A free, fully local clipping studio. Paste a YouTube link (or drop a video file), and it finds the moments worth sharing, cuts them, reframes them to vertical 9:16 and burns in captions.
Nothing is uploaded anywhere. Video work happens with ffmpeg and yt-dlp on your own machine, and
the model that picks your clips is either the built-in free local selector or your own API key.
Windows:
.\install.cmdmacOS / Linux:
./install.shThe installer builds the interface, packs the tool and installs the sudoclipper command globally.
It opens the app in your default browser the moment it is ready — no account, no config file to edit.
After that, sudoclipper from anywhere starts the studio again. Useful flags:
sudoclipper # start and open the browser
sudoclipper --port 4500 # pick a port
sudoclipper --no-open # do not open a browser
sudoclipper --doctor # report the state of ffmpeg / yt-dlpRequires the .NET 10 SDK (or runtime). Node.js is only needed if you want to rebuild the UI yourself.
SUDOCLIPPER needs two local binaries and downloads them once into its own folder
(%LOCALAPPDATA%\sudoclipper\bin on Windows, ~/.local/share/sudoclipper/bin elsewhere):
| Tool | Why |
|---|---|
yt-dlp |
downloads the source video and its caption track |
ffmpeg |
cuts, reframes to 9:16, burns captions, encodes |
If they are already on your PATH, those copies are used instead. Progress is shown in the UI and
in Config → Diagnostics. The download resumes where it stopped if the connection drops or the app is
closed mid-way, and only one setup ever runs at a time — even with two windows open.
Clipper — paste a link or drop a file. Choose how many clips, a length preset, and captions on or off. A live job panel shows Download → Captions → Pick moments → Render with the worker's log.
Memory — every clip rendered on this machine, with thumbnails, source, model, and playback. Search, open the folder, or delete a clip (which removes the files).
Config — provider and API keys, the model list, clipping defaults, output folder, diagnostics.
Pick a provider, paste a key, and the app lists every model that key can reach:
- Local (free) — no key. Scores transcript windows for hook language, questions, numbers and pacing.
- OpenRouter, OpenAI, Google Gemini, Groq, DeepSeek, Mistral — hosted models.
- Ollama — models already running on your machine, so nothing at all leaves localhost.
- Custom endpoint — any OpenAI-compatible base URL (LM Studio, vLLM, LiteLLM, Together…).
Keys are stored in %LOCALAPPDATA%\sudoclipper\config.json (plain text, local only) or read from
SUDOCLIPPER_<PROVIDER>_KEY environment variables. They are never sent anywhere except the provider
you selected.
yt-dlpdownloads the video (≤1080p) plus its subtitle track, converted to WebVTT.- Captions are parsed, de-duplicated and timestamped (auto-caption roll-up handled).
- The transcript goes to your model, which returns start/end, title, hook, score and a rationale. Long videos are chunked and merged by score.
ffmpegcuts each moment, builds a 1080×1920 frame (blurred background, centred source), burns a styled ASS subtitle track, normalises audio and writes a faststart mp4 plus a thumbnail.- The clip is recorded in Memory. The working directory is cleaned up.
Exports land in ~/Videos/SUDOCLIPPER/<video-title>/ by default (change it in Config).
- Moment selection needs a transcript, so a video with no English caption track stops with a clear message instead of guessing. Captions are fetched in a second, best-effort pass, so a rate-limited subtitle request never throws away a finished video download.
- Only the first attached file is used per run; links and files are mutually exclusive in the input.
- Downloaded sources are deleted after rendering; only clips are kept.
- Everything binds to
127.0.0.1, so the studio is not reachable from other machines.
dotnet run --project src/SudoClipper # API on http://127.0.0.1:4319
cd web && npm install && npm run dev # UI on http://localhost:5273 with /api proxied
cd web && npm run build # writes the bundle into src/SudoClipper/wwwrootsrc/SudoClipper/ .NET minimal API: pipeline, LLM clients, ffmpeg orchestration
web/ React + Tailwind + Motion interface
MIT licensed. "# SudoClipper" "# SudoClipper" "# SudoClipper"