C2PA soft binding and watermarks: what a trim or re-encode does

A trim can change the bytes of a file, and a watermark or manifest can depend on those bytes. C2PA 2.3 includes a soft_binding_assertion, and its spec page defines an invisible watermark as imperceptible information that can identify the asset or reference a manifest. What a given trim leaves intact is something to test on your own files, not assume.

Two trim modes on Sume

The Sume video trim docs describe precision. The default exact is a frame-accurate re-encode with libx264 and yuv420p. keyframe is a stream copy where the cut may start a GOP early, so you re-base against actual_start_seconds. Both return a new MP4 artifact; the source is untouched.

What the docs do not say

Neither the trim page nor the C2PA pages I read says whether a given watermark survives either mode. A stream copy keeps the compressed frames, a re-encode makes new ones. That is general video behavior, not a claim from a vendor page, so treat it as a hypothesis to test.

A test you can run

Take one clip with the marker you care about. Trim it once with exact and once with keyframe. Run your detector or manifest reader on both outputs and on the source, and write down the result per mode.

Each trim is $0.02 per job, so the test costs four cents. The result tells you which mode to use before a launch.

• Keep the source artifact; trim never changes it.
• Check actual_start_seconds on keyframe outputs.
• Record the mode you used with the job id.