YT Short Clipper v2

A Windows desktop app that turns a long-form YouTube video into ready-to-post 9:16 short-form clips — picking the highlights with an LLM, reframing to portrait, and burning in word-by-word captions.

Stack: React 19 + TypeScript + Vite · Tauri v2 (Rust shell) · Python sidecar (yt-dlp, FFmpeg, MediaPipe)

Windows only. The build scripts are PowerShell, the app ships as a portable zip with a WebView2 bootstrapper, and caption/hook rendering reads fonts from C:\Windows\Fonts. Nothing is macOS/Linux-ready today.

Status: beta. Expect rough edges and breaking changes between releases.

What it does

1. You paste a YouTube URL and upload your own cookies.txt (see YouTube cookies).
2. yt-dlp fetches the video and its original subtitle track.
3. An LLM reads the transcript and proposes highlight segments (58–120s by default).
4. Each segment is cut, reframed to 9:16, captioned, and written out as an MP4.
5. Optionally, clips are uploaded and scheduled through Repliz.

Features

• AI highlight detection from the video's subtitle transcript, via any OpenAI-compatible endpoint
• AI direction per video — free-text steering: name a time range, chase a recurring topic, or rule material out
• Output language for titles and hook text, defaulting to the video's own language
• Portrait reframe in three modes: face tracking (MediaPipe), centered on black bars, or centered on a blurred fill — the centered modes are dramatically faster since they skip per-frame face detection
• Word-by-word captions burned from YouTube's original subtitle track — no transcription API and no local Whisper model required
• Optional hook text overlays and logo/credit watermarks
• Direct upload to the Repliz social media platform

AI providers

One provider serves the whole app — highlight detection and title generation both — and is chosen once in AI Models. There are two ways to power it.

In-app AI. The app provisions its own wallet for this installation on first launch: no registration, no email, no API key to paste. Top up with QRIS from inside the app and the balance shows in the sidebar. The wallet is held by a key kept in the OS credential store (Windows Credential Manager, Keychain, libsecret) — never in a config file. There is no server-side recovery, so export that key and keep it somewhere safe: it is the only way to reach the balance from another machine, and nobody can restore it for you. Top-ups are not refundable.

Your own API key. Anything OpenAI-compatible works; the app ships presets for OpenAI, Google Gemini, Groq, ApiSmart, the web wallet at ai.ytclip.org, and a Custom/Local option for vLLM, Ollama, or similar. Your key is stored in the app's local data directory and is never sent anywhere except the provider you configured.

The two are separate wallets. A balance topped up at ai.ytclip.org is reached through the "YTClip AI (Web)" preset under your own API key, not through the in-app balance.

AI direction

Left empty, the model picks clips on its own. Filled in, the direction outranks the built-in selection principles:

• An explicit clock range (2:00 - 2:50) is used exactly — that clip is exempt from the usual 58–120s duration filter, and the run samples at a lower temperature so the instruction is actually followed.
• Array order is clip order, so "clip pertama dari …" lands first in the list.
• Anything the direction rules out is left out.

Custom system messages (in AI Models) can place the direction with a {user_direction} placeholder; without it the direction is appended at the end.

YouTube cookies

The app needs cookies from a logged-in YouTube session for yt-dlp to fetch subtitles and video. You supply your own — nothing is bundled.

1. Open YouTube in your browser and make sure you are logged in.
2. Install the Get cookies.txt LOCALLY extension.
3. Run it while on youtube.com and export cookies.txt.
4. In the app, click Upload YouTube cookies to continue and pick that file.

Notes:

• Don't log out of YouTube after exporting — that invalidates the cookies.
• Frequent 403s mean the cookies went stale; export a fresh file.
• A valid export contains SID, HSID, SSID, APISID, SAPISID, and LOGIN_INFO.
• cookies.txt grants access to your Google account. It is gitignored here, stored only in the app's local data directory, and passed to yt-dlp on your machine — but treat the file itself like a password and never share it.

Prerequisites

• Node.js v18+
• Rust latest stable
• Python 3.13+
• Windows 10/11 with PowerShell

Setup

# 1. Install Node dependencies
npm install

# 2. Install Python dependencies
py -m pip install -r requirements.txt

# 3. Download runtime binaries (FFmpeg, Deno)
#    Existing files are skipped — add -Force to actually re-download them.
powershell -ExecutionPolicy Bypass -File scripts/fetch-deps.ps1

# 4. Build the Python sidecar executable
npm run build:sidecar

FFmpeg must be a GnuTLS build (the fetch script pulls one from gyan.dev). Schannel builds hang forever on the byte-range requests used for section downloads.

Development

npm run dev          # Start Vite dev server
npm run tauri dev    # Start Tauri dev window

Python changes need a sidecar rebuild. src-tauri/binaries/ytclip-sidecar-*.exe takes precedence over the dev Python module, so edits under yt_short_clipper_core/ do nothing until you run npm run build:sidecar — the app keeps running the frozen copy without complaining. Delete that binary to fall back to py -m yt_short_clipper_core.sidecar for fast iteration. Frontend and Rust changes hot-reload as usual.

Updating yt-dlp

py -m pip install --upgrade yt-dlp   # 1. upgrade the package
                                     # 2. bump the floor in requirements.txt
npm run build:sidecar                # 3. re-freeze it into the sidecar

Check the yt-dlp changelog for a raised Deno floor while you are there (2026.06.09 raised it to v2.3.0). Deno drives remote_components for YouTube's JS challenges, so an under-floor binary breaks extraction quietly — and fetch-deps.ps1 will not replace an existing deno.exe without -Force.

Release Build

# One command: sidecar → frontend → Tauri build → assemble portable zips
npm run release

This produces two zips in src-tauri/target/release/bundle/:

• YTShortClipperV2__x64_portable.zip — full bundle for new users
• YTShortClipperV2__x64_update.zip — app + sidecar only, for existing users

See BUILD_GUIDE.md for the full release flow, version bumping (4 files), and the build-only-what-changed shortcuts.

Network calls & telemetry

Being transparent about every host the app talks to, since some of it is the maintainer's own backend rather than a third party:

No video content or transcript is sent to any ytclip.org host except the AI endpoint you pick — and that one only receives what you would send any AI provider. The installation ID is a random UUID generated locally on first run, not tied to any account.

Google Analytics is the one third party here. It receives the installation ID, the app version, and each finished clip's duration and reframe mode — plus, as with any HTTP request, your IP address, which Google uses for coarse geolocation. No video, transcript, title, or URL is sent. Builds without GA_API_SECRET set at compile time send nothing at all.

If you fork this, point these at your own backend or strip them out — see src/hooks/ and src-tauri/src/commands/analytics.rs.

Sidebar links

The external links at the bottom of the sidebar are served by api-v2.ytclip.org/api/public/v1/menu, so they can change without an app release. The four page routes above them stay hardcoded.

Icons are named by the API as kebab-case strings and resolved through the allowlist in src/config/menuIcons.ts — an icon must exist there before the server can use it. Responses are validated (https-only URLs, capped count and label length, bad entries dropped individually), cached, and fall back to DEFAULT_MENU_ITEMS in src/hooks/menu.ts when offline.

Sponsors

AI Provider Sponsor

Become a sponsor

Interested in sponsoring this project? Reach out at sponsorship@ytclip.org.

Legal

This tool automates downloading and re-cutting YouTube videos. Only use it on content you own or are otherwise licensed to reuse; downloading or republishing third-party videos may violate YouTube's Terms of Service and copyright law in your jurisdiction. You are responsible for what you process and post. The maintainer provides no warranty and accepts no liability — see LICENSE.

Not affiliated with, endorsed by, or sponsored by YouTube, Google, OpenAI, or Repliz.

License

MIT © Aji Prakoso