Synthesia’s AI Avatars Redefine PR and Journalism, Offering a Glimpse into the Future of Digital Communication.

When Alexandru Voica, head of corporate affairs at the pioneering video-generation startup Synthesia, extended an invitation to explore the newest addition to their public relations team this summer, the initial reaction was one of profound surprise. The "addition" was not a human hire but an interactive virtual avatar of Voica himself, meticulously trained to address common press inquiries about Synthesia’s operations and technological functionalities. This revelation came mere days after participating in a panel discussion where the ethics of AI-generated text in PR pitches were debated among professionals. However, Voica’s sophisticated avatar transcended mere text generation; it represented a significant leap, signaling what felt like the "final boss" in the evolving landscape of AI applications within public relations.

In September, Synthesia formally opened its new office space in New York, marking a strategic expansion from its original base in the U.K. The company stands as a prominent digital avatar startup, operating alongside other notable players in the field such as D-ID, HeyGen, and Colossyan. Synthesia’s rapid ascent in the tech industry is underscored by its impressive financial milestones: it achieved a staggering $4 billion valuation earlier this year and, in the preceding year, reported crossing $100 million in Annual Recurring Revenue (ARR). These figures highlight Synthesia’s significant impact and growth within the burgeoning AI-driven content creation market. The company’s success is rooted in its ability to leverage advanced AI to simplify and scale video production, making high-quality, personalized content accessible to a wide range of enterprises.

Synthesia’s core offering empowers enterprises to develop highly interactive training videos featuring AI avatars. These avatars serve as digital instructors, facilitators, and even practice partners, streamlining the creation of educational and corporate communication content. A recent product innovation, "Roleplay Sessions," further exemplifies this capability. This platform allows employees to engage in simulated scenarios, such as practicing sales pitches with an interactive AI avatar that not only responds dynamically to their input but also provides performance scores and constructive feedback. This moves beyond passive video consumption to active, personalized learning experiences, designed to enhance skill development and retention in a scalable manner. The platform offers a realistic and safe environment for employees to hone their communication and sales techniques without the pressure of human interaction.

During the new office opening event, the opportunity to create a personal AI avatar was presented. The decision was instantaneous and enthusiastic; the prospect of owning a digital twin, especially on a day when one’s outfit was particularly appealing and hair impeccably styled, was too compelling to decline. This personal experience offered a unique vantage point into the technology’s capabilities and its potential implications.

Prior to this direct experience, the concept of digital avatars elicited a general indifference, coupled with an underlying sense of their inevitable integration into everyday online life. Observations of individuals on platforms like Instagram leveraging their likenesses to generate social content had already hinted at this trend. The overall phenomenon was undeniably intriguing, a sentiment that now underpins the willingness to present one’s own digital twin. This instance marked a significant milestone, as it was the first time Synthesia had ever created a digital avatar for a journalist, or indeed for anyone outside of Alexandru Voica himself. The journalist’s interactive avatar was specifically trained on a detailed story concerning why venture-backed startups exhibit a higher propensity for fraud compared to their non-VC-backed counterparts. Consequently, the avatar’s responses are strictly confined to questions pertaining to the content of this specific article, demonstrating a deterministic conversational scope. This limitation highlights the current constraints and focused applications of such AI agents.

To facilitate the creation of this digital doppelgänger, a brief visit to a specialized mini film studio, discreetly located within Synthesia’s office, was required. Inside, numerous photographs were taken, capturing various angles and expressions, followed by a two-minute audio recording of the journalist’s voice. These inputs are crucial for the AI models to learn and replicate distinct visual and auditory characteristics, ensuring a high degree of likeness. Crucially, explicit consent was provided for the creation and deployment of these avatars, underscoring the ethical considerations involved in digital identity. From this meticulous process, "digital Dom" was brought into existence. Synthesia developed multiple versions: a personal avatar capable of flawlessly reading any given script, available both with and without glasses; and two interactive avatars, also with and without glasses, designed to engage in two-way conversations and respond to spoken queries. This versatility allows for different use cases and stylistic preferences.

The process for creating the interactive avatar involved selecting a specific article as its knowledge base – in this case, the aforementioned venture fraud story. Following this, one of Synthesia’s specialized teams undertook the intricate task of constructing the interactive avatar. Its operational backbone is a sophisticated fusion of several advanced AI models: a voice-to-text model, a sophisticated agentic language model, a text-to-voice model, and Synthesia’s proprietary video model. The avatar’s underlying technology stack predominantly utilizes Synthesia’s own robust video and voice models, developed through extensive research and engineering. However, the company also offers clients the flexibility to integrate alternative models from other leading AI research labs, including Cartesia, ElevenLabs, Google, or OpenAI, catering to diverse preferences and technical requirements for voice and language processing. Furthermore, enterprises retain control over data hosting, with options to host their avatars on their preferred cloud infrastructure or opt for Synthesia’s secure hosting services, providing flexibility and control over sensitive data.

Delving deeper into the technical architecture, the voice-to-text model serves as the initial processing layer, accurately transcribing spoken words from a user into textual data. This text is then fed into the agentic language model, which interprets its meaning, understands context, and formulates appropriate responses or takes specific actions based on the query, effectively acting as the avatar’s "brain." Subsequently, the text-to-voice model converts the generated textual response back into natural-sounding audio, imbued with human-like intonation and cadence, often mimicking the original speaker’s voice characteristics. Finally, Synthesia’s advanced video model choreographs the visual aspect, animating the avatar’s facial expressions and body language in synchronicity with the synthesized audio, creating a seamless and lifelike conversational experience that enhances user engagement. This intricate interplay of models is what gives Synthesia’s avatars their compelling realism.

Synthesia’s product portfolio is broadly categorized into three distinct offerings, each catering to different enterprise needs. Firstly, there is a comprehensive video-creation and distribution platform featuring classic avatars. Here, users simply input a script, and the chosen avatar delivers the content with professional polish, ideal for standard video production needs such as corporate communications, marketing videos, and e-learning modules. Secondly, the company offers an agentic platform known as "Sessions," which enables dynamic interaction with avatars through various formats, including surveys and immersive roleplay scenarios. This platform is particularly effective for training, simulations, and interactive educational content, allowing for personalized and engaging learning journeys. Thirdly, Synthesia provides an API platform, offering developers and enterprises the modular components of its video and voice models. This allows for their integration with other technological services, facilitating the creation of bespoke interactive avatars or entirely new types of AI-powered products tailored to specific business requirements, fostering innovation and customization.

The development of the journalist’s full suite of avatars by the Synthesia team spanned a couple of days. The initial interaction involved the personal avatars. A relatively generic script was entered to assess the fidelity and naturalness of the AI-generated voice. The script described the arrival of fall in New York, a favorite time of year. The resulting voice was remarkably accurate, a relief given that it successfully avoided reproducing any hoarseness present during the original audio sample recording. The precision in voice replication was a testament to the sophisticated audio processing capabilities of Synthesia’s models.

Upon showcasing the personal avatar to non-tech-savvy friends, their reactions were a mixture of fascination and unease, describing the experience as both "interesting and creepy." The subsequent demonstration involved the interactive avatar, which, by design, is deterministic. This means its responses are strictly limited to the information it was specifically trained on—in this case, the venture fraud story. When posed with questions like the journalist’s previous employment or current residential area in New York, the avatar consistently and politely redirected the conversation back to the parameters of the venture fraud narrative, demonstrating its programmed limitations.

Friends observed that the interactive avatar’s voice did not sound as precisely like the journalist’s as the personal avatar, and its visual likeness was also perceived as slightly less accurate. Nevertheless, the resemblance was sufficiently close to maintain an unsettling, albeit compelling, "creepy" factor, highlighting the subtle nuances that distinguish near-perfect replication from absolute fidelity. The journalist’s mother, however, offered a high compliment, labeling it "amazing." Both parents engaged with the model, attempting to ask questions "only they would know" about their child, testing its boundaries. Predictably, the model adhered to its training, repeatedly guiding them back to the venture fraud story. After these attempts, the mother lightheartedly quipped, "I don’t remember giving birth to two of you," highlighting the uncanny realism while acknowledging its artificiality.

This profound personal experience with advanced AI avatars inevitably sparked contemplation about the potential future trajectory of journalism. Could a future emerge where news broadcasts are routinely presented by AI avatars? One investor, when posed with this question, immediately offered a definitive "no," reflecting a strong preference for human news anchors. Indeed, current trends indicate significant public pushback against undisclosed AI-generated content, often termed "AI slop," which has permeated social media and various news-sharing platforms, raising concerns about authenticity and misinformation. However, others surveyed were less certain, prompting deeper questions: Could AI avatars serve to augment, or even potentially replace, human journalists in certain capacities, perhaps in routine reporting or data visualization? Would corporate leaders, such as CEOs, prefer to interact with an AI avatar of a journalist rather than a human one, perhaps for efficiency, perceived neutrality, or to control messaging more precisely?

The core appeal of a journalistic career lies in human connection, the craft of storytelling, and the intellectual pursuit of researching new topics. Yet, arguably the most fundamental pillar of journalism is trust. This intrinsic element of trust, deeply rooted in human empathy, critical judgment, and accountability, seems inherently unsuited for outsourcing to artificial intelligence. While AI can process data and generate text, the nuanced understanding of human affairs, ethical discernment, and the building of rapport essential for impactful journalism remain distinctly human domains.

Beyond the specific domain of journalism, the broader concept of digital self-cloning holds considerable appeal for various professional sectors. Imagine a scenario where the need to catch up on accumulated work after a holiday or vacation becomes obsolete, as a digital version of oneself could perpetually remain "on duty," diligently answering questions and handling routine tasks. This vision suggests a profound shift in work-life balance and productivity, potentially freeing human employees for more creative or strategic endeavors. Such applications could revolutionize customer service, technical support, and internal communications.

The evolution of avatar usage within corporate America remains an unfolding narrative. For the journalist who now possesses her own digital twins, the feelings are decidedly mixed. After the initial novelty subsided, a closer examination of the avatar after it ceased speaking revealed a subtle, almost unconscious, waiting period. Waiting for what, precisely, remained unclear. Perhaps a blink, a spontaneous new utterance, a smile, or simply a sign that it "knew" something beyond its programmed parameters – a hint of consciousness or independent thought.

Given that these digital twins are deterministic, such spontaneous actions are beyond their current capabilities. Their responses are strictly confined to their training data and programmed logic. However, this experience illuminated the ease with which one could potentially succumb to a "tinge of AI psychosis" if interacting with a nondeterministic avatar, one powered by a chatbot capable of unconstrained pontification and improvisation. The blurring lines between artificial and genuine intelligence could pose significant psychological challenges.

In discussions with an investor, the prediction was made that the journalist’s generation, Gen Z, might struggle to fully acclimate to digital twins. There is an undeniable "sci-fi" quality to these technologies, bringing elements previously confined to cinematic narratives into tangible reality. This rapid technological advancement can evoke a sense of uncanny valley or existential discomfort for those accustomed to a clear distinction between human and machine. Yet, it is noteworthy that digital avatars are perceived as significantly less jarring than humanoids. With an avatar, the ultimate safeguard remains the ability to simply log off if the interaction becomes unsettling or "weird," providing a degree of control and detachment not afforded by a physical robot.

Until that future fully materializes, the personal avatar stands ready, offering a concise rundown of the top stories featured on the site this week, a testament to the technology’s present capabilities and immediate applications in content delivery and information dissemination. The journey into the world of AI avatars is just beginning, promising both remarkable utility and profound questions about our digital future.

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.