LOS ANGELES, CALIFORNIA, JUNE 1st 2026 – Video Intelligence ships on Qencode. The first version takes the cross-team request that used to take days, every clip where X happens, across vision, audio, speech, and on-screen text, and answers it in seconds with timestamps marketing, legal, or rights ops can act on.

Video catalogs used to compete on what was tagged at ingest. The line is moving to what’s actually in the content. Standard video tooling has been treating vision, audio, and speech as separate problems, capping out on short clips, with no native primitive for finding moments across modalities at archive scale. Video Intelligence adds a unified retrieval-and-generation layer to the encoding pipeline, powered by TwelveLabs and integrated directly into our solution, so the same query reaches across video, audio, speech, and on-screen text in a single pass, on hour-long content, alongside the rendition.

What it is

Video Intelligence is a new capability you turn on inside an existing encode job. Once it runs, Qencode delivers a structured result to your chosen destination alongside the rendition, in one of five modes:

• Description generates a comprehensive description of the video content.
• Categorization tags scenes against your taxonomy (or a default category list).
• Moderation flags violations against your list of reasons (or a default).
• Custom runs a user-defined prompt against the video and returns the structured result.
• Search returns timestamped spans matching a text or image query, scoped to visual, audio, or transcribed speech.

What this changes for your team

• Marketing’s “find every clip where the new sponsor appears” walk-up gets answered while the meeting is still on, instead of going on next sprint’s queue.
• The licensor takedown call closes with a per-asset record, not a “we’ll get back to you on that.”
• A senior editor cutting a four-minute feature from forty hours of raw footage spends the morning describing the moments and the afternoon making the cut, instead of the other way around.
• A two-hour live broadcast becomes a same-day recap with scoring plays, sponsor moments, and a 90-second highlight reel timestamped in time for the morning push.
• Catalog activation works in a market that doesn’t speak English: a query in Portuguese reaches Portuguese-language footage, a query in Korean reaches Korean-language footage, without parallel indexes.
• Catalog ops gets new uploads pre-sorted into the team’s own taxonomy bins instead of someone else’s category vocabulary.

How it works

Turn on Video Intelligence in your encode job and pick one of five modes: description, categorization, moderation, custom, or search. For search, you provide a prompt and choose which layers to query against, whether visual, audio, or transcribed speech. The system runs the query across all chosen layers in a single pass and delivers the result to your destination.

Search results land as ranked, timestamped spans: each hit carries a start time, an end time, and a relevance rank. Description, categorization, moderation, and custom modes return their structured payload in the same response shape, at the same destination.

Description, categorization, moderation, and custom modes accept assets between 10 seconds and 2 hours, up to 2 GB. Search accepts assets between 4 seconds and 4 hours, up to 4 GB. See the tutorial for the full mode reference, resolution and aspect-ratio bounds, and example requests and responses for every mode.

Try it

Start a free trial. The first jobs are on us. Point the source at one of your existing assets and watch the result land in seconds.

About TwelveLabs

TwelveLabs is the world’s most powerful video intelligence platform, enabling machines to see, hear, and reason about video like humans do. From semantic search to automated summaries and multimodal embeddings, TwelveLabs empowers developers, enterprises, and creatives to unlock the full potential of video data across industries including media, advertising, government, security, and automotive. For more information, visit www.twelvelabs.io.