Reference

AI for video engineering glossary

150 terms with crisp definitions, aliases, and cross-references to the deep dives where each is unpacked. From YOLO and SAM 2 to VLMs, LangGraph, and diffusion — every model, tool, and primitive in the AI for Video Engineering course in one place. Search by name, filter by phase, or jump straight to a letter.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
ALL
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

No terms match your filters. Try a different keyword or clear filters.



Agentic AI

AI built to act autonomously toward goals over multiple steps, as opposed to generative AI that just produces content on request. The agent paradigm.

A

Agent loop

The repeating cycle an agent runs: think, act with a tool, observe the result, repeat — until the goal is met. The core mechanism of agentic systems.

A

AgentOps

The practice and tooling for running agents in production — evaluating, tracing, costing, and securing them — so autonomous systems stay reliable and safe.

A



AI agent

An AI system that pursues a goal by deciding its own steps — calling tools, reading results, and acting — rather than answering a single prompt.

A

AI avatar

A synthetic on-screen presenter generated from a real or invented face, driven by text or audio to speak and move. Used for video without filming.

A

AI content moderation

Automatically screening live or uploaded media for unsafe content — nudity, violence, abuse — so it can be blocked, blurred, or flagged at scale.

A

AI dubbing

Automatically translating and revoicing video into another language using ASR, translation, voice synthesis, and lip-sync, instead of hiring voice actors.

A



AI meeting assistant

A bot that joins calls to transcribe, summarise, and capture action items, also called an AI notetaker. A flagship real-time AI product category.

A



Anomaly detection

Flagging events that deviate from normal patterns — a fall, an intrusion, a defect — without a labelled example of every possible problem.

A

AR effects

Real-time augmented-reality overlays on a face or scene in video — filters, masks, beauty smoothing, gaze correction — built on face and pose tracking.

A



ASR (Speech-to-text)

Automatic Speech Recognition — turning spoken audio into written text. The base layer for captions, transcripts, voice commands, and meeting notes.

A



Attention

The mechanism that lets a model weigh which parts of the input matter most for each output. The core idea that makes transformers work.

A

AutoGen

Microsoft's framework for multi-agent conversations, where agents talk to each other and to tools to solve a task. An early, influential agent toolkit.

A

Background blur

Softening or hiding everything behind a person on a video call. Built on real-time selfie segmentation that separates foreground from background each frame.

B

Benchmark

A standard test set and score used to compare models objectively. Useful for shortlisting, but real product quality needs your own evaluation too.

B

Bounding box

The rectangle a detector draws around an object, given as corner coordinates plus a label and confidence score. The currency of detection output.

B

ByteTrack

A fast, accurate tracker that keeps even low-confidence detections to recover objects others drop. Strong default for crowded real-time scenes.

B



C2PA

An open standard for attaching tamper-evident provenance to media, recording how a file was made or edited. The technical basis for 'this was AI-generated' labels.

C

Claude

Anthropic's family of frontier language and multimodal models, used via API for reasoning, long-context work, vision, and agentic tool use.

C

Claude Agent SDK

Anthropic's toolkit for building autonomous agents that run their own tool-use loop, with file access and command execution. Formerly the Claude Code SDK.

C

Claude Code

Anthropic's agentic coding tool that works in the terminal, able to read a codebase, edit files, and run commands to complete development tasks.

C

Client-side ASR

Running speech recognition in the user's browser or device instead of a server, keeping audio local and removing per-minute cloud costs.

C



CLIP

A model that maps images and text into a shared space so it can tell how well a caption matches a picture. The base trick behind open-vocabulary vision.

C

Closed model

A proprietary model you reach only through a paid API; the weights stay private. You trade control and data residency for zero infrastructure work.

C

Cloud inference

Running models in a remote data centre reached over the network. Scales easily and uses big GPUs, but adds round-trip delay and per-call cost.

C

Computer-use agent

An agent that operates software the way a person does — viewing the screen, clicking, and typing — to carry out tasks across apps without special APIs.

C



Computer vision

The field of teaching software to extract meaning from images and video — detecting, classifying, tracking, and measuring what appears on screen.

C

Context window

The maximum amount of input — measured in tokens — a model can consider at once. It caps how much video, transcript, or document fits in one prompt.

C

Contrastive learning

A training method that pulls matching pairs together and pushes mismatches apart in embedding space. How CLIP links images to text.

C

Convolutional neural network

The classic image-processing network that scans for local patterns like edges and textures. Long the default for vision; efficient and still widely deployed.

C