Clip Pipeline User Guide — install, settings, and every step

1. Quick start

Windows: run the installer and follow the wizard. macOS: open the DMG, drag Clip Pipeline to Applications, then right-click → Open the first time (the build is unsigned).

The installer is self-contained: ffmpeg, Node, yt-dlp, and a portable faster-whisper engine are inside. You do not need Python, ffmpeg, or Node installed separately.

Install and launch. A new job is created automatically.
Get a free Groq API key. The full workflow needs one for script rewriting and AI moment scoring. Groq setup → Skip it only if you want the no-AI fallbacks.
Paste a video URL, or click Open Local Video to work on a file you already have.
Click Run All for the full pipeline, or run individual steps. Steps you do not want can be skipped.
Find the result in the job folder, or in the local export folder you choose in Settings.

Download for Windows Download for macOS

Get a free Groq API key Start here

The full workflow needs a free Groq API key. Script rewriting and YouTube Clipper’s AI viral-moment scoring both use Groq. Without a key, Clip Pipeline falls back to local Whisper, the transcript as the script, the built-in system voice, and fixed-length clips — usable, but not the full intended workflow.

Free tier Groq’s free tier is enough to start and does not require a credit card. Rate limits apply; the app can rotate extra keys if you add them later.

1. Create an account

Open console.groq.com. Click Continue with Google, Continue with GitHub, Continue with SSO, or enter your email and click Continue with email. Complete the verification step.

2. Open the API Keys page

After signing in, open console.groq.com/keys, or choose API Keys in the top navigation.

Groq Console API Keys page with the Create API Key button. — The API Keys page. Click **Create API Key**. Existing keys are hidden from this documentation screenshot.

3. Create the key

Click Create API Key.
Set Display Name to Clip Pipeline.
Leave Expiration at No expiration, or choose a rotation period you prefer.
Click Create. Copy the gsk_… key immediately — Groq shows it only once.

4. Paste it into Clip Pipeline

Open Clip Pipeline → Settings → Groq API key, paste the key, then click Save Settings. The field is at the top of Settings; see the API keys section. If you have more than one free key, add the extras under Extra Groq keys for automatic rotation.

Keep it secret The key is stored in the app settings file. Do not share it or commit it. If it leaks, revoke it on the Groq API Keys page and create a new one.

With the key in place, Rewrite Script and Find Viral Moments work. If you hit a rate limit, add another free key; the pipeline rotates to it automatically.

2. The main window

The main window. Top: source bar and run controls. Centre: the pipeline steps. Right: the live Activity Log. In this screenshot four optional heavy steps are pre-skipped, so Run All shows “4 skipped.”

Control	What it does
Single Video	Paste one video URL (YouTube, Instagram, Facebook, TikTok) and run the pipeline on it.
Entire Profile	Paste a creator profile or page URL to scan and bulk-download everything, then process the batch.
New Job	Starts a clean job folder. The current job stays in History.
Open Local Video	Uses a video already on your disk and skips Download Video.
Run All	Runs every enabled step in order. The label shows how many steps are skipped.
Activity Log	Live output from downloads, transcription, rendering and every other step.
Header links	YouTube Clipper, AI Images, History and Settings.

3. The pipeline

Clip Pipeline steps 4 to 6, showing Generate Voice, Remove Original Captions with a method selector, and skipped optional steps. — The step cards. Each has **Skip** and **Run**. Skipped cards turn amber and are excluded from Run All; manual Run still works. The simple caption removal runs on ffmpeg and needs no extra engine.

Step	What it does	Needs
1. Download Video	Downloads the source, merges video + audio, saves `source.mp4`.	Bundled yt-dlp + Node
2. Extract Captions	Transcribes the original audio with word timestamps.	Groq key or local Whisper
3. Rewrite Script	Rewrites the transcript into a tighter short-form script.	Groq key; otherwise the transcript is used
4. Generate Voice	Speaks the script with ElevenLabs, a local Qwen voice, or the built-in system voice.	ElevenLabs key, Qwen engine, or nothing
5. Remove Original Captions	Blurs, delogos or crops the caption band using ffmpeg.	Bundled ffmpeg only
6. Analyze Scenes	Finds scenes, crossfades and zoom moves.	Optional VSR engine
7. Remove Captions & Recreate Images	Rebuilds clean scene backgrounds and saves stills.	Optional VSR engine
8. Regenerate Images (AI)	Creates new artwork per scene.	Optional ComfyUI engine
9. Rebuild Video	Rebuilds the video from clean images with push-ins and crossfades.	Optional VSR engine
10. Assemble with New Voice	Merges the clean video and the new voiceover, retiming if needed.	Bundled ffmpeg
11. Attach New Captions	Burns word-synced captions onto the final clip.	Groq timestamps or local Whisper

Skip tip The heavy VSR/ComfyUI steps are pre-skipped when those engines are not installed. The core workflow still completes through steps 1–5, 10 and 11.

Transcript and Script boxes let you paste your own text. Paste a transcript and skip Extract Captions; paste a script and skip Rewrite Script; or skip Rewrite with a transcript present and the transcript becomes the spoken script.

4. Settings

Open Settings from the header. Everything below is stored in the app settings file and applies to the next job.

API keys and engines

Clip Pipeline Settings showing Groq API key, Kinetra Studio key, transcription engine, YouTube Data API key, clips per video, ElevenLabs key, voice engine, voiceover voice and Groq chat model. — Settings → API keys and engines. Keys are masked. Groq is required for the full workflow; every other key is optional. The local Whisper engine and built-in system voice provide the no-AI fallback.

Setting	What it does
Groq API key	Required for the full workflow: script rewriting and YouTube Clipper AI moment scoring. Also used for Groq Whisper transcription if selected. Free tier available. Without it, the app uses local Whisper, transcript-as-script, and fixed-length clips.
Kinetra Studio API key	Reserved for the AI Images engine. Kinetra key creation lives in Kinetra Studio → Account → API Keys.
Extra Groq keys	A pool of keys; the pipeline rotates to the next key when one is rate-limited.
Transcription engine	Groq Whisper (cloud, rate-limited) or faster-whisper (local) using the bundled engine.
Local Whisper model	`tiny` (~75 MB), `small` (~466 MB, default), `medium` (~1.5 GB), `large-v3` (~3 GB). Downloads once and is cached.
YouTube Data API key	Only for YouTube Clipper’s “Find viral videos to clip” discovery. Clipping a pasted URL does not need it.
Clips per video	1–30 clips in the YouTube Clipper. The same control is in the Clipper panel.
ElevenLabs API key	Cloud voiceover. Optional; the built-in system voice is free.
Voice engine	ElevenLabs, Qwen3-TTS (local GPU clone), or Built-in system voice (free, no key).
Voiceover voice	The ElevenLabs voice ID. The list loads from your account when a key is set.
Groq chat model	Model used for rewriting and moment scoring. Defaults to a current model; the clipper also tries fallback models.

Exports

Clip Pipeline Settings showing the Local export folder with Choose and Open buttons, plus a Google Drive export Coming soon notice. — Choose a local export folder. Finished pipeline clips and YouTube Clipper outputs are copied there automatically. Google Drive export is coming soon.

Choose opens a folder picker. Open opens the current folder. Leave it blank to keep clips only in the job folder. The folder is created if it does not exist.

Caption removal

Setting	What it does
Blur the caption band	Fast ffmpeg box blur. Safe default for most clips.
Delogo region	ffmpeg interpolation over the caption area. Best on simple backgrounds.
Crop the caption band off	Removes the band entirely; changes the output aspect ratio.
Rebuild / Still Clipper / GPU MAX / GPU inpaint / AI inpaint	Optional heavier engines. These require the VSR/torch or OpenCV setup and are not bundled; leave the step skipped if you have not installed them.
Burn new captions	Burns word-by-word captions timed to the new voiceover onto the final clip.
Region x / y / w / h	Caption band as fractions of the frame (0–1). Default covers the bottom ~27% where captions usually sit.
Script Rewrite Prompt	The system prompt used by Rewrite Script. Edit it to change tone, length or format.

Platform cookies

Export a cookies.txt from a logged-in browser, then use Import file or Paste for that platform. Cookie source can also read live cookies from a closed Chrome profile. Cookies are stored locally in the app data folder.

5. YouTube Clipper

YouTube Clipper panel with URL, clip length 30 seconds, Clips 8, Burn captions checked, colour picker, position slider at 42 percent, and the four clipper engine steps. — The YouTube Clipper. The controls row sets clip count, caption burn-in, caption colour, and vertical caption position. Clip length sets the target duration for each clip.

Control	What it does
Video URL	The YouTube video to clip. The title is fetched without an API key.
Clip length	15s, 30s or 60s. AI moments are clamped to this length; the fixed fallback uses it exactly.
Clips	1–30. The AI is asked for this many moments; remaining slots are filled with fixed clips if it returns fewer.
Burn captions	On: word captions are burned into both outputs. Off: clean exports with no captions.
Colour	Caption text colour. The default is white; the outline stays dark for legibility.
Position	Vertical caption position from 10% to 90% of the frame. Default 42%.
Clip it	Runs download → transcribe → find moments → cut clips.

The clipper engine shows four stages: Download Video, Transcribe, Find Viral Moments, and Cut Clips. Qwen is tried first for moment scoring; other models are fallbacks. If AI finds nothing, it cuts fixed-length clips rather than failing.

YouTube Clipper results showing completed engine stages, one viral clip card with a vertical preview, score 78, 19 seconds, and Vertical and Original buttons. — Results. Each moment renders as a vertical 9:16 clip and an original-aspect clip. Use **Vertical** / **Original** to open them, or **Open folder** for the whole job.

Find viral videos to clip (collapsible at the bottom) needs a YouTube Data API key. It is optional; pasting a URL works without it.

6. No-AI workflow

Clip Pipeline has a no-AI fallback path: the bundled local Whisper engine and the built-in system voice cover transcription and voiceover. These are fallbacks, not the full workflow — for script rewriting and AI viral-moment scoring, add the free Groq API key.

Transcription: with no Groq key, Extract Captions switches to local Whisper automatically. Download the model once; later runs use the cache.
Script: Rewrite Script falls back to the transcript when no Groq key is set. You can also paste your own script and skip Rewrite.
Voice: choose Built-in system voice in Settings. No key, no account; the OS voice speaks the script.
Captions: Attach New Captions uses local Whisper word timestamps when Groq is unavailable.
YouTube Clipper: no Groq key means fixed-length clip selection. No YouTube API key means use a pasted URL instead of discovery.
Caption removal: step 5 uses ffmpeg blur/delogo/crop. No VSR or ComfyUI needed.

What is optional Groq, ElevenLabs, the YouTube Data API key, VSR (GPU rebuild/inpaint), ComfyUI (AI image regeneration), and rclone/Google Drive. The app skips the heavy steps automatically when their engines are missing.

7. Files and troubleshooting

Where files live

Item	Location
Jobs	`Videos/ClipPipeline` (Windows) or `Movies/ClipPipeline` (macOS)
Settings	`%APPDATA%\clip-pipeline\settings.json` (Windows) or `~/Library/Application Support/clip-pipeline/settings.json` (macOS)
Exports	The folder chosen in Settings → Exports
API keys	In the settings file, in plain text. Treat the file as sensitive.

Common issues

Symptom	Fix
App still shows an old build	Close every Clip Pipeline window and relaunch. A running window keeps the old code until restarted.
Local transcription is slow the first time	The model downloads once. Use `tiny` or `small` for speed; watch the Activity Log for progress.
YouTube download fails (403)	Use the Update downloader banner when it appears; the app checks yt-dlp on launch. For logged-in videos, import cookies.
Groq `model_not_found`	The configured model was retired. The app migrates to a current default; moment scoring also tries fallback models.
Groq `json_validate_failed`	The model returned malformed JSON. The clipper retries and falls back to fixed-length clips.
Rebuild or AI Images fails	Those steps need VSR/ComfyUI, which are not bundled. Leave them skipped, or point `CLIP_PIPELINE_VSR_DIR` / `CLIP_PIPELINE_COMFY_DIR` at your installs.
macOS says the app is damaged	The build is unsigned. Right-click the app → Open the first time; if needed, remove the quarantine flag in Finder.

Download Clip Pipeline Back to the product page

Everything in Clip Pipeline, with screenshots.