Upload
Drop in a VOD or a recording, up to 4 GB. It streams straight to disk, so long files don't stall.
Trisphere · Lenz
Long streams in.
Short clips out.
Lenz turns a stream, a stream clip or a screen recording into vertical shorts. It points you at the moments worth cutting, frames the speaker for a phone screen, writes the captions and publishes to YouTube. It runs on our own server, so your footage never goes to an outside AI service.
01 How it works
Footage in.
A post out.
Every clip keeps up to three versions side by side: the original cut, a vertical one and a captioned one. You choose which one to post.
- Ingest Uploads · SheekBot clip links · A watched folder on your Mac · A token-protected upload API Up to 4 GB per file
- Highlights The audio track is scanned for loud, energetic stretches and turned into suggested clips you can trim Suggestions, never automatic cuts
- Reframe Landscape footage is cropped to 9:16 around the speaker's face, with a smoothed pan instead of jump cuts 1080 × 1920 out
- Captions Speech to text with word timings, split into short cues and burned into the picture The .srt file is yours too
- Publish Straight to YouTube with title, description and tags · Download for Instagram and TikTok YouTube uploads start unlisted
02 Editing
The parts that take the longest by hand.
Finding the moment, framing it for a phone and captioning it are the slow part of short-form. Lenz does a first pass on each and leaves the call to you.
Suggested clips from the audio
Lenz measures the loudness of the audio in two-second windows and ranks the stretches that stand out: the laugh, the shout, the clutch moment. Pick a suggestion and it fills in the start and end marks, ready to trim.
- Available Ranked suggestions with a score per clip
- Available Trimming in the browser, with your own start and end marks
- Roadmap Chat activity from SheekBot as a second signal
A 9:16 crop that follows the speaker
Face detection runs five times a second across the whole frame. The largest face wins, so a facecam beats a face on screen, and the crop eases toward it instead of snapping. With no face in the shot, Lenz uses a centre crop and tells you so.
- Available Landscape to 1080 × 1920 with the original audio
- Available Captions re-rendered onto the new crop automatically
- Roadmap Split layout: facecam on top, gameplay below
Short, word-timed captions
Speech is transcribed on our own server with word timings, then split into cues of about four words so they stay readable on a phone. They're burned into the vertical version, in the lower third, above the app's buttons.
- Available Burned-in captions plus an .srt download
- Available No transcription service or per-minute fees
- Roadmap Editing the transcript before it's burned in
03 Getting footage in
From wherever it was recorded.
Streams happen on one machine, screen recordings on another. Lenz takes footage from each without a special app.
SheekBot clips
Paste the link from a chat !clip and Lenz pulls the video. Only SheekBot clip links are accepted.
A watched folder
A small script on your Mac uploads every recording you save into one folder, once the file stops growing, then files it away.
Upload tokens
Scripts sign in with a token that can upload footage and nothing else. Revoke one any time; only a fingerprint of it is stored.
04 Specifications
The details.
| Input | MP4 · MOV · MKV · WebM · M4V · Up to 4 GB per file |
|---|---|
| Output | 1080 × 1920 H.264 with AAC audio · Original, vertical and captioned versions per clip |
| Highlights | Audio loudness in two-second windows, ranked into suggested clips |
| Reframe | Face detection five times a second · Largest face · Smoothed pan · Centre crop when no face is found |
| Captions | Whisper speech-to-text on our server · Word timings · About four words per cue · Burned in, plus .srt |
| Publishing | YouTube with title, description and tags, unlisted by default · Instagram and TikTok: download and post while their app reviews are pending |
| Ingest | Browser upload · SheekBot clip links · Mac folder watcher · Upload API with revocable tokens |
| AI | Speech and face models run on our own server. Footage is never sent to an outside AI service. |
05 Early access
Clip your next stream with it.
Lenz cuts Trisphere's own videos today. If you stream or post short-form and want in early, tell us where you record and where you post.
More from Trisphere: Aram · Bunnie · Vandelay · SheekBot · All products