The archive · AI & Models · Product decision · 2026
SigmaZ bets real-time interactive video replaces frozen pixels
SigmaZ AI generates video as live, editable visual code; one X demo drew ~80k plays in six hours and led to a Lanchi-led round.
SigmaZ AI
What the business is
SigmaZ AI is a startup building real-time interactive video from diffusion language models. It trains models to emit executable visual code that renders into live scenes, and it is bringing that capability to consumers through Tap8, an AI-native interactive video platform where viewers can inspect, ask about and remix anything on screen.
Starting capital：Several-million-dollar round led by Lanchi Ventures (蓝驰创投) with Yunqi Capital, Decent Capital and XiaoXiao Fund participating, announced in August 2026; the exact amount was not disclosed.
How it started
SigmaZ AI was co-founded by William Yang, a 2003-born Cambridge serial entrepreneur, and Derek Law, an Imperial College computer scientist who led visual-code training for Amazon's flagship model at the AGI Lab. From late 2025 the team started collecting preference-labeled data through consumer and B2B products, betting that interactive content would become a major new medium.
What happened
On 2026-07-22 the company published an X demo of its real-time interactive video engine, drawing roughly 80,000 plays in six hours and thousands of waitlist signups. In August 2026 it announced a round led by Lanchi Ventures with Yunqi, Decent Capital and XiaoXiao Fund. InfoQ reported a July internal benchmark in which SigmaZ's engine ranked first on quality, factual coverage and text accuracy against Fable 5, Seedance and Veo, and that its first product Tap8 was planned for a 2026 launch.
No ending yet — it is still running.
Background
SigmaZ AI is an interactive-video startup built around diffusion language models. Where conventional video models output a fixed pixel stream, SigmaZ trains models to generate visual code that renders into scenes whose elements remain live: viewers can touch, question, remix and alter what they see while it plays.
The company was founded by William Yang, a 2003-born serial entrepreneur with a Cambridge background, and Derek Law, who previously led visual-code training for Amazon's flagship model at the AGI Lab. The pair chose a code-pixel hybrid path: code carries structure, causality and interaction, while pixel diffusion handles texture and open-ended visual detail.
On 2026-07-22 SigmaZ published a demo on X that drew nearly 80,000 plays within six hours, topped X's Today's News that day, and brought thousands of waitlist signups plus more than 20 B2B intent partnerships within 24 hours. In August 2026 the company announced a several-million-dollar round led by Lanchi Ventures, with Yunqi Capital, Decent Capital and XiaoXiao Fund participating.
SigmaZ's first product, Tap8, is an AI-native interactive video platform positioned as a daily medium rather than a developer tool: users can click any element for an instant explanation, ask questions answered in video, or remix a clip into a personalized version. InfoQ reported the platform was planned for a 2026 launch, with a July internal benchmark placing SigmaZ's engine ahead of Fable 5, Seedance and Veo on quality, factual coverage and text accuracy.
What has to be true
- Pixel video has a structural ceiling: once rendered it is frozen, so changing one detail means regenerating everything, which makes interactivity and personalization impossible at scale.
- Diffusion language models decode in parallel rather than token by token, giving real-time editing speed, while code keeps a scene's structure addressable after generation.
- The self-improvement loop was part of the pitch: generated code is rendered and reviewed, then correction trajectories are distilled back into the model, compounding quality week over week.
- The X demo converted attention into pipeline: thousands of waitlist users plus 20+ B2B intents gave SigmaZ the data flywheel and revenue conversations a pre-launch lab would otherwise lack.
What can be applied
A demo can open a market before the product exists: one clip filled a waitlist and a cap table, but the bet lands only if real-time generation stays fast and cheap for daily use.
Aftermath
As of 2026-09-06 SigmaZ AI remains pre-launch: Tap8 is in waitlist mode with no public release date or revenue disclosures beyond the announced round and B2B intents. The round led by Lanchi Ventures and coverage by InfoQ, ijiwei and ai-bot.cn remain the reference points, and the open question is whether real-time interactive video can clear the engineering hurdle of latency and cost at consumer scale.
Sources
- 扩散语言模型驱动的实时交互视频模型：SigmaZ 完成数百万美元融资 (SigmaZ raises millions in funding for diffusion-LM real-time interactive video)
- 扩散语言模型驱动的实时交互视频生成公司SigmaZ AI，完成数百万美元融资｜蓝驰家族 (Lanchi family: SigmaZ AI raises millions)
- Tap8 – AI 原生互动视频平台，支持实时双向交互 (Tap8: AI-native interactive video platform)
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card