YouTube Clip Extractor
ViralMint is an open-source YouTube clip extractor and AI video clipper that leverages local Faster-Whisper AI to transcribe long-form videos on-device, identifies viral segments via LLM analysis, and exports 9:16 vertical shorts with automated captions. By processing heavy data locally, it eliminates per-minute cloud cost — no subscription, no watermark.
What the YouTube Clip Extractor produces
{
"tool": "ViralMint YouTube Clip Extractor",
"input": "youtube.com/watch?v=… URL",
"fetch_backend": "yt-dlp (1080p + auto-captions when available)",
"transcribe_backend": "faster-whisper int8 (word-timestamps)",
"clip_selector": "AI viral-clip picker (hook + retention scoring)",
"max_clips_per_video": 8,
"clip_duration_range_seconds": [15, 60],
"output_aspect_ratios": ["9:16", "16:9", "1:1"],
"captions_burned": true,
"caption_style_presets": ["viral", "classic", "bold"],
"watermark": false
}
Extraction logic in backend/services/clip_extraction_service.py.
See Whisper, ASS captions, and hook in the glossary.
ViralMint vs. Traditional AI Clip Extractors
| Capability | ViralMint (Hybrid Local) | Opus Clip / Submagic |
|---|---|---|
| Processing Cost | $0.00 (Local CPU) | $0.30 - $0.50 per minute |
| Transcription | Unlimited Local Faster-Whisper | Usage-capped Cloud API |
| Data Privacy | On-device (Zero video upload) | Server-side Upload Required |
| Automation | Full MCP / Claude Code Support | Web UI Only |
Technical Performance Benchmarks
ViralMint is optimized for consumer hardware. Our local transcription engine achieves near-real-time results without cloud latency:
What's a YouTube clip extractor?
A YouTube clip extractor takes a long video (typically 10 minutes to 3 hours — podcasts, interviews, lectures, earnings calls) and finds the 30-60 second moments most likely to perform as short-form clips on TikTok, YouTube Shorts and Instagram Reels. The naive way is manually: scrub the video, find a good moment, mark in/out, export, caption. The AI way is: a model reads the transcript, scores every candidate segment for "viral fit", and you pick from the ranked list. The good clip extractors handle the transcription, scoring, cutting, reframing to 9:16, and captioning in one pipeline.
ViralMint's Clip Studio runs the entire pipeline locally for the heavy parts (yt-dlp download, Whisper transcription with word-level timestamps, FFmpeg cutting and reframing, ASS caption rendering) and only routes the AI viral-clip picker to the cloud. That means a 90-minute podcast doesn't cost you the per-minute transcription fees Opus Clip and Submagic charge (making ViralMint a powerful **free Submagic alternative no watermark**) — it costs the same as a 5-minute one because the AI call is just over the transcript text, not the audio. Typical end-to-end cost on a 90-minute video is under 20 cents.
How the clip extractor works
- Paste a YouTube URL. Any public video. Works on long-form podcasts, interviews, lectures, earnings calls, news segments, channel back-catalog — anywhere the long-form content has discrete moments worth surfacing.
- Download. yt-dlp pulls the video locally to your desktop. Universal downloader works on YouTube, YouTube Shorts, TikTok, Douyin, Bilibili, X (Twitter), Instagram, Reddit, Vimeo, Twitch and 1800+ other sites.
- Whisper transcribes locally. faster-whisper runs on your CPU (int8 quantized) — no cloud cost, no per-minute pricing. Word-level timestamps mean every clip's caption sync is exact, not the 2-second drift typical of cloud transcribers.
- AI viral-clip picker scores segments. The transcript gets scored segment-by-segment for hook strength, tension structure, emotional peaks, and standalone interpretability. Cloud AI call (gpt-5.4-mini or Sonnet 4.6 depending on length); typical cost is a few cents per long video.
- Export. Top-scoring segments cut to standalone mp4s, reframed to 9:16 (with face-tracking when there's a person on screen), captioned word-by-word with the viral / classic / bold preset of your choice. Bulk export of N clips at once is a ZIP download.
Why ViralMint's clip extractor is different
Local Whisper transcription
faster-whisper int8 runs on your CPU — no per-minute transcription fee, no audio upload to a 3rd-party. A 90-minute video costs the same to transcribe as a 5-minute one.
AI viral-clip picker
Scores every 30-60s segment on hook strength, tension structure, emotional peaks, standalone interpretability. Top-N exported. You can override the picks if you have specific moments in mind.
Face-tracking reframe
16:9 → 9:16 reframe keeps the speaker centered. OpenCV face detector + smooth-track interpolation means the face never jumps even when the speaker moves across frame. Manual frame-by-frame fixup available if needed.
Word-by-word captions
Whisper word-level timestamps + ASS subtitle rendering. Three presets — viral (yellow highlight, Montserrat Bold 56pt), classic (Arial 42pt), bold (Impact 64pt green). Customizable per channel.
Universal downloader
yt-dlp handles 1800+ sites — YouTube, YouTube Shorts, TikTok, Douyin, Bilibili, X, Instagram, Reddit, Vimeo, Twitch, Dailymotion, Facebook. Same pipeline for non-YouTube sources.
Bulk export
Generate 5, 10, or 20 clips at once from a single source video. Bulk export bundles all clips into a ZIP with AI-drafted Shorts titles and TikTok captions ready to copy-paste when posting.
How it compares to Opus Clip and Submagic
Opus Clip and Submagic are the two best-known YouTube clip extractors — both run their entire pipeline (download, transcription, AI scoring, captioning) in the cloud and charge subscription pricing per month with caps on minutes processed. Opus Clip's Starter plan is $19/mo for 60 minutes; Submagic's Pro plan is $23/mo for similar limits. For high-volume users (50+ long videos per month), the subscriptions can be cost-effective; for everyone else, the per-minute charge structure penalizes occasional use.
ViralMint's hybrid local + cloud architecture inverts that economics. The download (yt-dlp), transcription (Whisper), cutting (FFmpeg), reframing, and captioning all run locally with no per-minute cloud charge. Only the AI viral-clip picker is a cloud call, and it's cheap because it runs over the transcript text, not the audio. Typical end-to-end cost on a 90-minute podcast is under 20 cents at ViralMint; comparable run on Opus Clip Starter consumes ~1.5x your monthly minute allotment. For comparison detail, see /alternatives/opus-clip/ and /alternatives/submagic/.
Frequently asked
How does ViralMint's YouTube clip extractor pick viral moments?
ViralMint's Clip Studio transcribes the video with Whisper locally (word-level timestamps, no cloud cost), then runs an AI viral-clip picker over the transcript that scores each 30-60 second segment on hook strength (does the opening line earn the next 5 seconds?), tension structure (problem → setup → payoff), emotional peak density, and standalone interpretability (does the clip make sense without the surrounding context?). The top-scoring segments get cut, reframed to 9:16, and exported as mp4s with word-by-word captions burned in. The whole pipeline runs from a single YouTube URL.
How much does the YouTube clip extractor cost?
The local pipeline has no per-video cost: downloading (yt-dlp), transcription (Whisper, local), clipping (FFmpeg), reframing to 9:16, and caption burn all run on your desktop. The AI viral-clip picker uses cloud AI billed per use via prepaid top-ups (new accounts get a daily starter allowance to evaluate) — typical cost per long video is under 10 cents. Compare to Opus Clip at $19/mo subscription or Submagic at $23/mo: ViralMint is pay-per-use, no subscription, and works on private/ unlisted videos you have access to.
Can I extract clips from a YouTube video I didn't film?
Technically yes — ViralMint's yt-dlp downloader works on any public YouTube URL, and the AI viral-clip picker doesn't care who filmed the source. But: respect copyright. Clipping someone else's video for fair-use commentary (reaction, criticism, education) is generally permissible; reuploading their footage as your own original short is not. ViralMint also supports public-domain footage, your own filmed content, and Creative-Commons-licensed videos. The tool is the same; the ethics are your call.
What about clip extraction from non-YouTube videos?
Same pipeline. yt-dlp supports 1800+ sites including TikTok, Douyin, Bilibili, X (Twitter), Instagram, Reddit, Vimeo, Twitch, Dailymotion and Facebook. Paste the URL from any of those and Clip Studio runs the same download → transcribe → score → export pipeline.
How long can the source video be?
ViralMint caps source videos at 2 hours (7200 seconds) for the download step — that covers most podcasts, interviews, lectures and earnings calls. Whisper transcription scales linearly with length; a 90-minute video takes 8-15 minutes to transcribe on a typical laptop CPU (int8 quantized base model). The AI viral-clip picker runs on the transcript text, so it's fast regardless of source length.
Extract your first clip
Sign up takes 30 seconds. Clip extraction ships in the open-source desktop app — the heavy pipeline (download, Whisper, FFmpeg, reframe, caption burn) is all local.