Runway Gen-3 Alpha (and Gen-2 since the 2024 update) embeds a C2PA Content Credentials manifest in every still export and exported frame, plus a Runway-branded XMP block. If you extract a frame for a thumbnail or composite, both signals propagate into the JPEG. GhostMeta detects the Runway signature in the APP11 segment or PNG caBX chunk, displays it, and strips it via canvas re-encode. The visible Runway logo overlay (if your tier shows one) is pixel-burned and not removable by metadata strip. When you extract a Runway still and drop it into a composite, the C2PA `c2pa.actions` and ingredient assertions chain forward into the new file, so a marketplace or stock reviewer can trace the AI origin even through an edit. Sellers on Adobe Stock, Etsy, and print-on-demand stores can get listings auto-flagged because the JUMBF box in the JPEG APP11 segment still carries the Runway `claim_generator` after upload. GhostMeta scans that box plus the Runway-branded XMP block and drops both in one canvas re-encode, so the delivered still carries no provenance assertion at all.
Said upfront: a visible watermark lives in the pixels and this tool does not remove it, and video is not supported. On a still image, however, GhostMeta removes EXIF, IPTC and XMP metadata, PNG text chunks and the provenance manifest, by re-encoding in your browser.
Runway is a video generation tool. Two things therefore fall outside GhostMeta's scope: video files, which are not processed, and visible watermarks burned into the image, which belong to the pixels rather than the metadata.
A visible watermark cannot be removed by cleaning metadata. It would require retouching pixels, which this tool does not do and does not claim to do. If that is your need, nothing described here answers it.
On a still exported from a project, cleaning keeps its point: EXIF, XMP, IPTC, PNG text chunks and any provenance manifest all go together in the re-encode.
That is the real use case: preparing a board, a thumbnail or a capture before publishing, without carrying along the edit history or tool information. Drop the file, read what it holds, download the cleaned copy.
If a visible watermark is your problem, a metadata cleaner is not the answer. Depending on the platform's terms, a paid tier removes the watermark at the source: that is the clean route, and the only one that does not degrade the image.
Retouching pixels to erase a mark is technically possible, but it falls outside this tool's scope and may breach the terms of the service that produced the image.
| Item | Nature | Handled |
|---|---|---|
| EXIF, XMP, IPTC | Metadata | Yes |
| C2PA manifest | Metadata | Yes, on JPEG and PNG |
| PNG text chunks | Metadata | Yes |
| Visible watermark | Pixels | No |
| Video file | Unsupported format | No |
No. The visible logo is burned into pixels. GhostMeta strips metadata only. Upgrade to a Runway tier without the visible watermark or use a video editor for pixel-level removal.
MP4 video metadata is separate from JPEG/PNG. GhostMeta v1 focuses on still images. For MP4, use a tool like ffmpeg with -map_metadata -1.
Gen-2 since mid-2024 yes. Older Gen-2 outputs may not — they pass through GhostMeta untouched if no manifest is found.
The JUMBF box holds a manifest store: a claim with `claim_generator` set to a Runway string, a signature with the certificate chain, and an assertion list. That list typically includes a `c2pa.actions` assertion (created, opened) and a content-binding hash unique to the file. GhostMeta removes the entire box, not just the generator label, so no single assertion survives.
Yes, unless you strip it. When a C2PA-aware editor opens the Runway still, it records the original as an `ingredient` assertion and chains a new manifest on top, so the AI origin propagates into your export. Run GhostMeta as the last step after editing, not before, to break that chain in the final JPEG or PNG.
In a JPEG export the manifest sits in the APP11 marker segment as a JUMBF box; in a PNG it lives in the caBX chunk. Runway also writes a branded XMP block (often in an iTXt chunk on PNG or an APP1 segment on JPEG). GhostMeta detects all of these and the canvas re-encode produces a file with none of them.
Partner : M.E Expert Serrurier