Share
Embed
Step into the Director's Chair! The Latest Trends and Expert Techniques in AI Video Generation
Google I/O Connected Shanghai 登壇レポート:AI動画生成の最新トレンドとビジネス実装
先日開催された「Google I/O Connected Shanghai」において、日本から唯一のGoogle Developer Expert (GDE) スピーカーとして登壇する機会をいただきました。本セッションでは「Step into the Director's Chair!」と題し、生成AIを用いた動画広告の自動生成トレンドと、それを実際のビジネスに導入するための実践的なアプローチについて発表を行いました。以下にそのハイライトをレポート
Fukku
October 31, 2026
Transcript
9 slides
-
Step into the Director's Chair! The Latest Trends and Expert
Techniques in AI Video Generation Yukari Fukushima Cloud AI&Cloud Data GDE Japan -
How to Create AI Movie? AI generate only here ↓
Need All tasks before generate movies. Too much costs and develop resources. Can Generative AI solve this challenge? -
WHAT WE'LL DISCUSS TODAY 01 What is ViGenAiR? An overview
of the unofficial 02 Analyzing OSS with AI Agents 03 Providing Value to Clients marketing OSS Open Source How we decoded the codebase The combination of Human x AI as a Software) released by a Google using an AI Agent-equipped IDE. 'translation layer' accelerates Engineer that uses AI for automated implementation. video ad conversion. © 2026 Dentsu Digital Inc. -
What is ViGenAiR? github.com/google-marketing-solutions/vigenair An OSS for automatically converting long-form
video ads into short-form videos using Gemini Vertex AI. ・Frontend: Angular Apps Script) ・Backend: Cloud Functions 2nd gen + GCS Vertex AI Workflow ① Upload ② AI Analysis Submit long-form video Gemini splits video and via browser audio by semantic units ③ Variant Generation ④ Rendering Automatically combines vertical, horizontal, multiple short-form square formats Export/Download in patterns Features Details Segment Splitting Audio and video to analyze semantic units to split Variant Generation Multiple short-patterns automatically proposes Multi-format Support Vertical 916) and square 11) formats to support Asset Generation Demand Gen for specifically text and images generation © 2026 Dentsu Digital Inc. -
Solution Summary Based on manually produced banners, two AI banners
created with VigenAir were distributed on YouTube Ads. As a result, an astounding positive response was achieved with a completed view rate of 52.6 to 55.3% on CTV and non-CTV screens, meaning one in two people watched to the end. Show Live Demo here -
Actions Taken with AI-Agent-Powered IDE Challenge: Rapidly Understand an OSS
Codebase to a Usable State What was done (1) Have the AI Agent Explore the Entire Codebase (3) Have the AI Agent Support Deployment to Cloud Environments Map out process flows and identify (2) Ask Questions in Natural Language like "Why does this UI work like this?" role delegation between modules OSS specs and implementation can environment to deeply understand diverge → Identify the true state the solution's operation personally with AI to resolve client questions Quickly build a prototype -
What Was Learned Insights Gained with AI Agent Prompt Structure
to Gemini Video Processing Flow Identified two key areas for natural language Understanding the processing flow allowed for optimization: "Scene Selection" and "ABCD Framework confident explanation of capabilities and limitations to Evaluation" the client → AI Agent pinpoints critical touchpoints within a vast → Previously, lack of time to understand others' OSS often led to ill-prepared client explanations codebase Reached "Understanding" from "Operating," enabling confident dialogue with the client © 2026 Dentsu Digital Inc. -
Providing Value to Clients Prompt Support Achievements Explained that prompts,
'Scene Reduction of Support Period Selection' and 'ABCD Framework Completed the assistance period utilize ViGenAiR for actual ad Evaluation', can be input, and byte for mastering ViGenAiR in about delivery, despite the support period. advised on wich what to write and two weeks. Successfully enabled the cliennt to how to write it for each. © 2026 Dentsu Digital Inc. -
Key Message for Today With the evolution of AI Coding
Agents, "Reuse" of coding assets such as OSS has become dramatically However, a piece is still missing. easier. Code can be read → Translated into business context The tool runs → Explained in the client's language OSS Code Assets) → 【Translator: Human × AI】 → Client Introduction. If there is a person × AI that can perform the "translation," the speed of introduction will accelerate even further. Summary ViGenAiR is a powerful OSS that uses Gemini to automatically generate short-form video ad variants. By analyzing code with AI Human x AI as a codebase "Translator" is Agents, "Operational Understanding" of the the key to accelerating codebase can now be client implementation. achieved in a shorter period than before. 谢谢!