Glossary
The words behind digital humans.
59 terms you will meet when you build, buy or compare AI video agents, explained in plain language.
The basics
Digital human
An AI that looks, sounds and talks like a person in real time. It hears what you say, answers in its own voice and reacts with a lifelike face, the way a person would on a video call.
What is a digital human?AI avatar
A computer-generated likeness of a person. Most AI avatar tools turn a written script into a finished video; a real-time avatar holds a live conversation instead.
Interactive avatar
Also: real-time avatar, streaming avatarAn AI avatar that responds live, streaming video of its face as it speaks rather than rendering a video file in advance.
Conversational video
Also: conversational video interface (CVI)A two-way video conversation with an AI, where both sides speak and react in turn. Tavus calls its product a conversational video interface.
AI video agent
A digital human that does a job: it answers questions, qualifies leads, books appointments or looks up orders, over video, voice or chat.
AI clone
Also: digital twinA digital human made of a real person, with their face, voice and knowledge, that others can talk to. Making one of anyone other than yourself requires their consent.
Make a LookalikePersona
The written description that tells a digital human who it is: its role, personality, what it knows, how it talks and what it must not do.
Stock face
Also: stock avatar, stock replicaA ready-made face a platform provides, for teams that do not want to use a photo of a real person.
One-photo avatar
Also: zero-shot avatarA digital human made from a single photo, with no video recording or training session. Lookalike works this way; some platforms need a minute or more of recorded footage instead.
Uncanny valley
The uneasy feeling people get from a face that is almost, but not quite, human. Natural eye contact, lip sync and listening behavior are what carry a digital human across it.
Deepfake
Media that puts someone’s face or voice on content without their consent, usually to deceive. A responsible digital human says it is an AI and is made with the depicted person’s permission.
Lookalike
A digital human made on lookalike.com from a photo, with its own voice, persona and knowledge, that people can video call, voice call or text.
Explore LookalikesHow a conversation works
Speech recognition
Also: ASR, speech-to-text, STTTurning spoken words into text as someone talks. It is the first step in every voice or video conversation with an AI.
Voice activity detection
Also: VADTelling speech apart from silence and background noise, so the system knows when someone is talking at all.
Endpointing
Also: end-of-turn detectionDeciding when a person has finished speaking, so the digital human replies without cutting them off or leaving an awkward pause.
Turn-taking
The back and forth of a conversation: who speaks when. Good turn-taking makes a digital human feel attentive rather than robotic.
Barge-in
Also: interruption handlingLetting a person interrupt while the digital human is speaking. It stops, listens and answers the new question.
Full-duplex conversation
A conversation where both sides can speak and listen at the same time, as people do, instead of strictly taking turns.
Backchanneling
The small signals a listener gives, such as nods, “mm-hm” and a change of expression, that show they are following along.
Large language model
Also: LLMThe AI model that writes each reply, following the persona and drawing on the knowledge it was given.
System prompt
Instructions given to the language model before the conversation starts. A persona is usually turned into a system prompt.
Guardrails
Rules that keep a digital human on topic and safe: subjects it must avoid, claims it must not make and when it must hand off to a person.
Bring your own LLM
Using a language model you choose or host yourself behind a platform’s face and voice, usually through an API.
Multimodal AI
AI that works across more than one kind of input or output, such as text, speech, images and video, in the same conversation.
Retrieval-augmented generation
Also: RAGLooking up the relevant passages in your documents before writing a reply, so answers come from your material rather than the model’s general knowledge.
Knowledge base
The documents, spreadsheets, images or website a digital human answers from. On Lookalike you can add PDFs, documents, spreadsheets, images or a website.
Hallucination
When an AI states something false with confidence. Grounding answers in a knowledge base and telling the persona what not to guess reduces it.
Voice cloning
Making a synthetic voice that sounds like a specific person from a sample of their speech. It should only be done with that person’s consent.
Lip sync
Matching mouth movements to speech, frame by frame. Poor lip sync is the first thing people notice in a digital human.
Talking-head model
Also: audio-driven animationA video model that animates a face from a photo or clip so it speaks, blinks and moves in time with an audio track.
Listening state
Also: idle animationHow a digital human looks while the other person talks: small nods, blinks and attention, instead of a frozen face.
Tool calling
Also: function callingLetting the language model take an action, such as checking a calendar or looking up an order, and use the result in its reply.
Model Context Protocol
Also: MCPAn open standard for connecting AI models to outside tools and data, so one integration works across many AI products.
Memory
What a digital human keeps from earlier conversations with the same person, such as their name or what they asked last time.
Visual perception
A digital human seeing through the camera or a shared screen and responding to what it sees, such as a held-up product or a document.
Delivery and pricing
Latency
The delay between when a person stops speaking and when the digital human starts answering. It is the sum of every step: speech recognition, the reply, the voice and the video.
Time to first frame
How long it takes for video of the face to appear after a call starts or a reply begins.
Cold start
Extra delay when a call needs a server that is not yet running. Platforms keep servers warm to avoid it.
WebRTC
The browser technology for live audio and video calls. Real-time digital humans stream to you over it.
LiveKit
An open-source platform for real-time audio and video built on WebRTC. Lookalike’s API sessions run over LiveKit, and many digital human platforms offer LiveKit plugins.
The Digital Human APIEmbed
Also: widgetA snippet of code that puts a digital human on your website. Lookalike’s is one line, and works with WordPress, Shopify, Webflow, Wix, Squarespace and Framer.
Add one to your siteRealtime API
An interface developers use to create digital humans and start live sessions from their own app.
The Digital Human APISDK
A software development kit: ready-made code libraries for building a platform into an app on the web or a phone.
Concurrency
Also: concurrent sessionsHow many conversations can run at the same moment. Many plans limit it, so a busy website may need a higher tier.
Per-minute pricing
Paying for each minute of conversation. Compare complete digital humans with each other, since face-only services leave out the speech and language costs.
Compare pricesFace layer
Also: avatar-onlyA service that only animates a face for a voice agent you already run. You pay separately for speech recognition, the language model and the voice.
White label
Running a platform under your own brand, with its name and watermark removed. Paid Lookalike plans remove the “Powered by Lookalike” watermark from embeds.
SIP and phone calls
SIP is the standard for connecting voice calls between phone networks and software. It is how a digital human answers a business phone number.
Uses and trust
Standardized patient
A person, or now a digital human, who plays a patient so medical and nursing students can practice taking histories and delivering news.
HealthcareSales role-play
Practicing a sales conversation with someone playing the buyer. A digital human can play the buyer on demand and push back like a real one.
SalesDigital concierge
A digital human that greets guests, answers questions about a property and the neighborhood, and passes requests to staff.
HospitalityVirtual receptionist
A digital human that answers a business’s website visitors or calls, answers common questions and books appointments.
Lookalike for BusinessInteractive memorial
A digital human of someone who has died, made with their family, that keeps their face, voice and stories so others can talk with them.
Remember a loved oneLikeness consent
Permission from a real person to make a digital human of their face or voice. Lookalike requires it.
Likeness takedown
A request to remove a digital human that uses someone’s likeness without permission. On Lookalike, the person or their representative can file one in the Legal Center.
Legal CenterAI disclosure
Telling people they are talking to an AI. Every Lookalike page says it is a digital human, not a real person.
HIPAA
The United States law that protects patient health information. Healthcare teams should confirm a vendor’s HIPAA terms, such as a business associate agreement, before sharing patient data.
SOC 2
An independent audit of how a company protects customer data. Many enterprise buyers ask vendors for a SOC 2 report.
Comparing platforms? See Lookalike vs Tavus, Lookalike vs HeyGen and the best Tavus alternatives.
See every term in action.
One call with Sophie shows lip sync, turn-taking and barge-in better than any definition.