P.56 Manufactured Capital (2)

Manufactured Capital Business Foundation

AI utilization expanding into content creation sites TBS is focusing on the utilization of AI in the develop- ment of proprietary technology. We aim to improve operational efficiency and enhance creativity at content creation sites. We will also strengthen and carefully mon- itor governance regarding the use of AI.

AI utilization in drama production

In the second season of “VIVANT,” we utilized Google Cloud generative AI in CG production. For example, in the fire scenes, realistic fire images were generated from original still-image materials using the Nano Banana AI image generation tool, which were then fed into the Veo 3 AI video generation tool to create base videos. Unique prompts were employed through- out these processes. Finally, the textures were refined by hand. This workflow has made it possible to produce high-quality CG footage for scenes such as backgrounds in a short space of time. アタリ

AI video analysis with Uzit

We developed Uzit, an application that analyzes video using generative AI. Until now, metadata related to video was produced manually, but Uzit automatically performs text conversion of video content, extraction of caption information, and transcription of audio, enabling the creation of metadata linked to video timestamps. Through the utilization of Uzit, we expect annual reductions of tens of thousands of hours of labor. In addition, this metadata has dramatically improved the accu- racy of video searches and has made it possible to effectively utilize vast amounts of previously dormant archive footage. It serves as a foun- dation for further enhancement of content quality and the creation of new value, such as via the secondary use of past assets. Creating base footage with generative AI and refining textures by hand for the final product Work previously performed by hand Video materials Text-format metadata Provisional output Text-format metadata Final version For the actual fire scenes that were broadcast, work that used to take two to three people about one week could be completed by one person in two to three days using Veo 3. By fusing generative AI, a powerful tool, with the CG techniques we have accumulated over many years, operational efficiency has been improved, and time is freed up to concentrate on scenes that require detailed handcrafted work, which contrib- utes to the further enhancement of content quality. Metadata creation work – self-check Checked by another person After Uzit deployment Video materials Timestamped metadata Provisional output Timestamped metadata Final version Automatically created by generative AI Checked/corrected by a human VIVANT: AI-Powered Scenes and Development Secrets

AI text checker TBS LUPE*

TBS LUPE is an application we developed independently by combining a long-accumulated typo database with AI, in order to check with high accuracy for typos, omissions, and errors in expression in captions, flipcharts, and text drafts. On-site staff had previously been occupied daily with vast amounts of text-checking work, but TBS LUPE has significantly reduced the burden of checking text. Recently, use cases have expanded to include checking of typical internal documents. We began sell- ing it externally in FY2026, and utilization outside the company is also growing. * Received a 2025 Institute of Image Information and Television Engineers Technical Progress Award (Progress & Development Award)

Governance regarding the use of AI

In conjunction with the expanding utilization of generative AI within the TBS Group, we have formulated Group-wide “Gen- erative AI Guidelines” in order to minimize the risks of infor- mation leakage and rights infringement. In addition, we have newly established the AI Strategy and Governance Committee with the aims of promoting innovation through AI technology, gaining social trust, and preventing rights infringement. These moves have put in place a system for ensuring safety while benefitting from the convenience of the technology. We have also joined the Japan Deep Learning Association (JDLA) as a support member. Through our collaboration with the association, we are rapidly incorporating the latest knowledge on AI talent development and AI ethics and governance, and are promoting responsible AI utilization across the entire Group. 56