PHOTO PRIVACY · TESTED RESULTS

What Is FBMD Metadata in Facebook Photos, and How Can You Remove It?

FBMD is a value that can appear in the IPTC Special Instructions field of a Facebook-sourced photo. In our tests, No Trace Share's current web processing engine detected and removed it from a public Facebook-origin test fixture without changing the decoded image pixels. Removing that value does not make a photo anonymous or prove what Meta uses the value for.

A photograph has two things worth examining before you share it: what the picture shows and what the file carries. A photo downloaded from a social network may have lost its original camera details and still contain information added during its journey. That is why checking only for a camera model or a location map can leave an incomplete picture of the file you are about to send.

This guide separates three questions: where FBMD appears, what a cleaner can do about it, and what the result actually establishes. We tested the deployed No Trace Share web engine with a public Brave fixture and controlled synthetic JPEGs. The results below describe those files and those cleaning choices. They are not a survey of all Facebook or Instagram download methods.

Where the FBMD value lives

In its 2019 investigation, IPTC reported that Facebook populated an Instructions field with a value beginning with FBMD. It observed that this value persisted when another user uploaded a downloaded copy. IPTC did not establish what Facebook used it for. That historical finding should not be stretched into a claim about every present-day download. [1]

For the JPEG structure tested here, the value lives in IPTC-IIM dataset 2:40, carried inside a Photoshop image resource in an APP13 segment. No Trace Share now displays recognized FBMD values as Facebook photo identifier (FBMD). The underlying field remains Special instructions (IPTC). Those words describe the metadata field. They do not mean that the instructions were written by the photographer or that their contents are safe to retain.

FBMD is not a visible Facebook tag on a person, a caption displayed next to a post, or a location pin in a photo library. Removing a person tag on Facebook does not edit this embedded field. Similarly, changing a downloaded file's name does not remove the field inside its bytes. Treat the attachment name, the picture, and the embedded properties as separate things to review.

Why the field is receiving attention again

Brave's November 12, 2026 announcement describes FBMD removal for desktop downloads and uploads, beginning with version 1.98. It also identifies further metadata coverage and mobile support as planned work. This guide does not independently test Brave's release configuration or whether the protection is enabled in a particular installation. [2]

The useful lesson is practical: a field added by a platform can deserve inspection even when familiar camera metadata is absent. The existence of an embedded value is observable. The platform's internal use of that value is a different question. Neither our cleaning result nor the presence of the field proves that Meta used it to identify a person or connect two accounts.

For someone preparing a copy to share, the immediate decision is narrower. If a platform-added instruction has no purpose in the copy you intend to send, you can choose to remove it and check the output. That decision does not require presenting a speculative explanation of the value as an established fact.

What we tested

On November 17, 2026, we first tested the processing worker fetched from the public No Trace Share website, then repeated all seven cases against the built v0.9.7 release worker. Both runs executed the inspection and cleaning interface locally in Node.js. We recorded the worker hashes; all seven output files from the release worker matched the earlier outputs byte for byte. This is an engine test, not a recording of someone clicking through the production interface or a new device compatibility test.

The first input was Brave's public FBMD test image. The introducing change describes the fixture as a Facebook-origin image resized while preserving metadata. We retrieved the repository fixture rather than downloading a new photo through our own Facebook account. Its file hash was f0b10009d892bf6d4a4e0c332dedef5e50ca3589b452b824e634699230000725. [3]

We also generated an original 320 × 200 color-pattern JPEG and inserted deliberately labeled test properties. One field contained an artificial FBMD-prefixed string; another contained a synthetic copyright notice. A further variant stored the fields in separate APP13 segments. These controlled files let us test decisions without using anyone's personal photographs or treating invented values as genuine platform identifiers.

Each output went through the engine's inspection again. Separately, a Python check examined the known IPTC record structure in the fixtures and decoded the input and output images with Pillow. We compared dimensions and every RGB pixel. This independent check helps distinguish an absent field from a changed label and a preserved picture from a merely similar-looking preview.

Results: default and selective cleaning

All seven cases met their expected outcomes. Decoded RGB pixels matched in every input-output pair.

Test case FBMD after cleaning Other result
Public Brave fixture, default clean Removed No supported findings remained
Synthetic FBMD and copyright, default clean Removed Synthetic copyright also removed
Synthetic file, explicitly keep copyright Removed Exact copyright notice retained
Synthetic file, explicitly keep FBMD instructions Retained Retained value remained visible in inspection
Synthetic file, replace instructions Original value removed Exact replacement text retained
Synthetic file with two APP13 segments Removed No supported findings remained
Synthetic JPEG with neither test field Absent before and after No supported findings introduced

The selective cases matter as much as the default result. A cleaner with retention controls should respect an explicit decision to keep a field. In our test, choosing Keep for the FBMD-bearing instructions did exactly that. Therefore, the claim supported by this work is that default cleaning removed the tested FBMD field, with selective removal also available. It is not that every possible combination of settings removes it.

Keeping copyright produced a different useful result: the output retained the exact synthetic copyright text while the FBMD instructions disappeared. That demonstrates field-level separation for the tested JPEG. It does not establish that every attribution field in every format can be preserved, and it does not determine who owns a real photograph.

How to check and clean your own copy

Start with the actual JPEG you intend to send. If it came through several apps, inspect that received or exported copy rather than an earlier original. Our tests did not establish whether a particular Facebook, Instagram, or messaging download route always includes FBMD. An absent value in one file is a result for that file.

  1. Open the No Trace Share browser cleaner and select the supported file. Processing of the selected file takes place locally in the browser.
  2. Review File Footprint for Facebook photo identifier (FBMD), the name used for a recognized FBMD value in Special instructions (IPTC). If the value begins with FBMD, you have found the type of field covered by this test.
  3. Use the default Clean Now workflow, or review individual choices if you need to retain specific information. Do not choose Keep for the FBMD-bearing instructions if your goal is to remove that value.
  4. Review the output inspection. A copyright field you intentionally retained should still appear; that is consistent with your choice rather than a reason to assume the cleaning failed.
  5. Download the generated copy and select that copy for sharing. For an additional check, inspect the downloaded file again before attaching it elsewhere.

Keep the original if you need its history or other properties later. Working with a separate output makes it easier to preserve an archive while making a deliberate decision about what leaves your device. Also read the selected attachment name before sending. A cleaned file is useful only if it is the file you actually attach.

What if you want to retain attribution?

Before cleaning, write down the exact credit you want the recipient to receive. A copyright notice, author field, and instruction field have different purposes, even when a viewer lists them together. Review the actual values rather than keeping every item in a broad category because one of them is useful.

In our synthetic test, we retained only the copyright notice and left the FBMD-bearing instructions set for removal. The final inspection showed the notice and no FBMD value. A professional sharing workflow should make both checks explicit: confirm the identifying field is absent, then confirm that the intended attribution survives in the output.

If the receiving publisher needs credit in a caption, include that credit in the agreed delivery instructions as well. Our experiment concerns embedded file properties. It does not test how another website imports credits, displays notices, rewrites metadata, or complies with a license. For a related worked workflow, see removing GPS while keeping copyright.

What removing FBMD does not establish

A successful field-removal test answers a byte-level question about a particular output. It does not show that every way of identifying an image has disappeared. An unchanged picture can still contain a face, a house number, a badge, a document, or another detail you did not intend to disclose. Inspect the visible content separately.

Our decoded-pixel comparison deliberately confirms that the tested cleaner did not alter the image. That preservation result also means you should not expect it to conceal recognizable content. If the concern is a visible identity clue, cropping or appropriate image editing is a separate task; it is outside the evidence established by these metadata tests.

The tests also do not demonstrate deletion of data a service already received, removal of existing posts, or prevention of future platform processing. If another app later exports or transforms the file, inspect the resulting copy again when the decision matters. The evidence here stops at the bytes produced by the tested worker.

Questions to ask when evaluating a cleaning claim

Ask which file and metadata structure were tested, whether default settings or selective choices were used, and whether someone inspected the output rather than only the input. A useful report should explain its sample's origin and distinguish a synthetic fixture from a newly downloaded platform sample.

Also ask what “unchanged” means. Identical file hashes would mean identical bytes, including unchanged metadata. A cleaned file normally has a different file hash. In this study, the preservation claim concerns decoded RGB pixels, which were equal even though the metadata bytes changed. Those are different checks serving different purposes.

Finally, look for a description of what remains uncertain. Here, the evidence covers one public Facebook-origin fixture and controlled JPEG cases, using the deployed web engine fetched on the stated date. It does not independently verify iOS or Android app builds, every JPEG layout, or Meta's internal behavior. Future changes to the deployed engine should be checked against these cases before reusing the exact result as current release evidence.

Check the file you are about to share

If your downloaded JPEG contains an FBMD-bearing Special Instructions field, the tested default No Trace Share web engine can remove that structure. You can also retain selected copyright information in the tested workflow. Check the resulting copy, honor any retention choices you made, and review what the visible image still reveals.

Inspect a file locally, or read the format coverage and safety limits before preparing a sensitive attachment.

Sources and method

  1. IPTC's Facebook metadata investigation, July 22, 2019. Historical platform observations; not a current universal behavior guarantee.
  2. Brave's November 12, 2026 announcement. Primary source for Brave's stated release scope; Brave itself was not tested here.
  3. Brave's public fixture at the pinned repository revision, and the introducing change. Third-party fixture provenance is attributed to its authors.
  4. No Trace Share engine experiment, November 17, 2026. Seven cases; v0.9.7 release worker (hash recorded in the test summary); saved input/output hashes, exact selections, reinspection results, and independent Pillow comparisons. The accompanying evidence package contains the reproducible test method. Native app behavior was not evaluated. Download the test summary (JSON).