You can turn the Hermes desktop app into a full video editing workbench that trims dead air, slices long recordings into clips, and even edits based on what is happening on screen, all from a single prompt. The heavy lifting runs locally with two free tools, FFmpeg and Melt, so most edits cost zero tokens. You can even kick off an edit from Telegram while you are away from your computer. Here is exactly how to set it up.
Core Concepts
- What FFmpeg and Melt each do, and when Hermes reaches for each one
- How Hermes detects and trims silence automatically, no timeline scrubbing required
- How visual frame analysis edits by what is on screen, not just gaps in the audio
- Why most edits cost no tokens, and what the roughly five-cent visual analysis actually buys you
- How to run edits from the desktop app or remotely from Telegram
- How Hermes auto-generates titles, descriptions, timestamps, and tags for each clip
Who does this apply to
Creators, marketers, and founders who record screen walkthroughs, tutorials, or product videos and are tired of manually cutting dead space and chopping long recordings into social clips. If you would rather film, walk away, and come back to finished clips, this is for you.
Your AI agent becomes the video editor
It means your AI agent becomes the editor. Hermes already has access to the files and folders on your operating system, so you point it at a folder, tell it what you want, and it trims, stitches, and renders the video in place. There is no separate editing app to open and no timeline to scrub by hand.
The workflow is dead simple: drop a recording into a folder, tell Hermes which folder it is working in, and let it run. This builds directly on the Hermes desktop app that no longer needs a terminal, which is what makes file-based editing feel native instead of technical.
The setup: two free tools and one optional upgrade
Three things, plus one optional upgrade:
- Hermes, installed on your machine
- FFmpeg, the engine for fast cuts on shorter clips (ffmpeg.org)
- Melt, part of the MLT multimedia framework, for longer recordings (mltframework.org)
- Optional: a vision-capable model like a Codex subscription or a cheap Gemini 2.5 Flash for visual frame analysis
Installing is as easy as giving Hermes a single command and letting it build the skill out. Once the skills are baked in, editing is just a prompt.
Want to follow along step by step? I built an interactive Hermes video editing dashboard that walks through the full setup and commands.
From experience: the one snag with Melt is timestamps. It took me a weekend to get it reading time correctly, because it can misread a value like 15.8 seconds as frame 15 instead of the actual time in hours, minutes, and seconds. Once that instruction is baked into the skill, you never have to think about it again, but it is the kind of detail that will quietly ruin an edit if it is missed.
Short clips are FFmpeg's job
For a short clip, Hermes probes the video first, detects the silences, and trims the dead air, with every command running behind the scenes. In a live walkthrough, it probed a 95-second clip, flagged it as short enough for the fast path, and detected the silent stretches automatically.
The result: a clip that started at 1 minute 35 seconds came back trimmed by about 20 seconds, with clean cuts where I had been standing around opening files instead of talking. One prompt, one sentence, a couple of minutes of background processing, and the trimmed file lands in the same folder as the original.
Longer recordings are where Melt takes over
This is the most important technical detail, and it explains why you want both tools installed. FFmpeg builds the entire edit in memory, every frame and every audio sample at once. For a short 720p clip that is fine. But screen recorders capture in variable frame rates, so the memory usage spikes and the edit can fall over on longer files.
Melt works differently. Instead of building one giant filter graph in memory, it builds a timeline where each segment is a separate clip reference with an in-point and an out-point. The timeline streams through the encoder sequentially, so there is no memory buildup. That is why FFmpeg is great for clips under a minute, while Melt is what you want for anything five minutes and up.
The payoff of installing both: in that same demo, FFmpeg started scanning a 120 FPS clip, hit a snag, and Hermes automatically handed the job off to Melt to finish. The two tools work in tandem, so you recover gracefully no matter what the source file throws at you. And you do not lose quality, the audio stays synced and the output is as clean as the recording.
Visual frame analysis edits by what's on screen
Silence detection tells Hermes when to cut, but it does not tell it whether the result is a clean cut. Picture a cooking video where you are mid-motion but not talking, like melting gelatin for a batch of gummies. Silence-based editing might cut right through the money shot.
Visual frame analysis solves that. Hermes maps the frames into a temporary folder and sends the images to a vision model (a Codex subscription, or a cheap model like Gemini 2.5 Flash) to actually watch the footage and decide where the clean cuts are. It is genuinely looking at each frame, not guessing from the audio. This is the only step that touches tokens, and it is what lets Hermes pull the best three or four clips out of a 40-minute recording, even stitching two moments together when that makes a better clip.
How much does AI video editing with Hermes cost?
The trimming, stitching, and rendering all run locally on your machine, so that work costs nothing in tokens. The only paid step is the visual frame analysis, and even that is cheap, around five cents a video on a model like Gemini 2.5 Flash.
That economy is the same reason I am bullish on agent workflows in general, the same way I broke down running multi-job AI automation pipelines for pennies a day. You can even orchestrate the whole thing from a free chat model, since the local editing does not need a premium model, you only route to a vision model when you want scene-by-scene analysis.
Can you edit videos from Telegram?
Yes, and this is the feature that changes how you work. Because Hermes lives on your computer and can be reached through Telegram, you can record a video (or have someone send you one), then message Hermes to edit it on your desktop while you are out. Film a walkthrough, go grab lunch, and message: "edit down the video in my XYZ folder into 10 clips." You come back to finished clips. No sitting at the keyboard cutting clip by clip.
From raw footage to social-ready clips
The last piece of the skill is metadata. Since these clips are headed for social, Hermes generates the package for you from the same vision analysis: three title variations, a description, timestamps, and tags. So a single 40-minute recording becomes a folder of trimmed clips plus ready-to-post metadata, all without opening a traditional editor. You might touch a couple of clips up at the end, but the grunt work is done.
FFmpeg vs Melt vs Visual Frame Analysis: which should you use?
| Tool | Best for | How it works | Token cost |
|---|---|---|---|
| FFmpeg | Short clips under ~1 minute, fast cuts | Builds the entire edit in memory at once | None (runs locally) |
| Melt (MLT) | Long recordings, 5 minutes and up | Streams a timeline of in/out points sequentially, handles variable frame rates | None (runs locally) |
| Visual Frame Analysis | Clean, scene-aware cuts and clip selection | A vision model watches the frames to decide where to cut | ~5 cents per video |
The short version: keep both FFmpeg and Melt installed so Hermes can fall back automatically, and layer in visual analysis only when you need scene-aware cuts rather than simple dead-air trimming.
Frequently Asked Questions
Do I need both FFmpeg and Melt installed?
You can get away with Melt alone, since it handles both short and long videos. But installing both lets Hermes use FFmpeg for speed on short clips and fall back to Melt automatically if FFmpeg hits a snag, so nothing breaks mid-edit.
Does editing this way reduce video quality?
No. The audio stays synced and the output is as clean as the original recording. Melt streams the timeline through the encoder rather than rebuilding everything in memory, so there is no quality loss.
Can Hermes edit video without any paid AI model?
Yes. The trimming, stitching, and rendering run locally and cost nothing. You only need a paid or subscription model for visual frame analysis, and even a cheap model like Gemini 2.5 Flash works for around five cents a video.
How long does a long edit take?
It varies by length and hardware. A 16-minute recording with about 40 seconds of dead air took roughly 20 minutes to render on Apple silicon in the demo, all in the background while other tasks ran.
If you want help building AI-driven content systems like this into your own workflow, that is exactly the kind of work I do at Jason Pollak Marketing.
About Jason Pollak
Jason Pollak is a marketing strategist with over 10 years of experience building campaigns for entertainment brands, artists, and businesses across music, film, television, eCommerce, and B2B SaaS. As Director of Marketing at Young Money Entertainment, he grew Lil Wayne's Facebook following from 10 million to 50 million and managed over 60 million followers across the roster. He also served as Paid Media Director at Horizon Media, launching major TV shows for History Channel, A&E, WWE, and Lifetime, and led film marketing for Utopia Distribution, generating over $10 million in revenue on a $200K media spend. Jason specializes in paid media, organic social strategy, email automation, SEO, content development, and AI-driven marketing systems. He holds a BA in English Literature from Binghamton University and a Masters in Media Studies from Brooklyn College. Learn more at jasonpollakmarketing.com.