Quickplay and TwelveLabs: Turning Live Broadcasts Into Social-Ready Clips in Minutes

Social media has cut the window to hold a viewer’s attention down to seconds, not hours. A moment airs, and within minutes, someone with a phone and a TikTok account has already clipped it, captioned it, and hit publish, beating the broadcaster who actually owns the footage to the punch. That's not a hypothetical embarrassment — it's lost views, lost ad revenue, and a viewer who now associates that moment with someone else's account.

Quickplay is the infrastructure broadcasters use to move content at scale: ingesting live feeds, running workflows, and getting finished clips out the door to every platform that matters. TwelveLabs’ models run inside that pipeline: video-native AI, built for temporal reasoning rather than adapted from other modalities. They watch the feed as it airs and understand what's happening in it well enough to know when a moment is worth clipping.

That combination is what turns “the moment already happened” into “the clip is already live.”

Everything described in this blog was demoed live in a webinar with Quickplay and TwelveLabs. If you’d rather watch the demo than read about it, the on-demand recording is available now.

The Real Bottleneck: Infrastructure, Not Content

The reason the social publishing race is so often lost isn't due to a shortage of content. It’s that the systems built to move it were never designed to move this fast. Regional desks, NRCS rundowns, and social publishing tools were built at different times, for different jobs, with no thought given to handing content off between them. Manual clipping compounds the problem — a captioned, trend-matched clip that's ready for social media still has to go through a person for review, and by the time they're done, the moment's window has often closed.

AI bolted onto that stack doesn't fix things, either. The math stops working the moment content volume outpaces what a team can watch, judge, and clear in real time. What actually closes the gap is scaling coverage without scaling headcount: running on the broadcast systems already in place and keeping the audience watching where they found the moment. The goal should be to build something that works for news today and sports or other verticals tomorrow.

Inside AI Studio Live

AI Studio Live runs TwelveLabs’ models directly against a live feed, turning broadcast content into something searchable, clippable, and distributable in near real time. Here's what that looks like in practice.

Scene Detection and Tagging, Running on the Live Feed

AI Studio Live analyzes a live feed as it airs, not after it’s archived. As the broadcast plays, Pegasus, TwelveLabs’ video language model, is making a judgment call that most systems can’t: where does this scene actually end, and where does the next one begin?

A scene break in a breaking-news segment looks different from one in a studio interview. Rather than applying a timer or a hard cut detector, Pegasus reasons over what’s happening in the moment. Each segment gets metadata, tagging, and a confidence rating attached as it’s created, so a producer can jump straight to a specific point in a broadcast still in progress and verify Pegasus’s read, rather than scrubbing through a timeline manually.

Search That Understands What's Actually Happening

Pegasus draws the lines; Marengo, TwelveLabs’ multimodal embedding model, makes each segment searchable. Most tools can only surface what someone manually labeled — a name, a location, a keyword typed in ahead of time. Marengo turns the underlying footage itself into a query target, so a search can match on what's actually happening in a moment.

A basic query — every moment a specific public figure, like the Prime Minister of Canada, is on camera — returns matches along with the reasoning behind each one, whether that’s a related figure discussing him or the PM addressing the media directly. Text recognition extends that to on-screen text, again with full contextual reasoning for why a given clip matched.

The more telling example is sentiment. A search for the moment that the Prime Minister “can’t stop laughing” returns exactly that clip, because Marengo inferred the emotion directly from the video. That’s the kind of moment a keyword search would never surface, because nobody would think to tag it in advance.

Live Clipping, From Mark-In to Distribution

Finding the moment is the easy part now. Turning it into something publishable used to require a full edit session — it doesn’t anymore.

Producers can search through transcripts, set in and out points on a segment, subclip it, and save it directly off the live edge with no separate export-and-reimport step. From there, clips can go straight to TikTok, Instagram, or YouTube, with Pegasus even generating suggested titles that can be edited before publishing.

Once individual clips exist, audience cohort analysis builds next-day distribution packs tailored to different segments. Real-time trend monitoring adds another layer on top by watching what's gaining traction and triggering action on the content pipeline.

Automated Workflows Built for Risk Tolerance

Not every piece of content needs the same level of human oversight, though. Low-risk, predictable segments, like a mall opening or a county fair, can be set to trigger fully automated publishing straight to owned-and-operated platforms. Higher-sensitivity content routes through configurable rules and logic, gating it for review.

AI Studio Live also integrates directly with newsroom control systems, pulling in rundowns and graphics information so that flagging and workflow execution happen against the systems teams are already running.

Getting It Live

Once an HLS feed is connected, AI Studio Live starts capturing and analyzing it right away, using commercial break markers to establish segment boundaries. The latency contract works out to segment length plus roughly five minutes, putting teams at a rudimentary 10 to 20 minutes behind air.

More customized integrations — getting closer to the live edge, or adding on iNews or AWS Elemental Inference — will take longer to stand up, scaling with technical complexity rather than content volume.

Scaling Beyond Live News

Archival content can be made searchable through AI Studio in the same way as live content, whether a library spans ten days or several decades. That depth pays off most when it’s paired with what's happening live: when a player scores a milestone goal, the same system can pull together and package their career highlight reel on the spot, so the clip is ready for social minutes after the goal happens.

Reframe that same infrastructure from news to sports, and the pattern holds. A show’s fanbase can be kept engaged between episodes, and an actor’s archive can build an awards-season push. Both work for the same reason the goal example does — the system understands what’s actually in the footage, not just where it’s filed.

This Isn’t a Demo — It’s Already Live

Quickplay and TwelveLabs are already running this pipeline in production across news, sports, and entertainment. Content that’s actually understood can be found, published, and repurposed without a growing headcount.

If you’d like to see what AI Studio Live powered by TwelveLabs looks like running on your own feeds, reach out to our sales team to start the conversation.