Background
The client deploys AI digital humans at scale in e-commerce live-streaming, smart customer service, and brand marketing; some digital humans are generated from real persons’ likeness and voice. As the business scaled, three obligations surfaced simultaneously—deep-synthesis filing, content labeling, and personality-rights authorization—directly affecting platform-approval passage and infringement complaints.
Key Challenges
- Deep-synthesis filing trigger: Generating content from real persons’ likeness/voice triggered deep-synthesis service filing.
- Labeling boundary: AIGC labeling (GB 45438-2025) requires explicit and implicit labels for generated content; digital-human labeling placement and retention were prone to non-compliance.
- Personality-rights gap: Real-person authorization covered only “real appearance/voice-over,” not “generating a digital twin for commercial use.”
- Impersonation & misleading: Unlabeled digital-human endorsement risked being deemed consumer misleading.
Geason’s Approach
- Filing & assessment: Managed deep-synthesis service filing and necessary generative-AI safety assessment.
- Labeling implementation: Designed explicit (UI notice) + implicit (metadata) dual-labeling and retention per GB 45438-2025.
- Personality-rights redesign: Rewrote portrait and voice-right authorizations to explicitly cover “digital-twin generation and commercial use” and sublicense scope.
- Advertising compliance: Designed explicit labeling and disclaimer boundaries for digital-human endorsement to avoid misleading findings.
The Non-Obvious Judgment
The most underestimated digital-human risk is the “purpose” gap in personality-rights authorization. Most real-person consents are for “appearance/voice-over,” not “generating a digital twin for long-term commercial use”—meaning even with the person’s consent, commercial digital-twin use may still infringe. The second point is labeling: GB 45438-2025 requires not “a watermark” but the dual obligation of explicit UI notice and implicit metadata retention, and it must be traceable. These two points determine whether a digital-human business can scale safely.
Key Outcomes
- Deep-synthesis service filing and necessary safety assessment landed with traceable records.
- Digital-human content achieved explicit+implicit dual-labeling and retention per GB 45438-2025.
- Portrait/voice-right authorizations redesigned to cover digital-twin generation and commercial use, avoiding the high-incidence infringement area.
- Explicit labeling and disclaimer boundaries for endorsement designed, with clear platform-approval expectations.
Client names and commercial data are anonymized. The rules, regulatory procedures and deliverable types described are as they occurred, not simplified.