Capability Details

Digital human

A unified management and operations platform for digital humans in venues and showrooms: multiple avatars, languages and publishing channels.

01 / Technology decisions

Three trade-offs you cannot avoid

Each has its own cost; here is which side we chose and why.

2D or 3D: does it look right, or does it move

Common practice3D modelling, aiming for high fidelity

Alternative2D real-person avatar, aiming for natural lip sync

Our approach2D as the default (narration), 3D as a supplement (interaction)

3D modelling and driving are expensive, and on a hall's large screen the more lifelike it gets the more easily it falls into the uncanny valley. A 2D real-person avatar is already enough for one-way output such as narration and announcements, at an order of magnitude lower cost. 3D is worth it only when users need to view from any angle or when body movement is required.

Where the voice comes from: TTS or human recording

Common practiceAll synthesised with TTS

AlternativeAll recorded with real voice actors

Our approachFixed content recorded by voice actors, variable content by TTS

TTS reads any text, but long sentences lose natural emotion and pauses. Fixed narration in a hall (welcome messages, exhibit descriptions) is better recorded with a real voice — it sounds good and stays stable; dynamic data (today's visitor numbers, live figures) is better served by TTS.

Where to deploy: cloud or on-premise

Common practiceCloud rendering, client only plays the stream

AlternativeRendered locally, fully deployed on on-site devices

Our approachNarration On-Premise, Rendering-Intensive Workloads in the Cloud

Cloud rendering saves local compute, but it dies without the network — and a hall's worst fear is having no connection on site. So narration-type digital humans must run locally; only scenarios needing high-precision real-time driving should consider the cloud.

02 / Option comparison

One problem, three approaches

There is no optimal solution, only solutions that fit the scenario.

Solution	Realism	Cost	Works offline	Best suited to
2D real-person avatar + TTS	Medium-high	Low	Yes	Showroom narration, announcements, guided tours
2D real-person avatar + recorded voice	High	Medium	Yes	Fixed narration scripts, promotional videos
3D virtual avatar + real-time driving	High (but prone to the uncanny valley)	High	Depends on the solution	Interactive Q&A, gamified experience

For showroom scenarios, 'works without network' comes first — that is a hard constraint.

What we do

Avatar library · Persona & voice · Knowledge base
Scripted Talking-Points Authoring
Multilingual · multi-channel publishing
Operational data and impact tracking

Discuss requirements

Delivery format

Deployment modelOn-Premise / Private Cloud / Intranet

Scope of deliverySystem + Documentation + Training

Maintenance supportRemote + on-site

Source Code DeliveryNegotiable

Related Cases

Proven systems

Client names are anonymized per confidentiality requirements.

Public venues / multimedia interaction

Public Science Venue in a Special Administrative Region

Background: A public science venue that needs interactive content covering food, weather, epidemic prevention, school and traffic topics.
Solution: Several custom interactive systems: food safety, lightning safety sand table, campus safety sand table, infectious disease prevention, school bus safety.
Technology: Touch interaction · sandbox projection · 3D scenes · central control backend · data collection
Outcomes: Unified delivery of interactive systems across themes, centralised management in one control backend, and exportable operational data.

Port / Agent + Consortium Chain + App

Large Container Port · Smart Safety Training Centre

Background: A container port ranked among the top in China by annual throughput. Its existing training was mostly classroom lectures and could not cover hands-on practice for high-risk operations.
Solution: Comprehensive safety training centre + training and assessment platform + mobile learning. 7 assessment systems, 9 software modules, a 3-node consortium chain.
Technology: FISCO BCOS consortium chain (PBFT) · transcripts anchored on-chain via SHA-256 · node.js back end · Android client · 3D model interaction
Outcomes: Training records verifiable on chain and tamper-proof transcripts; covers 55 equipment categories and 66 equipment ledger entries.

View all case studies

03 / Implementation path

How many steps from zero to launch

The order cannot be reversed: starting with write operations or multi-step orchestration makes errors very hard to trace.

Define scenarios and interaction depth

Is it one-way narration or two-way Q&A? For one-way, 2D plus recorded audio is enough; Q&A needs a knowledge base and speech recognition.

Record appearance and voice

2D requires filming a real person (multiple angles, multiple expressions); record the voice from the narration script, watching pace and pauses.

Connect the rig and lip sync

Convert text into a time-stamped phoneme sequence that drives lip sync and expression. The precision of this step decides whether it 'looks like it is talking'.

Connect video walls and central control

Digital humans usually sit under the showroom's central control, so they must be wakeable, switchable and stoppable from it. Define the interfaces up front; changing them later is painful.

On-Site Integration Testing

Lighting, noise and network conditions in the showroom all affect the result. Microphones and volume must be tuned on site; you cannot tune them in the office and then deliver.

04 / Key metrics

Verifiable Numbers

No adjectives; write the specifications.

Lip-sync error< 80ms　Practically imperceptible to the eye

First response1–2 seconds　Includes speech recognition and search

Avatar type2D real person / 3D virtual　Choose by scenario

Works offlineYes　On-Premise for Narration-Type Workloads

05 / FAQ

What technical leads ask most often

These are questions we have actually been asked, answered without marketing spin.

Can a digital human fully replace a human guide?

Fixed narration is fine; complex Q&A is not there yet. Its current role is to take over repetitive explanations and free the human guide for the parts that need a person.

Can the avatar be customised to a specific person of our choosing?

Yes, but it requires the person's authorisation and recorded material. For real people's likenesses, mind portrait rights and agree the scope of use.

How good is multilingual support?

Yes — Chinese, English, Japanese, Korean and more, with lip movement driven per language. Quality for less common languages depends on the phoneme library for that language.

Do we need dedicated GPUs?

An ordinary industrial PC is enough for a 2D solution; real-time 3D driving needs a discrete GPU. The exact configuration depends on the required rendering quality and frame rate.

Is it hard to change the narration later?

For the TTS parts you simply edit the text in the back office; the recorded parts have to be re-recorded. So our recommendation is: record the fixed copy, use TTS for content that changes.