Incredibly Fast Whisper

Optimized Whisper model with word-level timestamps, ideal for caption generation.

Property	Value
Model ID	`vaibhavs10/incredibly-fast-whisper`
Provider	Replicate
Type	Speech-to-text

Basic Usage

import { compose, captions, audioModel } from "@synthome/sdk";

const execution = await compose(
  captions({
    video: "https://example.com/video.mp4",
    model: audioModel("vaibhavs10/incredibly-fast-whisper", "replicate"),
  }),
).execute();

Why Use Fast Whisper

Word-Level Timestamps

Unlike standard Whisper which provides sentence-level timing, Fast Whisper gives precise word-by-word timestamps:

[
  { "word": "Hello", "start": 0.0, "end": 0.3 },
  { "word": "world", "start": 0.35, "end": 0.7 },
  { "word": "how", "start": 0.8, "end": 0.95 },
  { "word": "are", "start": 0.95, "end": 1.1 },
  { "word": "you", "start": 1.1, "end": 1.4 }
]

This enables:

Word-by-word highlighting
Karaoke-style captions
Precise timing for animated text

Speed

Significantly faster than standard Whisper while maintaining accuracy.

Best For

Caption generation: Word-level timing for professional subtitles
Word highlighting: TikTok-style active word effects
Karaoke: Sync text with audio precisely
Time-critical workflows: Faster processing

Caption Styles

TikTok Style (Word Highlighting)

captions({
  video: "https://example.com/video.mp4",
  model: audioModel("vaibhavs10/incredibly-fast-whisper", "replicate"),
  style: {
    preset: "tiktok",
    highlightActiveWord: true,
    activeWordColor: "#FFFF00",
  },
});

YouTube Style

captions({
  video: "https://example.com/video.mp4",
  model: audioModel("vaibhavs10/incredibly-fast-whisper", "replicate"),
  style: {
    preset: "youtube",
    wordsPerCaption: 8,
  },
});

Cinematic

captions({
  video: "https://example.com/video.mp4",
  model: audioModel("vaibhavs10/incredibly-fast-whisper", "replicate"),
  style: {
    preset: "cinematic",
    fontFamily: "Georgia",
    fontSize: 42,
  },
});

Full Pipeline Example

Generate a video with AI narration and auto-captions:

import {
  compose,
  captions,
  generateVideo,
  generateAudio,
  videoModel,
  audioModel,
} from "@synthome/sdk";

const execution = await compose(
  captions({
    video: generateVideo({
      model: videoModel("veed/fabric-1.0", "fal"),
      image: "https://example.com/speaker.jpg",
      audio: generateAudio({
        model: audioModel("elevenlabs/turbo-v2.5", "elevenlabs"),
        text: "Welcome to our channel! Today we're exploring AI video generation.",
        voiceId: "21m00Tcm4TlvDq8ikWAM",
      }),
    }),
    model: audioModel("vaibhavs10/incredibly-fast-whisper", "replicate"),
    style: {
      preset: "tiktok",
      highlightActiveWord: true,
      activeWordColor: "#00FF00",
    },
  }),
).execute();

This pipeline:

Generates speech audio from text
Creates a lip-synced talking head video
Transcribes with word-level timestamps
Adds captions with word highlighting

Incredibly Fast Whisper

On this page