ClipSpark AI

Streams and VODs

Find the highlights in a four-hour broadcast.

A long stream is mostly filler by design — that is what makes it a good stream. The problem is afterwards, when the six moments worth posting are somewhere inside four hours and nothing marks where. ClipSpark reads the whole recording and tells you where they are.

01

The problem is the runtime

Scrubbing does not scale

Reviewing a four-hour VOD at double speed is two hours of work for maybe six clips. Do that after every broadcast and the clipping is a second job rather than a side effect of the first.

Chat is not a clip finder

A spike in messages marks where the audience reacted, not where a self-contained thirty seconds starts and ends. Plenty of the best moments get no reaction at all until someone posts them.

Your memory is worse than you think

You remember the moment from hour one and the one just before you ended. The middle two hours are the part that quietly never gets clipped, and it is half the broadcast.

Raw stream pacing is slow

Live has room to breathe. Posted as-is to a vertical feed, the same thirty seconds reads as dead time, and the viewer is gone before the moment arrives.

02

What ClipSpark does with a long recording

Every minute is a candidate

The full runtime is analysed rather than sampled, which is the point on a recording this long. A moment three hours in competes on equal terms with one in the opening minute.

Ranked, with the reasoning

Each candidate arrives with its hook line and why it scored, plus the ones that lost. You are picking from a shortlist you can disagree with, not accepting whatever came out first.

Silence removed

Optional, and it earns its keep here more than anywhere else. Dead air is cut so the posted clip is tighter than the broadcast it came from, while the short pauses inside speech are left alone.

A batch per pass

One upload can return several clips instead of one, so a broadcast becomes a week of posts in a single job rather than one clip at a time.

03

Cutting for the vertical feed

9:16, 1:1 or 16:9

Vertical for Shorts, Reels and TikTok, square for feeds, or landscape when what you are actually making is a trailer for the next broadcast. You get a file to upload, not a scheduled post.

Gaming as the audience

One of the audience settings, alongside general short-form, storytelling and the platform-specific ones. It changes what counts as a strong moment before anything is cut.

Captions that survive reposting

Subtitles are optional, generated from the recording's audio, and burned into the picture — so they still work when the clip is downloaded and posted somewhere else.

Post copy per platform

Titles, descriptions, hashtags and several alternative opening hooks, generated for the destination and editable before you use any of them.

04

What to know before you upload

The file has to fit. Uploads go up to 2 GB, and your plan sets the maximum video length. A four-hour broadcast at source bitrate is well past both, so re-export the VOD at a lower bitrate — 720p H.264 is plenty for finding moments and for a vertical crop — or upload the section of the broadcast you care about rather than all of it.

Face tracking is for talking heads. Following the speaker reads mouth movement on camera, so it works on a facecam and does nothing useful for gameplay filling the frame. When it cannot tell who is speaking it holds a single framing rather than guessing, which on a game capture is the right answer anyway.

MP4, MOV, WebM or MKV. Credits are charged by length on a published table and only when a job finishes — a failure or a cancellation puts them straight back.

Start with last night's VOD

See what was in the middle two hours.

The free plan is enough to judge the cuts before you pay for any.

Create an account