Captions launches Mirage Avatar X, trained from 10 seconds of video
By Steven Van ·
The new avatar model aims to preserve micro-expressions and hold identity steady across long generations.
Captions introduced Mirage Avatar X, a new AI avatar model it's positioning as the new standard for the category. The pitch is identity, not just likeness: 'It looks like you. It moves like you. It sounds like you.' Avatar X claims industry-leading identity preservation, the most expressive AI avatars available, and support for both vertical and horizontal video.
Micro-expressions as the differentiator
Captions is drawing a specific line against other avatar models: not mouth-flap lip sync, but full-face and body performance. In a follow-up post, the team called out laughing, yawning, smirking, and gasping as the kind of subtle, involuntary detail that other avatar models struggle to preserve. Avatar X is built to keep those micro-expressions intact, and to generate audio-driven expression across the whole face and body rather than animating just the mouth, so a performance reads as natural and emotional instead of a still photo talking.
No drift over long generations
A common failure mode for AI avatars is drift, where identity or quality slips as a generation runs longer. Captions says Avatar X holds up across long-form, continuous generations with no quality degradation, so an AI twin stays consistent from the first second to the last rather than gradually losing likeness.
Ten seconds to create a twin
The onboarding bar is notably low: Avatar X can build an AI twin from just 10 seconds of video, capturing face, voice, and identity from a single short recording. Captions is calling it the fastest way to create an avatar, which matters for a platform whose broader pitch is speed, turning a script or prompt into a finished video without a shoot.
Where this fits for Captions
Avatar X is core to what Captions (built by Mirage) already does: AI avatars and digital twins sit alongside its script-to-video generation and AI Edit tools for finishing real footage. A stronger, faster avatar model raises the ceiling on how usable AI-generated talking-head content is for marketing, social, and product video, without a camera, a studio, or a reshoot when the script changes.
Sources: Mirage on X.