A Three-Way Dialogue on AIGC Creativity, Product, and Investment: Will the Future Pixar Be Born Inside an AI Company?

At the end of last month, OpenAI announced a stunning update that sent ripples across the internet: a brand-new image generation model integrated into GPT-4o. 4o doesn't just offer stronger editing and compositing capabilities — it also weaves deep textual understanding into image generation, replacing complex creative workflows that previously required stitching together multiple AI models and tools with a single line of text instruction. There's no doubt: we are witnessing the dawn of an entirely new visual era.

Since Sora's release as a starting point, AI content generation has undergone staggering development — from initial seconds-long clips to today's higher-resolution, more powerful and controllable long-form video generation with precise editing capabilities. AI is gradually becoming the core engine of content production.

In this episode of ZhenTalk, we've invited Barkley, a product manager at Luma.ai, a leading Silicon Valley video generation startup, alongside two of China's most active top AIGC creators: Haixin and A Wen. This is a cross-disciplinary conversation spanning creative, product, and investment perspectives, structured around three themes:

First, an overall observation and technical retrospective on the current AI video generation industry. How do frontier creators understand and leverage video generation tools? How will next-generation AI creative workflows shift with 4o's release? Where will the industry's next breakthrough come from? And how far is the AI video field from its AGI moment?

Guest Introductions

04:45 Haixin: Filmmaker turned AI content creator, using AI-generated video to build games

05:46 A Wen: PPT designer making striking collage-style animations with AI

07:17 Barkley: Product manager at Luma, formerly at TikTok

After 4o Image Generation's Release

11:38 Deconstructing image layers: generating transparent PNGs, unlocking productivity

13:26 Google Gemini vs. GPT-4o image generation experience comparison

20:13 How autoregressive models reversed previous diffusion model advantages

22:40 How far apart are pixel distance and semantic distance?

30:23 4o converts images into language — visual understanding matters as much as visual generation

32:15 Foundational model research ultimately drives change at the technology, application, and product layers

33:23 Domestic video models deliver real productivity gains; expectations for Sora were actually too high

Stunning Moments in AIGC Progress Over the Past Year

34:42 From Sora to Keling AI, from Midjourney to Google Whisk to GPT-4o

36:46 When unified models become powerful enough, workflows get replaced outright

38:22 What 4o still can't do: highly customized assets won't extend, face-swapping only recognizes celebrities, etc.

42:42 Building an "agent" for video generation may be premature

AIGC Isn't Just Cost Reduction — It's a New Art Form

45:05 Every model has different strengths; subscribing to all of them gets expensive fast

48:37 Complex shot composition and audiovisual language require sufficient data and training time

54:54 Production needs unmet today will likely be solved sooner than expected

57:20 For learning AI creation, go straight to primary sources and filter out information noise

59:57 Has life become happier since AI emerged? AI isn't just about efficiency — it's a new art form

01:03:02 AI video generation applications: response speeds in 3D animation, film, and advertising industries

What Is the AGI Moment for Video?

01:13:13 Single-point tools like rotoscoping are most vulnerable to AI disruption

01:14:24 Adobe is actually used more than before: AI handles the messy work, pros stitch it together

01:16:30 The future Pixar might be born inside an AI company

01:18:22 Creation is no longer a privilege reserved for the wealthy and powerful

---

ZhenFund's 21st cohort of ZhenInterns is now recruiting. Partners mentor directly, with dual experience in entrepreneurship and investing, plus conversion opportunities. Interested candidates may apply at zhenfund.jobs.feishu.cn

Luma AI is currently hiring for AI data roles — data PM/engineer and model infra positions are open, with remote work available from China. Self-referrals and recommendations welcome via email: barkley@lumalabs.ai

The puzzle game Haixin mentioned at the beginning, aka Rust Shanghai — Xiaohongshu

Also follow Haixin and A Wen and their future creations: Weibo, Xiaohongshu, Channels, and Jike: search "海辛Hyacinth" and "Simon_"; X: search "ring_hyacinth" and "simonxxoo"

Executive Producers: Jiamin, Zoe, Wendi

Post-production: Yanaego

ZhenTalk is a general business podcast produced by ZhenFund, where the investment team shares the latest hot topics and industry insights with leaders across various fields.

Founded in 2011, ZhenFund is one of China's earliest angel investment institutions. Since its inception, ZhenFund has been actively seeking out the most outstanding entrepreneurial teams and era-defining investment opportunities in artificial intelligence, chips and semiconductors, robotics and hardware, healthcare, enterprise services, new energy, cross-border expansion, consumer lifestyle, and beyond.

ZhenFund — Your First Stop for Entrepreneurship!