TwelveLabs raises $100 million to bring superintelligence to AI video models

TwelveLabs Inc., developer of generative artificial intelligence-based models that can understand videos like humans, today announced that We raised $100 million in seed funding to take our goals beyond simple understanding. overall intelligence.

The Series B round was co-led by NEA and NAVER Ventures. Amazon, Radical Ventures, Korea Investment Partners, Index Ventures, Quadrille Capital, and Red Bull Ventures also participated in the round. Today’s capital injection brings the company’s total funding to more than $207 million.

“Five years ago, we made a contrarian bet: The foundation of machine intelligence is not language, but recorded reality in motion,” said Jay Lee, CEO and co-founder. “Language is downstream from understanding. Video is the data to which understanding answers.”

TwelveLabs is building frontier foundational models capable of video understanding, and its technology brings real AI power to the field. The company worked from the ground up to build a multimodal model, a model that natively understands video, rather than just a large language model that processes it.

The company’s flagship products include the Marengo model family, including versions 3.0 and Pegasus 1.5, released late last year.

Maregno enables real-world AI embedding for video, audio, text, and synthesis. This means that AI models can understand and retrieve information at scale because they can parse different types of content and add it to machine-readable data structures, such as vector databases.

Pegasus works with previous models to convert video into structured data. It understands scene boundaries, entities, time segments, and what’s going on, allowing LLM to infer the entire visual information. This works similar to how large language models process large documents and images and summarize them in a markup language to make them easier to understand.

Today’s LLMs cannot consume videos all at once. They have to break it up into flashes (a series of screenshots) and use it to reason. TwelveLabs said it has built inference capabilities that natively understand trends over time by maintaining memory that persists between queries, rather than evaporating after each query. Deepen your model knowledge with intelligence that grows with each video.

The company said it intends to use these models to build systems that allow machines to analyze, search and manipulate footage, creating a new paradigm in video recognition.

Use cases for this type of technology span numerous workflows across industries such as security, advertising, sports, and automotive, where vast amounts of information reside within videos.

As part of the funding, TwelveLabs is deepening its relationship with Amazon Web Services. The company’s services have been available on AWS Marketplace since at least 2025, along with managed access to the underlying models through Amazon Bedrock.

AWS already TwelveLabs is a preferred cloud provider and with this investment, the company has signed a multi-year agreement to optimize video inference workloads on AWS Trainium. Chips. Additionally, new Frontier models will be released first on AWS.

Photo: Pixabay