Shizuku AI押注直播AI VTuber而非聊天机器人，打造全球最受喜爱的AI伴侣

做的是什么生意

Shizuku AI (San Francisco-based, Japan-born) builds AI companions and characters, starting with 'Shizuku', an AI VTuber who chats live with viewers in Japanese and English, sings, and reacts with a Live2D avatar across YouTube, Discord and X.

启动资金：Seed round led by a16z, announced February 10, 2026; amount not disclosed in primary sources

起因

In January 2023, while finishing his PhD at UC Berkeley, Akio Kodaira launched the AI VTuber 'Shizuku' on YouTube. She could chat with viewers in Japanese and English, sing, and respond in real time with a Live2D avatar. Kodaira ran dozens of streams and built a community of thousands of followers — proof, in his telling, that an AI character could feel present. His research base included StreamDiffusion, the real-time image-generation paper he led (90+ fps, 10,000+ open-source GitHub stars, published at ICCV 2025), an internship at Meta working on real-time video generation, and a research role at Luma AI.

经过

After roughly two years during which Shizuku was dormant, Kodaira relaunched her as 'Shizuku V2.0' and announced the formation of Shizuku AI in February 2026, with a seed round led by a16z — the first time the U.S. fund has led a round for a Japanese startup. The company is based in San Francisco but is building an AI lab in Japan, combining Bay Area research with Japanese character-design sensibility. Initial focus is expanding Shizuku's multilingual voice synthesis, richer conversational AI, and presence across YouTube, Discord and X, with a community that participates in her evolution.

结果

Still early stage: as of February 2026, Shizuku AI had closed its a16z-led seed round, relaunched Shizuku V2.0, and was hiring in Japan and beyond to build an AI-companion lab.

可借鉴之处

如果AI伴侣因缺乏真实对话数据而显得扁平，不要等待数据——让产品成为公开的直播角色，这样每场直播都既是内容又是训练信号。