AI clipping studio
One podcast in.
A week of clips out.
Claudevid transcribes your long videos, finds the moments people actually share, renders them as vertical clips with captions, templates and music — and posts them to every platform on schedule.
12:40
31:05
54:22
1:12:10
1:26:44
How it works
From link to published in three moves
Drop a link or a file
YouTube, Twitch VODs, webinars, raw uploads up to 4 hours. We pull the source at up to 1080p and probe it in seconds.
Audio-first analysis
We extract a tiny 16 kHz audio track, transcribe it, score every sentence locally (energy, hooks, pace) and make one compact AI pass to pick the winners. No frame-by-frame vision models burning money.
Render & publish everywhere
ffmpeg renders each clip with animated captions, your template and ducked background music. Connect YouTube, TikTok, Instagram, Facebook and X once — then publish or schedule in one click.
Templates
Five looks. Zero editing.
Every template is rendered server-side in a single ffmpeg pass — no browser rendering, no watermarked previews that lie to you.
TELLS YOU
Impact
Full-frame vertical crop with big karaoke captions. The classic viral look.
TELLS YOU
Blur Pad
Original framing preserved over a blurred, zoomed echo of itself.
TELLS YOU
Split Stack
Speaker on top, hypnotic B-roll below. Built for retention.
Letterbox
Cinematic black bars with a bold headline on top.
Focus Frame
Floating video card on a dark gradient stage. Podcast favourite.
Engineering
Cheap on purpose. Fast by design.
Most clipping tools push entire videos through vision models. Claudevid doesn't. The pipeline is built to spend as close to zero AI tokens as possible:
Audio, not frames
A 2-hour video becomes a ~14 MB mono Opus file before anything touches an AI model.
Local pre-scoring
Loudness peaks, hook keywords, speech pace and questions are computed on-device for $0.00.
One compact AI pass
Only the top ~55% of transcript lines, with integer timestamps, in a single call that returns every clip at once.
Heuristic fallback
No API key? The local analyzer still finds strong moments — zero tokens, zero cost.
pipeline
ffmpeg -vn → opus 16k mono
2h video → ~14 MB
whisper turbo transcription
~100x realtime
local scoring (rms + hooks)
energy · pace · questions
1 compact LLM pass
top 55% lines only
ffmpeg render × N clips
veryfast preset, one pass
≈ $0.05 per hour of source video with Groq Whisper + a small LLM. Not per clip — per hour.
Distribution
Post once. Land everywhere.
OAuth once per account. Claudevid handles token refresh, per-platform captions, hashtags, privacy levels and scheduled publishing — including YouTube's native premiere scheduling.
FAQ
Do I need my own AI API keys?+
For the hosted version, no — analysis is included in your plan. Self-hosting? Bring a Groq or OpenAI key, or run whisper.cpp locally and pay nothing.
Which platforms can Claudevid post to?+
YouTube Shorts, TikTok, Instagram Reels, Facebook Reels and X. Scheduling is supported everywhere; TikTok requires an audited app for public posts.
How long does a video take to process?+
Transcription runs about 100x realtime and each clip renders in seconds on our workers. A 1-hour podcast usually delivers clips in under 5 minutes.
Do clips have a watermark?+
Free plan clips carry a small claudevid.com mark. Creator and Agency plans render clean 1080×1920 files.
Your next 30 days of content are already recorded.
They're sitting inside videos you already published.
Start clipping free