Wenbo (Vito) Zhu

About Me

Hello! I’m Wenbo (Vito) Zhu. I’m the Agent Research Lead at HeyGen, a leading AI video platform, where I drive agent research across two product lines: Video Agent and HyperFrames.

Before HeyGen, I spent 4+ years at OpusClip as Founding AI Lead & Head of AI Research — joining a ~5-engineer team that failed and pivoted, then leading a 30+ cross-functional AI team delivering next-generation multimodal and generative video products.

Earlier, I was a Senior ML Engineer at ByteDance/TikTok (founding engineer of Gauthmath, 100M+ downloads) and a Research Scientist at Cloudwalk Technology, where I built a billion-scale face clustering engine deployed across 10+ cities (patented).

• Recognition: HeyGen was named to the Forbes AI 50 (2026) and Fast Company’s Most Innovative Companies (2026); both OpusClip and Gauthmath were recognized by Andreessen Horowitz (a16z) as Top 50 GenAI Apps.
• Leadership: I founded the Opus AI Research Team, the research team at OpusClip.
• Collaboration: I work with Prof. Xu Yang on multimodal video intelligence and robust vision-language alignment.
• Education: I hold a Master (OR) from UC Berkeley and an Undergrad (IE & Math) from Beihang University.

Flagship Products & Features

• [HeyGen] Video Agent – a creative engine that turns any idea into a ready-to-publish video, end to end
• [HeyGen] HyperFrames – an agent-native framework for creating and editing videos by vibe-coding
• [OpusClip] OpusClip – No.1 AI clipping tool in the market
• [OpusClip] ClipAnything – the first multi-modal clipping tool with prompting capabilities
• [OpusClip] Agent Opus – the first video agent tailored for social media
• [ByteDance] Gauthmath – the world’s first AI-based geometry solver with 100M+ downloads

Career

• [2026–Present] Agent Research Lead at HeyGen, leading Video Agent & HyperFrames
• [2022–2026] Served as Founding AI Lead & Head of AI Research at OpusClip
• [2020–2022] Joined ByteDance/TikTok as a Senior ML Engineer
• [2019–2020] Started as a Research Scientist at Cloudwalk Technology

Research Interests

• Multimodal Video Intelligence: understanding, reasoning, and editing for video content
• Agentic Systems: LLM-based planning, tool use, and evaluation frameworks
• Generative Media: automatic video repurposing and multimodal content creation

News

• [May 2026] 🚀 Joined HeyGen as Agent Research Lead, leading agent research for Video Agent and HyperFrames.
• [Feb. 2026] 🎉 One paper is accepted by CVPR 2026.
• [Jan. 2026] One paper is accepted by ICLR 2026.
• [Sept 2025] Two papers accepted by NeurIPS 2025.
• [May 2025] One paper accepted to ACL 2025.
• [Feb 2025] Two papers accepted to CVPR 2025, including One Highlight.
• [Jan 2025] One journal paper accepted to BIT
• [Dec 2024] Two papers accepted to AAAI 2025.
• [Nov 2024] Co-authored an article with Google AI team on Gemini Flash:
  “OpusClip achieves 30% cost savings with Gemini Flash” (Google AI Showcase).
• [Feb 2024] One journal paper accepted to IJHCI

Publications

Competition & Awards

• [Oct 2025] 🥈 2nd Place – Perception Test Challenge 2025 (Task 5: Hour-Long Video QA)
• [Aug 2025] 🥉 3rd Place – CVPR 2025 SoccerNet Challenge (Multi-View Foul Recognition)
• [Jun 2025] 🥇 1st Place – CVPR 2025 VidLLMs Challenge (Multilingual Video Reasoning)
• [Jun 2025] 🥈 2nd Place – CVPR 2025 VidLLMs Challenge (Complex Video Reasoning & Robustness)
• [Oct 2024] 🏆 Winner – ECCV 2024 Perception Challenge (Hour-Long Video QA Track)
• [Jun 2024] 🏆 Winner – CVPR 2024 LOVEU Workshop (Long-Term Video QA Track)
• [2020] 🥈 Kaggle ASHRAE Great Energy Predictor III – Silver Medal (Top 2%)

Services

Conference Reviewers

• IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2026
• European Conference on Computer Vision (ECCV) 2026
• AAAI Conference on Artificial Intelligence (AAAI) 2025

Powered by Jekyll and Minimal Light theme.