Turn GTA 5 and GTA RP streams into vertical clips

Why GTA clips break a normal auto-clipper

Every auto-clipper in this category was designed around a talking head: one person, centred, filling a 16:9 frame that crops cleanly to 9:16. A GTA stream is the opposite on all three counts. The frame is gameplay, so a centre crop keeps Los Santos and drops the minimap, the kill feed and the chat. The only face is a small webcam box pinned to a corner, so a face tracker either ignores it or, worse, latches onto a pedestrian NPC. And the source is not eight minutes long, it is eight hours. Those are three different problems and each one has its own answer below.

How the webcam inset layout works

The OBS layout almost every GTA streamer uses, gameplay full screen with the camera composited into a corner, is a single video file with two things in it. OpenShorts detects that geometrically rather than asking a model: it looks for a subject that is small, off centre horizontally and still between samples. All three filters are needed. A talking head sitting high in frame is still centred, so size alone is not enough, and a real person moves 300 pixels between samples where a pinned webcam box moves three to eleven. On our 48-video test corpus that detector found all five clips that had a webcam inset, with no false positives.

When it fires, the clip renders as INSET: the gameplay across the full width at the top of the 9:16 frame, and the webcam box cropped out and blown up to fill the bottom. You get the play and the reaction to the play, both legible on a phone, instead of one of them at 100 pixels wide.

When the gameplay itself is the point

Not every moment has a face worth showing. A chase, a heist finale or a five-car pileup means what it means across the whole width of the frame, and cropping to a vertical column deletes the half that explains it. OpenShorts measures how wide the meaningful content is and routes on that: content spanning more than 85% of the frame renders as WIDE, which keeps the full width intact over a blurred backdrop rather than side-cropping it. Content that leaves room beside it, a GTA RP scene playing out on one side of the screen for instance, gets stacked over the presenter instead.

Width is the gate rather than coverage because width is what survives measurement. A corner kill notification and a full-screen map both look "busy"; only one of them spans the frame and cannot be cropped.

How to clip a GTA 5 stream, step by step

1. Paste the VOD link (a Twitch export, a YouTube upload) or drop the local recording in. Multi-hour sources are the normal case here, not the edge case.
2. faster-whisper transcribes with word-level timestamps and PySceneDetect maps the cuts, which is what keeps a clip from opening mid-explosion.
3. Gemini reads the transcript against those boundaries and returns the 3 to 15 segments that stand alone best, 15 to 60 seconds each.
4. Leave the vertical layout on auto (dashboard, advanced options; "layouts": ["auto"] on the API). Auto is what enables the screen layouts, and the inset detector is chained behind them.
5. Subtitles are burned in from the word-level transcript. On gaming feeds this is not optional polish: the clips autoplay muted.
6. Download the clips, or post them straight to TikTok, YouTube Shorts and Instagram Reels from the dashboard or the API.

What an eight-hour stream costs to clip

Competitor terms checked 2026-07-27 on public pricing pages.

This is the arithmetic that decides the category, and it is worth doing before you pick a tool. Credit-metered clippers bill one credit per minute of the video you import, not per clip you keep. One eight-hour GTA RP stream is 480 minutes. Opus Clip's free tier is 60 minutes a month, Starter is 150 minutes at $15/month and Pro is 300 minutes at $29/month, so a single stream does not fit in any of them, and a streamer who goes live three times a week is importing roughly 6,000 minutes a month. Gaming is the category where per-minute pricing and the actual shape of the content are furthest apart.

OpenShorts comes in two editions and they are priced very differently, so it is worth being precise. OpenShorts (self-hosted) is free and open source under the MIT licence: Free and open source under MIT. You run it with Docker on your own machine and bring your own API keys. No watermark, no usage cap, no subscription. What it costs you is hardware and time: on a typical CPU an 8-minute video takes 5 to 8 minutes to process. OpenShorts Cloud is the hosted service: The same software with the compute and API keys covered. An NVIDIA GPU clips that same 8-minute video in about 50 seconds. Free plan is 20 minutes a month with a watermark and no credit card. Paid plans start at $12/month for 100 minutes with no watermark, up to $59/month.

What happens when nobody is talking

Most of this page assumes commentary, because the default picker reads the transcript: roleplay dialogue, heist banter and party voice chat are exactly what it is good at, and reading words rather than frames is why an eight-hour source costs about the same to analyse as an eight-minute one. A silent grind has no transcript to read, so OpenShorts does not use one.

It switches paths on its own, and it does not need to be told to. Footage with no audio track at all, and footage whose transcript comes back under 8 words or under 5 words per minute (music-only streams, a mic that was muted the whole session), both trip the same branch: the video itself goes to Gemini, which watches it and returns the same 3 to 15 moments in the same 15 to 60 second band as the transcript path. Everything downstream is identical, layouts and inset detection included. The one difference is that the clips come out without captions, which is correct rather than a bug: there is no speech to caption.

The one ceiling worth knowing This is the single stage that sends Gemini the footage instead of a handful of frames, and Gemini bills video at roughly 300 tokens per second. An hour of gameplay is around 1.08 million tokens, which does not fit a 1 million token context window, and there is no length guard in front of it: a silent eight-hour VOD will fail at the model rather than politely. So for silent footage, hand it the session or the segment you care about rather than the full stream. With commentary the ceiling does not exist, because the transcript path never uploads the video at all.

Whose footage can you clip?

Yours, and footage you have permission for. Your own streams and recordings, your RP server co-stars' VODs with their blessing, clients' channels you manage. Two separate rights questions apply to GTA clips and they have different answers: the recording belongs to whoever streamed it, and the game footage is covered by Rockstar Games' own policy on fan videos, which has historically permitted gameplay videos monetised through the platforms' standard ad programs. That is their policy and it can change, so check the current version rather than taking this page's word for it. Reuploading another streamer's clips without permission is the one case that is clearly not fine, and the platforms strike it.

Common questions

How do I make GTA 5 clips for TikTok?

Paste the stream VOD link into OpenShorts with the vertical layout set to auto. It transcribes the whole recording, has Gemini pick the 3 to 15 strongest 15 to 60 second moments out of what was said, reframes each one to 9:16 keeping the gameplay full width with your facecam enlarged below it, and burns in word-level subtitles. Clips download or post straight to TikTok, Reels and Shorts.

Can it handle a whole eight-hour GTA RP stream?

Yes, long sources are the design case. Moment scoring reads the transcript rather than the raw video, so an eight-hour VOD does not degrade selection the way it degrades a frame-by-frame approach. Processing time scales with length: the GPU-backed hosted edition clips about 8 minutes of source in 50 seconds, and self-hosted on CPU it is roughly 5 to 8 minutes of processing per 8 minutes of source.

Does it keep my facecam in the clip?

Yes, when the layout picker is on auto. A webcam box composited into a corner is detected geometrically (small, off centre horizontally, static between samples) and the clip renders as INSET: gameplay at full width on top, the webcam cropped out and enlarged underneath, so both are legible on a phone.

Does it work on gameplay with no commentary?

Yes, and it switches by itself. Footage with no audio track, or whose transcript comes back under 8 words or under 5 words per minute, goes down a vision pass instead: Gemini watches the footage and returns the same 3 to 15 moments, with the same layouts and inset detection after it. The clips come out without captions, since there is no speech to caption. The practical limit is length, because that pass sends Gemini the video rather than a few frames: give it the session you care about, not a silent eight-hour VOD.

Is it free for streamers?

Self-hosted OpenShorts is free and open source under MIT with no per-minute meter, which is the edition that makes sense when your sources are measured in hours: run it with Docker and bring your own Gemini API key. OpenShorts Cloud covers 20 minutes a month free with a watermark, and paid hosted plans start at $12/month.

Sources

• Opus Clip tier minutes and prices checked 2026-07-27 on their public pricing page.
• Inset detection and layout accuracy figures are our own measurements on a 48-video internal corpus, 2026-08.
• Silent-footage thresholds (8 words, 5 words per minute) and the vision fallback are in main.py; Gemini video token rates from Google's published pricing.
• Inset, WIDE and screencast layout implementations in the project source at github.com/mutonby/openshorts.

Related comparisons