How it works
Three steps, no account required to try it.
Load your file
Drop an image or video into the browser. It is read into memory locally — there is no upload request, which you can confirm in your network tab.
Review what was found
The engine walks the actual byte structure and lists every metadata segment it finds: C2PA manifests, EXIF blocks, XMP packets and text chunks.
Export a clean copy
Metadata segments are dropped and the remaining bytes are rewritten. Pixels are untouched, so there is no re-encode and no quality loss.
What this tool handles
Provenance inspection
See every embedded segment with its size and a readable preview before you remove anything.
Content Credentials removal
The C2PA manifest store is located and dropped cleanly, leaving a valid file behind.
Frame-accurate masking
Position the mask over the watermark and preview the reconstruction before exporting.
Private by architecture
No upload endpoint exists. Privacy here is a property of the design, not a policy promise.
Before / after comparison
A split view in the Studio lets you check the reconstruction against the original.
Bulk processing
Strip metadata across an entire batch of Sora stills in one pass.
What gets removed
| Format | Removed | Preserved |
|---|---|---|
| PNG | tEXt, iTXt, zTXt, eXIf, caBX (C2PA/JUMBF), tIME | IHDR, IDAT, IEND, iCCP colour profile |
| JPEG | APP1–APP15 (EXIF, XMP, IPTC, C2PA) | SOI, APP0/JFIF, ICC profile, scan data |
| MP4 / WebM | Visible watermark across every frame, container metadata | Audio track, full clip length (real re-encode) |
Frequently asked
What does OpenAI embed in Sora files?
Sora output carries a C2PA Content Credentials manifest that records the generator and generation details, plus a visible watermark on video. The manifest sits in a dedicated metadata segment which this tool parses and removes.
Will this reduce my image quality?
No. Metadata scrubbing rewrites the container around your image data — it drops the metadata segments and copies the pixel data through byte for byte. There is no decode and re-encode step, so the output is pixel-identical to the input. ICC colour profiles are deliberately preserved so colours do not shift.
Do my files get uploaded anywhere?
No. The scrubbing engine is JavaScript running in your own browser tab. There is no upload endpoint and no server-side processing queue, so your media never leaves your device. You can verify this by opening your browser network tab while you scan a file.
Can it handle the moving Sora watermark?
The mask is a fixed region per export, so a watermark that moves across the frame is not fully handled by a single mask. Metadata removal is unaffected and works regardless of what the watermark does visually.
Is it legal to remove watermarks and content credentials?
It depends entirely 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 or identity documents is prohibited by our terms and may be unlawful where you are. You are responsible for how you use the output.
Is there a free tier?
Yes. You can scan and inspect files without an account at all. A free account adds a daily cleaning quota, and paid plans raise that quota and unlock batch processing and API access.
Other tools
Generator-agnostic
Inspect any image byte by byte and remove its embedded provenance data — C2PA Content Credentials, EXIF records, XMP packets and PNG text chunks — whichever tool produced it.
Open toolGoogle Veo & Flow
Veo powers both the Gemini app and Flow, Google’s AI filmmaking tool — and everything they render carries a corner watermark plus SynthID provenance.
Open toolGoogle Gemini · Nano Banana
Gemini image generation attaches provenance metadata and a SynthID mark to everything it produces.
Open toolRunway Gen-4 & Aleph
Runway exports carry a corner watermark plus embedded provenance metadata that identifies the clip as AI-generated.
Open toolMidjourney V7
Midjourney PNGs carry text chunks that can include your prompt, job identifiers and generator attribution.
Open toolGPT Image & DALL-E 3
Images generated by DALL-E 3 through ChatGPT carry C2PA Content Credentials identifying them as AI-generated.
Open toolYour files never leave your device
Every scrub runs in your browser tab. There is no upload endpoint, no queue, and no copy of your media on our side — by architecture, not policy.