Podcast Atlas
Source note Episode guide Original audio Topics: Technology

No.225 对话 Yuri 尤栗打造者汗青：AI 虚拟偶像，越像真人越容易塌房？

Summary

This 三五环 episode has 刘飞 interviewing 汗青, founder of AI Talk and creator of Yuri / 尤栗 / 优瑞, about moving from celebrity-style AI dialogue videos into an original AI-native virtual idol. It connects AI-Native Virtual Idol / AI 原生虚拟偶像, AI Character Personality System / AI 角色人格系统, AI Creative Collaboration, Generative AI Music, and AI Video Production Workflow by showing that AI content value comes from personality, taste, workflow orchestration, and long-term narrative rather than from a single model or novelty window.

The durable contribution is a creator-side account of AI-native IP after the demo stage. 汗青 argues that AI can lower execution cost and open new formats, but the scarce asset is still human judgment: choosing the character, voice, look, song, story, copyright boundary, and degree of real-time interaction that will not collapse user trust.

Key Claims

  • 汗青 began paying attention to AIGC in 2022 through newsletters, Silicon Valley sources, Hacker News, and early Midjourney image generation.
  • AI Talk combined generated text, digital-human tooling, lip sync, and TTS into virtual conversations; an early Steve Jobs and Elon Musk video reportedly neared 10 million plays on WeChat Channels in one evening.
  • The early celebrity-dialogue format is treated as a training ground rather than the final business: by early 2024 the team decided to use authorized modern figures or public-domain historical figures and to shift toward original IP.
  • Yuri / 尤栗 / 优瑞 launched on 2025-06-03 in the source account, with Omni-Human in March 2025 described as the model window that made expressive AI singing performance more plausible.
  • Yuri’s first song/MV workflow used many domestic and foreign models rather than one end-to-end system, with Suno named for music generation and video-model progress from Sora, 可灵, and Seedance changing visual production.
  • The project treats character personality, content tone, visual assets, voice, music, clothing, props, and narrative continuity as compounding IP assets.
  • 汗青 says a virtual idol is not immune to “collapse”: once real-time 2C dialogue opens, an AI character can say inappropriate things, so Yuri’s consumer chat remains constrained.
  • The source argues that AI-native characters should not only imitate humans; Yuri should acknowledge being AI and express a perspective different from human role-play.
  • Both speakers frame AI anxiety as partly over-amplified by media and management KPI narratives; useful AI work still depends on direction, judgment, editing, testing, and final organization.
  • The business remains source-scoped: cash flow is described as healthy under modest ambition, but scale depends on finding clearer product, platform, hardware, or distribution leverage beyond mobile short-video formats.

Key Quotes

“吃百家饭” - Hanqing’s phrase for combining many model capabilities in Yuri’s production workflow.

“跟时间做朋友” - his argument for personality-based IP as a longer-term asset.

“不能回避自己是 AI” - the source’s boundary for AI-native character identity.

Connections

Contradictions

  • No settled contradiction found. The episode reinforces existing wiki distinctions between AI assistance, generated-media abundance, and human judgment.
  • The source title uses “Yuri / 尤栗”, while the body summary repeatedly uses “优瑞”; this ingest normalizes the project to Yuri / 尤栗 / 优瑞 and keeps the naming variance source-scoped.
  • Reported launch dates, view counts, cash-flow condition, model usage, copyright registration, and future narrative plans are creator-reported and remain source-scoped pending independent evidence.