photographs and specifications.

For a product, inspect shape, proportions, colour, material, labels, controls, connectors, packaging and included accessories. Check every repeated instance of the item.

For a service, inspect the setting, equipment, uniforms and implied capabilities. A generated image of a specialist machine can imply that the business owns or uses it.

Use a simple decision table:

Do not approve an image simply because most details are right. One incorrect feature may be the feature that motivates a purchase.

Inspect the whole scene

Next, look for visual problems at full size.

Check hands, faces, reflections, shadows, perspective, repeated objects and text. Look at the background as carefully as the main subject. Generated scenes can contain accidental signs, malformed logos or objects that change meaning.

Then view the image at the size customers will actually see. Fine print may disappear; a subtle disclosure may become unreadable; a crop may remove the product feature that explains the post.

If an image will be used in multiple formats, review each crop separately.

Review permissions and implied endorsement

Confirm that source photographs and other assets are cleared for the intended use. For identifiable people, verify that the proposed context is covered by the business’s permissions.

Do not assume that access to a photograph means permission to use it in an advertisement. Likewise, a generated person resembling a public figure can create an endorsement implication even when no name appears.

Review visible third-party logos and recognisable designs. When rights or permissions are uncertain, use an original or clearly cleared alternative rather than trying to hide the issue in a crop.

Compare image and caption together

Images make claims indirectly. A caption saying “our workshop” beside a generated workshop can mislead even if the caption contains no technical error.

Ask:

• Would a customer think this is a real place, event or result?
• Does the image suggest an accessory or service the caption does not mention?
• Does the caption explain any illustrative element that needs clarification?
• Is the image appropriate for the actual audience and offer?

The image and words must pass as a pair.

Give specific revision instructions

“Make it better” is difficult to act on. A useful correction names the problem, the source of truth and what must remain unchanged:

The generated lid has a handle our product does not have. Match the approved front photograph. Keep the background, lighting and camera angle unchanged.

Réson’s asset and revision workflow can help keep the original references connected to the draft. The person approving the image still needs to compare the result.

Save the approved version with its caption and intended placement. If either changes materially, review the pair again.

Use a clear pass, revise or reject decision

Without a decision rule, teams can spend too long polishing an image that should have been replaced. Separate cosmetic preferences from problems that prevent approval.

Pass means the image is accurate for its stated purpose and suitable for the approved placement. Revise means a specific, repairable issue remains, such as an unreadable label or a crop that hides the relevant feature. Reject means the central idea is misleading, the required permissions are unresolved or repeated revisions cannot preserve the product reliably.

For example, a hypothetical furniture retailer may receive an attractive room image in which a two-seat sofa has become a three-seat sofa. That is not a minor styling preference. The reviewer should require an accurate product depiction or replace the image with an approved photograph.

Record the decision in plain language. “Rejected: changed sofa size” is more useful later than “Not on brand.”

Keep a small review record

Save the image version, intended placement, references checked and the reviewer’s decision. Note any conditions, such as a caption explaining that a setting is illustrative. A teammate should be able to understand the approval without reconstructing the whole conversation.

When the image is reused, check what has changed. An approved crop for an editorial post is not automatically approved for an advertisement with a different claim. A new product release may also make an old image inaccurate.

This record is especially helpful in an iterative AI workflow. It shows which mistakes recur and which instructions actually improve the output. In Réson, keep revision feedback tied to the draft and make the correction specific enough to evaluate. Better feedback helps the next attempt; it does not remove the need to inspect it.

Read Réson’s help guide for the current asset and content workflow.

Related reading

• Refresh article cover images
• What Réson does