Under the hood
How a 4-hour stream turns into 40-second Shorts
Seven steps, fully automatic. The cheap steps read everything; the expensive steps only touch the few clips worth posting.
01 · THE LONG VIDEO
Paste the VOD link
Your full recording goes in: landscape 16:9, hours long, gameplay plus facecam plus chat noise. StreamToShorts reads its length first and charges 1 credit per started hour.
02 · LISTEN
Every word, every scream
Speech becomes text with a timestamp for every word. In parallel, loudness is measured against the surrounding minute, so a sudden yell scores high even in a loud game.
03 · FIND THE MOMENTS
Setup, peak, payoff
AI reads the transcript in 12-minute chunks, with the loud spots marked. For each moment it picks where the setup starts, where the peak hits and where the reaction lands, then scores it 0–100 for someone who never saw the stream. Duplicates and filler are dropped.
Why: a calm plan, a growing threat, then panic and a crash anyone understands in 3 seconds.
04 · LOOK BEFORE CUTTING
Vision only for the shortlist
Only the top clips get looked at: four frames each. AI confirms the moment is real, finds the facecam and the action, and writes three hooks that match what's on screen.
startbuildpeakendWho called the cops?!
How did a quiet drive end like this?
Watch without wincing. Go.
05 · RESHAPE TO 9:16
Landscape in, vertical out
The facecam box is cut out and placed on top. The gameplay is cropped around where the action is, not just the center, and fills the bottom. No facecam? The action gets a square crop over a blurred copy of the full frame.
06 · MAKE IT WATCHABLE
Captions, hook, punch, polish
- Word-by-word captions, the spoken word lit up
- Key words in color, bigger when they land
- Hook banner in the first 3 seconds
- Zoom punch and flash on the payoff
- Audio leveled so screams don't clip
- Fade-out ending that loops cleanly
UPSIDE DOWN
07 · THREE EDITS, ONE MOMENT