Artificial intelligence news.
Straight from the original announcements.

Snowflake Brings TwelveLabs Video Search and Analysis to SQL

November 18, 2026, 17:15By the AI News Bites Editorial TeamSource: Snowflake

Snowflake is partnering with TwelveLabs to let organizations search and analyze video through Snowflake Cortex AI Functions and SQL. The integration combines video-derived insights with existing structured data under Snowflake’s governance and security controls.

Key points
  • Snowflake is partnering with TwelveLabs to bring video AI to Cortex AI Functions.
  • Marengo creates 512-dimensional multimodal video embeddings.
  • Pegasus generates summaries, metadata, answers and scene descriptions.
  • Cortex Search enables natural-language searches for moments in video.
  • Marengo 3.0 and Pegasus 1.5 are now generally available.

Turning enterprise video into queryable data

Snowflake describes video as the most produced, most consumed and least understood data in the enterprise. Organizations hold marketing libraries, sports archives, security footage, training materials, earnings calls and other video collections, but much of that content remains difficult to query.

According to Snowflake, this leaves useful information in cold storage rather than in the analytics layer. Making video searchable could help marketers connect creative content with campaign performance, broadcasters work with archive footage, retailers create structured catalog data from product videos and compliance teams find relevant material more efficiently.

The company says video analysis has traditionally been split between computer vision, which can recognize objects in frames, and transcription, which captures speech but misses visual information. TwelveLabs is intended to provide a unified approach that considers visuals, audio and speech across time.

How Marengo and Pegasus work

TwelveLabs provides two models for the integration. Marengo is a multimodal encoder that converts video into 512-dimensional embeddings representing visual content, audio patterns and spoken language. It is designed to help locate relevant moments in a large video collection.

Pegasus is a video-language model that analyzes video and generates text, including summaries, structured metadata, answers to questions and scene descriptions. Snowflake presents the two models as complementary parts of a video intelligence stack: Marengo finds a moment, while Pegasus explains what it contains.

Videos can be stored on Snowflake stages, including internal stages or external stages on S3, GCS and Azure. A Cortex AI Functions call can generate embeddings for each video. Snowflake says the returned structure contains embeddings across visual, audio and transcription modalities, with each segment represented by a 512-dimensional vector. The transcription capability supports 36 languages.

Search, analysis and governance in Snowflake

Once embeddings are stored, Cortex Search can index them for natural-language queries. Users can describe the moment they want to find, such as a scene involving a ship burning at night, and retrieve matching video segments with fields including the video file and start and end times.

Snowflake says this approach reduces reliance on manual tags, keyword matching and maintained content taxonomies. Video-derived signals can remain in the same tables as campaign data, audience segments, content metadata and spending information, allowing those sources to be examined in a single query.

The integration also applies Snowflake controls to video-derived data stored in its tables, including row-level policies, role-based access and audit logs. For more advanced workflows, Cortex Agents can combine visual, audio and transcription signals through multi-vector search and custom ranking strategies.

Availability and an example from Challenger School

Snowflake says Challenger School’s marketing team built a footage-search tool with AI_MULTI_EMBED. The company says the tool surfaces relevant clips in minutes, replacing a manual process that previously took weeks.

“Using Twelve Labs' Marengo model through Snowflake's AI_MULTI_EMBED, our team can ask for footage in natural language and receive timestamps, transcript context, and Premiere review sequences linked to the original source files. This gives our editors a practical starting point for creating videos,” said Yashwanth Karumanchi of Research & Development at Challenger School Foundation.

Marengo 3.0 and Pegasus 1.5 are now generally available, according to Snowflake. The offering is aimed at organizations with content libraries, advertising creatives, broadcast archives, surveillance footage or training materials that they want to analyze through their existing Snowflake environment.

Related news

Tencent Hy4 goes live on Together AI with 1M-token context and image input
ModelsNovember 19, 2026

Tencent Hy4 goes live on Together AI with 1M-token context and image input

Together AI has made Tencent's flagship Hy4 model available through serverless and dedicated endpoints. Hy4 is the first model in Tencent's Hy family to accept images…

AWS details Social Department’s culture-aware video discovery system
Products and toolsNovember 18, 2026, 20:27

AWS details Social Department’s culture-aware video discovery system

Social Department has built QuickSearch, a set of single-action queries designed to help entertainment marketers find culturally relevant moments in large video…

For developers
For developersNovember 18, 2026, 23:44

Snowflake adds AI evaluation and optimization for Cortex Functions

Snowflake has introduced native evaluation and optimization tools for Cortex AI Functions, helping teams compare models and prompts against quality and cost requirements…