C2PA metadata in AI Assistant generative-content-credentials

On this page: Learn which AI Assistant actions attach C2PA metadata, what this means for images that combine several generative AI sources, and what carries over when your content moves between apps.

INFO
New laws are emerging around generative AI transparency, and Adobe is working to meet applicable requirements across jurisdictions. C2PA metadata are the provenance tool Adobe uses to meet the requirements of these laws.

C2PA metadata are durable, invisible metadata that record how a piece of content was created or edited. When you use AI Assistant in Adobe Journey Optimizer to generate or edit an image with generative AI tools, C2PA metadata are automatically attached to that image, no action is required on your part.

Actions that attach C2PA metadata cc-workflows

The following table summarizes when C2PA metadata are attached, based on the image action performed in AI Assistant.

Action
Description
C2PA metadata attached?
Use case example
Generate an image
Create a new image from a text prompt, from a reference image, or generate a similar image
Always. The image is generated by generative AI, so it always carries a fresh C2PA metadata.
A banner image for an email campaign is generated from a text prompt describing the desired visual.
Crop an image (center or smart crop)
Adjust an image to requested dimensions
Only if the source image already had a C2PA metadata. Cropping recreates the image’s pixels, which would normally erase that C2PA metadata, so AI Assistant reads it from the source image before cropping, then rebuilds and re-attaches it to the cropped result. Cropping itself does not add a new generative AI action, it preserves the existing one.
A generated banner image is cropped to fit a web page: the C2PA metadata is preserved through the crop.
An uploaded stock photo used as a push notification background is cropped to fit the screen: since the stock photo carries no generative AI action, no C2PA metadata is created.
Add a text overlay
Render generated text on top of a background image
Only if the background image already had a C2PA metadata. Rendering the overlay produces a new image from the background plus the text, which would normally erase that C2PA metadata, so AI Assistant reads it from the background image beforehand, then rebuilds and re-attaches it to the result. The overlay step does not add a new generative AI action.
A promotional headline is rendered as a text overlay on a generated background image for a landing page: the C2PA metadata from the background image is preserved.
Overlay images
Composite two or more images together
If any of the source images have a C2PA metadata, the combined image carries all of them, merged into a single C2PA metadata. Compositing produces a new image from the sources, which would normally erase those C2PA metadata, so AI Assistant reads each one before compositing, then builds a single combined C2PA metadata listing every source that contributed a generative AI action.
A generated product image is composited with a generated background for an email header: the result carries a C2PA metadata reflecting both generative AI sources.
Two uploaded brand photos are composited into one collage: since neither source carries a generative AI action, no C2PA metadata is created.

Content types and their scope cc-content-types

  • Images: Covered. C2PA metadata are attached when images are generated with generative AI, and preserved through cropping, text overlay, and image overlay operations performed by AI Assistant.
  • Text: Not applicable. Text-only outputs of AI Assistant, such as copy generation, translation, and brand alignment suggestions, do not require C2PA metadata.

What happens as your content moves cc-content-moves

C2PA metadata travel with the image file. When an image that was generated or edited with generative AI is downloaded or exported from Adobe Journey Optimizer, its C2PA metadata are preserved. Learn more about C2PA metadata.

Some ways of bringing images into your content, such as extracting an image from a PDF or from an embedded (base64) source, may not preserve the original C2PA metadata. In these cases, no C2PA metadata can be read from the source, and none is created for the result.

Additional resources

recommendation-more-help
journey-optimizer-help