What Are Digital Humans?

Digital humans are computer-generated characters designed to look, move, and communicate like people, and to do so interactively rather than as prerecorded animation. They are a specialized form of avatar, distinguished by an emphasis on human likeness in geometry, skin appearance, motion, speech, and social behavior. A complete digital human couples a rendered body and face with a control system that drives expression and gesture in response to a conversational partner or a live performer.

The field draws on computer graphics, computer vision, speech processing, and artificial intelligence, with a long history in visual effects and game production supplying the appearance side and research on embodied conversational agents supplying the behavioral side. A survey of more than five decades of work on realistic virtual humans, crowds, and embodied conversational agents traces how tools that once required a studio pipeline became accessible enough that individual developers can create and animate convincing characters. Terminology varies across communities: virtual human, synthetic human, and photorealistic avatar are used for overlapping ideas.

Appearance and Animation

Constructing the visible character begins with capture. Photogrammetry rigs, structured light scanners, and light stages record geometry together with reflectance properties, which is what allows subsurface scattering in skin and the specular response of eyes and hair to be reproduced under new lighting. Parametric body and face models fitted to that data, along with blendshape rigs and physically based muscle simulation, provide the deformation controls. Learned representations have shifted much of this pipeline: neural radiance fields and Gaussian splatting reconstruct animatable heads and bodies from ordinary video, and diffusion-based generators synthesize identities that were never scanned. Audio-driven models map speech directly to lip and jaw motion, which removes the need for hand-keyed visemes.

Behavior and Interaction

An animated shell becomes a digital human when something decides what it should say and do. Contemporary systems chain automatic speech recognition, a large language model for dialogue, text-to-speech synthesis, and a behavior planner that selects gaze targets, head nods, posture shifts, and hand gestures consistent with the utterance. Nonverbal timing matters as much as content, because turn-taking cues and backchannel responses carry much of the sense of presence. A study of photorealistic embodied conversational agents administering surveys found measurable effects of embodiment on response quality, engagement, and satisfaction, which is the kind of evidence used to justify the added system complexity.

Fidelity, Trust, and Human Digital Twins

Higher realism is not uniformly better. Mismatches between visual fidelity and behavioral fidelity tend to reduce perceived social presence, the effect commonly discussed as the uncanny valley following Masahiro Mori's 1970 essay, so designers often deliberately stylize a character whose behavior model is limited. A related strand of work builds digital humans that represent a specific real person rather than a synthetic identity. A survey on the definition and design of the human digital twin describes systems that mirror an individual's physiology, ergonomics, or task behavior for use in healthcare and industrial settings. Both strands raise questions of consent, likeness rights, and disclosure, since the same techniques that produce a helpful agent produce convincing impersonation.

Applications

Digital humans have applications in a range of fields, including:

• Film, television, and game production, including digital doubles and de-aging
• Customer service and virtual reception agents
• Healthcare, for patient education, therapy support, and clinician training
• Corporate training and language learning with simulated interlocutors
• Virtual reality and social telepresence, where users are represented by avatars
• Ergonomics and human factors analysis using biomechanical body models
• Broadcast and retail presenters, including virtual influencers