Doubao has now opened “real-person avatars” on mobile. After completing identity verification, recording a specified short video and adding selfies, users can have their avatars appear on camera and speak from written prompts. Our view: AI video is moving from “generating images” to “replicating personal likeness,” lowering the barriers to content production while making avatar credibility and authorization risks land more directly on individuals.
What this is
A real-person avatar does more than swap the face in an existing video. It first builds a digital likeness from a person’s facial features, voice and movements, then uses a model to generate new video. Doubao’s workflow requires the user to complete identity verification, record themselves following prompts, add selfies and enable “use avatar on camera” under “Video Generation.”
The result depends on the quality of the source material. Soft lighting, fully visible facial features, a stable voice and consistent-style selfies are more conducive to preserving similarity than heavy beauty filters. Before publishing, users must also set the scope of authorization, review the script and label the content as AI-generated.
Industry view
The key development is that Doubao has compressed a digital-human workflow that previously required professional shooting, editing and voice-over into a mobile app. Individual creators can now produce spoken-address, narrative and training videos inexpensively and at scale. This will expand the supply of AI video while shifting competition away from whether a platform can generate content at all and toward whether the likeness is stable and the delivery is credible.
The risks are equally clear. The more realistic a personal avatar becomes, the more likely it is to be used for unauthorized endorsements, fabricated statements or misleading content. Doubao currently requires users to create avatars of themselves and restricts uploads of other people’s faces or unauthorized likenesses. But entry-point moderation alone cannot cover the distribution chain after generation. Portrait authorization, training-data provenance and AI-labeling standards remain infrastructure the industry needs to build.
Impact on regular people
For enterprise IT: The production cycle for marketing, training and customer-service video may become shorter, but enterprises need formal systems for avatar authorization, content review and generated-content labeling. “Created by the person themselves” should not be treated as a complete compliance workflow.
For individual careers: People who can write effective prompts will be better positioned to handle video scripting, asset management and content operations. Understanding the business and delivering content consistently will be more valuable than simply mastering complex tools.
For the consumer market: Consumers will encounter more content featuring familiar faces, making it harder to distinguish footage shot by the person from an AI avatar. Beyond platform labels, whether major platforms can provide traceable generation records will determine the foundation of trust.