Big Idea

The argument in one line.

Pointing Claude at a library of saved DaVinci and Remotion templates, instead of asking it to invent a design from scratch, is what turns AI video editing into a reliable near-one-shot process rather than a slow trial-and-error loop.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You already record raw talking-head footage and want to automate the cutting and motion-graphics grind instead of editing by hand.
You're comfortable downloading a connector repo from GitHub and running setup through Claude, similar to a Claude Code style workflow.
You want to understand how template-based prompting produces consistent AI output instead of random one-off results.

SKIP IF…

You want a no-setup, point-and-click editing tool. This requires a connector folder, a template library, and prompt-writing.
You want AI to choose templates and pacing for you. In this workflow the editor still manually marks cuts and assigns templates himself.

TL;DR

The full version, fast.

The creator shows how he connects Claude to DaVinci Resolve with a free connector script, then hands Claude a raw talking-head recording plus one detailed prompt. Claude cuts the silences and bad takes, builds motion graphics through Remotion (HTML turned into video), and inserts AI-generated b-roll from a third model, all inside the live DaVinci project. The core mechanism is template-based prompting: every overlay instruction names a specific saved template to copy rather than describing a design from scratch, which is what makes the output consistent. A companion 'pre-editor' tool he built lets him mark cuts and assign templates visually before Claude ever sees the prompt. Manual tweaks and voice-dictated follow-up instructions round out the workflow.

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 – 00:36

01 · The Result

Cold open plays the finished, fully AI-edited intro before explaining how it was made, then states the premise: Claude did 100% of the edit inside DaVinci.

00:36 – 01:59

02 · Setup

Three requirements: Claude, DaVinci Resolve, and a connector folder of scripts (downloaded from GitHub) that lets Claude control DaVinci.

01:59 – 03:59

03 · Editing An Intro

Drags a raw, mistake-filled 2:24 talking-head recording into Claude along with a prepared prompt, then leaves it to run.

03:59 – 06:11

04 · The Three Tools

Breaks down how DaVinci (cuts, native text), Remotion (HTML-coded motion graphics), and an AI video model via Kai (generated b-roll) combine into one finished timeline.

06:11 – 07:31

05 · Manual Tweaks

Shows editing directly inside DaVinci for small changes, like a text color swap, because it is faster than prompting AI for trivial edits.

07:31 – 09:08

06 · Asking For Changes

Sends a voice-dictated follow-up prompt asking Claude to reposition a clip into a 9x16 picture-in-picture overlay at a specific timestamp.

09:08 – 10:14

07 · The Prompt

Walks through the initial prompt's structure: source file paths, an explicit keep/cut list, and per-overlay instructions that each reference a template.

10:14 – 11:24

08 · Templates

Argues that referencing a pre-built template library, rather than letting AI invent a design, is the single biggest factor in getting a reliable, near one-shot result.

11:24 – 14:35

09 · Pre-Editor Tool

Demos a custom app that transcribes the raw video, lets him mark cut regions and assign templates visually, then auto-generates the master prompt.

14:35 – 15:47

10 · Recap

Summarizes the three-tool pipeline and points to the free connector download and the paid mastermind for the pre-editor tool and template library.

Atomic Insights

Lines worth screenshotting.

Pointing an AI editor at a saved template, instead of describing a design in words, is what makes AI-generated motion graphics look consistent instead of random.
A 2 minute 24 second raw recording was cut down to roughly 38 seconds using nothing but a written prompt and Claude driving DaVinci Resolve directly.
DaVinci Resolve handles the literal cuts and native text elements, Remotion (HTML turned into video) builds graphics DaVinci can't natively construct, and a third AI video model generates any inserted b-roll.
Small manual fixes, like changing a text color, are often faster to do by hand than to prompt an AI for, even though the AI edit itself only costs a couple of cents in tokens.
Speaking a prompt out loud through a voice transcriber produces better AI output than typing it, because people naturally add more context when talking.
A picture-in-picture repositioning request, moving a talking-head clip into a 9x16 corner overlay, is specific enough that Claude can execute it as a single follow-up instruction on an already-edited DaVinci project.
The creator built a separate 'pre-editor' web tool that transcribes the raw footage, lets him mark red cut-regions and click a template for each section, and only then auto-generates the full prompt he pastes into Claude.
Even with a thorough, template-driven prompt, one edit needed a follow-up correction because Claude first tried to build an overlay in Remotion instead of directly inside DaVinci as instructed.
The video claims a mid-tier model was sufficient for the edit because the prompt itself was so detailed that a more expensive model wasn't necessary.
Every overlay in the demonstrated edit is tied to a specific named template drawn from the creator's personal library, not generated fresh by the model each time.

Takeaway

Constrain the AI with templates, don't let it improvise

WORKING WITH AI AGENTS

The reliability of this AI-driven editing workflow comes down to one habit: pointing Claude at a library of saved templates and precise cut instructions instead of asking it to invent an edit from scratch.

02Setup

Getting Claude to control DaVinci Resolve requires installing a separate connector, a folder of scripts that translates Claude's instructions into DaVinci actions, downloaded once from a shared repo.
The one-time setup is done by handing the connector folder to Claude and asking it to install everything itself, rather than manually configuring anything by hand.

03Editing An Intro

The entire cutting process starts by dragging the raw, unedited recording straight into the Claude conversation along with a single detailed prompt describing the desired edit.
A first-pass AI edit on a short clip can take anywhere from ten minutes to a couple of hours, so it is designed to run unattended while the editor does something else.

04The Three Tools

DaVinci Resolve handles the mechanical work: cuts, native text elements, and combining every other tool's output into one timeline.
Remotion exists specifically because DaVinci can't assemble complex multi-element mockups from scratch; since HTML is code, and AI is good at writing code, Remotion lets the AI build that graphic as a program instead of a hand-placed composition.

05Manual Tweaks

AI-driven editing doesn't replace direct manual control. Every element the AI edited is still a normal, editable DaVinci object that can be tweaked by hand afterward.
A trivial visual fix, like a text color change, is worth doing manually. Even a cheap AI edit adds up in time and token cost once you're making many small changes.

06Asking For Changes

Iterating on an AI-edited project works by giving Claude a specific timestamp and a precise instruction, such as resizing a clip into a 9x16 corner overlay at the 14 second mark, rather than a vague redo request.
Speaking a follow-up instruction through a voice transcriber rather than typing it produces a more detailed, better-followed instruction, because people naturally supply more context out loud.

07The Prompt

The prompt that drives the whole edit has three parts: where the source files live, an explicit list of which time ranges to keep or cut, and one instruction per overlay.
Each overlay instruction names a specific saved template to copy, then tells the AI which text to swap in, rather than describing the visual from scratch.

08Templates

Using a pre-built template library is described as the single biggest factor in getting a reliable, near one-shot result instead of needing several correction passes.
The template approach works because the AI's job shrinks to matching an existing example and swapping in new words, instead of inventing new design decisions every time.

09Pre-Editor Tool

A separate purpose-built tool transcribes the raw footage first, then lets the editor visually mark cut regions and assign a template to each remaining section before any AI prompt is written.
Fully automating template selection is described as the next unsolved step; for now, a human still manually decides which template fits each moment for a good result.

Glossary

Terms worth knowing.

Remotion: A framework that turns HTML and CSS code into rendered video, used here so Claude can write motion graphics as code instead of needing a pre-built animation asset.
Template-based prompting: Pointing an AI at a specific saved example to copy the structure of, rather than describing a desired result in words, so the output matches an established style reliably.
Pre-Editor tool: A custom app shown in the video that transcribes raw footage, lets the creator mark which sections to cut and which template to use for each overlay, then generates the master prompt handed to Claude.
Kai: A tool mentioned in the video for connecting to any image or video generation model through one API, used here to route a screenshot to an AI video model for a generated b-roll clip.

Resources

Things they pointed at.

00:44tooldavinci-ai-editor connector (GitHub) ↗

04:36toolRemotion ↗

05:53toolGoogle Omni (AI video model)

08:22toolCicero (voice-to-text tool)

08:57toolWhisperFlow (voice transcriber)

14:20toolKai (API router for image/video models)

00:40productApplied AI Mastermind ↗

Quotables

Lines you could clip.

00:00

“Everything that you see in this video was 100% edited using Claude and DaVinci.”

cold-open claim that states the entire premise in one line→ TikTok hook↗ Tweet quote

08:28

“You get much better results when you're actually speaking to AI rather than just typing it out... you yap a little bit more, and that actually does produce better results.”

counterintuitive, standalone tip about prompting AI by voice→ IG reel cold open↗ Tweet quote

10:03

“This is the one really critical thing which makes the outputs that you get with Claude and DaVinci much, much better. And that is using templates.”

names the video's central mechanism in one sentence→ newsletter pull-quote↗ Tweet quote

13:16

“We don't want AI to do all of the heavy lifting, right? We want to give it a template so that all it has to do really is just swap in the correct words.”

crisp explanation of why template-constrained prompting works→ newsletter pull-quote↗ Tweet quote

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.

Everything that you see in this video was 100 % edited using Claude and DaVinci. All I had to do was tell Claude exactly what I wanted and it would make those edits inside of DaVinci. It removes the silences, it removes the bad takes, it creates motion graphics, it adds text overlays, adds sound effects and loads more.

You do not need to know how to edit to do this, however you do need to know how to properly set it up and you do need a little bit of AI knowledge. But luckily for you I spent a load of time and a credits figuring it all out and that is what I'm going to share with you today in this video.

So buckle in, we are going to get into editing with Claude and DaVinci. Okay so to set it up you need three things. Obviously you need Claude, you need DaVinci and then you also need this special connector which is going to connect Claude to DaVinci.

Essentially, this is a load of scripts and it tells Claude what to use, what to interact with in order to speak to DaVinci and make edits to DaVinci. I've left a link down in the description where you can go and you can download this entire folder, which will have all of the setup in there. And then what you need to do after you download that is simply bring that into Claude and work within Claude with that.

Let me show you how to do that. So we're within Claude. We're gonna go down here to the bottom.

We're gonna go to open folder. Then what you want to do is you want to select the folder which you downloaded with the link down below. So here is mine here called DaVinci AI Editor and I'm going to open that up.

Now the first thing that we need to do is we need to make sure that everything is installed. So I'm going to say to Claude, please can you look at the contents of this folder and please can you set up everything that is needed in order to get Claude to speak to DaVinci. Let me know when that's done.

Okay, and we're going to send that off. And that is going to do all of the setup for us. All of that setup, all the instructions, all of the details are in that folder.

So it should be super, super easy. It might take a minute or two to set up. So let's come back to it when it's done.

Now, I've already set mine up because I've already been using it. So it comes back and says setup is already done. But you're also going to come back and say, okay, I've set it up.

If you do have any questions about this setup process. Just ask Claude. Everything is in that folder, which you give to Claude.

Okay, great. Now let's actually get on to making something in DaVinci with Claude. So what I'm going to show you is actually me editing the intro for this video.

So here is the video that I recorded. And if we watch it through, we can see I've made a load of like mistakes in it. There's like a load of pauses.

You can see there's a big point here where I don't say anything. Essentially, this is just completely raw and needs to be edited. This is two minutes, 24 seconds.

And I think the actual intro would be something like. 40 seconds so there's a lot that needs to be edited out of this raw video so let's come back over to claude let's create a new chat let's make sure that we are working within that davinci ai editor folder or whatever it is called on your computer then all we need to do is drag and drop that raw video file into claude and then what we need to do is simply tell it how we want the video to be edited.

Now, I have already prepared a prompt. We're going to come back to this prompt later because the prompt is like the secret sauce here. But let me paste this in.

You can see it's a big long prompt with lots of things mentioned in there. I'm going to send that off and it is going to go and edit everything for me inside of DaVinci. This is going to take a while depending on how big the edit is.

It might take between 10 minutes or it might take a couple of hours as well. And it's going to go and edit that completely. for you.

It might open up DaVinci, you might see DaVinci opening or closing. Don't worry, that is normal. Another thing which I didn't mention is you need to make sure you're using the right model.

Now because I've got a very thorough prompt, I don't need to use a super... powerful model. So Sonic 5 is going to be good.

You will see a difference if you're using Sonic 5 compared to Opus 5 compared to Fable 5 or Fable 5 .1. So let's leave it for 10, 20, 30 minutes to do its edit and we'll come back to it. Okay, that is now completed.

Let's go over and have a look at DaVinci. And here we go. Now I won't go through all of this because you've probably already just seen this intro, but you can see it's first cut all the bad takes.

This is only 36 seconds long rather than like two minute. 20 seconds or whatever the the raw video was you can see that we've got that overlay at the beginning we've then got some rotoscoping going on which is nice and some logos which is cool um we've got this bit typing in claude we've got this bit here which was created by google omni um let me talk about a little bit about the different tools that are used here so basically there are three tools which you used here.

The first one is raw DaVinci, just DaVinci by itself. That is how all the edits are made. So like all the cuts are made, like down here, you can see like these bits were cut together and you know, all the cuts.

And that also does things like adding in this text over here. You can see that inside of DaVinci, we've got this background that was added on top, which is just like this white dots, which I use in some of my videos. And then we've got the text elements added on as elements with inside of DaVinci.

So that's one of like the three things. which this process is using. The other one is ReMotion.

So ReMotion is how you create motion graphics. Essentially, it allows you to create whatever video elements that you like just by using HTML. Now, you might be asking, Matt, why are we using ReMotion as well as DaVinci?

Well, the reason is that in DaVinci, you can only create videos with elements, right? You've got to have all of those elements. So if we go to this part here, where we've created like a mock -up of Claude, We can't create that.

Well, you could create it, but we haven't created it in this instance with DaVinci because you would need all of the different elements yourself, right? You would need the text on the top. You would need this rectangle here as like the input thing.

You would need all of these like buttons down the bottom. We don't have those. However, if we create it in Remotion, Remotion takes HTML and turns that into video.

And HTML is code. And what do we know about AI? It's very good at writing code.

Hence why for overlays like this here. You can see there's actually a separate clip, which was created in Remotion. And then it's just overlaid it with inside of DaVinci.

And the same with this part here. This is fully made within Remotion. And the same here.

This is a Remotion clip as well. So we've got DaVinci doing all the cuts and some of the edits, like where there is text. And then we've got Remotion doing some parts.

And then this part here, so that was done using Google Omni. So essentially what I asked it to do, and I'll show it to you in more detail later when I show you the prompt, but essentially it's taking a screenshot of the editor as it's editing. And then it sent that to Google Omni, which is a AI video model.

And it said, take the screenshot and make it look like the... um video timeline is being edited and it's come back with that video and then claude has put that video clip from google omni inside of davinci so it is one uh continuous thing so there we go um let me show you what you can do which is really cool and that is because we're working with inside of davinci you can change anything how you would normally change anything inside of DaVinci.

So if I come in here and I want to change the color of, you know, this word, everything here, I can just simply select that, come over to color and just, you know, change the color. So you can make edits manually inside of DaVinci because there are some things when it comes to video editing, which it's much quicker to do yourself rather than get AI to do it.

So if we wanted to ask AI to change the color of that text there, it might take, you know, two minutes and it might take, you know, two cents worth of tokens, which is not a lot, but if you want to make loads of these little edits, it adds up in time, especially, and then also a little bit in credits. But mainly it's going to take way more time if you get AI to do all the little edits for you.

And also I'm going to mention here that this is quite a good output. You might not see outputs this good because of the way that I've created it. I've created a whole system around this, which I'm going to show you in just a second, of essentially how you can almost one -shot videos, right?

We're going to need some tweaks to this, so it's not entirely one -shotted, but... is pretty good, right? It's not far off.

Another great thing is that we can make changes to what we've created using Claude. So let's say that on this overlay here, when I'm talking about what it can do, I want to make a change so that the video, which is underlying this video here, right? So if I get rid of this, I want this video here to be like in the corner, like it's picture in picture down here.

So I'm going to grab the timestamp, which is 14 seconds. And I'm going to say to Claude, Great, can we now make an edit to this project?

On the overlay that appears at the 14 second mark, I actually want the A -roll bit of video, so the talking head video, kind of in the bottom right hand corner in a 9x16 format, kind of like picture in picture sort of thing, just for this section that appears at like the 14 second marks and goes a couple of seconds either side.

Please make this happen. Alright, I'm using my tool Cicero, which is free to everyone inside of the Mastermind, to speak to Claude. it is much, much better.

You get much better results when you're actually speaking to AI rather than just typing out. It's quicker, but also you include like more context. You like blab a little bit more, you yap a little bit more, and that actually does produce better results.

So a little tip for you there, use a voice transcriber, whether that's Whisperflow, whether that's one inside my mastermind, whatever you want to use, use a voice transcriber. Okay, let's give it a minute or so and let's come back and see how it's handled that edit. Okay, it's now done that.

It tried to create it in hyperframes first, so I had to go back in and tell it to make sure that you do. inside of DaVinci, so you still have to prompt it here and there to make sure it does the right thing. Let's go over and have a little look.

Here you can see it's edited this section, so I've got myself in the bottom right. Doesn't look particularly fancy, doesn't look like a great edit that I'm going to keep, but you can see you can then make changes, make tweaks with inside of Claude to make whatever changes that you want to make. But let's have a look at that initial prompt because that initial prompt is the part which really dictates how well the outcome is going to look and how many passes through it you need to do in terms of seeing where the mistakes are or seeing what things need to be edited, telling Claude to do that.

So let's go back and have a look. Okay, so here we are. This is the prompt that we gave it initially.

Let me break down a few parts for you. So we're telling it that we want to use Remotion to create some of these overlays that we're going to use within this project. We're then mentioning the source video and where it is within the computer.

And then we are first telling it which bits we want to cut, right? So we're taking it down from a 2 minute 24 video down to, what was it, 38 seconds or so. So we're telling it what bits that we want to keep and which bits that we want to cut.

How does it know which bits? I'll tell you about that in a second. Then we are telling it about the overlays and the edits, right?

So you can see the overlays here are mentioning templates. This is the one really critical thing which makes the outputs that you get with Cloud and DaVinci much, much better. And that is using templates.

If you've seen any of my other videos about editing with AI, this is essentially the way in which you make it reliable and good every time. And the way in which it can one shot something.

So it can do it on the first time of asking and doesn't, you know, you don't need to go in and tweak it loads of times, you need to be using templates, essentially. So you can see that all my overlays, I'm referencing a template, and I'm telling it what I want, right? So we've got overlay one, which is at right at the beginning, which if we have a look, is this first bit here.

So this what you're seeing here is from a template. And I'm simply telling Claude to use this template. But I want it to look like this, so the template.

I want it to be made in the DaVinci, not in Remotion, and use the same background, same font, swap the text for what is being said in this section of the video. So the two things which I'm essentially doing in this prompt is saying what to cut, what not to cut, and which overlays to use. Now you're probably asking, Matt, where do you get these cuts from?

What templates are you mentioning? This doesn't really make sense. Not a problem.

Let me show you what tool I'm using to create this initial prompt. Okay, so this is my pre -editor tool. And what this does is essentially creates that prompt for us and allows us to create that prompt.

Now, if you want access to this pre -editor, it is within my Applied AI Mastermind, which I've got a link to down below if you want to join it. You can just create your own. However, downloading what I've got here will save you 30, 40 hours.

You also get a load of my templates. So what you do is you first select the video that you want to edit. So you drag and drop in that raw video.

Then it will transcribe the video and you can see here the red parts are the bits which I've cut out. So you can see at the beginning I'm going, okay, okay, okay, making sure that the microphone is picking me up. And I've cut out a load of bad takes in this intro that I recorded.

So you can see all of those have been cut out in this first step, which is taking the cuts out. Then we move on to the overlays. Now you can see here that pretty much every word has an overlay.

That's more like that purple. denotes. And you can see that in the edit, pretty much everything does have an edit over it.

There's very few parts which don't have some sort of editing over it. That's usual for an intro, right? It's a little bit more heavier editing at the beginning.

So if we select one of these, so if we select this first one, you can see that this is the template that I am using, right? And I've got a load of templates built into this. Again, you can access all of it if you're inside of the mastermind or build something yourself, you can see that I've got a load of templates, all which fit my theme or my style.

If you've watched many of my videos before, you probably recognize a lot of these because I use these templates a lot within my videos, right? And you can see that what we did here was that we selected which template that I wanted to use for this initial part. So you can see that this part at the beginning looks very much like this template here, which is essentially the whole point, right?

We don't want AI to do all of the heavy lifting, right? We want to give it a template so that all it has to do really is just swap in the correct words.

This makes it much quicker, much easier, much cheaper. It makes it go wrong less often as well because essentially it's just changing the template, right? So all I'm doing is going through and I am selecting personally what things I want to appear.

Now, the holy grail is that AI can do it itself. And that's something which I'm working on that's coming soon and I'll share with everyone in the mastermind. AI can almost do that.

However, at the moment, choosing these edits yourself is the better way to do it if you want something that actually looks decent. So what I did was I went through this entire video and just like selected these different bits.

You can see here I've made different notes down at the bottom and selected different templates here. And I went through and I essentially listed all the changes that I want until I had everything selected. Let's have a look at this one here.

This one is... for this part here where it says, and it would make those edits inside of DaVinci. So for that section where I got Google Omni to create a little bit of AI video, my prompt that I used was use computer use to get a screenshot of the DaVinci editor and then send this to Google Omni via Kai.

Kai is just a tool where you can connect to any image or video model via API and generate that clip that shows the DaVinci editor being edited by AI. That's all I told it. And then it went off and created that for me.

And then once I decided on all those overlays and all the edits that I wanted within this. I came over here and it generated that prompt for me, which we saw within Claude, put it into Claude along with the video and DaVinci does the rest in collaboration with Remotion as well. So there we go.

This has been a breakdown of how you can get Claude to edit a video for you using DaVinci. We are also using Remotion as well to create motion graphics for the things that can't really be done inside of DaVinci. And we're also connecting to things like Google Omni to create some AI videos or to edit some footage.

to you know give some ai effects here and there if you want to do this yourself then you can download that file which i've linked down below in the description for free and then you can connect claude or whichever ai tool that you want to use to davinci and if you want to see all of my training upon ai editing ai content creation ai video marketing and you want access to the tool that i showed you then all of that is available within the applied ai mastermind which is also linked down below now if you've got the end of this video and you found it useful then you'll probably also like this video here so go ahead go and check it out I'm sure you'll find it useful as well that's it from this video like comment subscribe all that good stuff and I'll see you in the next one goodbye

The Hook

The bait, then the rug-pull.

The cold open is the video's own intro, already fully cut, motion-graphed, and captioned, played back before any explanation of how it was made. Only after that does the creator rewind to the raw, unedited recording and walk through the exact setup.

Frameworks

Named ideas worth stealing.

04:09list

The Three Tools

DaVinci Resolve — cuts, native text overlays, and the final assembled timeline
Remotion — HTML-coded motion graphics for elements DaVinci can't natively build from parts
AI video model via Kai — generated video/image inserts (e.g. an animated recreation of the editor screen)

Three specialized tools chained together, each handling the part of the edit it's best suited for, rather than one AI trying to do everything inside a single application.

Steal forany AI-driven creative pipeline that needs to combine structural editing, generated graphics, and generated media

09:35model

Prompt Structure (Source / Keep-Cut / Overlays)

Source video file path and Remotion project location
Explicit list of which timestamp ranges to keep vs. cut
One instruction per overlay, each naming a specific saved template plus the text to swap in

The prompt pasted into Claude has three parts: where the raw files live, an exact cut list, and per-overlay instructions that each point at a named template file rather than describing a design from scratch.

Steal forany agentic prompt where you want repeatable output instead of one-off generative results

CTA Breakdown

How they asked for the click.

VERBAL ASK

14:35product

“Download the free connector folder linked in the description, and if you want the pre-editor tool and template library, join the Applied AI Mastermind.”

Soft, split CTA: a free open-source download for anyone, paired with a paid-community upsell for the proprietary tooling (pre-editor app, template files) shown throughout the video.

MENTIONED ON CAMERA

00:44tooldavinci-ai-editor connector (GitHub) ↗

04:36toolRemotion ↗

00:40productApplied AI Mastermind ↗

Storyboard