AI Photo Retouching and Where It Goes Wrong

What automated retouching does well, the specific edits that cross from correction into misrepresentation, and how to set a house standard.

Automated retouching has moved from a specialist skill to a menu item, and the practical consequence is that a great deal of editing now happens without anyone deciding it should. Removing an object, extending a background, smoothing a surface, replacing a sky and cleaning a face are all one click operations, and each of them changes what the image asserts about reality.

The work that automation genuinely improves is the tedious and uncontroversial kind. Dust and sensor spots, stray hairs, cable removal, straightening, perspective correction, exposure and white balance recovery, and background extension to fit a different crop are all corrections that make an image match what a competent photographer intended. Doing these automatically saves hours and costs nothing in credibility.

The edits that require a decision are the ones that change a claim. Removing a defect from a product. Smoothing a texture that the customer will feel when they receive it. Changing the colour of a garment. Adding people to a room to imply a busier venue. Replacing a grey sky with a blue one on a property shoot. Each of these is technically identical to the corrections above and commercially different, because the customer can check them against reality.

Faces are the category that deserves the firmest house rule. Automated portrait retouching smooths skin, reshapes features and evens tone, and the cumulative effect is a person who does not look like themselves. For a staff photograph, a founder film or a testimonial this is a credibility problem rather than a flattery one. Peng et al. (2025) found that vocal cues shape credibility judgements in speech, and the visual analogue is well recognised in practice: an audience reads over processed faces as a signal that something is being managed.

The research on authenticity is directly applicable. Farooq and de Vreese (2026) documented how awareness of AI generation influences judgements of authenticity, and Kirk and Givi (2025) found that perceptions of AI authorship shape consumer responses to marketing communications and can produce negative reactions in some conditions. The implication is not that retouching should stop. It is that the audience is now primed to look for it, which raises the cost of being caught overdoing it.

Ecommerce is where the line has the clearest commercial consequence, because a returned item is a measurable outcome. An image that shows a colour, texture or finish the product does not have produces a return, a refund, a shipping cost and often a review. Kong and Lou (2026) found that visual appeal and visual congruence play distinct roles alongside social proof in how advertising is processed, and congruence is precisely the property that aggressive retouching destroys.

A workable house standard can be written in a paragraph and is worth having. Correct exposure, colour, straightness, dust and distractions freely. Do not alter the shape, colour, texture or finish of a product. Do not alter a person's face or body beyond what a make up artist would do on the day. Do not add or remove elements that change what the image asserts about a real place or a real event. Label constructed lifestyle imagery. Keep the unretouched original.

That last point is the practical safeguard. Keeping the original file, with a record of what was changed, means a dispute can be resolved in minutes rather than becoming a matter of recollection. It also allows a different crop or a different treatment later without repeating the work, and it protects the studio as much as the client.

Generative fill and background extension deserve a specific caution because they are so convenient. Extending a background to make a horizontal shot work vertically is a routine and harmless operation. Extending it to imply that a venue is larger, a crowd is bigger or a facility is more extensive than it is, is a different act performed with the same tool. The test is whether the extension changes what a viewer would conclude about the real world, not whether it looks convincing.

For a studio, the useful position to hold with clients is that retouching decisions belong in the brief rather than in the edit. Agreeing the standard before the shoot, writing it into the delivery terms, and applying it consistently across a set removes both the awkward conversation about a specific image and the inconsistency that makes a catalogue look unreliable.

References

Farooq, A., & de Vreese, C. (2026). Deciphering authenticity in the age of AI: How AI-generated disinformation images and AI detection tools influence judgements of authenticity. AI & Society, 41(1), 493–504. https://doi.org/10.1007/s00146-025-02416-5

Kirk, C. P., & Givi, J. (2025). The AI-authorship effect: Understanding authenticity, moral disgust, and consumer responses to AI-generated marketing communications. Journal of Business Research, 186, Article 114984. https://doi.org/10.1016/j.jbusres.2024.114984

Kong, J., & Lou, C. (2026). Beyond persuasion knowledge: Examining the roles of visual appeal, visual congruence, and social proof in influencer advertising. Journal of Retailing and Consumer Services, 88, Article 104502. https://doi.org/10.1016/j.jretconser.2025.104502

Peng, Z., Wang, C., & Jiang, X. (2025). On how vocal cues impact dynamic credibility judgments: Mouse-tracking paradigm examining speaker confidence and gender through voice morphing. Journal of Speech, Language, and Hearing Research, 68(11), 5261–5277. https://doi.org/10.1044/2025_JSLHR-24-00849