AnIntellectualBeing/GTA-Auto-Clipper
Media & DesignCapture GTA live streams locally, auto-detect highlight moments and render caption-burned 9:16 shorts.
What it does
GTA Auto Clipper is a local Python workflow that watches a live gameplay stream, scores its segments, and renders short vertical clips with burned-in pop-up captions. It searches YouTube live results for a query such as GTA 6, requires you to visually confirm the stream is real gameplay, then records in rolling chunks and ranks moments by live-chat reactions, falling back to audio and speech when chat is unavailable. Clips start a few seconds before the detected highlight (6 seconds by default).
The renderer fills a 9:16 canvas with gameplay, shifting the crop away from a persistent side facecam, or can fit the full source over a blurred background. Captions are one-line phrases of up to 3 words placed in the lower third of the visible gameplay panel; each word pops in yellow when spoken, earlier words turn white, and the next phrase replaces the previous one in the same spot. English speech gets English captions, Hindi speech gets romanized Hinglish, and unsupported or uncertain languages get no captions. Swear and sexual words are masked in captions while source audio stays unchanged.
Quick start
Prepare Python 3.10+, FFmpeg on PATH, and Node.js or Deno (yt-dlp needs a JavaScript runtime to read YouTube; Node is used when Deno is missing). OpenCV is needed for --crop face detection, and first-run speech-model downloads need a stable connection. In PowerShell:
python -m venv .venv
.venv\Scripts\Activate.ps1
pip install -r requirements.txt
For live search and optional chat scoring, set a YouTube Data API key without writing it into files:
$env:YOUTUBE_API_KEY = "YOUR_YOUTUBE_DATA_API_KEY"
python find_streams.py "GTA 6" --max 25 --json
The search returns candidates in viewCount order, which is not a live-viewer ranking; open a candidate and confirm actual gameplay is visible before capture. Then start a capture that runs until you press Ctrl+C:
python clip.py "https://www.youtube.com/watch?v=VIDEO_ID" --gameplay-confirmed --max-seconds 0 --clip-len 30 --lead-in 6 --top 2
Selected highlights are written as MP4 and JSON metadata to output/. Test on a local file first; test output goes to test_out/ and cannot be published:
python clip.py --file "C:\videos\gameplay.mp4" --test --clip-len 30
Key options include --moment-source (chat, combined, or audio), --layout (fill covering the 9:16 screen or fit over a blurred background), --spoken-language, --chunk, --min-score, and --output-dir, documented in the README.
Scheduling and publishing
Publishing is separate from clipping and uses Zernio. Connect a YouTube account and an Instagram Creator or Business account, then set ZERNIO_API_KEY, ZERNIO_YOUTUBE_ACCOUNT_ID, and ZERNIO_INSTAGRAM_ACCOUNT_ID as private environment variables. After reviewing the clip, mark its metadata "approved": true and preview the schedule:
python zernio_publish.py "output\gta-CLIP-ID" --scheduled-for "2026-11-18T12:00:00+05:30"
This dry run sends nothing; add --publish to schedule to both accounts. Unattended posting additionally requires GTA_UNATTENDED_POSTS=1 and --unattended. A post already accepted by Zernio can publish while the computer is asleep, but discovery and rendering still need an awake, connected machine. The publisher stores a post ID and idempotency key in the metadata; check the Zernio dashboard after an API failure before retrying. publish.py accepts the same arguments plus PUBLISH_* variable names, and GTA Auto Sender distributes posts to many channels.
Constraints
Rights and platform rules are on you: there is no creator allowlist, and you must obtain permission to record, edit, and publish footage. The project is an independent fan project, not affiliated with Rockstar Games, Take-Two, YouTube, Instagram, Meta, or Zernio. The YouTube API returns live candidates by view count rather than verified concurrent viewers.
Live capture requires the explicit --gameplay-confirmed flag. Captioning only covers English and Hindi (as Hinglish); other languages produce clips without captions. Chat-based scoring needs the YouTube API key, while audio and speech scoring works without it. Multi-stream, multi-agent setups are described in docs/MULTI-AGENT.md, and the optional AI workflow and prompt pack are separate, with the prompt file at GTA-CLIPPING-PROMPT.txt and the detailed workflow at AI_WORKFLOW.md. The code is MIT-licensed.
HotRepo's take
The repository entered HotRepo's new-project board on 2026-11-20, gaining 7 stars and 4 forks in the roughly 90-minute window from 04:46 to 06:16 UTC, reaching 107 stars and 41 forks overall. The fork-to-star ratio is unusually high for a small utility, suggesting people are taking copies to adapt rather than merely starring. It never recorded a best daily rank on the all-language GitHub board.
The design puts most of its differentiation in the scoring fallback (chat, then audio and speech) and the caption placement, while publishing is deliberately split behind Zernio and a dry-run-first command. That separation is a sensible fit for a single-stream, single-operator workflow; multi-channel use explicitly points to a companion sender project.
Written by AI from the project's README, updated . The project's own docs are the source of truth.
You might also like
- Bodila51/gtavi-claudecodeA Claude Code plugin for finding GTA 6 moments, cutting vertical clips and checking them against clipping campaigns.
- haziqahmedqasmi/PeakCutSelf-hosted tool that turns long videos and live streams into captioned vertical clips using your own AI key.
- gothamor/gta-vcs-recompRecompile the PS2 version of GTA: Vice City Stories into a native Windows, Linux or macOS build using your own disc.
- c22dev/muguetRun GTA V's web build as a native Apple Silicon app on macOS and iOS, with small-scale UDP multiplayer.
- 0marContreras/gta6recompAn experimental runtime shader and streaming optimization pack claimed for the PC version of GTA VI.