Do I need to label AI retouching?

No. Ordinary retouching, spot and blemish removal, color, and cropping are treated as normal editing and do not need a label, in the EU, New York, or the rest of the US. A real photo of a real thing is still a real photo.

Do I need to label an AI-generated background?

In the EU, lean toward a label if the scene could pass as a real place. In New York it is not required, because that law only covers AI-generated human performers, not backgrounds. The rest of the US has no general rule.

Does an AI voiceover need a disclosure?

If it clones a real person's voice, get consent first, which is a permission issue, not just a label. A generic AI voice leans toward a label for EU audiences. AI used only to translate a real person's real speech generally does not need a label.

Do fully AI-generated product images need a label?

For EU audiences, label a photoreal AI product shot that could pass as a real photograph. In New York and the rest of the US it is generally not required unless it misleads buyers. When unsure, label it.

Does My Ad Need an AI Label?

TL;DR: Most AI in ads does not need a label. What triggers a label is content that could pass for real, and even then it depends on where the ad runs. Today only two places make advertisers label AI: the EU (broad, from August 2, 2026) and New York (narrow, in force now, fake humans only). California's law targets the AI tools, not you. The rest of the US has no general AI-label rule. Below is the plain checklist. I am a photographer, not a lawyer, so this is researched guidance, not legal advice. When a case is fuzzy, label it.

Every week someone asks me a version of the same question. We used AI on this shot, do we have to say so? The honest answer is "it depends," but it depends on a small number of things you can actually check. So here is the checklist I use, case by case, with a yes or no you can act on.

If you want the law behind it, start with the EU AI Act, explained. If you want the mechanics of the label itself, see how to label AI content. This piece is the middle: does this specific shot need a label at all.

Which Places Actually Make You Label AI Content?

Four answers cover almost everyone:

The EU is the broad one. From August 2, 2026, any realistic AI-made person, place, object, or scene that could pass for real needs a label when the ad reaches EU audiences (EU AI Act, Article 50, 2026).
New York is narrow. In force since June 9, 2026, but it only fires on a fully AI-generated fake human in an ad. AI backgrounds, products, food, and scenes with no synthetic person are not covered (New York Governor's Office, 2026).
California does not make you label your ad. Its AI Transparency Act aims at the big AI tool makers, not advertisers, and it is not even active until August 2, 2026 (California Legislature, SB 942, 2026).
The rest of the US has no general AI-label rule. Two things still apply everywhere: get consent before using a real person's AI likeness or voice, and never use AI to mislead, which the FTC already polices.

So the practical read: if your campaign touches the EU, you have real labeling to do. If it only runs in the US, the bar is much lower, and mostly about fake humans and honesty.

How to Read This Checklist

For each case, two quick tests:

Could it pass for real? If a normal viewer could think it is a real photo or video of a real thing, it is in label territory. If it obviously reads as an illustration or a stylized render, usually not.
Where does the ad run? EU audiences pull in the broad rule. New York pulls in the fake-human rule. Everywhere else is lighter.

Verdicts below: Label it, No label needed, Gray, lean label, or Consent, not a label.

Editing a Real Photo

Retouching, spot and blemish removal, color, crop, sharpening. No label needed, anywhere. This is normal editing. A real photo of a real thing is still a real photo.
Removing a real object or person from the shot. Gray, lean label for the EU if the result still reads as a real scene. Not required in New York or the rest of the US.
Swapping the background to a different real-looking place, real subject kept. Gray, lean label for the EU. New York does not cover it (no fake human), and neither does the rest of the US.
A light AI background extension on a product shot. Usually fine. EU guidance treats small extensions in product ads as minor. If the extension turns into a whole invented location that looks real, label it.
Reshaping a body or changing skin with AI. Label it for the EU, it is a manipulation that reads as real. Not required elsewhere, though the FTC cares if it misrepresents the product.

Putting AI People in Your Ads

A fully AI-generated person who is not a real human. Label it. This is the one case the EU and New York both cover, and New York's law was written for exactly this.
A real, recognizable person's face or voice recreated with AI. Consent, not a label, is the gate here. California, Tennessee, Illinois, and New York all protect a real person's likeness, and permission comes first. For an EU ad, label it too, but do not skip consent.
A real model you photographed, with an AI-extended set behind them. No label needed in New York, the person is real. For the EU, only lean toward a label if the added scene reads as a real place. The performance belongs to someone who was actually on set.
De-aging or altering a real spokesperson. Consent first, since it is a real person. Label for the EU. New York treats this as a real human, not a synthetic performer, so it is a permission question, not a label one.

Products, Food, and Scenes

An AI-generated product or food image with an obvious stylized or CGI look. No "looks real" label needed, and no human means New York does not apply. The tool still adds its own hidden tag.
A photoreal AI food or product shot passed off as a real photograph. Label it for the EU, a realistic AI object counts under the broad rule. Not required in New York or the rest of the US, unless it misleads buyers.
A photoreal AI location or scene presented as real. Label it for the EU. New York does not cover it without a fake human. In the US generally, only an issue if it misrepresents something material.

AI in Video and Audio

An AI voice clone of a real person. Consent first, it is their voice. Label for the EU if used.
A generic AI voice that is not a real person. Gray, lean label for the EU. Not required elsewhere.
AI dubbing or translation of a real person's real speech. Generally no label. New York specifically exempts translation of a human performer, and the EU leans light unless it is lip-synced to look native.
An AI-generated spokesperson video. Label it. Covered by both the EU and New York.

Chatbots and AI Messages

A bot or AI assistant talking to your customers. Tell people it is AI. The EU requires it from August 2, 2026, and California's bot law already requires it when a bot drives a sale. Easiest to just always disclose.

Two Exceptions Worth Knowing

A fully manual edit with no AI at all. Outside all of these rules. The catch: many "magic eraser" and background tools are themselves AI, so this only holds if genuinely no AI touched the file.
A campaign already published before August 2, 2026 that just keeps running. You do not have to go back and re-label it. Anything you newly launch or relaunch after that date, label as normal.

The One Rule That Keeps AI Labeling Simple

Build to the strictest standard and you clear the rest automatically. The EU reaches the widest set of AI content and it binds you directly, so an EU-clean habit already covers New York and leaves the rest of the US with nothing extra to do on labeling. Two lines cover most of it: if a realistic AI person, place, object, or scene could pass for real, label it, and if you used a real person's face or voice, get consent. When a case is gray, label it. Labeling never creates liability. Omitting one can.

An AI-generated hospitality interior carrying a visible AI label, shown as a checklist example

Want the official mark to use? Download the official EU AI content labels (SVG and PNG), free to use, no attribution required.

The reason this stays simple for us is that most of what we shoot is real, then extended with AI on a known, documented path. We can tell a client exactly which assets were photographed and which were generated, so the label question answers itself. That is the backbone of how we run hybrid production. If you want that clarity built into your own campaigns, book a call and we will map it out before your next shoot, not after the ads ship.