A British startup today unveiled new AI humans that blur the line between the virtual and the real. Synthesia calls the digital beings “Expressive Avatars.” They promise the most realistic emotional expressions on the market.
Generated by an AI model that’s trained on footage of real actors, the avatars are built for video creation. Users simply enter a text prompt and the synthetic humans read them out on a screen.
The photorealistic renders are certainly impressive. But what makes them unique is their capacity to convey human feelings.
Using a technique called “automatic sentiment prediction,” Synthesia’s AI models infer the emotions within text. This determines the avatar’s tone of voice, body language, and facial expression.
The <3 of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!
“This is definitely the first iteration of avatars that can express emotions and understand the sentiment of the content,” said Victor Riparbe, Synthesia’s CEO and co-founder.
The avatars will also always generate entirely new and unique outputs. Feed them the same script twice and they will respond with two different performances. In Synthesia’s words, they’ve evolved from “digital renders” to “digital actors.”
At a product demo on Monday, TNW got to review their acting chops.
AI avatars take the stage
To this humble critic’s eyes, Synthesia’s avatars are the best GenAI actors to ever perform on the screen.
The combination of photorealistic faces, emotional gestures, expressive voices, and synchronised movements brings a new level realism to the market.
Given an upbeat script, the avatars delivered a smile and energetic tone. When fed sadder lines, they offered a sombre inflection and slower speech.
However, their performances still haven’t escaped the uncanny valley. Their main shortcoming is a tendency to slightly exaggerate their emotions. As actors, they’re closer to D-list soap stars than Academy Award winners.
Another drawback is that their movements are confined to the head, face, and shoulders. As long as that remains the case, they will probably only pose a threat to newsreaders.
Despite these limitations, the avatars have the potential to unlock new applications.