19 September 2026
Meet Pleofon: An AI Rough Cutter That Actually Listens
Most AI video editors start with a transcript. Pleofon starts with the recording itself. Built for long-form talking-head videos, essays, podcasts and other speech-heavy content, Pleofon handles the tedious first cut while keeping you in control of the edit.
There is a part of video editing that nobody puts in the thumbnail. You have already done the research. Written the script. Set up the camera. Recorded the video. Maybe repeated the same sentence six times because one word sounded wrong. And now you are staring at an hour of raw footage that somehow needs to become twenty minutes. Before the interesting editing can even begin, someone has to remove the failed takes, repetitions, awkward pauses, filler, silence and all the other material that was never supposed to make it into the final video. That is the problem Pleofon was built to solve.
The boring first cut
For long-form creators, the rough cut can take an absurd amount of time. Not because the decisions are always creatively difficult. Quite often, they are obvious. The difficult part is simply going through all of it. Again. And again. And again. There are already plenty of tools promising to automate this process. We tried many of them. The problem was not just whether they removed the correct material. It was how they cut it. A technically correct edit can still sound terrible. Leave a little too much silence and the pacing feels slow. Cut a few frames too aggressively and a natural breath disappears. Place the boundary slightly too late and the beginning of the next word gets clipped. A few dozen milliseconds can change whether an edit feels invisible or immediately sounds automated. That became the starting point for Pleofon.
What if the model actually listened?
Most automated editing workflows rely heavily on speech recognition. The recording becomes a transcript. Words receive timestamps. A system decides which words or sentences should disappear, and the video is cut around those timestamps. Transcription is useful. Pleofon uses it too. But it is not the foundation of the edit. Pleofon's custom models analyze the recording itself. They look at speech, pauses, fillers, retakes, silence and the rhythm of the person talking. The transcript can inform a decision. It does not make the decision. That distinction matters because editing speech is not only about understanding what somebody said. It is also about hearing how they said it.
Meet Guya 4
Guya 4 is the core cutting model inside Pleofon. Its job is deceptively simple: Find the right place to cut. Instead of applying one universal silence threshold to an entire recording, Guya evaluates individual boundaries. Sometimes a breath should stay. Sometimes a pause should be shortened. Sometimes the cleanest cut sits directly against the next word. Sometimes it does not. Guya was trained on real recordings paired with edited versions of those same recordings. That means it could learn from actual editing decisions rather than a collection of handcrafted rules about how long a pause is "supposed" to be. The result is a first cut designed to sound edited by a person.
And then there is Chase
Not every recording follows a script. Vlogs, podcasts, conversations and unscripted videos can contain several attempts at the same idea, tangents and entire fragments that technically make sense but probably do not belong in the final video. That is where Chase comes in. Chase joins Guya in Pleofon's Freeroam mode. Guya still handles the precise boundaries. Chase helps decide which larger pieces of the recording can disappear. You control how aggressively it works. That gives Pleofon two distinct workflows.
Scripted
Scripted mode takes a conservative approach. It removes obvious mistakes, pauses, fillers and failed takes while preserving nearly all of the actual content. It is designed for narration, video essays, documentaries and other material where the structure has already been decided. No actual script is required.
Freeroam
Freeroam is more adventurous. Guya and Chase work together to identify repetitions, alternate takes, tangents and other material that may not need to survive the first cut. A simple control lets you decide how aggressively the footage should be shortened.
We do not pretend the AI is perfect
This is an important part of Pleofon. Automated editing is not useful if fixing the automation takes longer than editing the footage yourself. So Pleofon was designed around the assumption that the models will occasionally make the wrong decision. If a cut is too tight, move it. If the model removed something important, restore it. If it left something behind, delete it yourself. Boundaries can snap to words and silence. You can change pacing globally. You can work directly on a traditional timeline or use text-based editing alongside it. The goal is not to remove the editor from the process, but to remove as much repetitive work as possible before the editor has to step in.
How much work does that actually save?
We wanted to measure that in a way that mattered. Not model accuracy, or other abstract score. Work. We gave the same raw recordings to several automatic editing tools and compared their results against a human edit. Then we counted how many corrections were required to turn each automated result into that human edit. On our harder four-minute test recording, Pleofon required 17 corrections. Gling required 36. Detail required 38. Eddie required 42. Descript required 48. On a longer 24-minute recording, Pleofon required 135 corrections compared with 303 for Gling. It is not a perfect benchmark for every creator, every language or every recording style. But it measures the thing we actually care about: How much work is still waiting for you after the AI is finished?
Local first
Pleofon's models run on your computer. That means you do not have to upload gigabytes of raw footage before analysis can even begin. There is no remote model processing every minute of your recording. Your source file stays where it is. This also means Pleofon can offer unlimited analysis in the Pro plan without turning every extra hour of footage into another cloud-compute bill. And if you do not participate in the Pleofon Improvement Program, your project media stays on your device.
Keep using your real editor
Pleofon is not trying to replace Premiere Pro, DaVinci Resolve or the rest of your editing workflow. It is trying to get you through the part before the interesting work begins. Once the rough cut is ready, you can render the finished media directly or export the timeline as XML and continue editing in your normal NLE. Add your graphics. Do your sound design. Color the footage. Spend your time on the parts that actually make the video better.
Pleofon is now in open beta
Pleofon is currently available for both Windows and macOS. The Free plan includes the complete application with up to 120 minutes of analysis every month. Pleofon Pro costs $19.99 per month and removes the analysis limit entirely. The models were trained primarily on Polish and English recordings, so results in other languages may currently vary. And this is still a beta. There will be bad cuts. There will be strange recordings we have never encountered before. There will probably be things we did not even know could break until somebody feeds them into Pleofon. That is exactly why we are releasing it. The next generation of these models should not learn from one editor, one voice or one style of video. It should learn how people actually edit.
Your first cut. Maybe your last.
Pleofon started with a very simple frustration: Why are we using increasingly powerful AI to generate more content while humans are still spending hours manually removing the most repetitive parts of recordings? The interesting part of editing is deciding how a video should feel. It is not finding the fifth failed take of the same sentence. That is the part we want Pleofon to handle. Download the open beta for Windows or macOS at pleofon.ai. Give it an hour of footage. And see how much of that first cut you never have to make yourself.