Mirage Studio: Your Hollywood actor in a browser

Captions AI’s omni-modal video model turns scripts into hyper-real actors

When artificial intelligence first tiptoed onto the video stage, the results were somehow robotic.

Avatars had an AI sheen, lip sync wobbled and went bad often, and subtle acting choices were a miracle. Now, Captions AI’s new foundation model, Mirage, looks determined to end the awkward phase.

What Makes Mirage Different?

1. Actors from scratch, literally
Feed Mirage a text prompt or reference image and it will conjure fully original humans, complete with lifelike skin textures that ditch the glossy CG look. The model builds your kindly librarian or neon-haired cyber-punk pixel-by-pixel, without the need of stock footage.

2. Micro-expressions
Real performers breathe, blink, cough, laugh, and occasionally flash that tiny eye-roll of disagreement. Mirage does too. The model was trained to reproduce fine-grained body language based on the audio, so your AI actor may shift from polite nods to sardonic eyebrow lifts.

3. Audio-driven magic
Upload a script or an audio file, and Mirage animates the entire scene, voice, visage, and setting, around it. No need for separate voice cloning or lip-sync passes; the model handles speech generation and facial dynamics in one go.

Let’s see how it works.