What is the difference between C2PA and a watermark?

C2PA is metadata that can be read and removed independently of image pixels. A watermark like SynthID is embedded directly into pixel values and cannot be removed by any metadata tool.

Remove AI Metadata & C2PA Credentials

Q: Will removing C2PA credentials prevent platforms from flagging my image as AI-generated?

Removing C2PA removes the metadata signal platforms read to auto-label content. However platforms may also use pixel-level AI detection algorithms that operate independently of metadata.

Q: Does removing AI metadata break the image?

No. For JPEG files the lossless stripper removes APP segments while leaving compressed scan data untouched. The resulting file is a valid standard image.

Q: Can I see what the C2PA manifest contains before removing it?

Yes. Ghoststrip shows every metadata field found including C2PA issuer and generation software before you download the cleaned file.

What is C2PA and why does it matter?

C2PA (Coalition for Content Provenance and Authenticity) is an open standard for embedding signed provenance data inside media files. It was designed to combat misinformation by letting platforms verify where an image came from and how it was created.

In practice, C2PA manifests are embedded by AI image generators — OpenAI's DALL-E, Adobe Firefly, Microsoft Designer, and others — to mark images as AI-generated. Platforms including LinkedIn, TikTok, and Meta read these manifests and label content accordingly. If you share an AI-generated image with C2PA credentials intact, the platform knows it's AI-generated before you've said a word.

🔏

C2PA manifest

A cryptographically signed block embedded in the image file containing provenance data — who created it, with what tool, at what time. Embedded by DALL-E, Firefly, Designer.

🎨

Stable Diffusion prompts

Many Stable Diffusion interfaces (AUTOMATIC1111, ComfyUI, Forge) embed the full generation prompt, model name, seed, and parameters into PNG metadata chunks.

🖼️

Midjourney job IDs

Midjourney embeds job IDs and generation parameters in image EXIF data — linking the image back to your Midjourney account.

⚡

JUMBF containers

The JPEG Universal Metadata Box Format container that wraps C2PA data in JPEG files — present in images from multiple AI platforms.

🤖

XMP AI markers

Adobe's XMP metadata format, used to carry AI-generation provenance data in TIFF, PDF, and other Adobe formats.

🔬

Platform-specific tags

Proprietary EXIF tags written by specific AI tools — Firefly writes to EXIF UserComment, some tools write to IPTC fields.

Important: pixel-level watermarks cannot be removed by any tool

Google's SynthID and similar pixel-level watermarks are embedded directly into the image's colour values at generation time — they are not metadata. No metadata cleaner can remove them because they are not metadata. Ghoststrip removes all metadata-based AI markers (C2PA, prompt chunks, job IDs) but cannot affect pixel-level watermarks. If a tool claims to remove SynthID, it is misleading you.

How Ghoststrip removes AI metadata

01 —

JPEG: lossless APP segment removal

C2PA manifests in JPEG files live in APP11 (JUMBF) segments. Ghoststrip's lossless JPEG stripper walks the marker stream and removes all APP segments — including APP11 — while leaving the compressed image data byte-for-byte identical. Zero quality loss.

02 —

PNG: chunk-level filtering

Stable Diffusion prompts, parameters, and C2PA data in PNG files are stored in tEXt, iTXt, zTXt, and caBP chunks. Ghoststrip applies an allowlist approach — only structural chunks (IHDR, IDAT, PLTE, IEND) are kept. Everything else is removed.

03 —

AVIF: canvas re-encode

AVIF files carry C2PA credentials in their ISOBMFF container. Ghoststrip re-encodes via canvas, which destroys the container metadata entirely and rebuilds a clean AVIF from the pixel data.

04 —

Diff view shows what was found

The report shows every AI credential found — C2PA issuer, generation software, job IDs, prompts — before removal. Download the clean file once you've confirmed what was stripped.

Which AI tools embed detectable metadata?

Metadata-based credentials — fully removed by Ghoststrip

✓ DALL-E / OpenAI C2PA manifest in JPEG APP11. Removed by lossless segment stripper.

✓ Adobe Firefly C2PA manifest + XMP markers. Removed from JPEG and PNG.

✓ Stable Diffusion Generation prompt, model, seed in PNG tEXt/iTXt chunks. All removed.

✓ Midjourney Job ID and parameters in EXIF. Removed for JPEG (lossless) and PNG.

✓ ComfyUI / A1111 Workflow JSON and prompt data in PNG metadata. Fully removed.

Pixel-level watermarks — cannot be removed by any tool

✗ Google SynthID Embedded in pixel colour values at generation. Not metadata — not removable.

Frequently asked questions

Will removing C2PA credentials prevent platforms from flagging my image as AI-generated?

Removing C2PA and other metadata-based credentials removes the signal that platforms read to auto-label content. However, platforms may also use their own AI detection algorithms that analyse pixel patterns — these operate independently of metadata. Ghoststrip removes the metadata layer; what happens at the platform's detection layer depends on that platform's own systems.

Does removing AI metadata break the image?

No. For JPEG files, the lossless stripper removes APP segments while leaving the compressed scan data untouched — the image is pixel-perfect. For PNG, only metadata chunks are removed, pixel data is preserved. The resulting file is a standard, valid image that opens in any viewer.

What's the difference between C2PA and a watermark?

C2PA is metadata — it lives in a structured field in the file that can be read, copied, and removed independently of the image pixels. A watermark (like SynthID) is a signal embedded directly into pixel values — changing specific colour intensities in a pattern that's invisible to the human eye but detectable by an algorithm. One is in the file's metadata layer; the other is in the pixel layer. They are completely different mechanisms.

Can I see what the C2PA manifest contains before removing it?

Yes. Ghoststrip's diff view shows every metadata field found, including C2PA issuer, generation software, and any other embedded credentials. You see the full report before downloading the cleaned file.

Does this work on images from AI video tools?

Ghoststrip processes still images only — JPEG, PNG, AVIF, WEBP, HEIC, TIFF, and GIF. Video files are not currently supported. AI video tools embed metadata differently and would require a separate processing pipeline.

Remove AI metadata from your images now

Free, private, instant. C2PA, Stable Diffusion prompts, Midjourney IDs — all removed. Zero quality loss for JPEG.

Strip AI metadata — no sign-up needed