AI Technology

We aim for sustainable growth through M&A and cross-group technology utilization.

Technology Team Vision

Emotion Beyond Technology

We do not consider technology itself to be the goal. The essence of entertainment lies in moving people's hearts and creating new surprises and emotions. Our Technology Team exists to maximize this value and deliver it to as many people as possible.

We are not bound by specific technologies or industry conventions. We constantly learn and explore new technologies, including AI, and continue to pursue methods that can truly create value. What matters is not using trendy technologies, but creating meaningful experiences for users and society. Furthermore, we are not satisfied with just research and experiments. No matter how excellent a technology is, it has no meaning unless it reaches people as value. We will continue to challenge ourselves to connect technology to business and shape new possibilities in entertainment. This challenge cannot be achieved alone. In the Technology Team, we share the future and goals we aim for across the entire team, and produce results by multiplying each other's knowledge and experiences.

Our strength lies in continuing to explore the possibilities of technology and entertainment, and maximizing our output as a team.

Our Technology

"Edge AI"
Solving the challenges of conventional cloud AI

Edge AI is a technology that executes AI processing in real-time on devices and on-site. Compared to cloud AI, it is less susceptible to communication delays and network failures, and achieves enhanced privacy protection, improved security, and reduced communication costs. In fields that require instantaneous judgment, edge AI plays an important role in enhancing safety and efficiency, supporting social infrastructure. Edge AI, which combines real-time performance, reliability, and safety, is increasingly valued as a foundational technology supporting the next-generation society.

Image of 'Edge AI' small-scale cloud server and device collaboration

High-accuracy local ASR
realized by audio signal processing and Transformers

This is a Japanese ASR technology that combines our audio signal processing technology and Transformer models to achieve high-accuracy speech recognition with a lightweight model size. While there are limited Japanese-compatible models that operate in edge environments, we achieved a high accuracy of 7.76% CER (Character Error Rate) in a comparison within the same environment. Independent of the cloud, it provides stable transcription even for confidential meetings and offline environments.

Traffic Control

AI traffic analysis is an effective means to solve the challenges of conventional surveys that rely on manual labor—such as quality variations, soaring labor costs, safety management burdens, and prolonged analysis times—and to realize more efficient and stable traffic volume surveys. It automatically detects the passage of vehicles and organizes the data into a format that can be used for traffic volume surveys, road planning, security planning, and inflow analysis around stores. It reduces the manual work of checking videos and leaves evidence data that can be re-verified.

Proprietary recommendation engine integrating VLM, time-series forecasting, and mathematical optimization

This is an AI recommendation system that aggregates and integrates purchase information, inventory information, product images, and trend information on the internet to propose optimal products to customers. In addition to embedding visual features of product images using Vision Language Models and multimodal similarity search, it captures shifts in demand and trends using time-series forecasting models, and derives the optimal product combinations considering inventory status and appeal conditions through mathematical optimization and search algorithms. By combining these, it rapidly selects "products that should be proposed now" from an assortment of tens of thousands of items. It has been introduced to the EC and store apps of the Shimamura Group, a major fashion retailer with over 2,000 stores, contributing to operational efficiency in product proposals and maximizing sales.

"Video/Image Generation Workflow" that enables countless variations from a single concept

A generative AI Agent rapidly generates video variations from a single material according to the medium and appeal. Simply by connecting nodes, everything from generation to super-resolution and export is completed, allowing immediate deployment in formats suited to each medium. By fully utilizing an AI Agent with image generation, editing, and reference functions, we achieve the "quantity" and "speed" required for events and advertisements while significantly reducing creation costs.

Image of the video/image generation workflow

LLM-as-a-Judge Equipped
Autofix

This is a security Autofix technology in which our company, which has conventionally engaged in automated testing, integrates generative AI (LLM) into the entire process of detection, repair, and verification. It expresses the detection results generated by the LLM as structured metadata such as CWE identifiers, file locations, severity, and repair guidelines, and provides this as explicit context to another LLM that performs the repair. Unlike conventional methods that stop at vulnerability detection, it automates everything from applying fixes to generating pull requests. (This technology was also presented at JSAI 2026)

Advertising Media
Edge Distribution System

This is a media platform capable of distributing advertisements in 420 luxury condominiums in the Tokyo metropolitan area (as of April 2023). Through AI signage, we repeatedly deliver information every day to approximately 200,000 affluent and executive individuals.

Image of video distribution system for condominium signage media

"High-speed image generation technology" that does not require large-scale computing resources

This technology eliminates the latent space conversion processing required by typical image generation AI, generating high-quality images quickly and at low cost. By completing the generation process—which conventionally required dozens of iterations—in a single step, it enables speedy generation even in lightweight environments compared to conventional image generation.

Image of high-speed image generation technology (generation speed comparison)

AI-Generated Animation
Music Video

This is an example of creating an original animated MV utilizing generative AI. We achieved a production flow that utilizes AI from planning to creative by producing music with a music generation AI and creating video by combining multiple generative AIs. This is a short version of the original song "Kotosirube" sung by our own character "Nyuko".

Realistically expanding brand appeal "3D LED Promotion"

This is a naked-eye 3D video technology that projects stereoscopic images that seem to jump out of the screen without special glasses. Through geometric processing, any images or videos are instantly converted into images and videos for 3D LEDs. It is also possible to generate the content itself. By realistically depicting the worldview of products and brands with deep visuals, we achieve memorable brand experiences at commercial facilities and event venues.

"AI Fashion Trend Analysis" based on massive data

By analyzing massive amounts of information—including over 20 million image data from social media and shopping sites—using AI, we provide fashion trend forecasts. Based on machine learning and deep learning, we automatically perform various correlation analyses and clothing categorization, and provide the analysis results of the latest trends visualized in an easy-to-understand manner along with a dedicated dashboard.

AI Technology

Emotion Beyond Technology

Our Technology

"Edge AI"
Solving the challenges of conventional cloud AI