How Virtual Digital Humans Are Shaping the Future of Entertainment and Commerce

Definition of Virtual Characters

Virtual characters, also called virtual digital humans, are a broad concept that can refer to various virtual personas across industries. The term dates back to the 1989 Visible Human Project by the U.S. National Library of Medicine, and in China it was introduced in 2001 during the 174th Xiangshan Science Conference as “digital virtual human”.

Form – Possess a human or humanoid appearance with specific facial features.

Movement – Exhibit human‑like behaviors, capable of expressing language, facial expressions, and body gestures.

Spirit – Have cognitive abilities to recognize the environment and interact with people.

These three elements form a progressive hierarchy for the completeness of a virtual character.

Industry Status

In recent years, virtual characters have entered e‑commerce, finance, film, gaming and other sectors with varying market sizes. For example, China’s virtual idol market reached ¥3.46 billion in 2020 and is projected to hit ¥6.22 billion in 2021.

The first virtual singer, Lin Mingmei, appeared in Japan in 1962, followed by the globally known Hatsune Miku created by Crypton. Technological advances—from CG creation to motion capture and AI‑driven pipelines—have continuously lowered production costs, improved realism, and made speech interaction more natural, driving rapid growth of the virtual character market.

Industry Chain

The virtual‑character industry chain is now well‑established and can be divided into three layers: foundation, platform, and application.

1. Foundation Layer

Provides basic hardware and software support, including display devices, optical components, sensors, chips, modeling software, and rendering engines.

2. Platform Layer

Encompasses hardware‑software systems, production‑technology service platforms, and AI capability platforms that supply the technical foundation for creating virtual characters.

3. Application Layer

Integrates virtual‑character technology into concrete scenarios, delivering solutions that empower various industry domains.

Industry Products

Only a few representative examples are listed.

IP‑Operation

Samsung Neon – AI‑driven digital human with realistic appearance and expressive emotions.

Lil Miquela – First CGI character used primarily for fashion‑brand marketing.

Hatsune Miku – World‑famous virtual singer from Crypton, with tools for user‑generated music.

CodeMiko – Independent Virtual YouTuber (VTuber) with a strong presence on YouTube.

Luotianyi – Popular Chinese virtual singer launched by Shanghai HENIAN.

Lu Ming – MiHoYo’s desktop virtual “girlfriend”, also active on B‑station.

Solution‑Based

VOCALOID – Yamaha’s singing‑synthesis technology used by many virtual singers.

MetaHuman – Unreal Engine’s 2021 tool for rapid creation of high‑fidelity digital humans.

Tencent IP Virtual Human, Xiangxin AvatarX, NetEase Youling – Enterprise‑focused suites offering end‑to‑end services, cloud AI algorithms, and extensive IP libraries for use in virtual anchors, teachers, customer service agents, directors, and marketing.

Game‑Focused

Zepeto, Shine Bright – Popular “dress‑up” games that sparked significant community discussion upon release.

Our Initiatives

At the beginning of this year, Alibaba’s Front‑End Committee launched a Virtual‑Character Working Group within the Interactive Graphics direction to share and research technologies and applications of virtual characters. Five teams contributed works covering gaming, video, and live‑streaming scenarios, aligning with current industry hotspots.

Gaming

Virtual characters are standard in games, especially those allowing players to customize avatars (“character‑creation” games).

1. Taobao Life

A lightweight game inside the Taobao mobile app that offers facial customization, dressing, beauty filters, photo capture, shopping, and home‑decor activities. Users can also export their avatars to other interactive games such as “Dou Dizhu”.

2. Yang Koala

A light interactive within the Kaola shopping app that lets users raise a virtual koala, featuring dressing and feeding mechanics built with Web technologies.

Video

Short videos featuring virtual characters provide consistent user experiences and generate incremental revenue. Production leverages motion capture, facial‑pose recognition, and AI‑driven generation to make virtual performances closely resemble real human behavior.

Live Streaming

The combination of live streaming and virtual characters is still exploratory; integrating two high‑growth industries into new commercial models is challenging. Virtual avatars appear in Taobao Live, Youku Live, and Ant Financial Live, sometimes acting as assistants, co‑hosts, or dedicated sales agents.

Our Research Topics

Art Production

Technology assists artists in both 2D and 3D pipelines. In 2D, tools can export code directly from Sketch. In 3D, technical integration is essential throughout the lengthy production process, leading to customized tools and plugins for different engines and business needs.

Expression and Motion

Realistic facial expressions and motions are crucial for user perception. We simulate skeletal and muscular interactions using Blendshape for facial animation and physics‑based bone rigs for motion, sometimes adding collision boxes to prevent clipping. When Blendshape count exceeds 50, a wide range of expressions can be reproduced.

Rendering and Style

Key rendering considerations include engine choice, material system, and cloud rendering.

Engines – Teams use Unity, self‑developed EVA Figure (Web, based on open‑source Hilo3D), Oasis3D (Web, open‑source), and ApisXEngine (cross‑platform).

Materials – PBR materials are standard; we have experimented with ultra‑realistic styles, exemplified by the sign‑language assistant “Xiao Mo” showcased at the Cloud Conference.

Cloud Rendering – In live‑stream scenarios we employ Unity rendering with streaming, collaborating with Alibaba Cloud and Node teams to explore Web‑engine + Puppeteer solutions for non‑real‑time cloud rendering.

Scene Orchestration

We have built a “director system” that enables rapid scene composition for various operational needs, such as scheduling virtual anchors in live streams, generating diverse game environments, or applying different visual templates to video scripts.

Big Data and Algorithms

Long‑term development of virtual characters relies heavily on big‑data‑driven algorithms, as facial expressions and motions are highly variable. Training models on large video datasets or key‑frame data collected during production is essential to overcome engineering bottlenecks and achieve desired quality.

D2 Preview

We will present a comprehensive talk titled “The Birth of Virtual Idols – Industry and Technology Exploration” at the D2 Front‑End Technology Forum on December 18‑19. All are welcome to join and discuss.