Free · 100% local · no account

Remove AI watermarks from your own files

Detect and remove C2PA metadata, Content Credentials, EXIF, XMP, and generator signatures from images, videos, and text. Everything runs in your browser.

Your files never leave your device.

$outwatermark-ai detect image.jpg
C2PA / Content Credentials: FOUND
producer: OpenAI
$outwatermark-ai strip image.jpg
verified: no provenance signal detected

Detects output from

OpenAIGoogleAdobeMetaStability AIMidjourney

Why it matters

Platforms strip metadata. Latent watermarks remain.

C2PA metadata can disappear during upload or re-encoding. Major generators now combine metadata with latent signals.

C2PA is removable

A C2PA manifest is a separate metadata block. Re-encoding, screenshots, or metadata stripping can remove it.

Two signals, one file

OpenAI uses C2PA and SynthID for images, while Google uses SynthID across Gemini, Imagen, and Veo.

Rules changed in 2026

EU AI Act Article 50 applies from August 2, 2026 and requires machine-readable marking of synthetic output.

The tool

Detect. Inspect. Clean.

A clear three-step workflow. Nothing is removed before you see what the file contains.

Detect

Add a file. The engine reads C2PA/JUMBF boxes, EXIF/XMP metadata, and signatures from OpenAI, Google, Meta, Adobe, Stability, Midjourney, and others.

Inspect

Review the decoded producer, claim generator, date, actions, metadata, and exact signal location.

Clean

Remove metadata without resampling. Optionally re-encode images to weaken latent signals, then verify the result automatically.

Signals handled

One AI watermark remover, multiple signals

C2PA / Content Credentials

Detect and remove signed JUMBF manifests in JPEG, PNG, WebP, AVIF, HEIC, and MP4 containers.

EXIF / XMP / IPTC

Find software fields, descriptions, provenance tags, digitalSourceType, and ai-generated markers.

Generator text fields

Read Stable Diffusion and ComfyUI parameters, including prompts, samplers, and models.

Latent signals

Controlled image transformations can weaken embedded signals such as SynthID, without an absolute guarantee.

Text signals

Analyze statistical repetition patterns. Text rewriting to neutralize AI text watermarks is coming soon.

Video and audio

Inspect C2PA boxes in MP4 and MOV containers and native container tags. Latent video support is planned.

The 2026 context

Why AI watermark detection matters now

August 2, 2026

EU AI Act Article 50 begins applying to machine-readable marking of synthetic output (source: EUR-Lex 32024R1689).

Dual signals

OpenAI adds SynthID alongside C2PA on images (help.openai.com, accessed August 19, 2026).

Growing demand

The most visible open-source project in this field passed 14,000 GitHub stars in eight days as of August 19, 2026.

FAQ

Common questions

Is it legal to remove AI watermarks from my images?

You may remove provenance metadata from content you created or have the rights to use. Removing copyright information from someone else’s work may violate the law, including DMCA Section 1202 in the United States. OutWatermark AI is for your own content only.

Does OutWatermark AI remove Google SynthID?

Purification uses controlled transformations and re-encoding to weaken latent signals such as SynthID. It cannot guarantee removal because SynthID is designed to survive transformations. C2PA, EXIF, and XMP metadata removal is complete and lossless.

Which formats are supported?

The browser analyzes JPEG, PNG, WebP, and top-level C2PA boxes in MP4, MOV, AVIF, and HEIC files. Image purification supports JPEG, PNG, and WebP.

Are my files stored on your servers?

No. Analysis and cleanup run entirely in your browser. Your files never leave your device, are never stored, and are never used to train a model.

Ready to take control of your content?

Free detection, no account required. See what is in your file before you act.

Try the free detector