Cue

My wife needed hardcoded subtitles with word highlights for real videos.

Most tools we tried were either paid, confusing, or missing the exact export behavior we needed.

The primary user is non-technical, so the flow had to stay simple.

Everything had to run locally for privacy, support RTL languages, and remain free to use.

I wrote a strict UX spec first, then broke delivery into milestone-sized tasks for Cursor.

I kept handoff notes and known issues in the repo so I could switch sessions without losing context or reopening solved debates.

Cue wraps Whisper and FFmpeg in a desktop workflow: transcribe, style, review, and export.

It supports word-by-word highlighting, style presets, and burned-in output so the result is ready to publish.

The interface is designed for clarity over feature noise: obvious next action, readable controls, and low-friction states for non-technical users.

It replaced paid subtitle workflows for our real use case while keeping videos local.

It turns strong open-source engines into a tool my wife can use without touching pro video editors.

A key AI miss was a docked left subtitles panel that looked good in spec but made the preview too small in reality.

I paused that direction and prioritized preview-first editing so usability wins over spec purity.

Detailed specs reduce wasted cycles, but real usage beats assumptions every time.

When UI behavior fails in practice, I update the plan and move on instead of forcing the original idea.