Big Idea

The argument in one line.

Full video editing can be automated in Claude by chaining six steps (rough cut, manifest-driven plan, motion graphics, independent QA, audio, render), and the style manifest is what makes the output feel like yours.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

A solo creator who edits their own talking-head videos in CapCut or similar and loses hours to cutting silences and bad takes.
A creator paying editors days and hundreds or thousands of dollars per video who wants to see what an automated pipeline can realistically replace.
A Claude Code or Claude skill builder looking for a concrete architecture for a multi-stage media pipeline with word-level timestamps and Remotion.

SKIP IF…

You want a ready-made tool to install today. The skill itself is sold inside a paid community, and this video only teaches the build outline.
You edit cinematic or heavily footage-driven work where talking-head cleanup and templated motion graphics are not the main job.

TL;DR

The full version, fast.

A creator shows a Claude skill triggered by the words "full edit" that turns a raw talking-head recording into a finished video with silences and bad takes removed, motion graphics, sound effects and music, reportedly for about five cents. He breaks the process into six steps: rough cut from a word-level transcript, a plan driven by a style manifest, motion graphics built in Remotion and Hyperframes, a check by a fresh agent, music and sound effects added last, then a final render. The manifest and the independent reviewer agent are the two ideas that do the most work, and he estimates 30 to 40 hours to build it yourself.

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 →

Chapters

Where the time goes.

00:00 – 00:55

01 · The promise

Claims a skill that rough-cuts, adds motion graphics, sound effects and music from one word, replacing days of editor turnaround.

01:01 – 04:30

02 · Live demo

Drags a messy one-minute video into Claude, types "full edit", and reviews the result, then requests a full-screen payoff graphic in the same chat.

04:30 – 05:44

03 · What you need to build it

A Claude subscription, a word-level transcription tool such as Deepgram, and optional image and video generation and image search APIs.

05:44 – 08:03

04 · Step 1: Rough cut

Transcribe with word-level timestamps, then remove silences and bad takes using rules for unfinished, flubbed and near-duplicate lines.

08:03 – 10:25

05 · Step 2: Plan with a video manifest

A style document guides the plan for motion graphics, timing, fonts, colors, subtitles and templates.

10:25 – 11:44

06 · Step 3: Build the motion graphics

Remotion and Hyperframes render the planned overlays, full-screen graphics, lower thirds and icons.

11:44 – 13:35

07 · Step 4: Independent QA loop

A fresh agent checks for overlaps, bad visuals and timing errors, looping until no issues remain.

13:35 – 15:39

08 · Steps 5 and 6: Audio and render

Music and sound effects go in after QA, either generated or pulled from an approved bank, then everything renders to one MP4.

15:39 – 17:23

09 · Wrap-up

Customization options, time estimate for building it yourself, and a pitch for the mastermind.

Atomic Insights

Lines worth screenshotting.

A single trigger phrase plus a dragged-in raw video is enough input when the editing style already lives in a saved manifest.
Word-level timestamps are the foundation of the whole pipeline, because every cut, overlay and sound effect has to land on an exact word.
A video manifest is a plan for the plan: it tells the model how to write the edit plan, covering pacing, fonts, colors, subtitles and template snippets.
Teaching the model what a bad take is takes tuning: unfinished sentences, mispronounced words and near-duplicate lines are the three signals.
The agent that built the video is poor at finding its own mistakes, so a fresh agent with no memory should review it.
Run QA in a loop until the reviewer finds nothing, and only then add audio, since edits made during QA would invalidate sound effect timing.
Music and sound effects can now be generated on the fly, but short-form work that rides a viral sound still needs a pre-approved bank.
Sound effect density is a style choice and belongs in the manifest, not in the model's guesswork.
Follow-up changes like making one graphic full screen work in the same chat with one prompt, without re-running the whole edit.
Common QA failures to check for are overlapping elements, badly drawn logos and animations whose timing breaks at render.
The builder estimates 30 to 40 hours to assemble and fine-tune this yourself, and says a full day of that belongs on the manifest.
Each project you run exposes new mistakes, and feeding them back into the skill is how it improves.

Takeaway

Automated editing works by layering small, checkable steps.

WHAT TO LEARN

The edit is good because every stage is separated, driven by a saved style document, and verified by a reviewer that did not make it.

Capture editing style once in a manifest covering cuts, fonts, colors, subtitles and templates, so every later run starts from your taste instead of generic defaults.
Get word-level timestamps first. Every cut, overlay and sound effect depends on knowing exactly when each word is spoken.
Define bad takes explicitly: unfinished sentences, mispronounced words and near-duplicate lines. The model needs these rules written out to cut reliably.
Plan before building. Deciding where graphics go, how long they run and how they animate makes the render step simple.
Use a fresh agent with no memory of the build to review the video, and loop until it finds no more errors.
Add music and sound effects only after QA passes, because timing changes during QA would break audio placement.
Decide per project whether audio is generated or drawn from an approved bank, since trending sounds for short-form cannot be generated.
Budget real time to tune it. About 30 to 40 hours of fine-tuning, with a full day on the manifest, is the stated cost of a good version.

Glossary

Terms worth knowing.

Word-level timestamps: A transcript that records not just what was said but the exact moment each word starts and ends, so cuts and overlays can be placed precisely.
Rough cut: The first editing pass that removes silences, mistakes and bad takes before any graphics, music or polish are added.
Video manifest: A detailed style document that tells the AI how to build an edit plan: cut spacing, fonts, colors, subtitle style and reusable templates.
Remotion: A framework for creating videos and motion graphics with code (React), which an AI can write and render.
Hyperframes: A code-based motion graphics tool used alongside Remotion for animation cases Remotion handles less well.
Deepgram: A speech-to-text API that returns word-level timestamps and offers free starter credit on signup.
Bad take: A section of a recording the speaker would not keep, such as an unfinished sentence, a flubbed word or a near-duplicate retry of a line.

The Script

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.

AI can now do all of your video editing for you with this new skill. All you have to do is drop your raw video into Claude, type this one word to trigger the skill, go and grab your coffee, and by the time that you come back, the entire video will be edited for you. It does the rough cuts, takes out the silences, take out the bad takes, puts in motion graphics, sound effects, music, everything.

Previously, I was paying editors thousands of dollars to edit my videos for me, and it was taking them, you know, days to get work back to me. And before that, I was spending hours and hours in front of the computer, in front of CapCut, editing my own videos.

But now with this new skill called the full edit skill, I literally just give it my video, ask it to edit it for me in my style, and it does everything for me. The skill is so crazy that I am a little bit worried for the video editors out there. However, in this video, I'm going to show you this skill in action.

I'm going to show you what it produces. I'm going to show you how to use it. And most importantly, I'm going to show you how you can build it yourself so you too can free up your time for video editing and get ai to do it so let's get into it Okay, so this here is the video that we are going to edit with AI.

You can see here I've got a lot of pauses in it. I also make some mistakes. There's big pauses as well here.

I'm trying to think of what to say. And we're going to get AI to edit all of this. Now, this is only a minute long, this video, but it is unlimited.

That's the length of video that you can get AI to edit. Now, I am with inside Claude. I'm going to grab that video and drag and drop it into Claude.

And I'm simply going to say full edit. That's all I have to write. I don't have to do anything else.

And I'm going to press enter. send it off and off we go now you'll notice that i'm using opus 5 .5 you will see a noticeable difference if you're using specifically opus 5 .5 compared to any other model even if you're using a really good model from another supplier like if you're using like one of the gpt6 models from openai they're still not as good as opus 5 .5 claude is currently the best one out there anyway um i'm gonna go make myself a cup of tea And we're going to come back when it's done and we're going to see what the output is.

Okay, couple two secure. And it is just finished editing this video and it's put it in my downloads folder. Let's have a look at it.

I've helped over 3000 businesses apply AI. And this is the biggest mistake that virtually every business makes. And that is AI is not actually that good at solving problems.

What AI is very good at is taking processes that you have already got working within your business, problems which you have already solved and simply automating them and speeding them up. And when you automate these processes, you can then free up your time to focus more upon solving problems yourself. And then once you've solved that problem, you simply hand it off to AI and you go and solve other problems within your business.

This is the number one way which I've seen over 3000 businesses profitably apply AI to their business. Okay, there you go. Pretty good.

You can see that it's completely edited out all of the silences, all of the bad takes. It has edited it in my style. You may be asking, Matt, does it always edit videos exactly like this?

No, I have trained the skill to edit videos specifically in my style with my brand assets, my colors, my fonts, everything like that. All the motion graphics here make sense. They look good and they actually add to the video as well.

Now, I don't think this really needs any changes to be honest with you, but let's say I did want to make a change. Let's say that at this point here, I want this payoffs. section to be full screen and I could just come over here and I can say looks good however i want the payoff section to be full screen i want the motion graphic to be full screen rather than just like a little bit in the top right hand corner please can you make that change and we'll send that off and it's going ahead and it's making that change and let me show you this is all in one chat like i'm not making any other changes there's no like trickery going on behind the scenes it is literally that one single prompt to produce that video okay that is now complete let's have a little look at that change and here we go we can now see if we go on to the payoff section that's now full screen run just a little thing in the top right hand corner so it does a really good job first time but if you want any changes you just prompt whatever you want and let me just emphasize again this is such an improvement to editing videos myself what we saw here is just a one minute video however when you expand this across you know 15 minute long videos and you're putting out several videos a week

and you're either doing the editing yourself or perhaps you're paying editors to do it, it saves you so much time or so much money if you're paying editors. It is honestly crazy. As you saw, it just took minutes to do and this probably cost me about five cent.

If you want this exact skill that I've used here and you're already inside of the Applied AI Mastermind, then simply come here. and you will find this skill to download if you're not already part of the mastermind but you want to join i'll leave a link to that down below you don't just get this skill but you get a load of other ai video skills as well and a load of education load of training and savings on ai tools and a load of other stuff go and check it out however i'm going to show you how this all works so you can build this yourself now it will take you maybe 30 or 40 hours to put everything together and to fine -tune everything so it's working properly but you can build it yourself but first what do you need in order to build this well first of all you need some sort of AI subscription.

I would highly recommend that being clawed because Opus 5 .5 is the best model out there at the moment when it comes to video editing and when it comes to motion graphics. Things like ChatGPT or Grok or even Gemini will work. They just won't be quite as good.

The other thing which you do need is access to a tool which can provide word level timestamps. I'll talk more about that later, but my favorite one to use is Deepgram. It's absolutely free.

You get like $200 for free when you sign up. I think 11 Labs also offer word -level transcriptions too. There are a few other optional tools that you can have as well.

One of them is access to some sort of AI video and image generation platform like Kai AI or FAL or Higgsfield or TopView. This will mean that you can incorporate AI -generated images and videos into your videos. And also another optional is a Brave search API that will allow you to get Google images and insert them into your videos.

Okay, so let's break down this process into six different steps or I guess there's seven steps because first thing you've got to do is record the video. That's step number zero, record your video. Then the absolute first step is taking that video that you have recorded and performing a rough cut.

Rough cut is where you take out all the silences and all of the bad takes. I have an awful lot of those within my videos and luckily AI takes them all out and that used to be one of the parts which would take me so much time previously. Now AI does it super quickly.

How does it do this? Well, first of all, it transcribes the audio. It takes the video file, gets the audio file from the video file, and then transcribes the audio file.

However, I mentioned this earlier, you've got to be using word -level timestamps. You've got to be getting a transcript of that video with word -level timestamps. What does this mean?

This means that the transcript you get back not only has all of the words which are said, but it also tells you when exactly each word is said. This is very important because if you are, you know, cutting out the silences, cutting out the bad takes, or later when we come on to the motion graphics, we need to know when everything is happening within the video so that we can make the cuts at the right point so we can insert the videos at the right points.

So the essential thing here is to make sure we're getting word level timestamps from our transcript. As I said before, the tool which I like to use is Deepgram. To set this up, you simply go to Claude, and you say, hey, I want to use Deepgram to transcribe audio to give me word level timestamps.

And then the great thing about AI is it will help you set that up. You can simply say, help me set this up, and it will help you do that. It will tell you exactly what you need to get, how you set up the account, how to get an API key, and it will allow you to set up your transcriber.

And then once we have that transcript with word level timestamps, AI will then be able to go through and highlight all of the silences. and all of the bad takes and cut them out. Now this is one of the areas where it does take fine tuning if you're building this yourself in order to get AI to know when is a bad take.

So if you start a sentence but don't finish it, that's probably a bad take. If you mess up a word and you don't pronounce the word properly, then that's probably a bad take.

If you have two sentences which are very, very similar to each other, but maybe you've just changed the word or changed the word order, then that is probably a bad take. I spent a while teaching AI exactly what bad takes were and what weren't bad takes. Once you've got this set up, it'll be able to do that full rough cut for you and take out all the silences and all of the bad takes.

That is step number one. Let's move on to step number two. This is creating the plan.

This is probably the most important part. And I have a secret tip here, which I haven't seen anyone else talk about, but it's one of the reasons why my system here seems to work much better than anyone else's system. And I don't mean to brag about that, but it just does.

And the reason for that is that I have what I call a video manifest. What this is, is essentially a plan for the plan. I am telling Claude.

how it should create the plan. And what's happening in the plan is it's essentially taking this video, which now has all of the silences and all of the bad takes taken out, and AI is going, okay, where should we include motion graphics? What should the motion graphics look like?

What colors should they be? How long should each motion graphic be? What should the motion be inside each one?

It's essentially doing all of the planning here. And it looks at this one manifest here in order to work out, okay, how should I create the plan? Now, what you're seeing here is my exact manifest from these videos here.

As you can see, it is quite in -depth. It's got a load of different sections which break down absolutely everything from how I like my videos cut, how much time I want to be between each different clip, the fonts and the colors and the spacing and everything like that that I like within the video. If I do want subtitles and what styles those subtitles should be.

And it's even got this section here where I actually include templates of certain parts of videos. So I say, okay, if we're ever talking about numbers, then these are some templates here which we can use. So this is all the information that Claude uses when it's creating our plan.

Now I've also got another skill which creates this manifest for you. If you've already got videos that you've recorded, Claude will look at your videos and say, okay, what style, how does this person like to edit their videos? And it will create that manifest for you.

If you haven't been editing videos already, then it'll ask you a series of questions and build this manifest for you. Or if you want to supply it with like reference images or reference videos that you like the editing style of, then it will look those and create an editing style and this manifest from those that's also one of the many other skills that i've got with inside the applied ai mastermind so we'll actually create this manifest for you this is the most important part if you're building this process for yourself this is where i would spend you know spend a full day going through adding everything that you can to this manifest so ai can create a really good plan of how to edit your videos okay so that is step number two creating a plan for how AI is going to edit your videos.

Then we come on to step number three, which is actually getting AI to create the edits and the motion graphics for your video. It's already got the plan. Now what it has to do is go out and actually create these overlays or these extra clips, which it's going to put in the video.

By the way, if you're finding this useful, then please do drop a like and subscribe if you want to see more videos like this. Now, how is it creating these motion graphics? Well, it's using mainly re -motion, but it's using a little bit of hyperframes as well where it needs to.

I've got a load of other videos talking about re -motion and talking about hyperframes. So if you want to learn more specifically about them, then you can go and check those out. However, to be honest with you, you don't actually really need to know how they work in order to get this to work.

If you are building your own system, simply say, hey, use a combination of re -motion and hyperframes to actually build the motion graphics for this video. And because we've done all of that planning in step number two, when it comes to step three, it's simply just creating them. It's going to take a little bit of time because it's got to render these out, but there's not really that much complexity to this step number three.

Okay, so now we've got those individual motion graphics created. Some of them are going to be full screen. Some of them are going to be overlays.

Some of them are going to be, you know, text along the bottom. Some of them are going to be like icons appearing, whatever it is. And also in step number three, it's compiled all of it, right?

And now what we need to do within step number four is to run a check on this video we've created to make sure that everything... is correct and there are no errors and again this is one of the areas where i spent a reasonable amount of time telling clause telling the system building it into the skill telling it which problems it needs to look out for so some of the common errors are overlapping of elements so where it may have multiple elements that said oh yeah we should have this here and this there but then sometimes they might overlap because it's just made a mistake All elements that just look terrible and don't actually make sense, right?

It's trying to create a logo, but it hasn't done it very well. And, you know, it needs to redo that. Or it could be timing, right?

That something comes in at the wrong time or that the animation that is planned actually when it renders, it doesn't render properly and the timing doesn't make sense. There's a whole range of these things that I ask it to look for. And as you build the system, if you're building it yourself, You will build projects, you will see mistakes, and you'll be able to feed back into the skill.

Hey, next time you build a video for me, look out for this and look out for this and look out for this. Now, one important part to do when we are running this checking process is to get another agent to spin up to check the video. This is because if you get the same agent to check the video that actually created the video, it often overlooks things I have found.

I think it's kind of like, you know, when you're back at school and you're on an exam and you think you've got everything right, you can't see any errors, but then if someone else looks over your paper, it's very easy for them to spot your errors, but it's almost impossible to spot your own errors. And it's the same sort of thing here.

If you spin up a new agent with essentially brand new context that has no memory of the video that's been created is actually much better at finding those mistakes. So I'd highly recommend using a new agent to check over your video.

Okay, so now AI has looked through the video, it has looked for any errors, and it has fixed any errors that it may have found. And it's going to keep looping over the checks and making changes until it can't find any more errors. Then what we do, step number five, this is where we build the music.

and the sound effects the reason that we need to do this after the checks is that if we do it before the checks and like something has changed with the timing or something has changed with the cuts or something has changed with what's happening then it will need to change the sound effects as well so what i have found is actually a better way to do it is make sure that everything is looking good to begin with after step four and then add the music and then add the sound effects now there are two ways to go about adding music and sound effects either you get ai to create the music and the sound effects on the fly or you have banks of music and sound effects which like you found online and that you've approved and you said yeah i like this music and i like these bunch of sound effects Now, previously, I've always used the approach of finding these myself and then just getting AI to look at the sound effects and look at the music and decide which ones it should insert into the video.

However, recently, AI has been improving with Opus 5 .5. So that actually now it's pretty good at creating the music by itself using something like 11 Labs. And the sound effects aren't too bad either that it can create itself.

So I think we're right on the cusp between getting AI to fully create all the sound effects and all the music itself compared to... selecting it yourself and then just getting AI to put it into the mix itself. However, if you are like editing short videos and you do want like a viral sound or viral music or a song in the background, then obviously you don't get AI to create that for you.

You download all of those songs that you may want to use and you just get AI to select one. So there are scenarios where you don't actually want AI to create everything there. You want to use a previously created asset that you've got a bank of.

And again, this part here does take fine tuning in terms of letting it know what sort of sound effects you want, how dense it should be. So do you want like loads of sound effects happening all the time?

Do you want to make it a little bit more subtle instead? And again, this is one of the things that is covered in that video manifest so that when it's going through and doing the planning, it will go, okay, we need a load of sound effects or it will say, okay, actually let's be a little more subtle and have fewer sound effects.

And then the last step after you got all of the music and all the sound effects done is to simply render it all together and to create this MP4 with your fully edited video. And there you have it. Those are the six steps.

That is everything that is doing in order to take a absolutely raw video, whether that's a one minute video, like I've shown you here, or whether that's an hour long video and completely editing it for you. There is a lot more here that I could talk about, such as, you know, if you want to include screen shares, then obviously you've got to provide this.

There is actually a way in which you can get Claw to record its own screen and put in screen shares of you showing like a website or something. but I won't go into that. Or of course, when you are prompting it initially, you don't just have to use that single word prompt.

You can say, Hey, I want this to happen when I'm talking about this. And I want this to happen when you're talking about this. So There is no limit to what you can do here.

If you go out and build this yourself, then you can customize it however you like. Or if you are part of the mastermind, you can take the skill that I have made here and you can just change it and edit it however you like. So it creates videos that are specifically for you in your specific style.

And you will find that if you are going from fully editing your videos yourself or using editors to using this process, you're going to save an ungodly amount of time. And actually, if you're someone like me, who's... fine at editing, but certainly not great, it might actually do a better job at editing than you do.

It does do a better job at editing than I can do myself, to be honest. So there we go. I hope you found this video useful.

If you have, please do drop it a like. If you want to see more videos like this, then subscribe. That should be everything you need to know in order to go away and build this skill for yourself.

But as I said before, if you want to save yourself 30, 40 hours, then you can join the mastermind and just grab that and a load of other resources with the link in the description. That is it for me. I hope you have found it useful.

If you have found it useful, then watch this video next because I think you might also find that one useful. That's it for me. See you later.

Bye -bye.

The Hook

The bait, then the rug-pull.

Drop a raw video into Claude, type two words, and come back to a finished edit. The video proves it with a one-minute demo, then opens the hood on the six steps that make it work.

Frameworks