Every media file you upload is automatically analyzed by MXT: Moments Lab's patented AI technology. This article explains what MXT detects and how it organizes your footage, so you understand what's behind your search results.
What is MXT?
MXT is Moments Lab's patented AI technology for indexing and understanding video. Unlike generic video AI, MXT can be tailored to your content: it learns to recognize your characters and talent, sorts media into your own taxonomy, and surfaces the types of moments your team actually looks for. It analyzes your footage to generate transcripts, detect faces and objects, and describe what's happening on screen, powering search, moments and sequences across the platform.
Shots and sequences
MXT breaks down every media file into two levels of detail:
Shot: the smallest unit inside a sequence, a short, continuous segment based on a camera angle.
Sequence: how MXT groups similar shots together based on what it detects on screen. A sequence is made up of those shots plus the transcript, where available.
This structure is what powers Moments view, where you can browse shots and sequences directly instead of scrubbing through a full media file.
Custom Moments
Custom Moments are customized sequences that work at scale for specific content types. You can work with the Moments Lab solutions team to set these up in advance. Custom Moments use a system prompt to instruct MXT to sequence a video file in a specific way, depending on the content type and your organizational needs. Examples include:
B-Roll
MicroSeries
Funny Moments
Trailer
Custom Moments can also be configured to generate tags from a predefined list of categories.
Transcript
MXT automatically transcribes the spoken dialogue in your media files. It can detect the spoken language automatically, or you can set a specific language in advance if you know what will be spoken — useful for content where more than one language appears in the same file. Transcripts can also be translated into up to 130 languages and exported for use in other systems.
The transcript powers dialogue search.
Soundbites
MXT highlights emotionally charged or quotable moments — perfect for short-form storytelling and social content.
Speaker detection
MXT diarizes transcripts, meaning it knows when one person stops speaking and another starts. It also analyzes lip movement to work out who's speaking. If that person has been trained in the Thesaurus, MXT assigns their name to the transcript automatically.
Labels
As MXT analyzes your footage, it automatically generates labels: tags that describe objects, actions and other elements it detects on screen. These labels are searchable as soon as they're created. You can also create labels manually for anything MXT doesn't catch on its own, such as team or stadium names specific to your content.
People
MXT can recognize people across your catalog once it's been trained to. Add a person's face to the Thesaurus, and MXT automatically recognizes them going forward in your moments and media file search results.
Labels and people are both managed from the Thesaurus. Learn how to train the Thesaurus for your characters and talent
Logo detection
MXT can detect both pictorial and typographic logos in your footage.
Pictorial logos (a brand symbol): contact our Support Team to set this up.
Typographic logos (a brand name spelled out): create a multimodal rule to generate a label whenever that word is detected through OCR — for example, the word "Nike" written out, rather than the Nike swoosh.
Media-level metadata
Beyond shots and sequences, MXT also generates metadata for the entire media file:
Media summaries: an AI-generated synopsis describing the media file as a whole.
Content classification: automatically sorts media files into your custom taxonomy, such as interview, press conference, documentary or social clip.
Custom insights: textual metadata fields generated from MXT's analysis — for example, to populate CMS or VOD fields, or to flag people, topics or brands mentioned even when they don't appear visually.
Good to know
If you add a person to the Thesaurus after MXT has already analyzed a media file, they might not be recognized on that file automatically. Tag them manually on the media instead, or reach out to our Support team.
AWS Celebrities: recognizes well-known public figures automatically during indexing, using a database of tens of thousands of names that Amazon maintains and keeps growing. You don't need to add these people to the Thesaurus yourself. Learn more.
Wikidata: link a person in the Thesaurus to their Wikidata entry to make every alias, nickname or translated name on that entry searchable too, not just the name you typed in. Learn more.
Need more help? Reach out to our Support team — we're happy to help.
UI not matching? Your platform may not be updated yet, let us know and we'll help.