TwelveLabs Brings AI Video Review and Synthetic Content Detection Into One Workflow
The new Compliance by TwelveLabs platform combines customizable content screening with NVIDIA authenticity checks, helping media teams identify footage that needs human review.
TwelveLabs, a leading video intelligence company, has launched Compliance by TwelveLabs, an AI application that screens video against regional standards and internal content policies while flagging potentially synthetic footage.
The platform brings content review and AI-generated video detection into a shared workflow for broadcasters, media companies, and entertainment organizations. It is the first application built on the company’s video intelligence platform.
From Footage to Findings
The system uses TwelveLabs’ Pegasus 1.5 model to analyze video and audio in context, checking material against rules defined by the customer. Potential issues can include violence, profanity, and brand-safety concerns.
Reviewers receive a prioritized queue of flagged moments, supported by timestamps and explanations. They can accept, reject, or annotate findings and export reports documenting the review. Teams can also modify detection thresholds and maintain different versions of their policies as requirements change.
“We surface potential violations and give reviewers the power to make decisions faster and easier than ever before.” — Jae Lee, CEO and Co-Founder, TwelveLabs
The platform supports more than 40 preconfigured regional and use-case rule packs. Customers can also create their own rules or generate them from imported PDFs, allowing the review process to reflect the requirements of a particular market or organization.
Checking for Synthetic Video
The integration of NVIDIA Synthetic Video Detector adds authenticity signals and confidence scores to the same review process. The detector assesses the probability that footage is AI-generated, helping reviewers identify material that warrants closer examination.
NVIDIA describes the technology as another source of analysis for editorial and media-integrity teams. Its scores support a judgment about authenticity; they do not establish the truth of an event or the accuracy of a claim made in a video.
For sports media operations, a potential application is screening externally supplied clips before they enter a broadcast or social publishing workflow. A questionable athlete video, for example, could require both an authenticity check and an editorial decision about whether it should be used. Combining those steps could help teams retain the evidence behind that decision.
Measuring the Work Saved
TwelveLabs reports an 80% reduction in reviewer time in internal evaluation. It also reports 92.5% agreement with a manual baseline across a 53-asset validation set and a 12-minute runtime for applying a 12-rule compliance pack to a two-hour video.
Those figures describe company testing, with results varying by content, rules, thresholds, and operating environment. Agreement with a manual baseline should not be read as a guarantee that the system will catch every violation.
The practical test for a broadcaster or rights holder is whether the tool reduces total review effort while reliably surfacing material that needs attention. That requires examining missed issues alongside false alarms and processing speed.
A pilot using previously reviewed footage would give a content team a useful comparison: which issues the AI identifies, which it misses, and how much work remains before an editor can approve publication.
Twelve Labs, Inc.
{{i4uQ1pNae}}
