Fully automated short-form content pipeline
Discovers top Twitch clips, scores and ranks them, reformats for vertical video,
adds subtitles and overlays, and publishes across 4 platforms — every day, zero manual input.
| YouTube views first 30 days |
Instagram views first 3 days |
videos uploaded fully automated |
platforms simultaneous cross-post |
Pipeline controls, autopilot triggers, and real-time clip recommendations.
Review, approve, and edit clips before they go out. Each card shows the thumbnail, score, streamer, and platform targets.
Track total views, best posting windows, and per-video performance. Scoring weights retrain automatically based on what actually gets views.
Every morning at 7am, the pipeline runs unattended:
Discover — Pulls hundreds of clips from the Twitch Helix API, filtered by game and streamer. Deduplicates against a history of 14K+ previously seen clips.
Score + Screen — Each clip is ranked by view velocity, duration, audio quality, and title keywords. Scoring weights are trained from actual upload performance — the system learns what gets views on each channel. Top candidates are then screened by Gemini for content quality.
Process — Clips are converted from 16:9 → 9:16 vertical format. The facecam is auto-detected and cropped to fill the top band. Gameplay is zoomed and repositioned with the HUD overlaid at the bottom. Each streamer has a calibrated layout profile (55+ profiles stored). Whisper generates word-level subtitles with karaoke-style highlighting. FFmpeg composites everything: subtitles, hook text, progress bar, and platform-specific CTA animations.
Publish — Clips are uploaded to YouTube with publishAt scheduling across peak hours, then cross-posted to Instagram, TikTok, and Facebook with per-platform throttling. Daily highlight compilations are assembled automatically with transition cards between clips.
Learn — After upload, the system tracks views and engagement. Scoring weights are periodically retrained so clip selection improves over time. The analytics dashboard surfaces best posting windows and top-performing streamer segments.
| Time | What happens |
|---|---|
| 7:00 AM | Fetch clips from Twitch, score and approve top candidates |
| 7:05 AM | Format, subtitle, and burn overlays (3 clips processed in parallel) |
| 7:35 AM | Upload 7 Shorts to YouTube, scheduled across peak hours |
| 7:40 AM | Build and upload daily highlight compilation |
| 8:00 AM | Marathon channel runs the same pipeline independently |
| 9:00 AM | Cross-post top performers to Instagram, TikTok, Facebook |
| Every 5 min | Release executor publishes any uploads whose scheduled time has arrived |
Runs via launchd cron on macOS. The web dashboard is for reviewing clips and monitoring performance — not required for daily operation.
| Pipeline + API | Python, FastAPI, SSE streaming |
| Video | FFmpeg via imageio-ffmpeg (libass, loudnorm, concat demuxer) |
| Speech-to-text | Whisper with word-level alignment |
| Content AI | Gemini for screening, titles, descriptions, captions |
| Database | SQLite in WAL mode with auto-migration |
| Frontend | React, TypeScript, shadcn/ui, Vite |
| Downloads | yt-dlp |
| Platforms | YouTube Data API v3, Instagram Graph API, TikTok, Facebook Reels |
pip install -r requirements.txt
cp .env.example .env # Twitch, Meta, TikTok, Gemini API keys
python -m clipper auth # OAuth setup for each platform
python -m clipper serve # API + web UI → localhost:8420All behavior is driven by config.yaml — streamer lists, scoring thresholds, daily caps, posting schedules, layout preferences. No secrets in config; credentials live in .env and OAuth token files.