CaptionForge

Upload a video or paste a public YouTube, Instagram or TikTok link; CaptionForge transcribes it with whisper.cpp, styles and optionally translates the captions, and renders them into the video with FFmpeg. A FastAPI control plane prices and tracks jobs, and a native worker on the Mac Studio does the heavy lifting.

The Problem

Most short-form video is watched without sound, so captions decide whether it is watched at all. Doing them by hand is slow, and online tools are expensive per minute and send every video to someone else’s cloud.

The Solution

A local pipeline with an API in front: transcription and rendering run on the studio’s own hardware, jobs are estimated and priced up front, and a doctor command reports exactly which media capabilities the host has.

What I Built

• →whisper.cpp transcription with multilingual models
• →Styled caption burn-in through FFmpeg and libass
• →Caption translation into target languages
• →Remote imports from public YouTube, Instagram and TikTok posts
• →Per-job estimates and pricing via the API
• →Nuxt dashboard, OpenAPI docs and safe media commands

Highlights

• →Transcribe → style → translate → render
• →Job estimates before you commit
• →Imports from YouTube, Instagram, TikTok

Outcome

• →Control plane and worker run on the Mac Studio; nothing leaves the building
• →Pre-launch — public service being prepared