Does Runway mark its video as AI-generated?

Runway is a leading AI video generation and editing platform (Gen-3/Gen-4 models), used for everything from short marketing clips to full creative productions. Video provenance marking is generally newer and less battle-tested across the industry than image marking, so Runway's practices here deserve their own scrutiny rather than assuming parity with image-generator norms.

Article 50(2) of the EU AI Act requires AI-generated or AI-edited audio, image, video, and text to be marked in a machine-readable format so it's detectable as artificially generated, via C2PA Content Credentials or IPTC provenance metadata, embedded in the file itself. This is a provider-side duty on the generative system, but it fails quietly in practice: content that leaves a generator marked routinely arrives on a published page unmarked, stripped somewhere in the ordinary pipeline of editing, compressing, and publishing.

Does Runway mark its output by default?

As of this review, Runway states it marks every generation: it has adopted the C2PA provenance standards and adds an invisible watermark to every Runway generation, retaining machine-readable evidence in the output that it is AI-generated. These are two distinct signals, metadata-based C2PA provenance (strippable on re-encode) and an in-signal watermark (more resilient, but not independently verifiable with third-party C2PA tools). Free-plan videos additionally carry Runway's visible watermark. This aligns Runway with the broader C2PA coalition (which includes Adobe, Microsoft, Google, OpenAI, Meta, and major camera/news organizations), giving Runway's own output a genuine machine-readable provenance signal at the point of generation.

What strips the marking before it reaches your published page

• Video re-encoding and transcoding, nearly universal before publishing (compressing for web, converting formats, adjusting for a platform's specs), is the single biggest risk for video specifically, far more common than for a static image
• Social media platforms typically re-process uploaded video entirely (re-encoding, sometimes re-hosting via their own CDN), which can strip embedded metadata as a routine part of that pipeline
• Video editing software (cutting, adding titles/music, color grading) can drop the manifest during export unless the tool is specifically C2PA-aware
• Screen recording or screen-capturing a Runway video, rather than exporting the file directly, produces an entirely new file with no relationship to the original's manifest

How to verify before you publish

• Check your final, published video file, not the raw Runway export, with a C2PA-aware verification tool, since video-specific checking tools are less mature/standardized than image checkers like Content Credentials Verify
• Test your actual publishing pipeline end-to-end: does the manifest survive your specific export settings, your video editor, and the platform you publish to?
• For video published to social platforms specifically, assume the manifest is stripped during upload unless you've confirmed otherwise with that specific platform
• If machine-readable marking can't be confirmed on your published version, use Article 50(4)'s visible-labeling path instead, an on-screen or caption-level "AI-generated" notice
• Keep the original, unedited Runway export as an internal provenance record even if a processed/published version loses its marking

The realistic compliance gap

Video is likely the least-reliably-marked media type across this entire research set, not because Runway's own marking is weak, but because transcoding is nearly universal before video gets published, social platforms, video editors, and format conversions routinely strip metadata that a static image pipeline might preserve. A business should assume any published (as opposed to raw-exported) Runway video needs independent verification, not rely on Runway's generation-time marking alone.

Common questions