How to Remove Flux AI Metadata: C2PA, EXIF and XMP Flags

By Calabi Labs Editorial Team · 2026-08-01

Flux images from Black Forest Labs usually have no visible watermark — the giveaway is invisible metadata baked into the file: a C2PA / Content Credentials manifest plus XMP and EXIF AI flags. Calabi strips that layer in one automatic pass and rewrites the file so it reads as a normal phone photo. Results vary by platform and source model; no tool can promise a platform won't flag you.

What actually flags your file

Because a Flux export often looks clean, people assume the file is clean. It usually isn't. The provenance lives in the metadata: C2PA / Content Credentials manifests stored as JUMBF atoms, XMP flags such as DigitalSourceType: trainedAlgorithmicMedia, and generator or tool tags naming the model or pipeline that produced the image.

These signals are exactly what platform detectors read. There's no logo to crop, so the metadata is the whole story — and it travels with the file when you upload, resize, or re-save it in most tools.

A raw AI export can also carry a bloated tag count compared with a real camera file. A genuine phone photo has a lean, structural EXIF set; an AI export often has far more. That mismatch alone is a signal.

How Calabi handles it

Calabi runs one automatic pass in three stages. It STRIPS the signals platforms scan for: the C2PA / Content Credentials manifest, the XMP AI flags like trainedAlgorithmicMedia, and the generator and tool tags. It INJECTS an authentic phone-capture identity — Make, Model, Software version, GPS coordinates, and a capture timestamp — using device profiles like iPhone 17 Pro Max, iPhone 16 Pro, Pixel 8 Pro, or Galaxy S24 Ultra. Then it VERIFIES with a forensic proof card from the same ExifTool scan platforms use.

In testing, Calabi reduced 18 JUMBF/C2PA atoms to 0, cleared 16 C2PA references, removed the trainedAlgorithmicMedia flag, and cut a raw AI export's 144 metadata tags down to about 94 neutral structural tags — all verified with ExifTool. For a no-watermark Flux file, this metadata pass is the entire job; there's nothing visible to crop.

Steps:

Upload the Flux image to calabilabs.com.
The pipeline automatically strips the C2PA, XMP, and EXIF AI flags.
It injects an authentic phone-capture identity into the file.
Read the forensic proof card to confirm the AI metadata is gone.
Download the cleaned file.

FAQ

My Flux image has no watermark — is there anything to remove?

Yes. The provenance is invisible: a C2PA manifest plus XMP and EXIF AI flags baked into the file. Detectors read that metadata, so an image with no visible mark can still be flagged. Calabi strips that layer.

Does Calabi change the picture itself?

No. It works only on invisible metadata and signals. It does not recolor, edit, or reconstruct any region — the image looks identical after processing.

Can I confirm the metadata was actually stripped?

Yes. Each run returns a forensic proof card built with ExifTool, the same scanner platforms use. It shows the C2PA atoms and AI flags reduced to zero before you download, so you're not taking it on faith.

Try Calabi free at calabilabs.com — 10 cleans, no card.