Why another subtitle tool?
Most auto-caption tools stop at transcription. They drop a block of text at the bottom of the screen and call it a day โ often covering a speaker's face or the very object being talked about.
AI Subtitle Studio treats subtitles as a design problem, not just a text problem. Our pipeline understands what is on screen and what is being said, then places captions so they never fight with your footage.
The pipeline
Every stage is an isolated, retryable service:
- Upload โ video and audio are extracted and stored as immutable artifacts.
- Transcribe โ Whisper (OpenVINO) produces accurate transcript segments.
- Align โ wav2vec2 assigns millisecond-precise word timestamps.
- Vision โ YOLO detects people, faces and objects on every frame.
- Scene graph โ a semantic layer merges transcript, timings and detections.
- LLM โ reasoning about emotion and emphasis picks styles and highlights.
- Layout โ a dedicated engine computes safe, collision-free positions.
- Export โ SRT, ASS, WebVTT or a burned-in MP4 / MOV.
The golden rule
The LLM never computes coordinates. It only performs semantic reasoning.
Positioning, collision avoidance and safe areas are the job of the layout engine, which knows the exact bounding boxes of everything on screen. That separation is what keeps results consistent and reproducible.
Try it
Open the editor, upload a clip and start styling. Pick a font, tweak the outline and watch the preview update live.