Content Credentials from Adobe Firefly and Photoshop: when they appear

If you took a lamppost out of a photo with Generative Fill and the exported JPEG now names Adobe Firefly in its Content Credentials, that is what Adobe’s own FAQ says will happen. Citing new laws, including the California AI Transparency Act and the EU AI Act, Adobe attaches C2PA metadata automatically whenever a “qualifying” generative tool creates or modifies content, and its FAQ answers the obvious question plainly: “this metadata cannot be disabled for qualifying content.”

When Adobe adds them, and what goes in

There are three routes, and they behave differently.

An image that is 100% Firefly output, from Text to Image for example, gets Content Credentials automatically. An image you shot yourself and then changed with a generative tool in an Adobe app such as Photoshop gets them under the newer rule, and Adobe lists what that metadata may carry: the application, the generative model used, whether the content was wholly generated or modified, the date and time of the generative use, and a unique content identifier. Adobe says traditional AI-powered features don’t trigger it, only generative ones.

The third route is optional. Photoshop’s own Content Credentials feature, still labelled Beta, records attribution and edit history when you turn it on, and the Export As dialog then offers three choices for JPG and PNG, the only formats that support it: Publish to Content Credentials cloud, Attach to file, or None.

One thing Adobe’s pages don’t give is a list of which Photoshop tools count as “qualifying”. The FAQ separates generative tools from “traditional AI-powered features” and stops there. For a photographer deciding whether a given retouch will mark the file, that is the question that matters, and we couldn’t find the answer on any Adobe page we read.

What users report from Photoshop exports

These are reports from Adobe’s community forum, not confirmed behaviour. In June 2026 a user posted that Photoshop had started “forcing Content Credentials on ALL my exports, regardless or whether or not any AI was used”, with the credentials naming Adobe Firefly; the poster argued that a file cut down to a single rectangle shape should not be marked as AI. No Adobe staff member replied. The community experts’ explanation was that the flag follows the document’s history, so a generative edit that was later deleted still counts, and a template or recycled document can carry it before you start. One expert tested it and could still refuse credentials at export. The reply marked as the answer said the healing and spot healing tools don’t use generative AI; another pointed out that some tools, the Remove tool among them, can optionally use Firefly.

An older thread from June 2024 shows where this ends up: a user who used Generative Fill to extend backgrounds and remove blemishes found Instagram labelling the photos. How Meta reads these files is covered in the Instagram AI info guide, so we won’t repeat it here.

Adobe’s terms say not to remove them

Adobe’s Generative AI User Guidelines, updated 15 May 2026, say that Adobe “may attach or publish Content Credentials for content created or modified with generative AI features”, and then: “You must not remove, alter, or disable any Content Credentials.”

Technically the manifest in a Photoshop JPEG or PNG is an APP11 segment or a caBX chunk like any other, and our tool strips it before changing the pixels, as the C2PA removal guide explains. Whether you should is a matter between you and Adobe’s terms, and the page on whether removing AI metadata is legal goes further into it. Our view is narrower and practical. If the only generative edit in a photo was a blemish or a stray bin, redo that one spot with the Spot Healing Brush (which, per the marked forum answer, uses no generative AI) and export again. You end up with a file that has no reason to carry a generative claim, rather than a file with a claim you deleted.

Stripping also doesn’t reach every copy. Adobe says Firefly credentials “may be stored in Adobe’s public Content Credentials cloud”, and Photoshop’s Publish option stores them there on purpose, “allowing recovery if they’re stripped from your content.” Adobe’s Inspect tool is what does the recovering.

TrustMark, and a gap we haven’t tested

TrustMark is Adobe’s open-source invisible watermark, built to carry an identifier that points back to stored Content Credentials. Its FAQ is unusually candid: the repository ships code for removing the watermark, and the FAQ says anyone can build a detector for it or remove it. We found no Adobe page saying that every Firefly image carries TrustMark, and we have not tested our Maximum disruption pass against Adobe’s check, so we make no claim either way.

The other gap is newer. Photoshop’s Generative Fill now offers partner models next to Firefly, including FLUX.2 pro and Gemini 3 with Nano Banana Pro. Google says all media from its tools carries SynthID. Whether a patch filled by the Gemini model brings a SynthID mark into your photo, and whether Gemini’s check would pick it up from one filled region, we don’t know yet.