AI-powered voice dictation that's so powerful it can replace your keyboard.
The Hook
“Today, we're releasing our state-of-the-art speech-to-text model called Willow Atlas-1.”
Product Reveal
“It is the most accurate human-powered transcription engine ever built.”
Problem Agitation
“We benchmarked it against every major model on the market: Whisper, Deepgram, ElevenLabs, and Assembly AI.”
Feature Teaser
“On the Open ASR Benchmark, our engine is significantly more accurate than the next best model. On real-world audio with background noise, that gap is even wider. Most models score between 5 and 7% word error rate on clean audio.”
Feature Teaser
“In production conditions, they fall to 10% to 15%. Our models hold at 1.2% word error rate on clean audio and 2.1% in real-world conditions. We outperform by a significant margin across the board, and especially for Asian and Asian-American accents. We've innovated on foundational model capabilities using state-of-the-art human operations infrastructure. We apply this directly to our dictation product, and beta users say it feels like someone is actually sitting there listening to every word you say and transcribing it in real time.”
Feature Teaser
“We could not have done this without our distributed processing team, who work around the clock. They're the backbone of this model, and we are actively hiring.”
Call to Action
“Our full benchmark report is on our website. Atlas-1 rolls out to all users of our Willow Voice dictation product today.”
Call to Action
“(No spoken dialogue — ambient room tone)”
Synthesizable in Remotion · key components of the deadpan-founder-launch template.
Bar columns interpolate upward from 0% height to benchmark values sequentially with logo stamps and percentage labels appearing at tops.
Use spring interpolation mapped to frame delays for each bar. Compute bar height from interpolate(frame, [delay, delay + 20], [0, targetHeight]).
Dashed border bounding box drops in over the winning bar with a callout label tag.
Trigger an opacity and scale spring when the final bar completes, rendering a dashed SVG rectangle and pill badge.
Synchronized hard angle switches between direct-facing camera and sharp 90-degree profile.
Wrap sequential video clips in Series.Sequence blocks at predefined script beat intervals.
Purpose. Establish serious technological breakthrough tone.
Execution. Frontal trio studio shot with minimal lower thirds.
Purpose. Substantiate claim with empirical graph data.
Execution. Crisp animated comparative bar chart.
Purpose. Reframe the high-tech model as literal human labor.
Execution. Cut to night-shift operations office b-roll.
Purpose. Deliver punchline constraints and link.
Execution. Centered typographic card detailing offshore operational hours.