Skip to content
Sign in
The law

SynthID, C2PA and content credentials: how to check and label AI photos

Updated · By Larpea

Label an AI photo whenever it could pass for a real one: use the platform's AI setting, add a plain mention such as (AI) to the caption, and leave embedded marks like C2PA Content Credentials alone. YouTube, Meta and TikTok all ask for it, and EU and French law add duties. This page is not legal advice.

The rule of thumb

If an AI image could be taken for a photograph of something real, say so, in the platform's own setting and in plain words. This guide belongs to is larping illegal. It sets the platforms' rules next to the law, as they stood on 18 October 2026, and it explains how to label. It does not explain how to remove a mark: nothing about honest content requires that.

What each platform asks

Platform rules are not law and change often. Each row is the platform's own page, read on the date shown.
PlatformThe rule as we read itRead onSource
YouTubeCreators disclose realistic altered or synthetic content in YouTube Studio (the AI use setting). Realistic means a real person's words or actions, a real event or place altered, or a believable scene that never happened. Beauty filters, colour fixes, animation and clearly unrealistic scenes need no disclosure. Repeated failure to disclose can bring labels applied by YouTube, removal or suspension from the Partner Program.support.google.com/youtube/answer/14328491
Facebook, Instagram and Threads (Meta)Meta shows an AI info label on content when it detects industry-standard AI indicators or when people disclose that they are uploading AI-generated content. Meta says the method is still evolving and may miss some AI-edited content. The page we read states no penalty.transparency.meta.com/governance/tracking-impact/labeling-ai-content
TikTokIn its newsroom post of 19 November 2025 TikTok says it requires people to label realistic AI-generated content, reads C2PA Content Credentials to label content made elsewhere, and is testing invisible watermarks so that labels survive re-uploads. The current Community Guidelines text could not be loaded by our tools.newsroom.tiktok.com/more-ways-to-spot-shape-and-understand-ai-content

The three pages agree on the core: a realistic AI image, video or sound is something to flag. They differ on how. YouTube makes the creator answer a yes or no during upload, and says it may label automatically from its own AI tools, from C2PA metadata or from internal detection, with the creator unable to override a label that came from C2PA metadata. Meta labels from industry-standard indicators or from self-disclosure and says its method is still evolving. TikTok says it requires realistic AI-generated content to be labelled and reads C2PA Content Credentials. All three, in other words, look at signals inside the file and not only at what you tick: YouTube and TikTok name C2PA, Meta speaks of industry-standard indicators.

Three kinds of label

  1. The platform's setting. The checkbox or toggle in the upload flow. It is the one a moderator sees first.
  2. Words in the post. A short mention in the caption or on the first frame, such as (AI) or AI-generated, written where people read before they react.
  3. Marks inside the file. C2PA Content Credentials are metadata that signal how a file was made, and can be inspected with the Content Credentials verifier. An invisible watermark is a mark hidden in the pixels. TikTok says it is testing such watermarks so that labels survive when content is re-uploaded or edited elsewhere, which tells you that metadata alone can be lost along the way.

Our own app is built on all three. According to its FAQ, My Larp Snap marks every result as AI-generated: the image model adds an invisible watermark (SynthID, for Google's model, the default), the file name starts with mylarpsnap-ai, and the free edit carries a visible AI mark. Larpcam makes parodies of travel vlogs and labels them as such. Neither makes the post honest on its own. The caption is still yours to write.

The law behind the labels

This is general information, not legal advice. Law not reviewed by a lawyer.

Law checked on .

Each text links to its official source.
InstrumentArticleWhat it coversMaximum cited by the textRead on
European Union: Regulation (EU) 2024/1689 (AI Act), definitionsArt. 3(4) and 3(60)A deployer is anyone using an AI system under their authority, except where it is used in the course of a personal non-professional activity. A deep fake is AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places or events and would falsely appear authentic or truthful.Not stated in the text
European Union: Regulation (EU) 2024/1689 (AI Act)Art. 50(2) and 50(4); fines in Art. 99(4)Providers of generative systems must mark outputs in a machine-readable format. Deployers of a system that makes a deep fake must disclose that the content is artificially generated or manipulated; for an evidently artistic, creative, satirical or fictional work the duty is limited to disclosing it in a way that does not hamper the work. The Regulation applies from 2 August 2026 (Art. 113).Up to EUR 15,000,000 or, for an undertaking, 3% of worldwide annual turnover, whichever is higher (Art. 99(4)); for SMEs, whichever is lower (Art. 99(6))
European Union: Regulation (EU) 2026/1744, amending Regulation (EU) 2024/1689Art. 1(39)(b), new Art. 111(4)Providers whose generative systems were on the market before 2 August 2026 must take the necessary steps to comply with Article 50(2) by 2 December 2026. The amending regulation was published on 24 July 2026.Not stated in the text
France: Code pénalart. 226-8Publishing, without the person's consent, visual or audio content generated by an algorithmic process that reproduces a person's image or voice, when its artificial nature is not obvious or not expressly flagged. Extended to algorithm-generated content by the law of 21 May 2024. Sexual deepfakes have a separate offence in art. 226-8-1. Official page not readable by our tools: wording confirmed from an extract of it. Re-read it before relying on it.1 year and a fine of EUR 15,000 (about $16,900); 2 years and EUR 45,000 (about $50,600) when published through an online public communication service
France: Loi n° 2023-451 of 9 June 2023 on commercial influenceart. 5Influence content with retouched images (slimmed silhouette, changed face) must carry the mention Images retouchées, and content with AI-generated faces or silhouettes the mention Images virtuelles. It targets people who promote products or services for consideration, not every poster. Official page not readable by our tools: wording confirmed from an extract of it. Re-read it before relying on it.Not read on the official text
France: Ordonnance n° 2024-978 of 6 November 2024rewrite of art. 5 of the 2023 lawRewrote article 5 to make the way the public is informed about retouched and virtual images more flexible and proportionate, according to the report to the President of the Republic. In force since 8 November 2024. Official page not readable by our tools: wording confirmed from an extract of it. Re-read it before relying on it.Not read on the official text

Amounts are in the currency of the text. Dollar equivalents use the European Central Bank euro reference rates of 16 October 2026, rounded; they are only to give a sense of scale.

The EU, as of 18 October 2026. Article 50 of the AI Act has two duties that matter here. Providers of generative systems must mark outputs in a machine-readable format; deployers of a system that produces a deep fake must disclose that the content is artificially generated or manipulated, with a narrower duty for evidently artistic, creative, satirical or fictional work. The Regulation applies from 2 August 2026, and Regulation (EU) 2026/1744 gives providers whose systems were already on the market until 2 December 2026 for the machine-readable marking alone. A person using an AI system in a personal non-professional activity is not a deployer under Article 3(4). Fines for breaking Article 50 can reach EUR 15,000,000 or, for an undertaking, 3% of worldwide annual turnover, whichever is higher; for SMEs, whichever is lower. The euro banknote rules are in the Europe section of the prop money guide.

France. Two separate rules. The influence law asks for the mentions Images retouchées and Images virtuelles on influence content, and article 226-8 of the Code pénal punishes publishing algorithm-generated content that reproduces a person's image or voice without consent when its artificial nature is not obvious or flagged. The Légifrance pages were not readable by our tools, so the rows say they come from extracts. See the France page, in French.

The United States. We read no general federal labelling duty for a personal post. Fraud and deceptive-practice texts apply to what a post is used for; see the US page.

A labelling habit that covers all of it

  1. Set the platform's AI option whenever the image looks like a real photo, even if it is only partly AI.
  2. Put the mention in the first line of the caption and, for video, in the first frame.
  3. Keep the original file, with its metadata, and the prompt, so you can show what it is.
  4. Do not crop, filter or re-save in a way that you intend to hide how it was made. The point is to say, not to disguise.
  5. If a real person's face or voice is in it, ask for consent first, or do not post.

If you are looking at someone else's photo, how to tell if a photo is AI gives the checks, and for staged lifestyle photos how to tell if a private jet photo is real. To make the honest version, AI photo editor prompts, fake vacation photos with AI, travel reels included and how to film a skit with a phone.

Questions

Do I have to label AI photos?
YouTube asks creators to disclose realistic altered or synthetic content, TikTok says it requires realistic AI-generated content to be labelled, and Meta labels from indicators or disclosure. In the EU, deployers of a system producing a deep fake, meaning users other than in a personal non-professional activity, must disclose it (AI Act, Article 50(4)). France adds its own rules.
What is the Made with AI or AI info label?
On Meta's platforms the AI info label appears when Meta detects industry-standard AI indicators or when people disclose that they are uploading AI-generated content. Meta says the method is still evolving.
What are C2PA and Content Credentials?
A cross-industry standard that embeds metadata signalling how a file was made. YouTube and TikTok say they read it. It can be inspected with a verifier, but an image without credentials proves nothing either way, because metadata can be lost.
What is SynthID?
An invisible watermark. According to the FAQ of our own Snap app, the Google image model it uses by default marks its output with one.
When does the EU AI Act labelling duty start?
The Regulation applies from 2 August 2026. Regulation (EU) 2026/1744 gives providers whose generative systems were already on the market until 2 December 2026 for the machine-readable marking in Article 50(2).
Does the law require a label on a personal post?
In the EU, a person using an AI system in a personal non-professional activity is outside the definition of deployer. Platform rules may still ask for a label, and French article 226-8 covers an unflagged fake of a real person's face or voice.

Keep reading