Stop dragging clips.
Start describing the edit.
You talk to an AI in your terminal like it is your editor — “cut the dead air, caption every word, drop b-roll on the keywords, render it” — and it does the timeline work for you. Because video is just data, and AI is great with data. Here is my exact setup. Free.
You type:
❯cut the dead air, caption every word, drop b-roll on the keywords, render it
▶ transcribing · cutting silence > 0.4s · captioning · done ✓
Video is just data. AI is great with data.
So you stop pushing clips around a timeline — and just say what you want the edit to do.
What you actually get
Each one used to be manual, fiddly timeline work. Now it is one sentence. Here is what is happening under the hood.
Cuts the dead air
It transcribes your audio, finds every silence longer than a threshold, and snips it out. The whole thing tightens up automatically.
silences over 0.4s removed · 2:47 becomes 1:58
Captions every word
Whisper transcribes word-perfect, then each word is placed in sync with your voice, in the caption style you set, burned right in.
word-by-word · bottom-center · synced to audio
B-roll on your keywords
Give it a keyword to clip list. Whenever you say the word, it cuts to the matching clip for a couple of seconds, your audio still underneath.
keyword spoken → clip inserted for 2s
Polishes & renders
Whooshes on scene changes, loudness normalized so it is never too quiet or clipping, soft fades, then exported and ready to post.
exporting → out/cut.mp4 ✓
The toolkit
Five tools, and two are optional. Claude Code plus ffmpeg plus Whisper alone gets you auto-cut and captions. Add the rest whenever you are ready.
Claude Code
FreeThe AI editor. It writes and runs the code that does the timeline work — you just talk to it.
claude.com/product/claude-codeWhisper
FreeAuto-transcribes your audio into word-perfect captions, and finds the silence to cut.
github.com/ggerganov/whisper.cppStart here 👇
Grab the first three. That combo already gives you auto-cut and captions — the part that used to eat your afternoon.
From zero to first edit
Install Claude Code
Open it inside a folder. That folder is your project — everything for this video lives in it.
Install ffmpeg
On Mac, run the command below. On Windows, download it from ffmpeg.org.brew install ffmpeg
Download a Whisper model
base.en is a great place to start — small, fast, and accurate enough.base.en
Drop in your recording
Put your raw footage into the raw/ folder so Claude can find it.raw/
Just talk to it
Open Claude Code in that folder and describe the edit. Steal a prompt below.
Keep the folder simple
my-video/ ├── raw/ ← your recording(s) go here ├── broll/ ← stock clips, memes, screenshots ├── audio/ ← music + sound effects ├── out/ ← finished exports land here └── CLAUDE.md ← your style, so edits stay on-brand
Put your preferences in CLAUDE.md — font, colors, “cut silences longer than 0.4s”, “captions bottom-center.” Claude reads it every time, so your videos stay consistent without re-explaining.
The exact prompts I use
Steal these word-for-word. Hit copy, paste into Claude Code, swap the file names for yours.
Cut the dead air + caption it
❯Take raw/recording.mov. Transcribe it with Whisper, cut every silence longer than 0.4 seconds, and add word-by-word captions synced to the audio. Export to out/cut.mp4.
The one step that used to take hours. Auto-cut plus synced captions.
Match b-roll to what I say
❯Whenever I say a keyword from this list, cut to the matching clip in broll/ for 2 seconds: [money -> cash.mp4, error -> glitch.mp4]. Keep my audio underneath.
It drops the right clip on your keywords, voice still playing.
Add sound + polish
❯Add a soft whoosh on every scene change, normalize the loudness to -14 LUFS so it is never too quiet or clipping, and add a 0.3s fade at the start and end.
Consistent loudness, clean transitions, tidy in and out.
Make a thumbnail
❯Generate 3 thumbnail options (1280x720): big bold title on the left, a screenshot on the right, dark background, one bright accent color. Save them to out/.
Three ready-to-pick thumbnails in seconds.
Fix anything
❯The caption at 0:42 is in the wrong spot - move it up 80px. Then re-render just that section.
It re-renders only the part you changed, not the whole video.
How to get great edits
Not literal ones. A few habits that make the difference.
Be specific
"Make it punchier" is weak. "Cut silences over 0.3s and speed the whole thing 1.05x" gets results.
You are the director
Claude is a fast, literal intern. It needs your taste. Review, then tell it exactly what to fix.
Save your style in CLAUDE.md
Font, colors, caption position, silence threshold. It reads them every time so videos stay on-brand.
It is reproducible
The edit is a script. Change one line, re-render the whole video in minutes.
Start small
Nail auto-cut plus captions first. Add motion graphics once that clicks.
Iterate out loud
Do not aim for perfect in one prompt. Say what is off, watch it fix, repeat.
Record two minutes. Paste prompt #1.
Drop a talking-head into raw/, paste the first prompt, and watch it cut the silence and caption itself. That single step used to take hours — and it is why I actually post now.
Made a video with this? Tag @bhuwanlabs — I want to see it. 🙌
Ready to ship
something great?
Book a free 30-minute call. You'll leave with a price range, a timeline, and an honest opinion on whether you need us at all. No pitch deck, no pressure.