Disney has been granted a US patent for an artificial intelligence system that would allow costumed theme park characters such as Mickey Mouse to hold real-time conversations with guests in the character’s own voice. The US Patent and Trademark Office granted the patent yesterday, about two and a half years after Disney filed it in May 2024, according to Disney fan site BlogMickey.
The patent, titled “Artificial Intelligence Based Character-Specific Speech Generation”, describes a system in which a human performer remains inside the costume while AI writes and speaks the character’s lines.
Disney has not said whether or when it will introduce the technology. WDW News Today, which reported the patent on Tuesday, said the grant does not guarantee the system will appear in Disney parks and that there is no timeline for it.
According to WDW News Today’s account of the filing, the costume would carry sensors and microphones to pick up what guests are saying and doing. An inward-facing camera, an eye tracker or both would read the facial expression of the performer.
A database would hold the character’s voice characteristics, including pitch, loudness, accent, rhythm and inflection, alongside its personality, archetype, such as hero, villain or sidekick, favourite words and catchphrases.
A large language model would take in the information from the microphones, sensors and cameras, consult the database and write a reply in the character’s style. A separate AI model would then voice that reply in the character’s voice through a speaker hidden in the costume, while actuators adjust the costume’s face to match the dialogue.
BlogMickey reported that the performer’s own expression would set the emotional tone of each line: a smile would produce a happy delivery, a smirk a smug one and a frown disappointment. The filing defines real time as a reply arriving without an unnatural delay, and allows for a human editor to review responses, the site said. It added that the system could also draw on background sounds, the time of day, the weather and nearby scheduled events.
The database could optionally include a history of a guest’s past interactions with a character. WDW News Today said that history would not store personally identifiable information such as age, gender, race or ethnicity.
Disney’s full-body “fur” characters, including Mickey and Minnie Mouse, do not usually speak during meet and greets, according to WDW News Today. BlogMickey said performers instead communicate through poses, gestures and physical comedy.
The company has tried pre-recorded speech before. Mickey spoke during his Magic Kingdom meet and greet for five years until 2018, WDW News Today said, and a talking version of Dug, the dog from Up, currently greets guests at Disney’s Animal Kingdom using pre-recorded phrases.
In the patent, Disney says guests would typically see interactions built on pre-recorded lines as “lacking spontaneity and immersiveness”, according to BlogMickey.
The filing does not name any specific character, park or attraction, BlogMickey reported. Its definition of a character covers cartoon characters, video game avatars, fictional people from film and television, and historical figures.
WDW News Today said Disney had previously filed two patents for AI characters that could form emotional bonds with guests using their memories.
BlogMickey noted that Disney named Karandeep Anand, the former chief executive of Character.AI, as its first chief technology officer in September, although the patent was filed well before that appointment.
Disney’s wider AI plans have not all gone ahead. A $1bn agreement with OpenAI that would have let people use Disney characters in AI-generated videos collapsed when OpenAI shut down its Sora video tool, and Disney has since signed a deal letting TikTok creators use its film and TV clips.
The patent arrives as companies across sectors assess how far AI will change frontline roles. A McKinsey report last year estimated that AI and robots could automate 57 per cent of US work hours, while finding that roles requiring physical presence and empathy are among the hardest to automate.
