Twitch clip maker: AI clips from your streams and VODs
First video free up to 60 min, then 20 min a month, watermarked, no card. No watermark from $12/month, or self-host free under MIT.
No link? Upload the file in the app instead. Sources need at least 45 seconds of material.
A three-hour Twitch stream is mostly waiting, travelling and loading screens, with a handful of moments worth posting. Paste the VOD link and OpenShorts reads everything you said across those three hours, picks the passages that stand alone, and cuts each one into a vertical clip: gameplay across the top, your facecam enlarged underneath, captions burned in.
One job returns the best 6 to 12 moments of what it is given. For a long stream, clip it in parts (one job per hour, or just the segment you know was good) and you get more candidates and spend fewer minutes.
Free plan: your first video clipped whole up to 60 minutes, then 20 minutes a month, with a watermark and no credit card. Paid plans start at $12/month with no watermark, and the self-hosted edition is free under MIT.
How the Twitch stream clipper works
1. Paste the VOD link
Open the past broadcast on Twitch and copy its address (twitch.tv/videos/...), or the link of a Twitch clip of 45 seconds or more. A channel page is the live channel, not a recording, so the box asks for the VOD instead.
2. The AI finds the moments in the stream
The VOD is transcribed with a timestamp on every word and scanned for scene cuts, and Gemini ranks every stretch of the talk for how well it would play to someone who was not watching live. Streams with almost no speech go to a vision pass instead.
3. Get vertical clips with your facecam
Each clip is reframed to 9:16: gameplay with a webcam box becomes gameplay on top and your enlarged facecam below, just-chatting segments follow your face. Captions and a hook are burned in; post to TikTok, YouTube Shorts and Instagram or download.
Twitch's own Clip button versus a stream clipper
Twitch already lets you and your viewers clip. Those clips are short, horizontal, made by someone who was watching at the right second, and they live on Twitch. They are great for the moment a chat explodes, and useless for the ninety minutes nobody happened to clip. A stream clipper starts from the other end: the whole VOD, read from start to finish by a model that has no favourite moment, cut into vertical clips sized for TikTok, Shorts and Reels.
That is also why the box asks for the VOD link rather than your channel. A
twitch.tv/yourname link is the live channel; the recording is at
twitch.tv/videos/ followed by a number, which you find under
Videos on your channel page. A Twitch clip link works too, as long as the clip
is at least 45 seconds long.
Long VODs: how a multi-hour stream is handled
Moment picking reads the transcript, not the frames, so a four-hour stream costs about the same to analyse as a forty-minute one, and selection does not degrade with length. The transcript is split into windows of about 90 seconds, every window is scored, and the best ten go on to be cut. That cap is the thing to plan around: whether the job is one hour or six, it returns its best 6 to 12 moments.
So for a long stream, clip it in parts. Clip only part of it lets you set a start and an end, and only that stretch is downloaded, analysed and billed. Run the first hour, the boss fight and the end-of-stream chat as three jobs and you get three sets of candidates. On OpenShorts Cloud the meter is minutes of source, so a full four-hour VOD is 240 minutes: more than the 100 minutes of the $12 plan and inside the 300 of the $29 one. Self-hosted, there is no meter at all, which is the edition that makes sense for a channel with a VOD every day.
One practical point about Twitch itself: past broadcasts are not kept forever. How long Twitch stores them depends on your account type, and the "store past broadcasts" setting has to be on for a VOD to exist at all. Clip a stream in the days after it, or export it, before it lapses.
Facecam, gameplay and just chatting: the layouts
A stream is not a talking head, and a centre crop of one keeps the middle of the map and loses you. With the vertical layout on auto (the default in advanced options), each clip is laid out by what is in it:
- Gameplay with a webcam box. The OBS layout most streamers use, gameplay full screen and a camera in a corner, is detected geometrically: a subject that is small, off centre and still between samples. The clip renders as INSET: the gameplay across the full width on top, the webcam box cropped out and enlarged to fill the bottom. Both stay readable on a phone.
- Full-frame action. When the meaningful content spans the whole width (a chase, a team fight, a map), the clip keeps the full width over a blurred backdrop instead of cropping half of it away.
- Just chatting and IRL. A face-filled frame is face-tracked, with a stabiliser that holds the frame still inside a safe zone instead of following every lean towards the mic.
- Two people on camera. A co-stream or a guest in the same shot is stacked, one person in each half.
If the tracker gets a scene wrong, the reframe editor lets you draw the crop yourself for that scene, or split it into two stacked regions, and the rest of the clip stays automatic. The GTA 5 clips page goes deeper into the inset layout and how it was measured.
Silent gameplay and music-only streams
The default picker reads what you say, which is exactly right for commentary. A grind with the mic muted, or a stream that is mostly music, has nothing to read, so OpenShorts does not try. When the transcript comes back under 8 words in total, or under 5 words per minute, the video itself is sent to Gemini, which watches it and returns moments in the same 15 to 60 second band. Layouts and facecam detection work the same; the clips come out without captions, because there is no speech to caption.
That vision pass has one ceiling worth knowing: Gemini counts video at roughly 300 tokens a second, so an hour of footage is around a million tokens, the edge of its context window, and nothing guards the length in front of it. A silent eight-hour VOD will fail rather than clip. Give the vision pass a segment, an hour or less, with "clip only part of it".
Chat, alerts and overlays
A Twitch VOD has no chat in it unless your overlay drew the chat onto the stream. If it did, the chat box is just part of the picture: in a facecam INSET layout it is usually cropped away with the rest of the corner, and in a full-width layout it stays where it was. Alerts and on-screen text behave the same way. Subtitles from the stream itself are noticed: when burned-in text sits along the bottom in at least 40% of the sampled frames, that band is trimmed before OpenShorts adds its own captions, so you do not get two layers.
Free tier compared
Competitor plans checked 5 October 2026 on their public pricing pages.
| Tool | Free plan | Paid from |
|---|---|---|
| OpenShorts Cloud | First video up to 60 min, then 20 min/month. Watermark, no credit card | $12/month, 100 min, no watermark |
| OpenShorts self-hosted | Free under MIT. No watermark, no cap; your hardware and API keys | $0 |
| Opus Clip | 60 min/mo. Watermarked, export within 3 days | $15/month |
| Klap | 1 video. To try it, no ongoing free tier | $14/month billed yearly |
| Quso | 75 min/mo. 720p, data kept 7 days | $19/month billed yearly ($29 monthly) |
| Vizard | 60 min/mo. 720p, watermark, exports up to 10 min | $14.50/month billed yearly ($29 monthly) |
For streamers the billing unit decides everything. A four-hour VOD is 240 credits on any per-minute meter, which is more than every free plan above and most entry plans. Klap's meter is clips rather than minutes, but it says its algorithm leans on speech, so quiet gameplay gives it little to work with.
What it does not do
- It does not clip live. It works from the VOD after the stream.
- It does not read Twitch chat to find hype moments; it reads what is said and, for silent footage, what is shown.
- It does not return dozens of clips from one long job. Each job returns its best 6 to 12; clip long streams in parts.
- It does not clip a silent VOD longer than about an hour in one go.
- It does not fetch subscriber-only VODs. Download and upload them instead.
- Free clips carry a watermark. Upgrading removes it from clips already made.
Whose streams you can clip, and the music question
Your own VODs, and streams whose owner said yes: editors working for a streamer, members of a clipping programme the streamer runs, co-streamers with permission. The VOD belongs to whoever streamed it, and the game footage is covered by the publisher's own policy on videos of its game. Music is the trap: a song playing during the stream travels into the clip, and TikTok and YouTube match audio far more aggressively than a live stream ever does.
Common questions
- Is this Twitch clip maker free?
- The hosted free plan clips your first video whole up to 60 minutes, then 20 minutes a month, watermarked, no credit card. A multi-hour VOD is longer than that, so on the free plan pick the stretch you want. Paid plans start at $12/month. Self-hosted OpenShorts is free under MIT with no meter, which is the edition that suits streamers with hours of VODs.
- Does it work with a whole Twitch VOD?
- Yes. Paste the twitch.tv/videos/ link of the past broadcast. Moment scoring reads the transcript rather than every frame, so a long VOD is analysed the same way as a short one. One job returns the best 6 to 12 moments of whatever it is given, so for a long stream it usually pays to clip it in parts.
- Can it clip my stream while I am live?
- No. It works from recordings: paste the VOD once the broadcast has ended, before Twitch removes the past broadcast.
- Will my facecam stay in the clip?
- Yes, with the vertical layout on auto. A webcam box composited into a corner is detected by its geometry (small, off centre, still) and the clip renders with the gameplay full width on top and the facecam cropped out and enlarged below.
- What about gameplay with no commentary?
- When the transcript comes back with under 8 words, or under 5 words a minute, OpenShorts switches to a vision pass where Gemini watches the footage and picks the moments. Those clips have no captions. That pass has a practical ceiling of about an hour of footage, so give it the segment, not an eight-hour silent VOD.
- Can I turn Twitch clips into TikToks?
- A Twitch clip of 45 seconds or more can be pasted as a source and comes back reframed to 9:16 with captions. Shorter clips are under the minimum; for those a static crop in the free video to 9:16 converter is the quicker route.
- How many clips do I get from a stream?
- Each job returns up to about a dozen: for any source longer than 12 minutes the AI is asked for 6 to 12. A four-hour stream clipped in four one-hour parts gives four times as many candidates as the same stream clipped in one job.
- Does it post to TikTok and YouTube Shorts?
- Yes, from the dashboard: YouTube Shorts and Instagram Reels are published directly, and TikTok receives the clip as a draft in your TikTok inbox to finish in the app.
- Can I clip another streamer's VOD?
- Only with their permission, for example as an editor or as part of a clipping programme they run. The VOD belongs to whoever streamed it, and the platforms remove reuploads when the streamer reports them.
- How does it compare with Opus Clip or Klap for Twitch?
- Both bill in ways that hurt on long streams: Opus Clip one credit per imported minute, from $15/month, and Klap per clip, from $14/month billed yearly, with a speech-reliant picker (checked 5 October 2026). OpenShorts Cloud also bills source minutes, but lets you clip just a stretch, and the self-hosted edition has no meter at all.
Sources
- Competitor plans: our comparison data, checked 5 October 2026. See OpenShorts vs Klap and OpenShorts vs Opus Clip.
- Silent-footage thresholds (8 words, 5 words per minute), the vision fallback and its length ceiling:
main.py; inset detection:camera_inset.py; burned-subtitle trimming:burned_subtitles.py. Source: github.com/mutonby/openshorts. - OpenShorts Cloud plan minutes: pricing, checked 5 October 2026.