gen‑ai.news
← Back
Video

Nearly half of test subjects mistook Tavus' AI video avatar for a real person on a one-minute call

Nearly half of test subjects mistook Tavus' AI video avatar for a real person on a one-minute call

Tavus has introduced a new AI system called Griffin, positioned as the first model purpose-built for real-time human interaction over video. Unlike earlier AI avatars that were largely one-directional or limited in responsiveness, Griffin processes multiple social signals at once - facial expressions, tone of voice, and physical gestures - and responds within a live video call context.

The benchmark that has drawn the most attention is from a Tavus-conducted study: 48 percent of participants believed they were speaking with a real person after a one-minute call with Griffin. That figure stands in sharp contrast to what previous systems achieved, which topped out at roughly two percent. While the study comes from Tavus itself and has not yet been independently verified, the gap between old and new is wide enough to suggest a meaningful shift in the technology's capability.

The underlying challenge in building convincing real-time video avatars has always been multi-layered. Latency, visual fidelity, and the ability to read and respond to conversational cues all have to work together without perceptible delay. Earlier systems tended to fall apart on at least one of these axes, producing responses that felt scripted or visually mismatched with natural human cadence. Griffin appears to address the integration of these signals more cohesively than what came before.

The practical implications span a range of industries - customer service, telehealth, sales, and companionship applications among them. However, the same capabilities that make Griffin useful also raise straightforward concerns about disclosure and consent. If nearly half of people in a controlled study cannot identify an AI interlocutor within a minute, the question of how and when users are informed they are speaking with a machine becomes more pressing. How platforms and regulators respond to that question will likely shape where this technology can be responsibly deployed.

Enjoy this story? Get the next one in your inbox.

Twice a week: the most important stories in generative image and video AI, distilled into a 2-minute read.

Free. Unsubscribe any time. No spam, ever.

Your next read

A Quick Look at How Adobe’s New AI Assistant Could Streamline Your Workflows in Premiere
Video

A Quick Look at How Adobe’s New AI Assistant Could Streamline Your Workflows in Premiere

Adobe introduced a conversational AI assistant for Premiere Pro at its annual MAX conference, letting editors use natural language prompts to handle organizational and assembly tasks directly inside the application. The feature draws on Firefly generative credits, meaning it carries an ongoing cost tied to usage. Whether it meaningfully reduces repetitive work for real-world editing projects remains to be seen outside of controlled demos.

Everything You Need to Know About Adobe Premiere 27.0 and Its Latest Improvements
Video

Everything You Need to Know About Adobe Premiere 27.0 and Its Latest Improvements

Adobe has released Premiere Pro 27.0 following Adobe MAX 2026, bringing a dedicated Color Mode workspace, several AI-driven audio tools, and a range of workflow improvements for video editors. The update also introduces new effects, HSL masking for color work, and native support for NVIDIA RTX Spark on Windows on ARM. Here is a breakdown of what has changed and what it means for editors day to day.

Video

Best AI Video Generation Models in 2026: Elo, Price per Second, Open Weights

The AI video generation landscape has shifted considerably over the past year, with native audio, longer clip lengths, and open-weights models now firmly part of the competitive field. A November 2026 ranking from Artificial Analysis's blind human-preference Elo arena pairs quality scores with API pricing to give a clearer picture of where each model stands. The result is a leaderboard that looks notably different from what it did in late 2025.