Baron App, Inc. d/b/a Cameo v. OpenAI, Inc., et al. is a case in the Northern District of California that pits a company supplying bespoke celebrity greetings against an AI company that provides AI-generated celebrity videos.
The plaintiff has a registered trademark for CAMEO and uses it in connection with having live actors record videos at the request and at the expense of the user. OpenAI’s video creation tool Sora also has a “cameo” feature in which the user can use licensed images of celebrities to make “deep fake” messages of their own. Cameo brought a lawsuit in the Northern District of California seeking a temporary restraining order (TRO) against OpenAI to prevent trademark infringement and dilution. Here, OpenAI was asked to show cause for why a TRO should not have been issued.
OpenAI claimed that it did not infringe because it did not use “cameo” as a trademark, based on the fact that it was not presented in a “stylized way.” Rather, OpenAI said its use of cameo was merely descriptive of “a brief appearance.” In response, the court said visual appearance is not the only thing that contributes to whether something is intended to be used as a trademark — and that “cameo” was actually used as the name for the service in OpenAI’s marketing materials.
Cameo sought to show that OpenAI’s use of “cameo” threatened irreparable harm to its reputation by associating the word with the offensive AI “slop” for which Sora is known (or infamous), including videos of famous people, like Dr. Martin Luther King, Jr., doing abhorrent things. The fact that these deep fakes often lacked any unrealistic features aggravated the situation, further suggesting that real celebrities had participated in the infringer’s service, Cameo said.
OpenAI claimed that videos made with Sora were easily distinguishable from Cameo’s videos because they were watermarked with the Sora name and logo, but the court found that didn’t exist on all of the videos. The court went on to characterize the watermark as a disclaimer and pointed out that disclaimers are generally insufficient to prevent consumer confusion. Rather, a consumer’s initial confusion is enough to signify infringement.
One of OpenAI’s surveys found forward confusion — consumers who confused OpenAI’s service for Cameo’s — in 10% of respondents. Generally, 10% or more is sufficient to demonstrate actual confusion. Perhaps because the results were so close to the forbidden threshold, OpenAI then sought to deduct 3% from the “net confusion rate,” claiming that 3% of respondents in any survey either get confused or try to please the person asking them questions. The court strongly countered this argument by essentially calling the 10% rate an underestimation given the multiple examples of documented, actual confusion. (The court found a different survey by OpenAI, which found 0% confusion, was not persuasive because of the methodology used.)
The defendant then suggested that it was engaging in fair use of the word cameo, but the court disagreed with defendant’s assertion that its use of “cameo” did not constitute a trademark. It pointed to a Second Circuit case in which Oprah Winfrey’s use of the phrase “Own Your Power” was deemed a trademark even though it wasn’t registered. (In that case it came down to consistent use in association with Winfrey’s approval and sponsorship.)
Finding a likelihood of success, and weighing the potential adverse consequences to OpenAI, the court granted the preliminary injunction. One of the reasons the court found that OpenAI was not likely to be harmed by the preliminary injunction was the fact that it had already stopped using “cameo” in response to the TRO.
It didn’t seem fair to me that OpenAI’s compliance with the TRO should be used against it. However, the relative strength of the plaintiff’s mark and OpenAI’s self-defeating surveys outweighed that factor. “Discounting” a survey by 3% seemed dubious if not a desperate attempt to avoid a finding that relevant consumers thought of the plaintiff when seeing “Cameo” used in connection with celebrity videos.
