Disney's Autonomous Star Wars Anzellans Pick Their Own Lines With AWS
Disney and AWS built three Star Wars Audio-Animatronics — the Anzellans — that pick their own pre-recorded lines and movements in real time, with no puppeteer, debuting at AWS re:Invent 2026.
Why it matters
- Disney Experiences announced the three-character Anzellans prototype on November 18, 2026, built by Walt Disney Imagineering R&D and AWS.
- The characters — Vilo Vele, Kado Bre'a and Jibu Bre'a — select pre-recorded dialogue and animations in real time rather than generating speech live.
- They appear at the Customer Showcase at Caesars Forum during AWS re:Invent in Las Vegas, November 30–December 4, 2026.
- Unlike Disney's BDX droids, which Disney said on April 3, 2025 relied on a behind-the-scenes puppeteer, the Anzellans respond autonomously.
- AWS already provides public-cloud infrastructure for Disney+; the Anzellans extend the partnership to real-time physical character intelligence.
Disney has built three Star Wars Audio-Animatronics characters that choose their own dialogue and movements in real time, with no puppeteer and no fixed performance loop. Disney Experiences announced the prototype on November 18, 2026, describing a joint effort between Walt Disney Imagineering R&D and AWS featuring a trio of alien characters called the Anzellans: Vilo Vele, Kado Bre'a and Jibu Bre'a.
The characters are scheduled to appear at the Customer Showcase at Caesars Forum during AWS re:Invent in Las Vegas, which runs November 30–December 4, 2026. That venue matters: re:Invent is AWS's flagship cloud conference, and the demo doubles as a public proof point for the idea of "physical AI" — software that perceives the world and acts in it, not just on a screen.
What do the Anzellans actually do?
According to Disney, the characters perceive guest activity in real time. They respond to actions such as waving or speaking, and they react to one another as well as to visitors. The design goal is an encounter that feels alive rather than a show on rails.
The mechanism is a two-tier decision system. Fast-decision models handle quick reactions, while larger models assess the broader situation. Disney says each component is optimized for latency so the characters can keep up with the pace of a real conversation with a guest.
Samira Panah Bakhtiar, AWS general manager for Telco, Media & Entertainment, Games, Advertising and Sports, described the prototype as physical AI, with the characters perceiving their environment, reasoning about events and responding dynamically. In plain terms: the robots sense, think and answer.
Do the characters generate their own words?
No. This is the detail that separates the Anzellans from a chatbot in a costume. The system intelligently selects pre-recorded audio and plays it with dynamic animations, so the characters choose among recorded lines rather than generating dialogue live.
That split defines the division of labor between the two companies. Imagineers wrote and designed the dialogue and personality quirks, while AWS supplies the real-time decisions that bring those choices into an encounter. Disney controls the words; the software controls which words get used, and when.
The approach is a deliberate hedge against the risks of live language generation. Everything the Anzellans can say was authored in advance, which keeps the characters on-brand and avoids the possibility of an improvising robot saying something Disney never approved.
How does this extend the AWS–Disney relationship?
AWS already serves as a public-cloud infrastructure provider for Disney+. The Anzellans extend that relationship beyond streaming infrastructure, using AWS's cloud and AI capabilities for the real-time intelligence of physical characters that guests can meet.
For AWS, that's a meaningful expansion. Moving from hosting video delivery to powering embodied characters in a theme park pushes the company's AI stack into territory with strict latency demands and unpredictable, crowd-filled environments — a far cry from a data center serving static content. The re:Invent demo gives AWS a live showcase of that capability in front of its core enterprise audience.
How is this different from Disney's BDX droids?
Disney has explored autonomous characters before, and the company's own track record frames the Anzellans. In Walt Disney Imagineering's autonomous-characters overview, Disney says its Vyloo figures used algorithms and simple sensors to determine behaviors such as eye contact, head movement and sleep. Animation blending, computer vision and AI supported interactions in the queue for Guardians of the Galaxy – Mission: Breakout! at Disney California Adventure.
The company's Automatronics project went further, exploring free-roaming characters that could detect their surroundings and decide where to go and how to interact. A prototype named Jake was tested in Tomorrowland at Disneyland Park.
The closest comparison, though, is the BDX droids Disney described on April 3, 2025. Those free-roaming robots learned movement and balance through reinforcement learning in simulation and belonged to a modular Robotic Character Platform built around expressive motion and character-driven design. But they weren't fully autonomous performers. Kyle Laughlin, senior vice president of Walt Disney Imagineering Research & Development, said in that account that a behind-the-scenes puppeteer brought out nuances in the droids.
The Anzellans cut that human out. They select responses autonomously, with no one puppeteering or controlling them. That's the technical line Disney has now crossed: from robots that move on their own but perform under human guidance, to characters that decide their own responses end to end.
What is the real technical shift here?
The Anzellans change the engineering problem itself. The task moves from choreographing a repeatable show to selecting an authored response under changing guest conditions. Instead of a fixed loop a guest can watch twice and predict, the system faces a fresh situation every time someone waves, speaks or simply stands there.
That's harder than it sounds. A repeatable show is deterministic — the same inputs, the same output. An autonomous character has to weigh what a guest is doing, what its fellow characters just said, and which pre-recorded response fits, all fast enough to feel like a conversation rather than a pause. The latency optimization Disney cites is what makes or breaks the illusion.
The stakes reach beyond Star Wars. If the model works, it points toward theme park characters — and potentially retail, hospitality and museum installations — that can hold unscripted-feeling encounters without the staffing cost of a hidden operator per figure. It also gives Disney a way to scale personality: the character library is fixed and controlled, but the performance never repeats.
What should watchers look for at re:Invent?
The key question for the re:Invent demonstrations is whether that autonomy can make encounters feel spontaneous while keeping the characters' words, gestures and personalities within the boundaries Imagineers created. Guests in Las Vegas from November 30 to December 4 will effectively be stress-testing that balance in person.
The Anzellans won't be generating sentences — but if choosing among authored lines in real time feels genuinely alive, Disney will have shown that autonomy and control can coexist in a walking, talking character.
Source: unite.ai
More from Sophie Lindqvist
Show full bio
Staff writer covering regulation, policy and their impact on artificial intelligence and machine learning at AI In Context.
749 articles
Related articles
- Disney and Adobe build brand-specific AI for parks and ESPN
- Dynamic Creatures and Hasbro Team Up on Character Robots
- Adobe becomes Disney's Official AI Innovation Partner in multi-year alliance
- Disney and OpenAI Strike Deal Bringing 200+ Characters to Sora
- Adobe turns Firefly into an agent-driven AI creative studio