Twelve Labs brings natural language video search to Frame.io V4

Anyone who has worked in post-production knows the drill: spending hours scrubbing through gigabytes of raw footage just to find that one wide aerial shot of a coastline.

Twelve Labs just brought its multimodal video understanding models directly into Frame.io V4, turning the popular review platform into a genuinely smart video search engine.

By hooking into Frame.io V4 Custom Actions, editors can now analyze, tag, and search their media archives without ever leaving their existing workspace.

Semantic search meets your daily timeline

The integration relies on two of Twelve Labs' proprietary foundation models: Marengo 2.7 and Pegasus 1.2.

Marengo 2.7 transforms video, audio, and speech into vector embeddings. This powers semantic natural-language search, meaning you can type descriptive prompts instead of relying on file names or manual tags.

It also introduces image-to-video search. You can drop in a reference photo and instantly retrieve visually matching takes or B-roll across your projects.

Meanwhile, Pegasus 1.2 takes care of metadata generation. Rather than just listing objects in a frame, it interprets narrative context, themes, and emotional tone, automatically populating custom fields in Frame.io.

There is also an automated compliance feature. If footage violates brand rules or policy guidelines, the AI flags the exact frame with timestamped comments directly in the review thread.

Why it matters

We think this is exactly where practical video AI needs to live—inside the tools creators already use every single day.

Instead of asking production teams to upload massive media files to yet another standalone web app, Twelve Labs piggybacks on Frame.io's established pipeline. A right-click to index and search removes nearly all user friction.

For enterprise marketing teams, newsrooms, and production houses sitting on thousands of hours of archival footage, this makes unorganized assets immediately usable again.

The real question moving forward will be computational cost. Indexing deep multimodal video vectors across hundreds of terabytes is computationally heavy, and teams will have to weigh API costs against manual cataloging time.

Still, as Adobe cements Frame.io as the creative industry's collaboration backbone, native video intelligence like this sets a high bar for what digital asset management should look like.