Have you ever watched an AI-generated video and felt something was… off? Characters are somehow always talking off-camera as the narrator bumbles on in the background. As I found out last week working with RunwayML, Hailuo and Kling AI, lip syncing for AI video is still a hard problem in AI. I had't realised it up until that point, but AI video was functionally mute.
This is the best I could muster with Kling AI's lip sync model just last week:
At least, AI video was functionally mute until February 3rd, 2025, when a team of AI researchers led by Jianwen Jiang at ByteDance released details and samples from their new OmniHuman-1 multimodal VGM model.
The company behind TikTok may have just cracked one of the hardest nuts in AI video generation: creating realistic, synchronised speech that doesn’t make your brain scream “fake!”.
Here's a sample of OmniHuman-1 synthesised Einstein footage:
What makes OmniHuman-1 special isn’t just its lip-syncing prowess. The system generates holistic human motion - facial expressions, head movements, and even body gestures - all driven by audio input. It’s like having a professional actor respond naturally to every word, except this actor lives in the cloud and can perform in infinite variations.
ByteDance achieved this by leveraging their massive advantage: data. With TikTok’s vast library of human expressions and movements, they’ve trained a system that understands the subtle dance between sound and motion in a way previous models could only dream of.
What This Means for the Future
We’re standing at the edge of a new era in digital content creation.
The ability to generate truly convincing human performances on demand could revolutionize:
Marketing campaigns (imagine personalized video messages that actually feel personal)
E-learning (teachers who speak every student’s language perfectly)
Customer service (digital avatars that communicate with natural warmth)
Entertainment (custom content that adapts to viewer preferences in real-time)
The Bottom Line
While other AI video tools have made impressive strides in generating scenes and effects, OmniHuman-1 tackles the final frontier: making our digital clones feel genuinely human. For businesses looking to stay ahead in digital communication, this isn’t just another AI update - it’s the beginning of a new chapter in how we connect with audiences through video.
The days of awkward, mute AI videos may finally be behind us. And for business leaders looking to leverage AI in their communication strategies, that’s something worth talking about.
Want to stay ahead of the AI curve? Subscribe to The Lodestone newsletter for weekly insights on how AI is transforming business and marketing.
Beyond Chatbots: How AI Agents Are Taking Over The Workforce
The future of work isn’t just automated – it’s intelligent, proactive, and surprisingly human-like. While most businesses are still grappling with basic chatbots, innovative companies are deploying AI agents that can handle complex interactions with unprecedented autonomy and effectiveness.
One of the services people are using is Relevance AI’s AI agent workforce, a company that’s pushing the boundaries of what’s possible in AI automation.
The Rise of AI Agents: More Than Just Another Chatbot
Traditional chatbots are like one-trick ponies – they can handle simple, predefined tasks but often stumble when conversations get complex. AI agents, powered by advanced Large Language Models (LLMs), represent a fundamental shift in capability. They can:
Handle complex workflows across multiple channels
Provide truly 24/7 support without human intervention
Deliver personalized experiences based on customer data
Proactively engage before issues arise
Communicate in multiple languages
What Makes Relevance AI Different?
While many platforms offer AI solutions, Relevance AI is one of the platforms that has gone all-in on “agentic AI” – AI that can truly act on behalf of your business. Their platform offers:
No-code deployment: Build and manage AI agents without extensive technical expertise - by leveraging and clone your teams subject matter experts with AI.
True autonomy: Agents can handle end-to-end customer interactions
Customisable workflows: Adapt agents to specific business needs
Enterprise-grade scalability: Handle growing customer service demands
Here's an example of an outbound sales workflow automated with Relevance AI's Bosh BDR Sales agent:
Fully-integrated with your existing systems, these AI agents are capable of making autonomous decisions.
What’s Next?
As AI technology continues to evolve, we’re likely to see even more sophisticated applications of AI agents across different business functions. The question isn’t whether to adopt AI agents, but how to implement them effectively to stay competitive in an increasingly automated world.
Want to learn more about implementing AI agents in your business? Visit Relevance AI’s platform page for detailed information and use cases.
This week in AI
The AI lip-sync race heats up: While ByteDance’s OmniHuman-1 grabbed headlines, Google has begun gradually rolling out a new version of its own text-to-video AI model, Veo 2. They are focusing on high-quality visuals and cinematic effects. The race is on to see which company can truly perfect the end-to-end AI video pipeline.
OpenAI released Deep Research on Monday, introducing an AI-powered agent within ChatGPT that autonomously conducts multi-step internet research. The tool browses diverse sources, analyzes data, and generates expert-level reports with citations in 5–30 minutes—compressing hours of human effort into tasks like literature reviews, market analysis, or product comparisons. Designed for professionals in finance, science, and engineering, it runs on OpenAI’s o3 reasoning model and is currently available to ChatGPT Pro users. Early tests show it synthesises 20+ sources into structured insights but may struggle with nuanced fact-checking.
