How to Remove a Midjourney Watermark

If you generate images with Midjourney, the “watermark” you are most likely dealing with is not a logo in the corner. It is a C2PA provenance manifest — machine-readable data that records how the image was made and that it came from Midjourney. This guide explains what that signal is, how to verify it, and how to remove it from an image you own.

Does Midjourney add a watermark?

Midjourney has joined the C2PA ecosystem — the Coalition for Content Provenance and Authenticity, the same standard Adobe and OpenAI use. Images generated on the platform can carry a signed C2PA manifest, often surfaced to users as Content Credentials. The manifest records the producer, the actions taken (for example that the image was created), a timestamp, and an instance identifier.

This is provenance, not a visible mark. Unlike a stock-photo logo overlaid on a picture, a C2PA manifest is an invisible block of metadata stored inside the file. That distinction changes everything about what “removing” it means, and about whether it is legal to do so.

In Midjourney’s web app you may see a small “Content Credentials” indicator on an image, but on most other platforms you will see nothing at all: the manifest stays invisible until a tool reads the file. That is why verifying an image means opening the file itself, not looking at what surrounds it.

How C2PA stores the manifest

Understanding where the signal lives makes the removal step much less mysterious. A C2PA manifest is not written into the pixels; it is stored as a separate block — technically a JUMBF box — attached to the image container. In a JPEG that block sits in an APP1 segment, and PNG and WebP use their own equivalent chunks. The image data and the manifest are side by side, not fused together.

That structure has two consequences. It is why a manifest can be removed without touching image quality: you are deleting an attached block, not re-encoding the picture. And it is why the manifest often disappears on its own — platforms that re-encode uploads for the web routinely strip metadata blocks. A Midjourney image can come back from a social platform with its C2PA manifest already gone, which is exactly why a detector that also checks for latent signals, and is honest when it finds neither, is more useful than one that only looks for the box.

How to check a Midjourney image

Before removing anything, it helps to confirm what is actually in the file. OutWatermark AI reads the file structure directly and shows you the evidence.

When you drop a Midjourney image into the detector, you get:

  • A visual stamp on the preview — ⚠️ AI WATERMARK DETECTED when provenance is present, ✅ NO WATERMARK DETECTED when it is not.
  • The decoded manifest — the generator or producer name, the recorded actions, the timestamp, and the instanceID. You can read the provenance claim field by field instead of trusting a single word.
  • Signal badges marking which families were found — C2PA, EXIF, XMP, or text chunks.
  • A location map showing exactly where the manifest block sits in the file’s bytes, with its offset and length.

This matters because it lets you distinguish the C2PA manifest (removable) from any latent signal (not directly detectable, see below).

How to remove the signals

Once you have confirmed the image is yours, there are two levers:

Surgical cleanup — lossless. The C2PA manifest, EXIF, XMP, and IPTC fields live in blocks separate from the image data. OutWatermark AI removes those blocks byte-for-byte, leaving the pixels untouched. There is no re-compression, so there is no quality loss. This works on JPG, PNG, and WebP.

Purification — weakening latent signals. If you are also concerned about an invisible pixel-level signal such as SynthID, the tool offers controlled re-encoding at three strengths (light, standard, aggressive). This is deliberately described as weakening, not removing: SynthID is designed to survive mild transformations, so no tool can honestly guarantee it is gone.

Before any download, the tool asks you to confirm that you own the content or have the rights to modify it.

What cannot be removed

Two honest limits apply to Midjourney images:

  • Visible watermarks are out of scope. If an image has an actual logo or overlaid text — for example from a stock provider, or a third-party mark — OutWatermark AI does not remove it. It handles provenance signals, not visible protection.
  • Latent signals are not directly detectable. SynthID and similar signals are embedded in the pixels and cannot be read from the file structure the way C2PA can. Purification can weaken them, but it cannot certify removal. Also remember that the absence of a detectable signal does not prove an image is clean — metadata can be stripped by a platform while a latent signal remains.

What the law says

If the image is yours — you generated it and you hold the rights — removing its provenance metadata is about controlling your own data. That is the use case OutWatermark AI is built for.

Two boundaries still apply. First, under Article 50 of the EU AI Act (Regulation (EU) 2024/1689), synthetic content must be labelled as such where required; removing the technical watermark does not remove that labelling obligation. Second, removing provenance from images you do not own can violate copyright law — in the United States, 17 USC 1202 covers removal of copyright management information. The compliance overview lays out both with sources.

A quick check before you publish

If you are about to post or deliver a Midjourney image, the whole routine takes three steps:

  1. Open the free detector and drop in the image. The preview shows a stamp and the decoded manifest.
  2. Read what it found — the producer, actions, timestamp, and any EXIF or XMP signatures.
  3. Clean it if it is yours. Confirm ownership, then download the cleaned copy, or re-encode to weaken a latent signal.

The point is not to hide anything from anyone; it is to know exactly what your file carries before it goes somewhere you cannot control.

FAQ

Does Midjourney add an invisible watermark? Midjourney has adopted C2PA, so generated images can carry a signed provenance manifest (Content Credentials). This is invisible metadata, not a visible logo.

Can I remove the C2PA manifest from my Midjourney image? Yes — for an image you own. OutWatermark AI removes the C2PA and related metadata blocks surgically, without re-compressing the pixels. You confirm ownership before download.

Is the tool free and local? Yes. Detection and cleanup run in your browser, free, with no account and no upload. Your image never leaves your device.

Does it remove the visible watermark or SynthID? No on visible watermarks — the tool does not handle logos or overlaid text. For SynthID, it offers purification to weaken a latent signal, but cannot guarantee removal.

Check and clean your Midjourney image for free →

For more on image provenance, read our page on removing AI watermarks from images.