1. Home
    2. ·
  2. Cases studies
    3. ·
  3. Klyvio.ai

AI platform that turns any video link into a finished short-form video Klyvio.ai

An AI-powered content platform that takes one to five source video links (YouTube, TikTok, Instagram) and automatically produces a finished, vertical short video with an AI voice-over and animated word-level subtitles. Creators submit a link, track generation progress in real time, and manage their usage through a credit-based subscription system — turning a multi-hour manual editing process into a few minutes of automated work.

Klyvio.ai

Tasks

1

Define the technologies and processing pipeline for turning long-form source video into a finished short-form video, and provide an estimation for the project

2

Build the architecture for an event-driven, multi-stage AI processing pipeline

3

Integrate AI services for speech-to-text, vision/frame scoring, script generation, and text-to-voice behind swappable provider interfaces

4

Implement automatic clip selection, vertical video stitching, and final video composition with animated subtitles

5

Build a real-time job tracking system so users can follow generation progress

6

Implement user authentication, a credit-based economy, and subscription/plan management

7

Implement the provided design for the public-facing site and user dashboard

Public-facing site and user dashboard Public-facing site and user dashboard

Tech Stack

TypeScript Node.js Fastify
Nuxt 3 (SSR) Drizzle ORM PostgreSQL
BullMQ Redis FFmpeg
Remotion OpenAI Google Gemini
Whisper ElevenLabs MinIO / Amazon S3

Approach

Researching available AI providers for speech-to-text, vision scoring, script generation, and voice synthesis, and designing swappable interfaces so providers can be changed without reworking the pipeline
Designing an eight-stage, event-driven pipeline (ingest → analysis → script → voice → selection → stitch → render → publish), with each stage running as an independent, retry-safe worker process
Building the backend on Fastify with PostgreSQL for job and asset state, and BullMQ/Redis queues to coordinate the pipeline stages
Integrating Whisper for transcription and vision models for key-frame scoring to identify the strongest moments in the source video
Generating voice-over scripts and captions with an LLM, converting them to speech, and composing the final video with FFmpeg and Remotion
Implementing the credit-based economy, subscription plans, and authentication (JWT + Google OAuth)
Building the SSR frontend in Nuxt 3 for the public site and real-time job-progress tracking

Result

A working AI video-generation platform that takes a creator from a source video link to a finished, publish-ready short in minutes instead of hours. The pipeline was taken from concept to a deployed MVP in about one week using an AI-accelerated development workflow, and the swappable-provider architecture means the platform isn't locked into a single AI vendor as models and pricing change. Klyvio.ai is live and in active use by content creators repurposing long-form video into short-form content.

Reviews