AI video avatars – a guide for news media

Editor’s note: for more useful coverage of AI for news media, subscribe to Alberto Puliafito’s newsletter course

AI video avatars aren’t just a futuristic concept – they’re here, stepping onto the digital world and also onto the newsroom stage. These digital stand-ins can help us change the way stories are presented and also save time: they can present news, captivate audiences, and streamline production, all without a human in front of the camera. But are they more than a passing curiosity? Can they find a real place in journalism, bringing value and connection rather than slipping into the realm of technical gimmickry? And what kind of ethical thoughts do they require?

In this piece, I’ll try to bring you into the role of both general AI avatars and personalised avatars, like the one I crafted with HeyGen, to explore their potential, limitations, and practical uses for newsrooms keen to navigate this new territory while steering clear of the so-called “uncanny valley.”

Please remember that I’ll be forced to talk about products that I pay for: I had to make choices. These choices are not recommendations, I do not receive any kind of money from the companies producing the tools I’m mentioning. I use them paying for the subscriptions, to ensure my independence as an early adopter, journalist and analyst. There are alternatives, of course, but it’s almost impossible for me to mention all of them and I need to explain the process and the tools I used to be accurate.

First of all: what is the uncanny valley?

The uncanny valley is a concept used to describe the unsettling feeling people experience when they see an artificial representation (such as a robot or digital avatars) that looks almost – but not quite – human. Coined by Japanese roboticist Masahiro Mori in the 1970s, the term refers to the phenomenon where an artificial figure becomes more lifelike, gradually eliciting more empathy and acceptance, until it reaches a certain threshold. At this point, the similarity is close but imperfect, creating a feeling of discomfort or even repulsion.

In this “valley,” the near-human likeness reveals subtle imperfections that make the avatar seem eerie or unnatural, such as slightly rigid movements, odd facial expressions, or lifeless eyes.

For AI avatars in journalism, avoiding the uncanny valley is crucial.

If an avatar’s appearance or mannerisms feel unnatural, it risks distracting or alienating the audience – undermining the authenticity and trust that journalism depends on

Let’s define AI avatars

AI avatars are digital characters created using generative artificial intelligence tools to visually present information or perform video tasks that would traditionally require a human presence. These avatars can be designed to look and act like real people, complete with facial expressions, gestures, and speech, making them a powerful tool for conveying information in an engaging way. Built with machine learning, natural language processing, and computer-generated imagery, AI avatars deliver pre-written scripts or even respond to basic user inputs, simulating a live, human-like presence.

In journalism, AI avatars provide a flexible way to present news stories, create explainer videos, and share updates in a consistent, branded style. They can read scripts, answer frequently asked questions, or guide viewers through complex information, all while fitting the newsroom’s preferred look and tone. By automating video presentations, AI avatars allow journalists to produce video content more efficiently, potentially reaching broader and more diverse audiences with fewer resources.

As “virtual presenters,” AI avatars can deliver content around the clock, in multiple languages, and across various platforms, giving newsrooms a versatile tool to enhance digital engagement without replacing the unique qualities of human journalists.

Please, let me introduce myself

In May 2024, I decided to create my own AI avatar using HeyGen, following their specific guidelines to ensure high-quality, lifelike results. The process began with a five-minute, high-definition video recording of myself, paying attention to details like lighting, framing, and expression as recommended by HeyGen. To meet HeyGen’s verification requirements, I had to read a randomised phrase on camera, which serves to confirm the user’s identity and authenticity before the avatar creation process can begin. This step is essential for compliance and security, ensuring that each avatar truly represents its creator.

During the setup, I initially experimented with a wider shot that included my hands, intending to add more expressiveness. However, I soon realised that HeyGen’s avatar technology still had limitations with hand gestures – movements didn’t always match the timing or meaning of the spoken words, making them feel somewhat disconnected. To address this, I chose a close-up, head-and-shoulders framing that focused on facial expressions, where the tool was more accurate and natural.

To enhance the realism, I integrated ElevenLabs’ API, which allowed me to use a generated version of my own voice: I prefer the results with ElevenLabs, for voice cloning. This setup added an extra level of authenticity, enabling the avatar to sound as much like me as it looked. (In a later section, I’ll dive into the details of working with ElevenLabs and how this technology fits into the broader landscape of AI voice generation.)

With my avatar complete, I was ready to experiment with real-world applications. My first test was at a live event, where I projected the avatar on stage to deliver a presentation in Italian. I pretended I wasn’t there, letting the avatar do the work while I observed from off-stage. The second time, I used it to introduce a lesson in Colombia, this time in Spanish. This cross-language flexibility is one of the most intriguing aspects of AI avatars – creating a seamless experience for multilingual audiences. You can watch both videos on YouTube if you’re curious about the results.

These tools are rapidly improving, with developers continually working to enhance everything from facial expressions to more accurate gestural synchronisation. While I chose HeyGen for this test, there are alternative platforms with similar functionalities. Again, this is not an endorsement of HeyGen specifically; I mention it because it’s the tool I have personally used and evaluated.

Before diving into these tools, always take the time to review their terms of service (TOS). These documents outline aspects like usage rights, privacy standards, and distribution limits. With AI avatar tools evolving so quickly, it’s important to stay informed on both the possibilities and the ethical responsibilities, ensuring these tools serve their purpose without compromising authenticity.

The role of AI avatars in journalism: what they bring to the table

These technologies provide several potential benefits for journalism and they are interesting.

Creating traditional video content often requires a dedicated studio setup, equipment, and professional staff, which can be cost-prohibitive for smaller newsrooms. AI avatars, however, offer a cost-effective alternative to evaluate, allowing journalists to produce professional-looking video presentations without the need for physical recording spaces or costly equipment.

For newsrooms dealing with high volumes of content, AI avatars enable rapid video creation, maintaining a consistent presentation style across videos. This can streamline workflows, particularly for repetitive announcements, short updates, or explainer videos. The quick turnaround also enables journalists to respond to breaking news with swift video updates.

Newsrooms aiming to reach diverse, global audiences can use AI avatars to produce multilingual video content quickly and affordably, or different video versions of their content for different audiences (i.e.: the elders, children, young adults…). Instead of relying on traditional voiceover talent or translation services, AI avatars can be programmed to speak in different languages, opening up new possibilities for international reach and audience inclusivity.

Although these advantages are true and measurable, you always try to test them and to show if they are appreciated by the audience.

The challenges of AI video avatars: considerations and constraints

Journalism relies on trust, and viewers often form connections with journalists over time. AI avatars, however, can struggle to deliver the subtle expressions, tones, and improvisations that human journalists bring to their stories. This raises questions about whether an AI avatar can establish the same level of credibility and empathy as a human presenter, particularly for sensitive or emotional content.

Transparency about the use of AI avatars is essential to avoid misleading viewers. Audiences should know that they are watching an AI-generated presenter rather than a live journalist. Additionally, ethical questions arise about the extent to which AI avatars may inadvertently impersonate real people or dilute the human element in news reporting. Newsrooms must establish guidelines to ensure transparency and avoid any potential deception.

AI tools typically rely on cloud-based systems, raising concerns about the storage and potential misuse of content, particularly if avatars are crafted to replicate real journalists. Newsrooms should be cautious about data privacy and work only with reputable providers who can ensure content security. Moreover, what happens if a journalist changes a newsroom or does not want to be impersonated anymore by his or her avatar? 

A practical guide to using AI avatars

For newsrooms interested in experimenting with AI avatars, here’s a practical outline for getting started:

• Define the purpose and scope
  Not all content suits AI avatars. Identify segments of your reporting that can benefit from AI avatars – such as explainer videos, brief news summaries, or multilingual content – and decide where human journalists are essential.

• Experiment with script creation
  Platforms like HeyGen typically require scripts that the avatar will read. Craft scripts that match the tone and purpose of each video. Remember, AI avatars deliver content in a straightforward style, so avoid complex phrasing or jargon that might reduce clarity. Writing for an avatar is a skill to improve. You can use LLMs like ChatGPT to help you craft the script for a video, for example. 

• Customise the avatar’s appearance and tone
  AI platforms often allow for customisation of avatars to align with your newsroom’s branding and desired look. This includes selecting outfits, adjusting speaking tone, and modifying accents to enhance relatability for specific audiences.

• Implement quality control and consistent review
  While AI avatars are efficient, their scripts should be reviewed to ensure they convey the intended message accurately. Regular review helps catch any inconsistencies or errors in pronunciation, phrasing, or delivery, which can occasionally occur with AI-driven video tools. Always check the result before publishing a video.

• Be transparent and establish ethical guidelines
  Decide guidelines with your newsroom, first of all about the kind of avatar you are going to use. Inform your audience when an AI avatar is presenting news, and explain the choice to use one, where relevant. Establish internal guidelines for when AI avatars are appropriate and ensure they adhere to your newsroom’s ethical standards.

In the end, AI avatars may serve as useful tools to support journalism rather than as replacements for human journalists. They can help free journalists from repetitive tasks, allowing them to invest more time in investigative work and creative storytelling that resonates on a human level. With careful integration, AI avatars could become valuable additions to a newsroom’s digital toolkit.

Source of the cover photo: generated by ChatGPT, DALL·E

---