# AI background, cut-out, retouching: when does a product photo need an AI label?

> Cut-outs, AI backgrounds, retouching, upscaling: which AI edit of a real product photo needs a label under the EU AI Act, and when AI MODIFIED fits.

URL: https://easyaiactlabels.com/en/guides/ai-background-cutout-retouching/

[Guides](/en/guides/) /Practice

# AI background, cut-out, retouching: when does a product photo need an AI label?

Last updated: 1 October 2026 Responsible: Martin Jäger Created with AI 9 min read

Short answer

A real product photo that was only retouched, cut out, upscaled or placed in front of an AI background for design reasons usually needs no AI label: the Commission’s guidelines of 20 July 2026 name such edits as usually of minor relevance to authenticity (para. 116). A label becomes necessary as soon as the AI adds something that can be taken as real and is not, such as a scene that misrepresents size or use, a person in the picture or a product redrawn by the AI; AI MODIFIED usually fits then. If the image shows the goods better than you deliver them, no label helps: the image has to change.

Most AI images in online stores are not invented from scratch but edited photos: cut out, placed in front of a new background, retouched, upscaled. Whether such a photo needs an AI label is not decided by the technology under Article 50(4) of the EU AI Act (Regulation (EU) 2024/1689), but by whether the result is a deepfake. This guide classifies the usual edits along the European Commission’s guidelines of 20 July 2026 (C(2026) 5054 final) and flags where the classification is an interpretation of its own. Whether an image gets a label is your decision; the table shows where the official examples draw the line. The basics are in the guide on whether you [have to label AI images in your online store](/en/guides/ai-images-labelling-obligation/), and [Article 50 of the EU AI Act](/en/guides/eu-ai-act-article-50/) has its own guide.

## What matters for edited photos?

If you use a ready-made AI tool for store images, you are a deployer (Article 3(4) AI Act) and must disclose under Article 50(4) that a deepfake has been artificially generated or manipulated. The machine-readable marking in the file under Article 50(2) is owed by the provider of the tool, not by you. The Digital Omnibus (Regulation (EU) 2026/1744) did not change Article 50(1) to (6). If an agency or freelancer edits the images for you, see [creating AI images for store clients](/en/guides/ai-images-agencies-labelling/).

Under Article 3(60), a deepfake is AI-generated or manipulated image content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful. Two clarifications in the guidelines matter for product photos (para. 113): objects expressly include consumer goods, and authenticity covers the actual appearance or use of objects. The assessment is objective and does not depend on an intention to deceive; a high degree of photorealism makes a deepfake more likely but is not decisive on its own (para. 114).

The guiding question is therefore: does the edited image still show your product as it is and as it is used?

## What counts as standard editing?

Para. 116 of the guidelines names AI edits that are usually of minor relevance to authenticity: editing background details, lighting adjustments, colour correction, noise reduction, cosmetic adjustments. For product advertising it expressly lists colour correction, background extensions of existing content, adjustments or replacements of backgrounds for clearly aesthetic purposes, compositions and arrangements of existing products, and re-scaling. Each depends on context and impact in the specific case.

A second list is in para. 92, there for the provider obligation under Article 50(2). As standard editing it names, among others, removing dust spots, deleting backgrounds that are visible in the original file, sharpening, edge image completion and pixel filling for a different aspect ratio. As substantial changes it names removing, replacing or inserting objects or persons in a way that changes the meaning and substance of the image, and composite images that modify the representation of objects. For your disclosure under Article 50(4), these examples are a pointer, not an official classification (interpretation of this guide).

## Which edit needs a label?

AI edit of a real product photo

Label?

Basis

Dust, lint, sensor spots removed; lighting, colour, sharpness corrected; the goods look as delivered

Usually no

Para. 116; para. 92

Classic cut-out: background removed, product unchanged

No

Para. 92 (deleting backgrounds visible in the original); para. 116

Background extended for a different aspect ratio

Usually no

Para. 116 (background extension); para. 92 (pixel filling)

AI background for design reasons (colour field, studio, mood) behind the unchanged product

Usually no

Para. 116; guidelines’ negative example: a real product against an AI background, as long as the ad does not mislead about the product

Upscaled without new detail

Usually no

Para. 116

Several real product photos arranged into a set

Usually no

Para. 116 (arrangements of existing products)

Real product in a realistic AI scene that misrepresents size, a characteristic or the use

Yes, AI MODIFIED

Guidelines’ positive example: an AI product image that can mislead as to appearance, characteristics or use; misleading advertising as well

AI scene with a person, a hand or a model

Yes, AI MODIFIED

Para. 113: realistic AI persons

AI tool redraws the product from your photos into a scene

When in doubt yes, AI MODIFIED

The AI determines how the product looks (interpretation of this guide)

Generative cut-out that invents edges, fringes or hidden parts

Borderline, AI MODIFIED when in doubt

Para. 116 does not name the case (uncertain, as of 1 October 2026)

Generative upscaler that visibly invents new detail (texture, pattern, lettering)

Borderline, AI MODIFIED when in doubt

Interpretation of this guide, as for the generative cut-out

AI colour variant of a real photo

Borderline; yes if the shade differs from the goods

Para. 116 names colour correction, not generating a variant

Real defects or signs of wear retouched away, accessories added that are not supplied

A label does not help, the image has to change

Positive example (more appealing or of better quality than in reality); misleading advertising

Two rules of thumb (an interpretation of this guide, not an official requirement):

-   If the AI only changes the surroundings of a product shown unchanged, para. 116 usually applies. If it changes how the product itself looks or what it appears to be suited for, the image needs at least AI MODIFIED.
-   For photorealistic borderline cases: label when in doubt. The Wettbewerbszentrale, the German industry body against unfair competition, recommends exactly that (guide version 2.0, as of 29 July 2026); no court decision on AI images in online retail was known as of 30 September 2026.

## When does the new background become a problem?

The background itself is rarely the problem; what it says about the goods can be. The guidelines’ negative example has a condition: a real product against an AI background is not a deepfake as long as the ad is not likely to mislead about the product’s actual representation, characteristics and use. Typical tipping points (interpretation of this guide):

-   **Size**: the proportions of the scene make a lamp, a bag or a piece of furniture look larger or smaller than it is.
-   **Use**: the scene shows a use the product is not suited for, such as a light transitional jacket in a snowstorm or an indoor-only lamp in the garden.
-   **People**: once a hand or a model is added, the image needs a label under para. 113 anyway.

A related example from the Commission for AI MODIFIED: authentic photographs of an empty apartment furnished using AI (icon page, page version 24 September 2026). What this means for room scenes, virtual staging and 3D renderings is covered in the guide [AI images in furniture and home stores](/en/guides/ai-images-furniture-home-store/).

## AI MODIFIED or AI GENERATED?

For edited photos, AI MODIFIED almost always fits. The Commission reserves AI GENERATED for content fully generated by AI with no human-created elements apart from prompting, and AI MODIFIED for pre-existing, human-made content partially modified with AI into a deepfake (icon page, page version 24 September 2026). Your product photo is such an element. If hardly anything of the source photo is left, choose AI GENERATED, the stronger disclosure.

## What is in the file?

Some tools record in the provenance marker whether an image was generated or edited. On the Easy AI Act Image Labels test bench on 1 September 2026, Canva wrote the IPTC value compositeWithTrainedAlgorithmicMedia for an own photo with an AI background, which IPTC defines as augmentation, correction or enhancement using a generative AI model. A test image from the Shopify image editor (“Generate background”) carried no metadata at all on the stored original on the same day. Other cut-out tools have not been measured. For the Shopify editor in detail, see [Shopify Magic and the EU AI Act](/en/guides/shopify-magic-ai-act/).

The marker says that AI was involved, not whether a label is needed. The free [AI image checker](/en/ai-image-checker/) suggests AI MODIFIED for the Canva photo because the marker reports an edit; if the unchanged product only stands in front of a designed background, the image can still go without a label. Conversely, an image without a marker can need one. An app such as Easy AI Act Image Labels reads markers on the original and suggests labels; the classification remains your decision.

## When does a label not help?

An AI label only says that AI was involved. If the image shows the goods differently from what you deliver, it can be misleading, labelled or not: Article 50(6) AI Act leaves other transparency obligations unaffected, and the guidelines make clear that the disclosure obligation does not permit misleading advertising (para. 129).

-   **Germany**: Section 5(2) no. 1 UWG lists, among the main characteristics of goods, their nature, execution, accessories, usability and condition; pictorial representations count as statements (Section 5(4) UWG).
-   **Austria**: Section 2(1) of the Austrian UWG covers statements about the nature and the main characteristics of the product. The AI Service Desk at RTR expressly lists colour correction, noise reduction and background extensions in product ads as examples that are not deepfakes (FAQ, retrieved 1 October 2026).
-   **Switzerland**: Article 3(1)(b) of the Swiss UWG covers incorrect or misleading statements about one’s own goods. The AI Act only reaches a Swiss store if the image is used in the EU (Article 2(1)(c) AI Act).

“Image may differ” or “illustrative image” are no AI disclosure, and they do not fix a discrepancy. The same applies to “serving suggestion” in food stores; more in the guide [AI images in food and drink stores](/en/guides/ai-images-food-drink-store/).

## How to proceed

1.  **Sort images by edit**: only retouched or cut out, new background, AI scene, person in the image, product redrawn by the AI.
2.  **Check against the goods**: do colour, size, details, accessories and use match what you deliver? If not, change the image.
3.  **Classify** along the table; you label photorealistic borderline cases with AI MODIFIED.
4.  **Label on the image**, visible without a click, in every place where the image appears (Article 50(5), para. 143), with a text alternative for screen readers.
5.  **For ads and marketplaces**, create a copy with the label burned in, for example with the free [AI Label Generator](/en/ai-label-generator/); the original stays without a label. When the label has to go into the image is covered in the guide [AI label: overlay or burned into the image?](/en/guides/ai-label-overlay-or-burned-in/)
6.  **Keep originals and decisions**: the file from the tool in its native format, plus tool, edit and reasoning per image. The [AI Act check](/en/ai-act-check/) shows where your store stands overall, based on your own answers.

Which image types beyond edits need a label is covered in the guide [Which images in your store need an AI label?](/en/guides/which-images-need-a-label/)

## Frequently asked questions

### Do I have to label a product photo with the background removed?

Usually not. Para. 92 of the Commission’s guidelines of 20 July 2026 counts deleting backgrounds that are visible in the original file as standard editing, and para. 116 names replacing backgrounds for clearly aesthetic purposes as usually of minor relevance. It is different if the tool invents edges, fringes or hidden parts of the product: that is a borderline case, and when in doubt you label it AI MODIFIED.

### Is “image may differ” enough as an AI disclosure?

No. The phrase does not say that AI created or changed the image, and it points to a different problem: if the goods differ from the image, that can be misleading, in Germany under Section 5 of the Act against Unfair Competition (UWG), and an AI label does not cure that either. The same applies to “illustrative image”.

### Does AI upscaling count as an edit that needs no label?

Usually yes: para. 116 of the guidelines names re-scaling images in product advertising as likely to have only a minor impact. If a generative upscaler visibly invents new detail, such as texture, patterns or lettering on the product, treat the image like a generative cut-out: a borderline case, AI MODIFIED when in doubt. That is an interpretation of this guide; the guidelines do not distinguish between the two kinds.

### What if the Shopify image editor generated the background?

On an unchanged product photo, that is a background swap that usually needs no label (para. 116), as long as the new scene does not suggest anything false about the product. Do not expect a provenance marker: a test image made with “Generate background” carried no metadata at all on the stored original on 1 September 2026. The classification therefore follows from the image, not from the file.

### How does the AI Service Desk in Austria see it?

The AI Service Desk at RTR lists colour correction, noise reduction and background extensions in product ads as examples that are not deepfakes (FAQ, retrieved 1 October 2026). The AI Act applies directly in Austria; whether an image deceives about the goods is assessed there under the Austrian UWG (Section 2), not the German one.

## Sources

1.  Regulation (EU) 2024/1689 (AI Act), CELEX 32024R1689, Article 3(4) and (60), Article 50. [publications.europa.eu](http://publications.europa.eu/resource/celex/32024R1689) (accessed 1 October 2026)
2.  Regulation (EU) 2026/1744 (Digital Omnibus on AI), CELEX 32026R1744, Article 1 point 20 (Article 50(1) to (6) unchanged). [publications.europa.eu](http://publications.europa.eu/resource/celex/32026R1744) (accessed 1 October 2026)
3.  European Commission, Guidelines on the implementation of the transparency obligations under Article 50 (Annex to C(2026) 5054 final, 20 July 2026), paras. 92, 113, 114, 116, 129, 143 and the example lists on Article 3(60). [ec.europa.eu](https://ec.europa.eu/newsroom/dae/redirection/document/131215) (accessed 1 October 2026)
4.  European Commission, EU Icons for labelling AI-generated content (page version 24 September 2026). [digital-strategy.ec.europa.eu](https://digital-strategy.ec.europa.eu/en/policies/eu-icons-labelling-ai-generated-content) (accessed 1 October 2026)
5.  IPTC NewsCodes, Digital Source Type: compositeWithTrainedAlgorithmicMedia. [cv.iptc.org](http://cv.iptc.org/newscodes/digitalsourcetype/compositeWithTrainedAlgorithmicMedia) (accessed 30 September 2026)
6.  German Act against Unfair Competition (UWG), Section 5 (German). [gesetze-im-internet.de](https://www.gesetze-im-internet.de/uwg_2004/__5.html) (accessed 1 October 2026)
7.  Austrian Federal Act against Unfair Competition 1984 (UWG), current version, Section 2 (German). [ris.bka.gv.at](https://www.ris.bka.gv.at/GeltendeFassung.wxe?Abfrage=Bundesnormen&Gesetzesnummer=10002665) (accessed 1 October 2026)
8.  Swiss Federal Act against Unfair Competition (UWG), SR 241, Article 3(1)(b) (German). [fedlex.admin.ch](https://www.fedlex.admin.ch/eli/cc/1988/223_223_223/de) (accessed 30 September 2026)
9.  RTR, AI Service Desk (KI-Servicestelle), FAQ, question on the deepfake concept (German). [rtr.at](https://www.rtr.at/rtr/service/ki-servicestelle/faq/FAQ.de.html) (accessed 1 October 2026)
10.  Wettbewerbszentrale, guide to labelling AI-generated content, version 2.0, as of 29 July 2026 (German). [wettbewerbszentrale.de](https://www.wettbewerbszentrale.de/wp-content/uploads/2026/07/2026_07_Leitfaden_KI_generierte_inhalte_version2.pdf) (accessed 1 October 2026)

This article is general information, not legal advice. Whether an image needs a label is your decision; for your specific case, please ask a lawyer.

Read next

[01 · Obligation Do I have to label AI images in my online store?](/en/guides/ai-images-labelling-obligation/) [17 · Practice Shopify Magic and the EU AI Act: What Applies to You?](/en/guides/shopify-magic-ai-act/)

I find the AI evidence. You decide with one click.

[View on the Shopify App Store](https://apps.shopify.com/easy-ai-act-image-labels?utm_source=easyaiactlabels&utm_medium=website&utm_content=artikel)