09:38100x Engineers · Tutorial
Claude Edited This Video (And I'll Teach You How)
A six-step checkpoint workflow for getting Claude to produce motion-graphics video edits, from style references to a production-ready cut.
October 20thThe argument in one line.
AI video editing works when you split the job into small, checkable steps, save each as a reusable skill, and keep a human reviewing the result instead of asking for a finished video.
Read if. Skip if.
READ IF YOU ARE…
- A YouTuber or solo creator who edits their own talking-head videos and loses hours to cutting retakes and dead air.
- Someone already using Claude or ChatGPT who wants a concrete, repeatable workflow for scripts, rough cuts and motion graphics.
- An editor using DaVinci Resolve or CapCut who wants to know where AI can take over the first rough cut safely.
SKIP IF…
- You want a one-click tool that outputs a finished, polished video with no review.
- You need step-by-step setup instructions for a specific app; the video shows ideas and demos, not full tutorials.
The full version, fast.
Instead of asking Claude for a finished video, split editing into six jobs: scripting, editing, design, motion graphics, review and feedback, and save them as a skill that improves with use. Whisper gives word-level timestamps so Claude keeps the last clean take and FFmpeg cuts the rest, producing an editable CapCut or DaVinci timeline. Build a beat map before any graphic, preview drafts at half size, make motion graphics adjustable with sliders, and have Claude check its own work frame by frame. A first rough cut drops from 30 to 50 minutes to about five, but a human still verifies everything.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Lines worth screenshotting.
- Don't ask AI for a finished video; give it six separate jobs one at a time and save them as a skill.
- A skill is a saved set of instructions for one job, and it gets better each time you use it.
- Whisper timestamps every word, which lets Claude keep your last clean take of each line and cut the rest with FFmpeg.
- Have AI build a real editable timeline in CapCut or DaVinci, not a flat video you can't change.
- Without a definition of good, AI motion graphics default to boring PowerPoint-style slides.
- Screenshot editors you admire and ask Claude what makes their text and transitions work before it designs anything.
- A beat map lists every idea in the script and what's on screen; changing a plan is nearly free, changing a finished graphic is not.
- Adding timing sliders to a motion graphic composition lets you tweak stagger, speed and glow without regenerating anything.
- Code-generated motion graphics preview far faster than pixel-generated AI video.
- Duplicate the timeline and restyle the graphics to match a page's fonts and colors so the video looks on-brand.
- Save hand-tuned animation values into a preset so Claude follows the recipe instead of improvising each time.
- Render drafts at half size while giving notes, and only do full-quality renders after approving the whole cut.
- Build one graphic as a webpage first, storyboard the rest, then turn it into video to save time and tokens.
- Make Claude review its own work by sampling frames, checking for repeats, late text, graphics on your face and lip-sync drift.
- Write down the three mistakes you always catch in your own videos and hand that checklist to the AI.
- A rough cut that took 30 to 50 minutes can drop to about five, but the value is offloading boring tasks, not replacing an editor.
Split editing into small checkable jobs
WHAT TO LEARN
AI editing pays off when you give it narrow jobs, a plan before any graphics, and a review checklist, while you still verify the final cut.
- Give the AI six separate jobs (script, edit, design, motion, review, feedback) and save each as a skill so it improves with every video.
- Transcribe with word timestamps so the AI can keep your last clean take of each line and list everything it removed.
- Ask for a real editable timeline in CapCut or DaVinci so any tweak stays possible after the AI finishes.
- Show the AI examples of designs you admire and have it write a beat map before building any graphic.
- Expose timing and style controls as sliders so you can adjust motion graphics without regenerating them.
- Preview at half size and prototype graphics as a webpage or storyboard first to save time and tokens.
- Hand the AI your personal list of recurring mistakes and have it check frames the way a person would.
- Treat the result as a first rough cut that saves 30 to 50 minutes, and keep verifying facts and b-roll yourself.
Terms worth knowing.
- Skill
- A saved set of instructions that teaches Claude how to do one job, such as editing video, and can be improved over time.
- Whisper
- A free speech-to-text model that transcribes audio and gives a timestamp for every word, which makes cutting by transcript possible.
- FFmpeg
- A free command-line tool for cutting, converting and assembling video and audio files.
- Beat map
- A plan listing every idea in a script and the graphic or visual shown on screen while it is spoken.
- Rough cut
- The first pass of an edit that removes bad takes and dead air before polish, graphics and color.
- Preset
- A saved bundle of animation values, layouts and style rules that Claude reuses so each video follows the same recipe.
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
The captions, the motion graphics, the cards flying out from behind me. I didn't touch an editing timeline once. This entire video was edited by Claude.
Not just the intro, but the entire YouTube video you're watching right now. So here's the part that you can copy today. I stopped asking Claude for a finished video and I gave it six jobs, one at a time.
Scripting, editing, design, motion graphics, review, and feedback. Then I saved all six into a skill. A skill is a safe set of instructions that teaches Claude how to do one job and mine gets better every time I use it.
So try this. Ask Claude for the hooks and format from three videos in your niche that beat their channel's usual views. Save them in a library and then have it write your scripts.
the same way. When the script is done, you read it to the camera, which is exactly what I'm doing right now. Next comes the footage and raw footage is a mess.
You trip over a word, restart a sentence or sit in silence. So give Claude your raw footage and ask it to transcribe it using Whisper. It's going to create a timestamp for every single word that you said, and now Claude understands how to do the edit.
It's going to keep your last clean take of every line and cut the rest using another free program called FFmpeg. So try this, film one short take with a few mistakes on purpose, give quad the file and ask it to transcribe it with whisper, cut the retakes and list everything it removed. What's cool is that this isn't just an AI video that you can't change.
It's built in a real CapCut timeline, meaning that we can make any tweak we want. Do not use computer use or open CapCut. Build the timeline by writing the project files directly through file access to my Mac.
And you can see that this is the fully edited timeline. It's got all the sound effects. If we click on a clip, it's applied key frames so that we get zoom on our clips and it's added in text files.
It's also found a music file directly from our computer. Another thing you can do inside of roomie is set up these layouts. So when I click on layout, I can choose to just show my face.
I can choose to show myself in the corner as a round circle in the corner. I can do just the screen. So there's all these different layout choices.
Again, pretty easy, but this is also something that ChatGPT can choose for us. And in fact, what I've done is right here, I've actually given instructions to ChatGPT. We can listen to that.
Hey, Barumi, when I'm talking about the AI adoption, create an infographic for that. It doesn't need specific numbers, but just something showing adoption over time going up, like a bar graph or a line graph. So what I've done is I've recorded the video.
And then while I'm recording the video, I'm actually giving instructions to ChatGPT on how I want the video edited. That way I don't have to type it out. And ChatGPT knows that when I say, hey, Barumi, it's going to remove those instructions and follow those instructions.
If you ask for graphics with no plan, you get boring PowerPoint slides. Claude does that because nobody told it what good looks like. So pick videos from editors you love and tell Claude to take screenshots of them.
Then ask it what makes them work. It will notice things like how text moves in and how one scene transitions to the next. Once you have the design, Claude now needs a game plan called a beat map.
A beat map is just a list of every idea in your script and what's on screen while you say it. So make cloud build that map before it builds any graphics because changing a plan costs almost nothing and changing a finished graphic costs a lot. So now instead, each idea gets one graphic and it stays up long enough to read.
Now, I mentioned a little bit ago that you can basically customize the timing of different assets and different projects that you have. And so in order to do that, all you got to do is say add timing sliders to composition I'll say called motion styles showcase, and your cloud code is automatically going to find different properties that you can basically customize to speed up certain elements.
So for example, obviously, the phone fading in, maybe the floating of the phone, maybe a glow in the background, but everything I just mentioned, you can also specify for cloud code to go ahead and create properties out of that as well. And okay, that was pretty quick. It basically added seven different controls on the right hand side of the panel, we can see here that we have the seven different things that I was showing you.
So first, we have full stagger, which obviously we can customize yeah so that's basically much slower fading of the phones obviously if i change that to zero all phones will pop up at once if i obviously set it to 40 it'll be one individual phone fading in and notice how quick it is especially in this render because this isn't actually generating pixels every single time the code is making it so much faster and we can preview this in insane speed so yeah i do think code is the future for these air editing and ai generation tools but yeah regardless you can see here we have a bunch of different little settings like the floor speed of the phone let's come back across and look at all these titles and graphics that it made didn't give it any guidance but let's take a look at some of them so mike asked me to record a quick video for this page all right solid animated in it's definitely talking about what it is i didn't give it the context it knows from understanding the transcript all right it's timed to what i'm saying this is just awesome and so fast they're not matching our brand in any way shape or form but i didn't ask it to or give it any guidance around that but the key points the key messaging and everything it's all good it looks awesome so the next thing obviously i tested was to come back here to chat gpt and
I say, okay, those are great. Can we duplicate the timeline? So I don't want you to replace what you've done.
I just want to, you know, let's make a whole new version of this timeline and a new set of titles and graphics that match the theme and the branding of the information page that this video is going to be hosted on. So we have the webpage here, which is the page here. So we can start to see the types of fonts and everything that are used.
So we again have something that matches the colors, the theme, the fonts, the branding. Like this is... awesome there's the 15 85 so the other one didn't look bad without context without guidance but this matches the brand this matches the page that it's going to go on and this to me is just the coolest thing, the time to create something that looks decent, that is human led.
This isn't AI slop. That's me. That's my thoughts and opinions.
But I'm not needing to do the grunt work and edit this down, remove the bad takes and whatever. I can focus on just presenting and doing the bit that I do. So we come through here, all the graphics and everything.
the zoom ins even put a call to action at the end build your own version apply for the beta like all of this stuff is on brand is timed up is perfect basically everything about the creative aspects of it i actually baked into the preset if you look in the default youtube preset that i've built out in my package there's all these files here that talk about basically exactly how i like my youtube video edited and that's actually how this video was edited that you're watching right now is with the youtube default preset i basically custom built a lot of the animations and the graphics by hand and i dialed them in by hand because only i know how to do them and then i just had claude basically save the values that way it doesn't need to creatively come up with these things from scratch every time we do a video it's basically just following the recipe and even kind of roasts me in a professional polite way and basically tells me that there's one content gap to flag the take promises a framework but ends after introducing strategic themes without a worked example or closing call to action
even the fact that it's told me that it's missing something and there is an opportunity to improve the video is very powerful it's understanding what i'm saying but then it's also understanding the frames and the images that's in the video and like is it perfect it's not 100 perfect of course i think that there's some b -roll changes you've made that don't really fit for example the shot at 1 minute 44 i think it would be better if you use motion graphics on an or an image there can you please create a visual to represent this use apple visual style so this is not the best prompt in the world i think it's pretty average but let's see what it actually generates map out the business's goals the pain points of the audience and create content that fits into content formats that work on social media this would have taken me hours to make and to source all these different things and get the motion right and stuff like even it's getting the motion right now which is like
Interesting. I also like to keep my drafts small, so while I'm giving notes, Claude will render everything at half size so the drafts come out much faster. Full quality renders only happen after I approve the whole cut.
So try this. Ask Claude to build one graphic as a webpage first, then turn the rest into a storyboard. Look at it, fix it, and only then turn it into video.
This will save you a ton on time and tokens in the long run. Step five is arguably the most important part. Have Claude check its own work.
So here's how I do it. Claude checks 10 frames every second, and there are six mistakes that it looks for. An animation that plays twice, a graphic that flashes for a split second, text that lands too late to read, a graphic sitting on my face, a word I said two times in a row, and my lips drifting away from my voice.
Make Claude check like a person, not like a computer. So try this. Write down the three mistakes you always catch when you watch your own videos and hand that checklist to Claude to watch out for.
And the reason that I'm not doing this directly through Claude code. is that i have found that no matter how good my prompt is there is still little things here and there that i do need to update and change when it comes to repeated sentences or some stuff getting cut out it's definitely not perfect i mean it does a pretty good job of getting the majority of it done but I do like to verify everything that I'm saying on the video to make sure that it's accurate.
How much time does it really save? I can now say that for my long videos, videos like this, the one you're watching right now, I am using DaVinci Resolve and ChatGPT to do the first rough cut, rough edit, and a few other basic things that I always do first when I start editing a video. And not only is it...
faster than if I would do it myself, but the result also looks like as if I would do it myself. And so something that used to take 30 minutes up to 45, 50 minutes is now done in five. So that's absolutely huge.
It just works for my workflow at least, you know? I know it cannot replace an actual editor yet anyway, but the real benefit wasn't just pressing a button and getting a perfect video. No, that's not what I want also.
It was getting help with the boring tasks. For me at least. That's where the value is of letting AI edit your video.
Let me know what you think. Have you tried AI editing? Has it actually saved you time?
Saved you money? Saved you something else?
The Hook
The bait, then the rug-pull.
The whole video you're watching was edited by Claude, and the host says he never touched a timeline. What follows is the six-job workflow behind it and how to copy it today.
