Zypherius
The speaker drifts out of frame
A 16:9 shot cropped to 9:16 loses whoever moves. Face tracking drives the crop across the clip instead of freezing it in the centre.
Built for talking heads and long episodes
Your episode, cut into clips people finish
A podcast is the hardest thing to clip by hand: an hour of two people talking, three moments worth posting, and no visual cue to find them. Zypherius reads the transcript instead of the waveform, so the cut lands on what was said, not on where somebody laughed loudly.
One job costs 10 credits — the same for a 12-minute solo episode and a 58-minute interview.
What it does for a podcast specifically. The whole episode is transcribed word by word, then scored for moments that hold attention — a claim, a story, a disagreement. Silence and filler words are removed, the 9:16 crop tracks the face so a guest who leans back is not cropped out, and subtitles are burned in word by word because most of the audience will watch the clip on mute.
The four things that break podcast clips
Dead air and filler
Pauses and filler words are cut before the render, so a 40-second answer becomes a 30-second clip that keeps its meaning.
Nobody has the sound on
Subtitles are burned in and highlighted word by word in the language spoken, so the clip works on mute in a feed.
Guessing which clip to post
Every clip carries a Hook Score from 0 to 99 with a one-line reason, so the order you publish in stops being a coin toss.
A realistic episode run
Upload the episode as it came out of your recorder — up to 60 minutes and 500 MB, in MP4, MOV, AVI, WEBM, MKV or M4V. Anything past 30 minutes is analysed in overlapping windows, so a strong moment at minute 51 is found as reliably as one at minute 4.
A typical hour-long interview returns two to four clips. That is not a limitation dressed up as a feature: candidates that overlap each other by more than half are dropped, and what is left is what the transcript supports. Two clips cost 10 credits, three or more cost 15.
When a clip is nearly right, open it: retype a misheard surname, tap a pause to cut it, rewrite the hook line at the top. Re-rendering costs nothing, so the third attempt is as free as the second.
Video podcast or audio only?
This tool cuts video. If your episode is a video recording — a camera, a Riverside or Zoom session, a stream — everything above applies. If it is audio only, there is nothing to crop or track, and you would be posting a static frame with subtitles, which the pipeline is not built to fake.
Two speakers in one frame work well. Separate speaker tiles in a grid work too, though the crop will favour the face it can track most confidently rather than switching between tiles on every line.
Podcast-specific questions
How long can an episode be?
Up to 60 minutes and 500 MB per upload. Longer episodes need to be split before uploading.
Does it handle two people on camera?
Yes. Face tracking picks the speaker it can follow most reliably and moves the 9:16 crop across the clip. It does not cut back and forth between two tiles line by line.
What language are the subtitles in?
The one being spoken. The transcriber detects it, and hooks and captions are written in that same language rather than translated into English.
How much does an episode cost to clip?
10 credits for a job returning one or two clips, 15 for three or more. Credit packs start at EUR 5 for 100 credits, and credits do not expire.
Can I get a 16:9 version for YouTube as well?
Yes — pick 16:9 or 1:1 instead of 9:16 before the job runs. Face tracking is applied on the vertical crop, where it matters.
How long do I have to download the clips?
Seven days. After that the file is deleted and the interface shows an expired state instead of a dead link.
Try it on the video you were about to clip by hand
Ten credits are waiting in a new account — one full clip job, no card, no watermark on what comes out.