Lightroom Content Credentials: why a real photo gets an “AI info” label

Why a real photo can say “AI info”

Meta’s own post on its approach says it adds “AI info” labels when it detects industry-standard AI image indicators or when people disclose AI use. It also says that some content with minor modifications using AI, such as retouching tools, carried those indicators and was labelled “Made with AI”, which is why the name changed to “AI info”. Its September 2024 update says content detected as only modified or edited by AI tools has the label moved to the post’s menu, while content detected as generated by an AI tool keeps the label on the post (Meta, April 2024, updated through 2025). Meta’s Facebook help page adds that photo editing tools that use AI to change the size or colour of an image may add such a signal (Facebook Help Center).

So the label is a reading of the file. If the file says “a generative tool touched this”, it can say so even when the picture is mostly a real photograph.

Where the signal lives in the file

Two places matter, and they can be in the same JPEG.

• A C2PA manifest (Content Credentials). A signed record inside the file, stored in JPEG APP11 segments, that can list actions such as “opened” and “edited”, each with a source type. See what Content Credentials are.
• An IPTC DigitalSourceType value in XMP. A plain field naming how the image was made. The IPTC vocabulary includes these values:

From the IPTC Digital Source Type vocabulary, read 2026-10-04. Which of these a platform treats as “AI” is the platform’s choice; we only know what Meta says in the posts above.

What Adobe says about Lightroom exports

Adobe’s help page for Lightroom (Content Credentials in Adobe Lightroom, last updated 25 Aug 2026 when we read it) says Content Credentials are applied when you select Apply Content Credentials in the export dialog. Under Preferences > Export you pick a storage method: publish to the Content Credentials cloud, attach to the files, or both, and you can also choose not to include them. The page also says that the Edit in Photoshop round trip is not supported by Content Credentials, so a photo moved between the two apps needs an export with credentials at each step.

That is the first fix, and it is free: decide the setting before you export. This page is for the files you have already exported, or that came out of another editor.

Fix order

1. Be honest about the edit. If a generative tool changed what the picture shows (filled in an area, extended the frame), the file’s statement is true and your audience may deserve to know. Disclose it on the post if the platform or the law asks.
2. Set the export before you export. In Lightroom, choose not to include Content Credentials if you do not want them in the file.
3. For files already exported: inspect one, then run AI only mode on the folder. It removes the manifest and the AI source type and leaves your copyright and camera data.
4. Upload the cleaned copy and keep the originals. A post that is already live is not changed by cleaning another file.

What we tested

We built six JPEGs (3000×2000, Nikon-style EXIF, orientation “rotate 90”, sRGB profile, IPTC copyright and by-line, XMP creator, and a Lightroom Classic name in the Software and CreatorTool text). Two carry a C2PA manifest that we signed ourselves with a test certificate, recording an “opened” and an “edited” action with a source type. Three carry only an XMP source type. One is a control with a camera-capture source type. They are synthetic: not exported from Lightroom, Capture One, or Topaz, because those programs are not on the test machine, and no real person or place is in them. We ran each through this tool in headless Chrome, then checked the downloads with ExifTool 13.25 and the C2PA reader (c2pa-python 0.38). Every download decoded to the same pixels as its original.

Test 1: AI only on a manifest that records a generative edit

Test 2: the same file in All metadata mode

Test 3: the AI source type in XMP, no C2PA

What the tool does not judge

Rows 4 to 6 of the table above are worth reading. The tool looks for the source-type field and removes it whatever the value is. In our run the XMP block was removed for the value meaning noise reduction and for the value meaning a plain camera capture, exactly as for the generative-editing value. It does not tell you whether a given value should trigger a label. Use the Before list to see which value your file holds, and the IPTC table above to read it.

Other editors: Capture One, Topaz, Photoshop

Meta’s help text speaks of photo editing tools in general, not of a list of programs, and we have not run Capture One or Topaz exports through anything. Whatever you use, the check is the same: export the file you will upload and inspect that file. If it shows a C2PA row or an AI source type, that is what a platform can read. If it shows neither, cleaning will not change a label that comes from somewhere else. For Photoshop, see Photoshop Generative Fill metadata, which also covers the layer you deleted but whose history stayed.

What this does not do

• It does not edit a post that is already live. Clean the file, then upload it again; see why Instagram can still show AI info.
• It does not remove a Content Credentials cloud record. Adobe describes the cloud copy as separate from the file.
• It does not touch pixel-level marks. SynthID and similar signals are in the image, not in a field.
• It was not tested against Instagram, Facebook, or LinkedIn. We did not post anything, so we cannot say whether a given platform will drop a label for a cleaned file.

Where to go next

For the general question of what a label can and cannot be fixed with, read remove an AI label from a photo. For Instagram, the owner page is remove Instagram AI info, with the older name covered at Made with AI. LinkedIn shows a different icon: LinkedIn Content Credentials. For the manifest itself, see remove C2PA and check for C2PA, and for handing a whole set to a client, clean photo metadata before delivery and XMP and IPTC.

Questions

Does Instagram label every photo edited in Lightroom?

No. Meta says it adds the label when it detects industry-standard AI signals in the file, or when someone self-discloses. Whether a Lightroom export contains such a signal depends on the tools used on it and on the export settings, so inspect the actual file instead of assuming.

Why would a real photo carry an AI signal at all?

Editors that use AI tools can record that in the file. Meta’s own post says content with minor modifications using AI, such as retouching tools, carried industry-standard indicators and was labelled “Made with AI”, which is why it renamed the label “AI info”. The file is stating a fact about one edit, not that the whole picture was generated.

Is it legitimate to remove those fields?

It depends on the facts. If a generative tool changed what the picture shows, removing the fields does not remove your responsibility to say so where the platform or the law asks you to. If the signal comes from something that does not change the content, such as noise reduction, many photographers feel the label misleads. That judgment is yours.

Will the label disappear after I upload the cleaned file?

Not necessarily, and we did not post anything to a platform to test it. A platform can also label from a self-disclosure or from its own detection, and a post that is already live is not changed by cleaning another copy. See why Instagram can still show AI info after a clean file.

What does AI only keep, and what does All metadata remove?

In our tests AI only removed the C2PA manifest and the AI source type but kept the camera model, lens, exposure, artist and copyright strings, IPTC, orientation, and the colour profile. When the AI source type sat in the XMP block, the whole XMP block was removed, so XMP-only fields such as the creator went with it. All metadata removed every one of these except orientation and the colour profile.

Does the tool decide whether the AI source value is really AI?

No. It looks for the DigitalSourceType field and removes it whatever the value is. In our tests the values for generative editing, for algorithmic enhancement such as noise reduction, and even for a plain camera capture were all handled the same way.

Did you test real Lightroom, Capture One, or Topaz exports?

No. The files on this page are synthetic and were built to look like such exports: a camera-style JPEG with a signed C2PA manifest recording an edit, or an AI source type in XMP. We do not have those programs on the test machine, so we make no claim about what a given version writes. Inspect your own export with the tool to see what it holds.

Does this remove the credential Adobe may hold in the cloud?

No. Adobe says Content Credentials can be published to a cloud record separate from the file. This tool rewrites only the local image.

Related guides

• Remove an AI label from a photo: what actually works
• Remove Instagram AI info metadata
• Photoshop Generative Fill metadata
• What LinkedIn Content Credentials are
• Open the Remove AI metadata tool on the homepage if you want the general cleaner.
• Short Instagram upload workflow if you only need the four-step version.