Trends · 5 November 2026 · updated 17 November 2026

Content Credentials (C2PA): should your CGI campaign carry them?

Content Credentials are a signed, tamper-evident record attached to an image or video that says how it was made and edited. They are based on an open standard from the Coalition for Content Provenance and Authenticity (C2PA). As far as we could find, no rule requires a brand to attach them to its own campaign images. They still matter for CGI, because platforms such as Instagram and Google read them when deciding whether to show an AI label, and some editing tools add them automatically. This article explains how that works. It is not legal advice.

What are Content Credentials?

A Content Credential is a small, signed package of information that travels with the file. The C2PA calls it a manifest.

Inside the manifest are assertions: statements about where the file came from, which tools were used, what was changed, and whether AI was involved. The whole package is signed with a digital certificate. According to the C2PA, the signature relies on standard techniques such as SHA-256 hashes and X.509 certificates, so any change to the file or the record breaks the signature and shows that something was altered.

Three points are easy to miss:

  • It is voluntary. The C2PA describes the standard as free to use and royalty-free. Nobody has to adopt it.
  • It records, it does not judge. Provenance alone cannot tell you whether content is true, as the C2PA explainer also points out. It only shows where the file came from and what happened to it.
  • It can be removed. The data can be stripped from a file. The standard allows a credential to be linked to an invisible watermark or a fingerprint, so the record can be found again.

How would a CGI render be described?

As a human-made digital image, unless a generative AI tool changed it.

Content Credentials and photo metadata use a shared list of terms from IPTC, the news-media standards body, to describe where an image comes from. The terms that matter for CGI:

IPTC termWhat IPTC says it meansWhere a CGI image fits
digitalCreationMade by a person with tools that are not generativeA render built and lit by artists from a 3D model
algorithmicMediaMade purely by an algorithm, without training dataA pattern computed from a formula, such as a fractal
compositeWithTrainedAlgorithmicMediaEdited or extended with generative AI, for example inpainting or outpaintingA render with an AI-filled background or extended edge
compositeSyntheticA mix of elements, at least one generative AIA composite with AI-generated props or people
trainedAlgorithmicMediaCreated by an AI model trained on captured contentA fully AI-generated image

The right-hand column is our reading, not an official mapping. It shows why the steps in your studio’s pipeline matter. One generative step can move a render from the first row to the third.

Why might a CGI image get an AI label?

Because a tool in the chain wrote AI information into the file, and the platform read it.

Meta said in February 2024 that it would use the “AI generated” information in the C2PA and IPTC standards to label images from several major AI tools. In July 2024 it renamed its “Made with AI” label to “AI info”, because images with only minor AI edits, such as retouching, were being labeled in a way people did not expect. In September 2024, Meta said it would move the label to the post’s menu for content it detects was only edited with AI tools, rather than generated.

Google has said that when an image contains C2PA metadata, its “About this image” feature can show whether it was created or edited with AI, and that its ad systems are starting to integrate C2PA metadata, with the goal of using those signals to inform how it enforces its policies.

On the tool side, Adobe announced in April 2024 that Firefly-powered features in Photoshop, such as Generative Fill, automatically attach Content Credentials. So a single generative fill during retouching can leave an AI record on an otherwise fully rendered image. Our article on whether CGI needs an AI label under the EU AI Act covers the legal side of the same question.

Are Content Credentials required by law?

Not for brands, as far as we could find in October 2026. The rules we found fall on the makers of AI tools and on large platforms.

  • EU AI Act. Article 50(2) requires providers of generative AI systems to mark their output in a machine-readable way. Recital 133 of the Act lists techniques such as watermarks, metadata and cryptographic methods for proving provenance, which is the kind of thing Content Credentials do. According to the Commission’s guidelines, the Digital Omnibus gives providers of generative AI systems already on the market before 2 August 2026 until 2 December 2026 to meet this marking rule. The visible disclosure for deep fakes is a separate duty for the person who uses the tool.
  • California. AB 853 amended the California AI Transparency Act and made its duties for covered generative AI providers apply from 2 August 2026. From 1 January 2027, large online platforms, defined as those with more than two million monthly users, must detect provenance data that follows widely adopted standards, let users inspect it, and not knowingly strip system provenance data.

That second point is worth noting. From 2027, credentials in your files are more likely to be shown to viewers, at least by the large platforms the California law covers.

What should you agree with your studio?

Treat provenance as part of the delivery spec, like color profiles and formats.

  1. Ask for a tool list per deliverable. Which tools touched the image, and did any of them generate or extend visible content?
  2. Ask what metadata the delivered files carry. Have someone open a delivered file in a Content Credentials viewer before it goes to media teams.
  3. Decide whether you want credentials at all. A brand may want a signed record showing that its imagery was made by people from the real product, or it may prefer files without extra metadata. Both are legitimate choices, but make the choice on purpose.
  4. Agree on who handles labeling. If a generative step is used, decide in advance who decides on a visible label and its wording.
  5. Keep the source trail. Keep the project files, so you can show how an image was made if anyone asks, with or without credentials.

Whichever studio you work with, from a general render house to a luxury and beauty specialist such as Not Important Studio, put these questions in the brief rather than after delivery. Our guide on what to put in a brief for a CGI studio has room for them.

What are the limits?

Content Credentials show where a file came from. They do not make an image honest, and they do not replace other rules.

A rendered product shown better than it really is can mislead whether or not it carries a perfect provenance record. Advertising standards on exaggeration and misleading claims apply regardless. And because credentials can be stripped or never added, the absence of a credential says nothing about how an image was made. For beauty and luxury brands, the stronger protection is still an image that stays true to the product, as we describe in our article on perfume bottles in CGI.

Frequently asked questions

Is a Content Credential the same as a watermark?

No. A Content Credential is signed data attached to the file. The standard can link it to an invisible watermark or a fingerprint, so the record can be found again if the data is stripped, but those are separate techniques.

Does a Content Credential prove an image is true?

No. It shows who signed a record of how the file was made and whether the file changed since. Provenance alone cannot tell you whether the content is true, and the C2PA explainer makes the same point.

Does a Content Credential replace an AI label under the EU AI Act?

No. Machine-readable marking under the AI Act is mainly a duty for the makers of generative AI tools. A visible disclosure for deep fakes is a separate duty, covered in our AI Act article.

Do we need to add credentials to old campaigns?

We found no rule that asks brands to do so. If old images already carry credentials from the tools that made them, it is worth knowing what they say.

Sources