Inside Synthesia's AI Avatars: From PR Tools to Digital Twins

Overview

Synthesia, the UK-based video generation startup valued at $4 billion, has expanded its capabilities beyond simple avatar creation into the realm of interactive digital twins. By combining advanced voice synthesis with agentic language models, the company is enabling enterprises and individuals to create personalized AI representatives that can engage in dynamic, two-way conversations.

This evolution marks a significant shift from static PR tools to complex digital interfaces capable of understanding context, answering questions, and maintaining brand consistency. The technology, which recently allowed a TechCrunch journalist to create an interactive avatar of themselves, highlights the growing sophistication of generative AI in professional and creative workflows.

The Evolution from Static Avatars to Agentic Twins

For several years, AI avatar platforms like Synthesia, D-ID, and HeyGen have been primarily used for creating passive video content. Users would input a script, and the avatar would recite it with realistic lip movements. However, the latest developments represent a qualitative leap toward agentic platforms that can listen, process, and respond in real-time.

"This seemed like the final boss of using AI in PR."

When Alexandru Voica, Synthesia’s head of corporate affairs, deployed an interactive avatar of himself, the intention was to handle common press inquiries autonomously. Unlike previous iterations, this avatar could understand natural language queries and generate relevant responses without human intervention.

The technology stack behind these interactive avatars is complex. It involves a pipeline where voice-to-text models transcribe user speech, language models analyze the intent and formulate answers, and text-to-voice engines synthesize audio. Finally, Synthesia’s proprietary video models animate the avatar’s facial expressions and movements to match the generated speech.

Technical Architecture and Customization

Building a digital twin requires more than just a likeness; it demands a specialized infrastructure that balances creativity with control. Synthesia offers three distinct product types to cater to different enterprise needs.

Video Creation Platform: A traditional tool where users type scripts for avatars to read, suitable for training videos and marketing content.
Sessions (Agentic Platform): An interactive environment where employees can practice sales pitches or customer service scenarios with AI avatars that provide feedback and scoring.
API Platform: A developer-focused solution that allows companies to integrate Synthesia’s video and voice models with other technologies to build custom interactive experiences.

Customization options are extensive. Enterprises can choose between Synthesia’s native models or integrate third-party providers such as ElevenLabs, Google, OpenAI, or Cartesia. Furthermore, companies retain full control over hosting, choosing to keep data on their own cloud infrastructure or utilizing Synthesia’s hosted services.

In a recent demonstration, a journalist created both personal and interactive versions of their digital twin. The personal version was deterministic, reading a provided script about autumn in New York. The interactive version, however, was trained exclusively on a specific article regarding venture capital fraud, ensuring it remained within its knowledge boundary.

Practical Applications and User Experience

The practical implications of this technology are already visible in corporate settings. Companies are using Roleplay Sessions to help employees prepare for difficult conversations, such as negotiating sales deals or handling customer complaints. The AI avatar acts as a simulated counterpart, offering immediate feedback on tone, clarity, and response effectiveness.

"My mom called it 'amazing,' which is pretty high praise from her."

During the creation process, subjects undergo a brief recording session involving facial photography and voice sampling. In the test case, the resulting interactive avatar successfully redirected off-topic questions back to its training material. When asked personal questions unrelated to the story, the avatar maintained its persona without breaking character, demonstrating a high level of contextual adherence.

While some observers found the realism unsettling, others appreciated the utility. The technology proved robust enough to handle playful attempts to trick the model, such as questions about the creator's family history, by consistently steering the conversation back to the defined scope.

Implications for Journalism and Media

The ability to create accurate, interactive digital twins raises profound questions about the future of news consumption and production. If an AI can replicate a journalist’s voice, appearance, and explanatory style, what role will human reporters play in an automated media landscape?

Some investors remain skeptical, citing the current backlash against AI slop that has flooded social media and news aggregators. Audiences may resist consuming news from avatars, preferring the authenticity of human interaction. However, others argue that avatars could augment journalism rather than replace it, handling routine information delivery while humans focus on investigative depth and ethical judgment.

The possibility of CEOs interacting with AI avatars of journalists instead of human correspondents introduces new ethical considerations. It challenges existing norms about consent, transparency, and the nature of truth in media. As the technology matures, the industry will need to establish clear guidelines on when and how digital twins are deployed.

Future Outlook

Synthesia’s move into interactive digital twins signals a broader trend in the AI industry: the shift from generation to interaction. As models become more capable of understanding nuance and maintaining long-term context, the line between static content and dynamic agents will continue to blur.

With a valuation of $4 billion and annual recurring revenue exceeding $100 million, Synthesia is well-positioned to influence how businesses leverage human-like AI. The coming months will likely see increased adoption across sectors such as education, corporate training, and customer support, as organizations seek to scale personalized interactions without scaling headcount.

The question is no longer whether AI avatars can look and sound like humans, but whether society is ready to trust them as reliable sources of information and companions in digital spaces.