Best AI character hosts for malls in 2026
Every avatar platform converses now. The market map that actually matters: which host advertises when the screen is idle, runs on the screens you already own, and keeps the AI model and the data on-site.
An AI character host is a branded, on-screen character that advertises when a venue screen is idle, converses with guests when approached, and learns from every interaction. It is the layer that sits on the screens a mall already owns and turns them from a poster into a host.
6 AI character hosts for malls in 2026
Every platform here can converse, holograms included. The separation is the venue-media job around it: advertising on idle screens, running on the screens you already own, keeping the AI model and the data on-site. On that job the list narrows to one.
Conversation is assumed. These four columns are the job.
In 2026 almost all of these converse. The criteria that actually separate them for a mall operator:
Advertises when idle. Does the screen earn attention on a quiet concourse, or sit dark until a guest triggers it?
Owns the data it learns. Does the venue keep first-party interaction signal it controls, or does the intelligence stay inside a vendor tool?
Runs on your existing screens. Does it run on the screens a mall already operates, or does it need proprietary hardware or a bespoke kiosk?
Runs on-venue (local AI model). Does the AI model run locally on-site so guest data stays in the venue and the country, or does every interaction round-trip to a vendor cloud? For GCC operators, this is a data-residency question.
Deliberately excluded: Soul Machines
Soul Machines, long a leader in photoreal digital humans, entered receivership in February 2026 (KPMG appointed as receivers) and states it can no longer provide normal services. It is not a live option for a 2026 build, so it is excluded rather than ranked.
Why the idle-screen column carries weight
In-store retail media is on track to reach USD 1 billion by 2028, and retail media has now passed TV ad revenue for the first time (Broadsign, Retail Media: In-Store Report 2025, September 2025). A screen that only converses leaves the idle-time advertising value sitting on the wall.
How the six compare
| Platform | Live conversation | Advertises when idle | Owns data it learns | Runs on existing screens | On-venue AI model | Fast self-serve setup |
|---|---|---|---|---|---|---|
| Performit LivePremium venues, full loop | ✓ | ✓ | ✓ | ✓ | ✓ | ✕ |
| HypervsnHolographic avatar | ✓ | ✓ | ~ | ✕ | ✕ | ✕ |
| D-IDReal-time agents at scale | ✓ | ✕ | ~ | ~ | ✕ | ✓ |
| UneeQReal-time service avatars | ✓ | ✕ | ~ | ~ | ~ | ~ |
| SynthesiaScripted video + agents | ~ | ✕ | ✕ | ✕ | ✕ | ✓ |
| HeyGenAvatar video + LiveAvatar | ✓ | ✕ | ✕ | ✕ | ✕ | ✓ |
Talking is solved, holograms included. Read across the venue-media columns, because that is where five of the six stop. Hypervsn comes closest, and stops at the hardware, the cloud and the data. UneeQ is the only rival that can go on-premise, and even then it does not advertise on idle screens.
The last column is the honest trade. The self-serve tools are quicker and cheaper to switch on, and Performit Live is not one of them. It is a built deployment, which is why it fits a premium estate and why a self-serve video tool fits a quick content need. Pick the column that matches the job you are buying for.
Why on-site processing matters in the GCC
For most of the field, every guest interaction round-trips to a vendor cloud abroad. For a GCC operator, that is a procurement problem before it is a technical one.
The AI model runs on-venue, so the data stays in-country
Performit Live runs a local AI model on-site. Guest interaction data stays in the venue and in the country, which matters where data-residency rules and operator policy require it. That local architecture is also what makes the owned-data loop real: the intelligence is held by the operator, not routed through someone else servers. UneeQ can go on-premise as an enterprise option. The rest are cloud only.
Where each one stops
Clean facts, then an honest limit. Every mark in the table above is verified against the vendor live product page.
Performit Live (Lumière AI)
Runs Lumière AI, its own character engine, on-venue. One screen advertises when idle, converses when approached, and learns from every interaction to sharpen the next impression. The AI model runs on-site, so guest data stays in the venue and, for GCC operators, in-country. Signage shows. Staff talk. Neither remembers.
In production across three continents. Deployments include a flagship character host for Cloak & Wand (Alice), a real-time avatar with JC Decaux in Paris (Amira), and character hosting inside a top-tier KSA retail group across two premium mega malls. The characters converse in multiple languages.
Where it stops: it is a built deployment, not a self-serve signup, and it is not a tool for spinning up a library of scripted avatar videos. Both are a different product.
Hypervsn
The SmartV Digital Avatar adds real-time voice conversation in over 50 languages, driven by a cloud ChatGPT integration. It is the closest thing here to a full venue character, and it advertises when idle because it is a display.
Where it stops: it runs only on Hypervsn own holographic hardware, not your existing screens, and the AI model is cloud-based. Scaling means buying more holographic units, not switching on more screens. For the full hardware breakdown, see the AI character vs hologram comparison.
D-ID
D-ID V4 Expressive Visual Agents deliver real-time, LLM-connected conversation with low latency and high-fidelity lip sync, used on retail kiosks. A capable real-time avatar layer, delivered as a cloud API.
Where it stops: it is a cloud conversational agent, not an on-venue venue-media system. It does not advertise on idle screens or hand the operator an owned data loop.
UneeQ
UneeQ does live conversation well and is the one platform here besides Performit Live that offers on-premise and private-cloud deployment with geographic data residency. Its centre of gravity is the service desk, not the retail concourse.
Where it stops: it is not designed to earn attention on an idle mall screen or run retail-media advertising across a venue.
Synthesia
The strongest scripted-video generator, now adding Interactive Avatars and Video Agents for real-time training and product flows, in enterprise rollout through 2026. Its home is content and internal use cases, delivered from the cloud.
Where it stops: its interactivity is built for products and training, not for advertising and engaging guests on physical venue screens.
HeyGen
HeyGen generates avatar video quickly and now offers LiveAvatar, a real-time WebRTC stream you connect to your own LLM. Strong for building conversational avatars into digital products, delivered as a cloud service.
Where it stops: no idle-screen advertising loop, no on-venue model, no venue-estate deployment, and the guest data stays inside the build.
Match the tool to the job
A full venue-media host (Performit Live)
If your screens sit idle between guests and you want them to advertise when quiet, converse when approached, and hand you data you own, running on-venue on the screens you already have.
A holographic avatar (Hypervsn)
If you want spectacle plus conversation at a flagship spot and can install proprietary holographic hardware, and cloud processing is acceptable.
An on-premise service avatar (UneeQ)
If your need is a data-resident digital agent at a service point, and advertising and estate-wide venue media are not part of the brief.
A video-and-agent tool (Synthesia, HeyGen)
If you mainly need to produce avatar content, with interactive flows for training or product rather than the venue floor.
Questions buyers actually ask
What is an AI character host?
Can these platforms actually hold a conversation?
Does an AI character host run in the cloud or on-site?
Why does on-site processing matter for a GCC mall?
Can an AI character host run on our existing mall screens?
Does it work in more than one language?
What does an AI character host actually capture?
Sources
Market data: Broadsign, Retail Media: In-Store Report 2025 (Colin Lewis, Sept 2025). Vendor capabilities verified against live product pages, July 2026: HeyGen LiveAvatar, Synthesia Interactive Avatars, D-ID AI Agents, UneeQ deployment docs, Hypervsn SmartV Digital Avatar. Soul Machines receivership: NZ Herald / New Zealand Gazette, February 2026.Related comparisons
AI character vs Hologram
Hardware spectacle versus a three-mode venue system that advertises, converses and learns.
AI character vs Digital signage
Static screens versus a character host that earns attention when idle and answers when approached.
Lumière AI for shopping malls
The venue-intelligence approach for mall operators, on the screens you already own.
Lumière AI for museums
Curator-governed character hosts for permanent visitor programmes, in multiple languages.
See a venue host close the loop, live
Twenty minutes. We will show Lumière AI running in production across all three modes, on-venue, and walk through what a deployment looks like on the screens you already own.