Pre-Publication V3. Neural Avatar (Synthetic Avatar Human, Generative Digital Avatar Human) A human likeness generated wholly through artificial intelligence and deep neural networks, intended as a visual digital representation for creative, interactive, or performance applications, without requiring traditional geometric pipeline inputs. This may be either a Synthetic Performer or Digital Replica.
Deliberation Summary:
(1) Clarify the term to ensure it matches the intended definition. (2) Align the terminology with established industry usage while keeping the definition technically accurate, including whether references to a performer or to a human should be removed so the definition covers non-human applications. (3) Remove reference to “human” as unnecessary given context.
Pre-Publication V2. A human likeness generated wholly through artificial intelligence and deep neural networks, intended as a visual digital representation for creative, interactive, or performance applications, without requiring traditional geometric pipeline inputs. This may be either Synthetic Performer or Digital Replica.
Deliberation Summary:
(1) Question if the term refers to an attempt to recreate a specific person's intellectual likeness for exploitation, and that the relationship between this term, synthetic output, and performance was unclear; alternative names were proposed. (2) Wording was long and closer to promotional descriptions, given that it describes generating an image or video of a person using the same underlying technology defined elsewhere. (3) Consider clarification of where the movement originates and whether it is present in the training data; question on whether the entry duplicates the separate entry for a synthetic performer. (4) Detail about replication matters; the entry should lead with established industry terminology, usage differs across existing agreements, and definitions should distinguish generated performance that is not identifiable from performance that is identifiable, including where one person's appearance is applied over another performer's movement that remains recognizable as that performer's.
Pre-Publication V1. A human likeness generated wholly through artificial intelligence and deep neural networks, without requiring traditional geometric pipeline inputs such as photogrammetry scans, rigged meshes, or actor-specific capture sessions. Neural avatars are capable of synthesized movement, facial expression, speech, and lip synchronization, and may operate in real-time or rendered contexts. Unlike scanned or modeled digital humans, the underlying representation is learned rather than constructed encoded in neural network weights rather than explicit geometry. Output fidelity ranges from stylized to photorealistic depending on implementation. Neural avatars may be entirely novel synthetic identities or generalizable human-like characters not based on any specific individual, distinguishing them from AI-driven digital doubles of real people.