Home / Blog

Product & news

Meet DEVA-1: UNITH's most lifelike conversational AI Avatars

Meet DEVA-1: UNITH's most lifelike conversational AI Avatars

Meet DEVA-1: UNITH's most lifelike conversational AI Avatars

Learn what DEVA-1 changes for Digital Humans, where the face fits in a conversational AI stack, and how to test it on your own content during early access.

Erianys Vergara

14 minutes

Key Takeaways

  • DEVA-1 is our new real-time visual AI model. It generates a Digital Human’s face live from streaming audio, with lip sync, while the conversation happens.

  • It brings depth, volume and realistic lighting, full-face expression and natural micro-movements to our Digital Humans.

  • DEVA-1 sits alongside our Interactive Avatars, so you can choose the right one for each conversation. Both use the same SDK and API foundation.

  • We’re opening access in stages, starting with a select group of customers, and pricing stays per conversation, not per minute.

A new hire asks about parental leave on their first day. A grounded knowledge model pulls the answer from your internal policy, and an expressive Digital Human delivers it in real time, with natural lip sync and warm micro-expressions.

DEVA-1 is the generative face engine behind that conversation. It generates the Digital Human’s face live from streaming audio, while the conversation happens.

Language models made accurate answers possible, but how an answer is delivered matters just as much. Text is easy to skim. A human face holds attention and builds trust. The face has always been central to UNITH’s Digital Humans, and DEVA-1 makes it more realistic and expressive.

What is DEVA-1?

DEVA-1 is UNITH’s real-time visual AI model for conversational AI avatars. It brings depth, volume and realistic lighting, richer facial expression and subtle, natural micro-movements to every conversation, rendered live from the first word to the last.

Capability

With DEVA-1

What it means in practice

Rendering

Depth, volume and realistic lighting

Avatars look natural and crisp without appearing flat or distorted.

Realism

Richer, more natural facial movement

Enhances viewer trust and engagement during critical interactions, such as executive communications or client-facing roles.

Facial expression

Full facial micro expressions

Allows the Avatar to convey subtle emotions, like empathy or professionalism, matching complex or sensitive conversations accurately.


How does DEVA-1 work?

DEVA-1 turns streaming speech into a live video of a Digital Human’s face. It controls how the face looks, moves and responds in real time through a continuous, 5-step audio-to-video pipeline:

[ Streaming Audio ] ➔ [ Motion Parameters ] ➔ [ 3D Head Pose ] ➔ [ Real-time Frame Render ] ➔ [ WebSocket Stream ]

  1. Streaming audio inputs: Speech audio arrives in small, continuous chunks as it is generated, rather than as a pre-recorded file. The face can start moving before the full sentence is spoken.

  2. Audio-to-motion translation: An audio-driven model translates sound into facial motion parameters. These set how the lips, jaw, cheeks, brows and eyes move at each fraction of a second.

  3. 3D spatial posing: The parameters pose a 3D head representation, so depth and realistic lighting are calculated on the fly.

  4. Frame rendering: Each posed instance is rendered instantly as a video frame.

  5. Continuous delivery: Frames stream to the browser over a persistent WebSocket connection and stay in sync with the audio as the conversation flows.

Lip sync and natural expression

  • Lip sync (visemes): Matches mouth shapes to distinct audio phonemes. Groups of similar-looking sounds (like p, b, and m) map to shared visual mouth shapes called visemes.

  • Facial expression: Coordinates the rest of the face, including the brows, eyes, cheeks and natural head movement, to convey emotion.

The conversational presence stack

Many avatar pilots fail because teams measure the wrong things. Steering committees judge whether the avatar knows the answer. End users judge whether talking to it feels natural and worth their time.

To keep these concerns distinct, we evaluate Digital Human programs across five layers:

  1. Knowledge: The LLM and grounded content, such as policies, product documentation, and sales playbooks.

  2. Dialogue: Turn-taking, latency, interruption handling, and conversational flow.

  3. Voice: Speech quality, accent, language coverage, and voice cloning.

  4. Face (DEVA-1): Live rendering, depth lighting, lip sync, and micro-expressions.

  5. Delivery: Interface embedding across intranets, LMS platforms, or apps, along with backend tool integrations.

An avatar program is only as strong as its weakest layer. Good content needs an engaging face to hold attention, and a lifelike face needs grounded content behind it to stay credible.

You can refine a prompt in minutes, but facial rendering depends on what the platform can do. That is the layer DEVA-1 improves.

Why does a conversational AI avatar need real-time rendering?

Human conversation is fast and sensitive. Research published in PNAS shows that across global languages, human speakers consistently keep turn-taking gaps brief and avoid overlapping speech (Stivers et al., 2009).

People watch an avatar closely in those short pauses, so how a Digital Human looks while it listens is part of the conversation. Pre-rendered video plays back a finished script. A live-streamed avatar is generated as the conversation happens.


Video model

Process and workflow

Best used for

 

Pre-rendered video

Script Finalized ➔ Video Rendered ➔ Static Playback

Static marketing clips, linear training videos

Live streamed Avatars (DEVA-1)

Real-time Audio ➔ Live Motion Synthesis ➔ Dynamic WebSocket Stream

Real-time engagement, open Q&A, branching workflows

What can businesses use DEVA-1 for?

Teams use DEVA-1 Digital Humans to build trust, help people retain information and add a human touch to automated conversations. Here are five high-impact use cases.

1. HR and employee onboarding

  • The challenge: Traditional onboarding relies on static PDFs and fragmented HR check-ins. Repetitive inquiries about benefits or PTO consume valuable HR bandwidth while leaving new hires with an impersonal introduction.

  • DEVA-1 deployment: An interactive First-Day Guide in the employee portal. It welcomes new hires, guides them through document submission and answers policy questions in real time.

  • Visual advantage: Sub-300ms facial expressions and warm lighting foster immediate psychological safety.

  • Key metrics: Reduction in routine HR ticket volume and faster time-to-productivity through guided checklist completion.

2. E-learning and corporate training

  • The challenge: Long-form videos and text modules suffer from low completion rates and poor retention. Static platforms cannot adapt to a learner's pace or address off-script questions.

  • DEVA-1 deployment: A real-time AI Tutor inside your Learning Management System. Learners can interrupt at any moment to clarify complex concepts, with answers drawn from the course knowledge base.

  • Visual advantage: Eye contact and subtle micro-expressions maintain active focus, turning passive viewing into a one-on-one tutoring session.

  • Key metrics: Higher course completion rates than static video, and domain expertise that scales without recording new videos.

3. Sales enablement and interactive role-play

  • The challenge: Sales coaching demands extensive manager time. Text-based role-play fails to prepare reps for the non-verbal cues of high-stakes discovery calls.

  • DEVA-1 deployment: Simulates buyer personas, such as a skeptical procurement officer, who push back, challenge value propositions and interrupt mid-pitch.

  • Visual advantage: Shows subtle cues like hesitation, skepticism or interest, so reps learn to read facial reactions and adjust their tone on the fly.

  • Key metrics: Onboarding and quota-readiness time reduced by weeks, backed by objective analytics on objection handling, pacing, and filler word usage.

4. Customer support automation

  • The challenge: Multi-step workflows like identity verification and account setup see high drop-off rates when they rely on web forms or basic text chatbots.

  • DEVA-1 deployment: A Virtual Concierge that guides users through complex forms, explains technical requirements and resolves input errors live.

  • Visual advantage: Natural eye-tracking and reassuring head nods ease anxiety during sensitive verification or recovery steps.

  • Key metrics: Significant reduction in form abandonment, plus diversion of routine Tier-1 calls away from human contact centers.

5. Healthcare and patient guidance

  • The challenge: Complex pre-procedure care plans or intake forms can overwhelm patients, leading to missed appointments or incomplete medical histories.

  • DEVA-1 deployment: A Patient Guidance Concierge on web portals or clinic kiosks that gathers structured health data and walks patients through care routines.

  • Visual advantage: Calm, empathetic expressions encourage candid disclosure and ease patient stress.

  • Key metrics: Marked reduction in missed appointments and pre-op preparation errors, alongside a lighter administrative burden on frontline clinical staff.

How do developers integrate DEVA-1?

DEVA-1 uses the same API and SDK foundation as the rest of the UNITH platform, so teams can deploy it with the technical stack they already have. That matters because, for many enterprise teams, developer capacity is a bigger bottleneck than avatar technology.

Integration architecture


Interface

Role and purpose

 

REST API

Handles administrative request-and-response tasks like Avatar management and setup.

WebSocket Streaming

Powers low-latency, bi-directional live video streaming during active chats.

React SDK

Embeds DEVA-1 inside modern React web applications.

Web Component

Drop-in, reusable UI element for quick placement on any HTML page.

Webhooks

Connects event-driven workflow automation tools like Zapier, Make, and n8n.

Upgrading existing Avatars to DEVA-1: Contact us and we’ll provision a DEVA-1 version of your current avatar and update the embed configuration.

How does DEVA-1 compare with pre-rendered AI video?

When you evaluate Digital Human platforms, compare how each one generates video, not just its feature list. The main difference is pre-rendered generative video versus live-streamed avatars:


Criterion

Pre-rendered generative video

Live streamed Avatars (DEVA-1)

 

How it works

Video compiled from finished script

Real-time face generation driven directly by live streaming audio

Best used for

Static marketing clips, linear lectures

Interactive Q&A, open dialogue, guided workflows

Core advantage

High offline render quality

Low latency, unscripted conversational flexibility


How do I get early access to DEVA-1?

Language models brought accuracy to automated answers. DEVA-1 delivers those answers through a lifelike face with natural depth, expression, and movement, generated live as conversations happen.

Early access is open in stages for select enterprise teams. You can test DEVA-1 with your own content, in your own languages, with the teams who will use it every day.

Request Early Access to DEVA-1 or speak with our team to plan your custom pilot evaluation.

A new hire asks about parental leave on their first day. A grounded knowledge model pulls the answer from your internal policy, and an expressive Digital Human delivers it in real time, with natural lip sync and warm micro-expressions.

DEVA-1 is the generative face engine behind that conversation. It generates the Digital Human’s face live from streaming audio, while the conversation happens.

Language models made accurate answers possible, but how an answer is delivered matters just as much. Text is easy to skim. A human face holds attention and builds trust. The face has always been central to UNITH’s Digital Humans, and DEVA-1 makes it more realistic and expressive.

What is DEVA-1?

DEVA-1 is UNITH’s real-time visual AI model for conversational AI avatars. It brings depth, volume and realistic lighting, richer facial expression and subtle, natural micro-movements to every conversation, rendered live from the first word to the last.

Capability

With DEVA-1

What it means in practice

Rendering

Depth, volume and realistic lighting

Avatars look natural and crisp without appearing flat or distorted.

Realism

Richer, more natural facial movement

Enhances viewer trust and engagement during critical interactions, such as executive communications or client-facing roles.

Facial expression

Full facial micro expressions

Allows the Avatar to convey subtle emotions, like empathy or professionalism, matching complex or sensitive conversations accurately.


How does DEVA-1 work?

DEVA-1 turns streaming speech into a live video of a Digital Human’s face. It controls how the face looks, moves and responds in real time through a continuous, 5-step audio-to-video pipeline:

[ Streaming Audio ] ➔ [ Motion Parameters ] ➔ [ 3D Head Pose ] ➔ [ Real-time Frame Render ] ➔ [ WebSocket Stream ]

  1. Streaming audio inputs: Speech audio arrives in small, continuous chunks as it is generated, rather than as a pre-recorded file. The face can start moving before the full sentence is spoken.

  2. Audio-to-motion translation: An audio-driven model translates sound into facial motion parameters. These set how the lips, jaw, cheeks, brows and eyes move at each fraction of a second.

  3. 3D spatial posing: The parameters pose a 3D head representation, so depth and realistic lighting are calculated on the fly.

  4. Frame rendering: Each posed instance is rendered instantly as a video frame.

  5. Continuous delivery: Frames stream to the browser over a persistent WebSocket connection and stay in sync with the audio as the conversation flows.

Lip sync and natural expression

  • Lip sync (visemes): Matches mouth shapes to distinct audio phonemes. Groups of similar-looking sounds (like p, b, and m) map to shared visual mouth shapes called visemes.

  • Facial expression: Coordinates the rest of the face, including the brows, eyes, cheeks and natural head movement, to convey emotion.

The conversational presence stack

Many avatar pilots fail because teams measure the wrong things. Steering committees judge whether the avatar knows the answer. End users judge whether talking to it feels natural and worth their time.

To keep these concerns distinct, we evaluate Digital Human programs across five layers:

  1. Knowledge: The LLM and grounded content, such as policies, product documentation, and sales playbooks.

  2. Dialogue: Turn-taking, latency, interruption handling, and conversational flow.

  3. Voice: Speech quality, accent, language coverage, and voice cloning.

  4. Face (DEVA-1): Live rendering, depth lighting, lip sync, and micro-expressions.

  5. Delivery: Interface embedding across intranets, LMS platforms, or apps, along with backend tool integrations.

An avatar program is only as strong as its weakest layer. Good content needs an engaging face to hold attention, and a lifelike face needs grounded content behind it to stay credible.

You can refine a prompt in minutes, but facial rendering depends on what the platform can do. That is the layer DEVA-1 improves.

Why does a conversational AI avatar need real-time rendering?

Human conversation is fast and sensitive. Research published in PNAS shows that across global languages, human speakers consistently keep turn-taking gaps brief and avoid overlapping speech (Stivers et al., 2009).

People watch an avatar closely in those short pauses, so how a Digital Human looks while it listens is part of the conversation. Pre-rendered video plays back a finished script. A live-streamed avatar is generated as the conversation happens.


Video model

Process and workflow

Best used for

 

Pre-rendered video

Script Finalized ➔ Video Rendered ➔ Static Playback

Static marketing clips, linear training videos

Live streamed Avatars (DEVA-1)

Real-time Audio ➔ Live Motion Synthesis ➔ Dynamic WebSocket Stream

Real-time engagement, open Q&A, branching workflows

What can businesses use DEVA-1 for?

Teams use DEVA-1 Digital Humans to build trust, help people retain information and add a human touch to automated conversations. Here are five high-impact use cases.

1. HR and employee onboarding

  • The challenge: Traditional onboarding relies on static PDFs and fragmented HR check-ins. Repetitive inquiries about benefits or PTO consume valuable HR bandwidth while leaving new hires with an impersonal introduction.

  • DEVA-1 deployment: An interactive First-Day Guide in the employee portal. It welcomes new hires, guides them through document submission and answers policy questions in real time.

  • Visual advantage: Sub-300ms facial expressions and warm lighting foster immediate psychological safety.

  • Key metrics: Reduction in routine HR ticket volume and faster time-to-productivity through guided checklist completion.

2. E-learning and corporate training

  • The challenge: Long-form videos and text modules suffer from low completion rates and poor retention. Static platforms cannot adapt to a learner's pace or address off-script questions.

  • DEVA-1 deployment: A real-time AI Tutor inside your Learning Management System. Learners can interrupt at any moment to clarify complex concepts, with answers drawn from the course knowledge base.

  • Visual advantage: Eye contact and subtle micro-expressions maintain active focus, turning passive viewing into a one-on-one tutoring session.

  • Key metrics: Higher course completion rates than static video, and domain expertise that scales without recording new videos.

3. Sales enablement and interactive role-play

  • The challenge: Sales coaching demands extensive manager time. Text-based role-play fails to prepare reps for the non-verbal cues of high-stakes discovery calls.

  • DEVA-1 deployment: Simulates buyer personas, such as a skeptical procurement officer, who push back, challenge value propositions and interrupt mid-pitch.

  • Visual advantage: Shows subtle cues like hesitation, skepticism or interest, so reps learn to read facial reactions and adjust their tone on the fly.

  • Key metrics: Onboarding and quota-readiness time reduced by weeks, backed by objective analytics on objection handling, pacing, and filler word usage.

4. Customer support automation

  • The challenge: Multi-step workflows like identity verification and account setup see high drop-off rates when they rely on web forms or basic text chatbots.

  • DEVA-1 deployment: A Virtual Concierge that guides users through complex forms, explains technical requirements and resolves input errors live.

  • Visual advantage: Natural eye-tracking and reassuring head nods ease anxiety during sensitive verification or recovery steps.

  • Key metrics: Significant reduction in form abandonment, plus diversion of routine Tier-1 calls away from human contact centers.

5. Healthcare and patient guidance

  • The challenge: Complex pre-procedure care plans or intake forms can overwhelm patients, leading to missed appointments or incomplete medical histories.

  • DEVA-1 deployment: A Patient Guidance Concierge on web portals or clinic kiosks that gathers structured health data and walks patients through care routines.

  • Visual advantage: Calm, empathetic expressions encourage candid disclosure and ease patient stress.

  • Key metrics: Marked reduction in missed appointments and pre-op preparation errors, alongside a lighter administrative burden on frontline clinical staff.

How do developers integrate DEVA-1?

DEVA-1 uses the same API and SDK foundation as the rest of the UNITH platform, so teams can deploy it with the technical stack they already have. That matters because, for many enterprise teams, developer capacity is a bigger bottleneck than avatar technology.

Integration architecture


Interface

Role and purpose

 

REST API

Handles administrative request-and-response tasks like Avatar management and setup.

WebSocket Streaming

Powers low-latency, bi-directional live video streaming during active chats.

React SDK

Embeds DEVA-1 inside modern React web applications.

Web Component

Drop-in, reusable UI element for quick placement on any HTML page.

Webhooks

Connects event-driven workflow automation tools like Zapier, Make, and n8n.

Upgrading existing Avatars to DEVA-1: Contact us and we’ll provision a DEVA-1 version of your current avatar and update the embed configuration.

How does DEVA-1 compare with pre-rendered AI video?

When you evaluate Digital Human platforms, compare how each one generates video, not just its feature list. The main difference is pre-rendered generative video versus live-streamed avatars:


Criterion

Pre-rendered generative video

Live streamed Avatars (DEVA-1)

 

How it works

Video compiled from finished script

Real-time face generation driven directly by live streaming audio

Best used for

Static marketing clips, linear lectures

Interactive Q&A, open dialogue, guided workflows

Core advantage

High offline render quality

Low latency, unscripted conversational flexibility


How do I get early access to DEVA-1?

Language models brought accuracy to automated answers. DEVA-1 delivers those answers through a lifelike face with natural depth, expression, and movement, generated live as conversations happen.

Early access is open in stages for select enterprise teams. You can test DEVA-1 with your own content, in your own languages, with the teams who will use it every day.

Request Early Access to DEVA-1 or speak with our team to plan your custom pilot evaluation.

Keep reading

Two employees measuring ROI of Avatar investment

Guides & strategy

How to measure conversational AI ROI

Measuring conversational AI ROI is not just about interaction volume. It depends on being able to attribute value to channels, campaigns, and touchpoints. Many enterprise AI deployments work in practice, but fail at renewal time because teams cannot clearly prove their impact. Attribution helps companies understand which website, campaign, or internal platform generated each interaction and what real usage looks like. Traceability matters for marketing, IT, and leadership because it connects activity to business outcomes. Features like Custom Data Tags help turn promising pilots into sustainable, defensible programs during budget reviews.

Written By: Nia

8 minutes

Guides & strategy

Why businesses need a Digital Human

Most organizations already use AI, but few have scaled it. In McKinsey's 2025 State of AI survey, 88 percent of respondents say their organization uses AI in at least one business function, yet nearly two-thirds have not begun scaling it across the enterprise. A Digital Human is an AI-powered character with a realistic face and voice that holds real-time spoken conversations, using the knowledge and AI you choose. It closes the gap between what AI can do and what people actually use, because it is clearer, more engaging and easier to trust than a text box. UNITH Digital Humans run in open dialogue, knowledge-base or plugin mode, and connect to 7,000+ apps through Zapier, Make.com and n8n. The fastest returns come from repeated conversations: onboarding, training, customer support, interviews and lead generation.

Written By: Nia

6 minutes

AI avatar helping human with roleplay

Training & HR

AI avatars roleplay for employee training

AI avatars roleplay gives employees a safe way to practice real conversations before high-stakes situations. Compared with traditional roleplay, AI-powered simulations are easier to scale, repeat, personalize, and measure. Roleplay with interactive avatars is valuable for sales, customer service, onboarding, leadership, and compliance training. Conversational AI makes training feel more realistic because learners respond in natural language instead of following static scripts. Platforms like UNITH AI help businesses build practical, human-like training experiences that are easier to deploy across teams.

Written By: Nia

9 minutes

Next step

Put Digital Humans to work in your business.

Put Digital Humans to work in your business.

Conversational AI solutions built around your business objectives, delivered and managed by our team from day one.