We just released QVAC TranscribePsy, a speech-to-text model small enough to run inside a phone app. It covers 25 European languages on a laptop and 8 on a phone, and on an iPhone 17 Pro Max's Ne
The Hook
“Every day, we talk to our devices. But before they can do anything with what we say, they have to write it down.”
Problem Agitation
“That's why we built TranscribePsy. So he doesn't have to ask her to repeat it. Nothing leaves the phone, so nothing has to wait.”
Product Reveal
“10 minutes of class, written down in under two seconds on Apple neural engine model loaded.”
Feature Teaser
“Your conversations are always yours.”
Feature Teaser
“TranscribePsy runs on just 254 million parameters. And this is where we're going. A robot that moves the moment you finish the sentence.”
Feature Teaser
“TranscribePsy is how the QVAC Psy models hear you. We're working toward all of them running together on your own devices and with no connection.”
Feature Teaser
“In our tests, it did 50% less mistakes than other models its size. On a benchmark of eight European languages, it beat models three times bigger, and it was the fastest on every phone we tried.”
Call to Action
“TranscribePsy. Your voice stays with you.”
Synthesizable in Remotion · key components of the cinematic-ondevice-ai-manifesto template.
Multi-column bar charts animating upward sequentially with latency callout badges popping at peaks.
Map numerical values through spring() scaling height from 0 to target with staggered frame delays.
Audio wave visualizer animating while typography appears character-by-character on simulated mobile screens.
Use Math.sin driven height arrays wrapped in flexbox combined with string slice text reveals.
Purpose. Ground voice technology in emotional, human reality.
Execution. Airplane window silhouette and private dictation.
Purpose. Demonstrate health, culinary, and legal security applications.
Execution. High-contrast vignettes showing offline device screens.
Purpose. Validate privacy claims with benchmark superiority.
Execution. Animated comparative bar charts showing WER and latency.
Purpose. Leave viewer with human connection and URL callout.
Execution. Wedding speech resolution leading to minimalist neon logo.
Objective
Launch of a local-first medical translation AI device.How the recipe adapts
Replace diverse daily vignettes with hospital and emergency clinic scenarios, featuring live medical terminology transcription cards and privacy verification badges.Scene durations (5 scenes, 28s)
Key Remotion components
Introducing Griffin, the first AI to pass the video Turing test
The most emotive and controllable TTS model

Always-on agents built to handle everything