Podcast Atlas

Updated · 4 episodes · 4 shows · 4 source notes

concept Topics: Technology

AI Public Likeness Generation

Definition

AI public likeness generation is the use of AI systems to generate, transform, imitate, or commercially deploy a person’s face, body, public profile, celebrity identity, or recognizable visual signal in ways that raise consent, privacy, authorship, endorsement, and consumer-confusion questions.

Current Synthesis

Public likeness has become a rights problem across film production, social apps, entertainment labor, and ecommerce. 贾樟柯’s film discussion shows that generated candidate people can raise portrait-right questions before release. The Meta Muse Image case shows how public usernames and social profiles can become promptable likeness surfaces by default. Film and dubbing examples add performers, blended voices, copyright, and proof. In commerce, altered faces, masked faces, cropped faces, nickname hints, live-room decoration, product categories, and voice substitution can still make consumers infer a celebrity or creator endorsement.

The current judgment is that likeness governance is not solved by a single label or opt-out toggle. The hard questions are consent default, authorization scope, audience or consumer interpretation, technical proof, platform pre-review, appeal for false positives, and whether generated or altered identity signals are being used to borrow trust from a real person.

Key Claims

  • Public-profile availability does not automatically mean consent to become a generative prompt target.
  • Film and entertainment workflows can raise likeness issues before a finished work reaches audiences.
  • Opt-out controls shift discovery and settings burden onto the person whose likeness may be generated.
  • Voice, face, body, nickname, account, and setting cues can combine into recognizable likeness even when no exact copy appears.
  • Commerce raises an endorsement-specific harm: consumers may buy because they infer a celebrity, creator, or brand relationship that does not exist.
  • Technical provenance and evidence matter because generated or blended likeness can be denied, altered, cropped, or routed through multiple tools.
  • False positives matter because people can look similar, models can misidentify them, and lawful authorization may not be visible to the platform.

Evidence

Counterevidence & Qualifications

The sources do not provide a unified legal doctrine across jurisdictions. The social-app case is a reported product-policy concern, the film cases are practitioner reflections, and the Douyin ecommerce claims are platform/legal-explainer evidence rather than court-adjudicated infringement findings.

What Changed

  • Migrated the page to the synthesis-v1 concept schema.
  • Added the Douyin ecommerce branch, extending public likeness from film/social generation into AI-enabled commercial impersonation and platform pre-review.

Sources

4 source notes across 4 shows
  1. EP277 对话贾樟柯（下）：我没有背叛真实世界，我只是在寻找电影的新可能 Talk三联
  2. E234｜未来实拍电影还存在吗？与导演陆川聊聊AI给影视人的恐惧与自由 硅谷101
  3. OpenAI's GPT-5.6 release raises questions about White House control over new models Marketplace Tech
  4. No.224 用 AI「偷」明星声音和形象侵权卖货，直播间该怎么管？ 三五环