Cygames establishes an AI company to recruit AI Vocalists! Uma Musume team develops proprietary models to avoid copyright issues

《Uma Musume: Pretty Derby》 developer Cygames has established Cygames AI Studio to develop game animation-specific small models to avoid ChatGPT copyright disputes. They are recruiting AI singing masters, reinforcement learning engineers, and language model engineers to train using internal group data. The parent company CyberAgent is promoting AI-driven Japanese dramas like 《TOKYO Miko Ninja》, and the Japanese gaming industry is actively integrating AI.

Why Cygames Chose to Develop Small AI Models In-House

Famous for 《Uma Musume: Pretty Derby》, Japan’s major game company Cygames announced the formation of the subsidiary “Cygames AI Studio.” The company’s goal is to develop AI and provide tools to establish a cyclical system that expands creators’ creativity, rather than just tools to improve work efficiency. They will base their development on the experience accumulated by their teams, aiming to create AI technology that creators can use with confidence.

Technologically, Cygames AI Studio is developing small-scale language models (SLM) and potential diffusion models (LDM) for entertainment purposes such as games and animation. Since large language models (LLMs) like ChatGPT and Gemini absorb a vast amount of works, raising copyright and intellectual property concerns, Cygames’ decision to develop their own AI models is likely to avoid such disputes.

This strategic shift to in-house models is very logical. When companies like OpenAI are sued by multiple publishers and artists over training data sources, businesses using these general-purpose large models also face indirect legal risks. If AI-generated content in games is found to infringe copyrights, developers could face lawsuits and removal from stores. By training only on their own IP and licensed data, Cygames circumvents this issue from the source.

Besides 《Uma Musume》 mobile game and related animations, Cygames also owns IP such as 《Saga Idol is Legend》, developed in collaboration with MAPPA, and original TV animations like 《Post-Apocalyptic Hotel》. These proprietary contents provide rich and clean data sources for training dedicated AI models. Instead of relying on external large models that carry copyright risks, investing in building their own AI infrastructure may be a wiser long-term strategy.

Small models have other advantages over large models: lower computational costs, faster inference speeds, and easier optimization for specific domains. For game development, models do not need to answer all questions but only generate character dialogues, story texts, or art assets in line with the game’s style. Small, domain-specific models may perform on par or even better than general large models for these tasks.

The Talent War for AI Singing Masters and Engineers

Meanwhile, Cygames AI Studio is actively recruiting. Currently, they are hiring AI image generation experts (aka AI singing masters), reinforcement learning engineers, and language model engineers. Job descriptions indicate that AI engineers will use internal group data for training and fine-tuning models to ensure the generated content meets entertainment industry standards for quality, rights, and safety.

Cygames AI Studio’s Three Core Positions

AI Singing Master (Image Generation Expert): Proficient in Prompt Engineering, capable of guiding AI to generate art assets matching the game’s style

Reinforcement Learning Engineer: Develop AI opponents and NPC behavior models to enhance game interactivity

Language Model Engineer: Train and fine-tune proprietary SLMs to generate character dialogues and story texts

Core Requirement: Use internal group data for training to ensure clear rights and content safety

The title “AI Singing Master” has sparked discussion in Japan’s gaming industry. Prompt Engineering was originally a technical role, but Cygames uses the more fantastical term “Singing Master,” which aligns with gaming culture and enhances the position’s appeal. In game and animation production, AI Singing Masters need to deeply understand the work’s style, character settings, and visual language to guide AI via precise prompts to generate suitable content.

Especially in image generation, Cygames AI Studio seeks talent capable of improving models for low latency and compatibility across various devices, indicating plans to deeply integrate self-developed AI into game production. This could mean AI not only assists in early concept design but also directly integrates into game engines for real-time content generation during gameplay. Such real-time AI generation requires high-level optimization, with models capable of running quickly even on mobile devices.

Parent Company AI Japanese Drama and Industry Chain Reactions

Not only is Cygames actively deploying AI, but its parent company CyberAgent has already expanded generative AI applications in film and entertainment. In early January, Japanese TV stations and CyberAgent aired the AI-assisted produced TV drama 《TOKYO Miko Ninja》, though main roles were still played by real actors and models. 《TOKYO Miko Ninja》 was created by CyberAgent’s “Extreme AI Daiba Studio” in collaboration with AOI Pro, involved in the 《8th Exit》 movie, with VFX by TREE Digital Studio.

The launch of this AI Japanese drama marks a significant breakthrough for Japan’s entertainment industry in AI applications. While main characters are still live-action, backgrounds, special effects, and some supporting roles may be AI-generated. This hybrid production approach reduces costs and explores feasible paths for AI in film and TV.

Japan’s gaming industry is shifting from observation to active AI adoption. At CEDEC 2025 developers conference, many major companies showcased AI application examples. Capcom revealed that during development of 《Monster Hunter: Wilderness》, they used models like Google Gemini and Imagen 2 to generate ideas for in-game posters, stickers, and other art assets, accelerating concept design.

SEGA has established an internal AI environment, using local AI to filter inappropriate player comments in online games, improving community management efficiency. Cygames has also used AI to analyze player social media trends. Although some generative AI applications still face copyright concerns, as Masahiro Sakurai, producer of 《Kirby》, stated, game development has entered a stage where AI must be actively utilized to change workflows. Balancing efficiency improvements with maintaining high-quality experiences will be a core challenge for future game companies.