How a voice becomes a space
From speech recognition to multi-surface LED output, every stage runs on our own pipeline.
What happens in 30 seconds
Scroll to move through each stage.
Concept stillListen
Speech-to-Text (STT) turns what the visitor says into text.
Understand
A lightweight language model converts the request into a scene, while a safety filter screens out inappropriate input.
Draw
Generative AI creates a scene image that fits the venue's theme.
Animate
The still image becomes living, moving video.
Accelerate
Real-time inference optimization keeps waiting time short.
Surround
The scene plays seamlessly across multiple LED surfaces at once.
Core technology
Concept stillGeneration pipeline orchestration server
Ties speech recognition, the language model, image and video generation and output into a single managed flow.
Copyright C-2026-047235
Concept stillReal-time rendering client engine
Plays generated scenes seamlessly and simultaneously across multiple LED surfaces.
Copyright C-2026-047234
Concept stillAutomatic character animation
Builds a skeleton for photos or 3D models automatically and generates motions such as walking and running in real time.
Copyright C-2026-047236 · Patent pending 10-2026-0188398
Concept stillPersona engine
Learns from a figure's records to give evidence-based answers, synchronized with voice and facial expressions.
Persona AI
Before and after
Drag the handle to compare.
Safety & operations
On-premise GPU edge server
Generation happens on site with no external cloud. Visitors' voices and photos never leave the venue.
Content safety filter
An allow-list-based filter and output review keep operation within public-space standards.
Remote, unattended operation
Remote monitoring lets us check status and swap content without on-site staff.
R&D roadmap
Commercial
- Live MetaCube
- Lumi Town
- Persona AI
Advancing
- Multi-person motion interaction
- More seasonal themes
- Faster generation
Research
- Photoreal space reconstruction with 3DGS
- Multilingual voice dialogue
3DGS: 3D Gaussian Splatting — reconstructing 3D spaces from real photographs
