2026-11-19 · 5 min read

Captions for Reaction Videos: Two Voices, One Screen

A film strip of reaction video scenes, the middle frame showing the captioned word REACT
This image was generated with AI.

Reaction videos are a duet: the source clip plays, the creator talks over it, and the viewer is following both at once. That format lives or dies on comprehension, and comprehension is the first thing to go when someone watches muted in a feed or the commentary overlaps the original audio.

Why reactions need captions more than most formats

  • Two audio threads compete. When the reactor talks over the clip, even viewers with sound on can lose a line. Captions keep the commentary anchored.
  • A large share of discovery viewing happens muted or near-muted in feeds. No captions means the reaction reads as a person making faces at nothing.
  • Fast cuts between source and reaction make context fragile. On-screen text carries the thread across the cut.

What to caption

Caption the commentary first: it is the original content and the reason the viewer is there. If the source clip has important dialogue, caption that too, ideally in a visually distinct style so the two threads do not blur. A different position or color per voice is enough.

A workflow that scales

Recording the reaction in one take and captioning after the fact beats trying to subtitle live. With caption.sh, captioning the finished edit is one POST per video:

curl -X POST https://api.caption.sh/v1/videos \
  -H "Authorization: Bearer $CAPTION_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"source_url": "https://example.com/reaction-cut.mp4",
       "options": {"position": "bottom", "max_words": 4}}'

The response is a job id; poll GET /v1/jobs/{job_id} until it reports done, then fetch the finished file from /v1/jobs/{job_id}/result, which redirects to the video. Results are stored for 24 hours, then deleted. Videos up to 10 minutes per job, $0.10 per minute prepaid, and a 429 just means wait and retry.

One accessibility caveat

Burned-in captions are pixels: viewers cannot toggle, resize, or translate them. Wherever the platform supports caption tracks (YouTube does), upload the track as well and keep the burn for the places tracks do not survive, like feed previews and reuploads.

Back to all articles