Meta Leaves Chinese Startup to Build Visual AI Memory Brain with $8M Seed Funding}

Author Introduction: Sheng Juyi, PhD student at Peking University, researching robot manipulation skills; Wang Ziyi, Li Peiming, master's students at Peking University, focusing on video understanding; Liu Yong, professor at Zhejiang University, specializing in autonomous robots; Liu Mengyuan, assistant professor at Shenzhen Graduate School of Peking University, studying human behavior and robot learning.

Amid the global AI talent war, especially with Meta recruiting three top Google IMO gold medalists, the industry is watching closely. While Zuckerberg pushes to revive Llama, the entrepreneurs leaving Meta are already making waves.

Meta's top scientists have founded Memories.ai, which has just completed an $8 million seed round led by Susa Ventures, with participation from Samsung Next, Fusion Fund, and other notable investors.

Memories.ai has achieved a major breakthrough in large model research, targeting the 'memory loss' problem in AI systems by creating a powerful 'memory brain' for visual models.

The Most Powerful Brain

Large models are often criticized for their 'goldfish memory,' lacking the ability to remember past scenes and understand their context, similar to a limited FIFO stack that hampers deep scene and dynamic understanding, especially in video tasks.

To solve this, Memories.ai introduces a revolutionary visual memory layer—Large Visual Memory Model (LVMM)—which transforms traditional video analysis from fragmentary to continuous, enabling AI to:

• Retain context permanently: Moving from isolated frames to understanding causal event chains.
• Recognize temporal patterns accurately: Tracking faces, objects, and behaviors over time.
• Perform intelligent comparison analysis: Quickly detecting changes and anomalies.

This platform converts raw videos into searchable, context-linked databases, endowing AI with human-like continuous learning and creating the 'strongest brain' for visual understanding. This milestone significantly advances AI's video comprehension and practical applications.

LVMM not only sets new benchmarks in multiple visual understanding tasks but also offers a new perspective on complex visual data retrieval and analysis challenges.

Its outstanding performance in video classification, retrieval, and question answering demonstrates its broad application potential, especially in visual memory retrieval, handling large-scale content searches efficiently and effectively.

Huge Potential

LVMM's core advantages include:

• Unlimited time span: Capable of processing and remembering videos spanning months or years.
• Deep contextual understanding: Not just object recognition but understanding causal chains and temporal patterns.
• Efficient retrieval and analysis: Transform raw videos into searchable databases for instant access.

The team is collaborating with partners across sectors:

• Security: Significantly improves surveillance video retrieval, searching months of footage in seconds.
• Media & Entertainment: Enables instant search of specific scenes or visual elements across decades of content.
• Marketing: Analyzes millions of social videos for trends and insights.
• Consumer Electronics: Brings powerful visual memory capabilities to next-gen mobile devices, with Samsung among early partners.

Dr. Shen, co-founder and CEO of Memories.ai, emphasizes: "Human intelligence stems from rich, interconnected visual memories. Our mission is to endow AI with deep contextual awareness to build a safer, smarter world."

Easy Interaction

To demonstrate LVMM's capabilities, the team has open-sourced core technology via APIs and launched an interactive web app, allowing users to upload videos or access existing libraries for fast, precise content retrieval and analysis.

Using millisecond-level retrieval engines, the system excels in video frame-level referencing, enabling multimodal deep analysis of videos. Additionally, several demo agents showcase practical applications:

• Video Creator: An interactive video creation assistant based on the world's first visual memory model, generating high-quality multi-clip videos from simple prompts.
• Video Marketer: An AI marketing tool analyzing TikTok trends, viral openings, and top influencer strategies for social media marketing.

These demo agents are not just technical showcases but also explorations of future applications. If your company is overwhelmed by video data, or you're a researcher fascinated by visual memory and AI, or simply curious about AI's evolution with memory—visit https://memories.ai.