Video Clipper: long video to platform-ready shorts

What it does

Send it a long video and it returns short, platform-ready clips - not a random trim, but the moments it can defend choosing.

How it picks. It builds a 1-second loudness profile of the audio track and detects visual scene changes, then scores every candidate window in the file on sustained loudness plus how many scene changes fall inside it. Loud, dense windows usually mark a topic switch, a punchline or a reaction. Every clip comes back with its score and the measurement behind it, so you can disagree with a choice and see exactly why it was made.

What you get per clip, in three shapes at once:

• 9:16 1080x1920 - Reels, TikTok, Shorts
• 1:1 1080x1080 - feed
• 16:9 1920x1080 - YouTube

Nothing is cropped away: the frame is fitted inside a blurred bed of itself, so vertical output keeps the whole picture instead of cutting off heads.

Output format. A Markdown report with the start timecode of every clip, its length, the reason it was chosen, and a direct download URL per aspect ratio. Clips are MP4 (H.264/AAC, faststart) - ready to upload.

Input

{ "input": { "video_url": "https://example.com/talk.mp4", "clips": 3, "length": 30 } }

video_url must be a direct link to an MP4/MOV/MKV/WebM file. clips (1-5, default 3) and length (10-90 seconds, default 30) are optional.

FAQ

Does it transcribe speech and pick on meaning? No, and it will not pretend to. It has no speech model, so it selects on loudness and visual change. That finds energy peaks, reactions and topic switches reliably; it cannot tell you which quote is the most quotable. If you need semantic selection, add a transcript and say so in the brief.

What is the maximum source length? Roughly two hours. Processing is roughly real-time divided by twenty, so a 30-minute video takes a few minutes.

What platforms are the aspect ratios for? 9:16 for Reels/TikTok/Shorts, 1:1 for feed, 16:9 for YouTube. All three are produced for every clip.

Does it add captions or music? No. It cuts and reformats only. Nothing is added, so nothing can be wrong.

What if the video has no audio? It falls back to scene-change density alone and says so in the report.