Celebrity face tests reveal gaps between image generation and permission

A familiar face can appear in an AI image even when the user supplies no photograph. In a November 12 test, The Paper had three people try four image generators with seven celebrities. It reported different restrictions for the Chinese and Hollywood figures tested in Jimeng. The immediate question for anyone making a short drama or advertisement is whether a successful generation also supplies permission to use the person’s likeness.

A separate dispute gives that question a concrete setting. Red Star News reported on November 12 that a hearing notice listed November 10 for Zhao Liying’s portrait-rights case against Hongguo’s operator, Beijing Bimo Liuxiang Technology, Beijing Douyin Information Service and an individual. The outlet said Hongguo had not answered its inquiry at publication. It also recounted an earlier complaint from Yang Zi’s studio over an AI short drama. Synthetic celebrity faces and objections to their use clearly predate this latest test.

The distinction is easier to understand without technical vocabulary. Imagine an illustrator drawing a recognizable actor from memory. The lines are newly drawn, but viewers may still identify the subject. Likewise, creating fresh pixels does not necessarily create a fictional identity. The Paper’s experiment describes particular prompts and conditions, not a universal result for every version of each tool. It is evidence of what the testers observed, rather than a complete map of the products’ safeguards. The creative process and the right to publish its result remain separate questions.

China’s Supreme People’s Court addressed that separation in a September explanation of its guidance on AI disputes. It said using AI to produce and use or publish an identifiable digital likeness without consent implicates personality rights, including portrait rights. Its explanation also described injunctions as an available response when the relevant conditions are met, with courts able to require users and service providers to act against further harm. That framework focuses on identifiable people and subsequent uses, rather than treating the absence of a copied photograph as the end of the inquiry.

For producers, a digital cast does not remove the need to resolve permission. For audiences, recognizing an actor’s face does not establish that the actor performed in the scene or endorsed the product. Those are practical distinctions even when an image is obviously synthetic. As generation becomes a small part of the production workflow, the harder work may sit around it: obtaining permission, explaining who participated and handling complaints. A tool can shorten the path from an idea to an image. It cannot make another person’s agreement appear simply by completing that image.

Original source published: The Paper, November 12, 2026, 21:48 Beijing time / 8:48 AM EST.