Teek

Open-source AI video clipping, built as a self-hosted alternative to OpusClip, Veed.io, caption.com.

This repo is based on the SupoClip repo (https://github.com/FujiwaraChoki/supoclip), which is a self-hosted alternative to OpusClip.

... because good video clips shouldn't cost a fortune or come with ugly watermarks.

OpusClip charges $15-29/month and slaps watermarks on every free video. SupoClip gives you the same AI-powered video clipping capabilities - completely free, completely open source, and completely watermark-free, while still providing you with a hosted version, that doesn't cost the same amount as your mortgage.

Teek vs OpusClip

Feature	Teek	OpusClip
Caption styles	Multiple styles (classic, hype, retro, minimalist, and more)	Limited presets, paid tiers unlock more
Clip aspect ratio / cutting ratio	Configurable per clip	Fixed 9:16, no control
Automatic face detection	Yes — MediaPipe + OpenCV fallback, no cloud dependency	Yes — cloud-based, behind paid plan
Review clips & adjust duration	Yes — trim start/end in the editor	Basic trim, paid plans only
Review clips against original video	Yes — side-by-side with source video timeline	No
Subscription / plan management	Self-hosted, no subscription required	$15–29/month, watermarks on free tier
Transcription providers	Whisper (local), AssemblyAI, or import your own SRT	Proprietary only, no external provider
RTL language support	Full RTL support via ASS subtitle engine	Partial / broken in most modes
AI clip selection model	Any LLM (OpenAI, Anthropic, Google, ZhipuAI, …)	Fixed proprietary model
Watermarks	None	Always on free tier
Open source	Yes (AGPL-3.0)	No
Self-hosted	Yes	No

Screenshots

What This Repo Contains

• backend/: FastAPI service for transcription, clip analysis, and video rendering
• frontend/: Next.js app for task creation and clip management
• docker-compose.yml: local full-stack orchestration

60-Second Local Start

cp .env.sample .env
# edit .env and add one model provider key (OPENAI_API_KEY or GOOGLE_API_KEY or ANTHROPIC_API_KEY or ZAI_API_KEY)
# transcription defaults to local Whisper; set TRANSCRIPTION_PROVIDER=assemblyai only if you want remote transcription
./start.sh

Then open:

• Frontend: http://${APP_HOST}:${FRONTEND_HOST_PORT} (default http://localhost:3000)
• Backend API: http://${APP_HOST}:${BACKEND_HOST_PORT} (default http://localhost:8000)
• API docs: http://${APP_HOST}:${BACKEND_HOST_PORT}/docs (default http://localhost:8000/docs)

Documentation Map

• Quick start and troubleshooting: QUICKSTART.md
• Canonical config reference: docs/config.md
• Local URL/port mapping reference: docs/local-host-mappings.md
• Agent/project-state guide: AGENTS.md
• Claude compatibility guide: CLAUDE.md
• Backend-specific development notes: backend/README.md

Model And Runtime Defaults

• Default LLM: openai:gpt-5-mini
• Preferred model env var: LLM
• Legacy model env var still accepted: LLM_MODEL
• Preferred Whisper env var: WHISPER_MODEL_SIZE
• Legacy Whisper env var still accepted: WHISPER_MODEL
• Frontend local runtime: Node.js 20+ / npm 10+ (frontend/.nvmrc)

Backend Entrypoints

• Docker/default path: src.main_refactored:app
• Legacy local path: src.main:app

Current Risks / Known Gaps

• Runtime quality depends on model/provider behavior and prompt consistency.
• No lightweight model regression/eval suite is currently documented.
• Repository is actively changing; keep docs aligned via docs/config.md.

License

AGPL-3.0. See LICENSE.