Frequently asked questions

Embedded metadata segments. In PNG that means tEXt, iTXt, zTXt, eXIf, tIME and caBX (the JUMBF box holding a C2PA manifest). In JPEG it means the APP1–APP15 marker segments carrying EXIF, XMP, IPTC and C2PA data. Image pixel data, and the ICC colour profile, are preserved.

No. The engine is JavaScript running in your own browser tab. There is no upload endpoint and no server-side processing queue. You can verify this by opening your browser network tab while scanning a file — you will see no request carrying your data.

No. The container is rewritten and the compressed pixel data is copied through byte for byte. There is no decode and re-encode step, so the output is pixel-identical to the input. This is the main advantage over tools that simply re-save the image.

Because removing it visibly shifts colour on wide-gamut images, which is a quality regression nobody expects from a metadata scrub. In JPEG this required extra care, since ICC profiles and C2PA manifests can share the APP2 marker — we check the payload identifier so the profile survives and the manifest does not.

PNG and JPEG can be rewritten losslessly in the browser. WebP and video containers such as MP4 and WebM can be inspected for generator signatures, but browsers cannot re-encode them efficiently, so those export a single frame as PNG instead of a full file.

Partially, and we would rather be precise about it. The Studio can mask and reconstruct a watermark region on a video frame, and metadata inspection works on the container. But full video re-encoding is not practical in a browser, so video export produces the current frame as a PNG. For a re-encoded clip you need a desktop editor.

No. Removing the manifest removes one declared signal. Classifier-based detection, pixel-level and steganographic watermarking, and platform-side provenance records are separate mechanisms this tool does not touch. Any tool claiming otherwise is overstating what it does.

No. This is a metadata tool. Watermarks encoded into the pixels themselves are a fundamentally different problem and are outside what the engine attempts.

Yes. The Batch Scrubber in the workspace queues multiple files and strips metadata from each in turn, giving you a cleaned copy of every one. It runs locally like everything else.

Not to inspect files or to use the free scrubber — those work with no sign-up at all. An account adds a daily cleaning quota, saved history and access to the workspace tools.

There is a permanently free tier with a daily quota. Paid plans raise the quota and add batch processing and API access. Current prices are on the pricing page, and you can cancel at any time.

Annual plans are fully refundable within 14 days of the charge, no justification needed. Monthly plans can be cancelled at any time, with access running to the end of the paid period. Full detail is on the refund policy page.

Yes, on the Agency plan. It exposes the same scrubbing engine over REST so you can call it from a render pipeline or backend service. See the developer docs for the endpoint and payload format.

It depends on the media and your intent. Removing marks from work you own or are licensed to edit is normally fine. Stripping provenance data to pass AI-generated media off as authentic in journalism, elections, legal evidence, academic work or identity documents is prohibited by our terms and may be unlawful where you are. You are responsible for how you use the output.

For an account: your email, a salted password hash, and subscription status. Theme and tool preferences live in your browser local storage and are never sent to us. Your media is never received at all. Full detail is in the privacy policy and DPA.

That is a large part of why it was built this way. Because no file leaves your device, using PurifyAI does not involve disclosing client media to a third-party processor. If your client requires a signed DPA covering your account data, contact us and we will provide one.