Media Intelligent Engine
A media engine that reasons about a video stream while it packages it — deciding which regions of a frame can safely carry payload, then embedding protection and forensic attribution inside the picture itself rather than alongside it as metadata.
The problem
Everything that protects a video stream today sits outside the picture. DRM wraps the bitstream, entitlement lives in a manifest, and attribution rides along in metadata. All three survive exactly as long as the container does. Re-encode the file, screen-record the player, or point a phone at a monitor, and the protection is gone while the picture — the thing that actually has value — travels on perfectly intact.
That gap is where leaked premium sport ends up. The frames survive; everything that said who they belonged to does not.
The approach
Put the mark where the value is. If attribution lives in the pixels, it inherits the picture's own durability: whatever transformation preserves the image well enough to be worth stealing also preserves the mark.
The difficulty is that a frame is not uniformly hospitable. Flat gradients show any modification, and hard edges destroy it. So the engine runs scene understanding first and selects carry regions — areas with enough texture and motion to hide a payload perceptually while retaining it through compression. Only then does it embed, and packaging happens in the same pass rather than as a separate stage.
Where it stands
The architecture and the carry-region selection are worked out; this is active, unpublished work rather than a finished system, and the numbers that would make it a claim — survival rates through specific re-encode and recapture conditions — are not measured yet.
Related
FluxMark takes the same question to surfaces that will not hold still — recovering a mark from clothing in motion by unwrapping the deforming surface into a stable domain first.