29:59Danny Why · Tutorial
Claude Code Just Changed CapCut Forever
A screen-recorded build of a full 3D motion graphic inside CapCut, using a Claude Code-generated asset and an image-to-video AI to carry it past the timeline's edge.
August 21stThe argument in one line.
Connecting ChatGPT's Codex agent to DaVinci Resolve over MCP lets it transcribe a raw recording, cut out pauses and repeated takes on its own, then hand off to a second plugin that generates synced motion graphics with no manual editing.
Read if. Skip if.
READ IF YOU ARE…
- A solo creator who edits their own raw talking-head footage and wants to see whether an AI agent can actually make first-pass cuts unattended.
- Anyone already using Claude Code, Codex, or another coding agent who wants a concrete MCP-to-creative-app example beyond writing code.
- A YouTuber curious about automated motion graphics generation synced to a script, even without touching DaVinci Resolve.
SKIP IF…
- You need a finished, ship-ready edit today. This is an early live demo with visible rough edges, such as mismatched card copy, not a workflow ready to hand off.
- You're looking for a DaVinci Resolve tutorial in the traditional sense. No manual editing techniques are taught here.
The full version, fast.
A YouTuber connects ChatGPT's Codex agent to DaVinci Resolve Studio through an MCP server, then asks it to edit a raw, uncut talking-head recording. Codex transcribes the audio inside Resolve, uses that transcript to spot pauses and repeated takes of the same line, and cuts a new timeline automatically, including taking direct mouse control of the desktop at one point. A second MCP connection to Higgsfield lets the same agent generate on-brand motion graphics, a blue gradient background with bold white text, and drop them onto the cut timeline synced to what's being said. The result is a rough but real end-to-end automated edit with zero manual cutting.
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 →Where the time goes.
00:00 – 00:42
01 · Cold open: agent editing live
The creator opens with ChatGPT already mid-edit inside DaVinci Resolve, stating the video's whole premise as a direct command he gave the agent.
00:42 – 01:40
02 · Setting up Codex to control an app
Switches ChatGPT to the Codex section, sets approvals to 'approve for me', and selects the GPT-6 Astra model at medium effort before opening a video editor with its own MCP server.
01:40 – 02:32
03 · The raw, uncut recording
Creates a new DaVinci Resolve project and drops in one long unedited talking-head take, pointing out the pauses and false starts still baked into the footage.
02:33 – 03:51
04 · Connecting ChatGPT to DaVinci Resolve
Asks ChatGPT to find and verify the Resolve MCP server; it checks the connection and management tools, then confirms it's connected.
03:52 – 05:01
05 · First edit prompt: cut the pauses
Prompts the agent to remove empty space and repetition from the timeline; Resolve begins transcribing the audio as the first step of the cut.
05:02 – 06:25
06 · How the agent decides where to cut
Explains that the agent matches its transcript against the audio to find retakes and dead air, and narrates its own reasoning about the footage before acting.
06:26 – 07:35
07 · The reveal: it already made every cut
Zooms out on the timeline to find the agent has finished cutting the entire recording while the creator wasn't watching closely.
07:36 – 09:33
08 · Second live demo with mouse control
Runs the edit again on new footage while filming his own screen with a phone, showing the agent literally moving the mouse and cutting clips in real time.
09:34 – 10:56
09 · Bringing in the Higgsfield MCP plugin
Introduces a second connected tool for motion graphics, walking through installing the Higgsfield MCP plugin from a link in the description.
10:57 – 12:56
10 · Generating and placing motion graphics
Prompts the agent to generate transparent, on-brand motion graphics and full b-roll clips and drop them onto the already-cut timeline, synced to the script.
12:57 – 13:59
11 · Reviewing the finished edit
Plays back the automated edit, calls out one graphic whose text doesn't match its point, and closes by pointing to an on-screen video about monetizing a channel.
Lines worth screenshotting.
- Codex connects to DaVinci Resolve as an MCP server, giving a chat agent the same editing controls a human editor has over the timeline.
- The agent finds cut points by transcribing the raw audio inside Resolve first, then matching pause length and repeated phrasing against that transcript.
- Setting Codex's approval mode to 'approve for me' is what lets it make cuts without a manual confirmation click for every action.
- Raising the model's effort level to Ultra was the one setting changed before the agent successfully edited a 25-minute raw recording unattended.
- The agent doesn't just plan the edit, it briefly takes literal control of the mouse and cursor to execute actions inside DaVinci Resolve.
- When a raw take has the same line repeated three times, the agent's job is to keep exactly one full delivery per point and discard the rest.
- A second MCP connection, this time to Higgsfield, lets the same chat agent generate transparent motion-graphic overlays and full b-roll clips, not just cuts.
- Motion graphics generated by the agent stayed visually consistent because the creator specified one fixed style, blue gradient background and bold white text, in the prompt.
- The agent times its generated graphics to the spoken script automatically, without being told where each individual graphic should land.
- Not every generated element lands correctly. One card's on-screen text didn't match the intended message even though the background graphic looked polished.
What an AI agent controlling your editor actually does
AI Editing Agents
Handing a timeline to a chat agent over MCP works today for mechanical cuts and on-brand graphics, but it still needs a fixed style and a human doing quality control.
02Setting up Codex to control an app
- Before an AI agent can control a creative app unattended, you set its approval mode to auto-approve and raise its effort level, not just point it at the app.
- Picking the model version, here GPT-6 Astra, mattered enough that the creator called it out by name rather than trusting whatever was already selected.
03The raw, uncut recording
- Raw talking-head recordings carry visible thinking pauses and false starts on purpose, since those are exactly the markers an editing agent uses to find cut points.
- Recording one long unbroken take, rather than manually splitting into short clips, gives a transcript-based cutting agent enough continuous context to work with.
04Connecting ChatGPT to DaVinci Resolve
- Connecting an agent to a creative app is a one-time step: ask it to find and verify the app's MCP server, then it stays connected for future sessions.
- The agent confirms the connection back in plain language before attempting any edit, which is the signal to trust it with real footage.
05First edit prompt: cut the pauses
- The agent transcribes the raw audio first and only starts cutting once it has that transcript, so cut accuracy depends entirely on transcription quality.
- A vague instruction like 'remove the pauses and repetition' worked because the underlying task, matching audio gaps to a transcript, is mechanical rather than creative.
06How the agent decides where to cut
- When the same line is recorded three times, the agent's default is to keep one full clean delivery and discard the rest, mirroring how a human editor picks takes.
- The agent reasoned about the footage out loud before cutting, which is worth checking against your own footage's actual issues before trusting the output.
07The reveal: it already made every cut
- A zoomed-in timeline view can hide dozens of new cuts happening in real time, so zoom out before judging whether an agent's edit actually worked.
- Tighter cuts translate directly into more watch time, which is the real business reason this kind of automated cutting matters for a YouTube channel.
08Second live demo with mouse control
- Some agent actions require literal control of your mouse and screen, not just tool calls behind the scenes, and the OS will prompt you to confirm that access.
- Watching an agent work in real time, rather than only reviewing the finished result, is the fastest way to catch a bad cut before it compounds.
09Bringing in the Higgsfield MCP plugin
- Motion graphics generation is treated as a separate connected tool, not a built-in feature of the editing agent, so mixing and matching AI tools is the norm here.
- Installing a new MCP plugin is typically a browser click-through, selecting the target chat app and clicking add plugin, rather than a technical integration step.
10Generating and placing motion graphics
- Specifying one fixed visual style up front in the prompt, background color, text color, weight, is what kept a batch of AI-generated graphics visually consistent.
- The agent syncs generated graphics to the spoken script without being told the exact timestamp for each one, inferring placement from the transcript it already built.
11Reviewing the finished edit
- Not every AI-generated element is usable as-is: one graphic's on-screen text didn't match its point even though the visual design looked polished, so a review pass is still required.
- Deleting and regenerating a single graphic you don't like is trivial, which matters more for whether this fits a real workflow than the generation quality itself.
- The creator's own hesitation about posting the result is itself a signal: fully automated edits are usable, but not yet good enough to skip review.
Terms worth knowing.
- MCP (Model Context Protocol)
- A protocol that lets an AI chat agent connect to and control an external application, such as a video editor, the same way it would use any other tool.
- Codex
- OpenAI's coding-agent product inside ChatGPT that can be pointed at local applications and files, not just used to write code.
- Approve for me
- A Codex permission setting that lets the agent execute actions automatically instead of asking for confirmation before every step.
- Effort level
- A setting that controls how much reasoning a model applies to a task; higher effort trades speed and cost for a more thorough result.
- Higgsfield
- A separate AI generation service with its own MCP plugin, used here to produce motion graphics and video clips that the chat agent then places inside the editor.
Things they pointed at.
00:42toolChatGPT (Codex, GPT-6 Astra model)
01:40toolDaVinci Resolve Studio
Lines you could clip.
00:00
“Holy sh... Check this out. ChatGPT, I want you to connect to DaVinci Resolve and edit this video for me. Make the cuts, the transitions and then add motion graphics to it.”
06:36
“Oh my God, it made all the cuts. No way. Look, it made all the cuts already.”
09:14
“This is the first time I let ChatGPT control my computer.”
09:54
“We can actually use Higgsfield's MCP and then have ChatGPT create motion graphics for us and automatically add them inside DaVinci Resolve and sync them to our audio and visuals.”
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.
Holy sh... Holy sh... Check this out.
ChatGPT, I want you to connect to DaVinci Resolve and edit this video for me. Make the cuts, the transitions and then add motion graphics to it. Look at it go.
I'm not touching it. Look at the timeline. It's literally editing in real time.
It's making cuts in real time. Sorry, I didn't even position my camera right because I'm literally way too excited. So, everyone on the internet right now is talking about ChatGPT 6 Astra.
But I've seen no one talk about how you can actually connect it to a video editing program like DaVinci Resolve and have it make cuts for you and add footage to your video and literally edit the video for you. As you've seen from the intro of my video, ChatGPT can now literally edit the video for you. This is absolutely insane.
And in this video, I want to teach you how you can make ChatGPT edit the videos for you. So here's how you can have ChatGPT edit videos for you. First of all, you want to have ChatGPT on your computer.
Once you have ChatGPT on your computer and you open it, over here on the left where it says ChatGPT, I want you to click on this and select codecs because we want to be on the codecs section of ChatGPT. Then down below over here where it says approve for me, for you, it should probably say ask for approval. So you want to change this to approve for me.
This way, we don't have to always approve commands ChatGPT is making on our computer. Then over here where it says ChatGPT Astra, I don't know what model you have selected, but make sure you select ChatGPT 6 Astra. And for this, we can reduce the F word to medium to save credits.
Then what I want you to do is open your video editing program that also has an MCP to it. For example, DaVinci Resolve has an MCP server to it, which means we can connect ChatGPT to DaVinci Resolve. So what I'm going to do...
Inside DaVinci Resolve over here, as you can see, I'm going to create a new project over here and I'm going to name it subscribe to Denny Y, okay? And I'm going to create a project. Now over here inside the project, what I'm going to do is I'm just going to drag and drop my clip.
So I'm going to take my video clip with myself over here and I'm simply going to drag and drop it over here in the media. And now if I take my video and I add it on the timeline over here, we will see that my video has no cuts.
So I'm going to play this video for you so you understand that this video has no cuts. This is a raw video. So let me play the video over here.
As you can see, I'm not talking, I'm preparing myself, I'm looking at the camera and now I start. Right? If you look below over here to the audio, you'll see that there's no cut.
If I play again, here I stop, I look at the camera, I think what I'm about to say next. So that's how I record a video for YouTube. There's a lot of cuts, there's a lot of mistakes, and I have to make those cuts myself.
And it takes a lot of time. But now with ChatGPT, you can have ChatGPT make the cuts for you, which is absolutely insane. So let me show you how to do it.
I'm going to leave the video on the timeline over here. I'm going to open ChatGPT over here. I'm going to activate the vocal message.
And the first thing I'm going to do is ask ChatGPT to connect to DaVinci Resolve. Hey, ChatGPT, can you connect to DaVinci Resolve? DaVinci Resolve has an MCP server and I want you to connect to it because in the future, I want to send you commands such as edit the video for me.
That's it. I'm going to send this message to ChatGPT and within one or two or three minutes, ChatGPT should search for the MCP server and it should connect to it. As you can see, I'll check whether DaVinci Resolve MCP server is available and how it's configured.
And it's basically going to do everything for you. You don't have to do nothing. You don't need to know how to use...
the terminal, right? Because before we had to use the terminal and send those commands over here in the terminal ourselves. But now, ChatGPT is going to do everything for you.
Connected and verified with DaVinci Resolve. After this, what you want to do is restart Codex. So just close ChatGPT and open ChatGPT again.
So I just closed ChatGPT and opened it again as you can see over here. And what I'm going to do is I'm going to change the effort level to this time I'm just going to increase it to ultra to see what happens. And over here, I'm going to again use the voice message and tell it to edit this video for me, to make the cuts.
Hey ChatGPT, I have DaVinci Resolve Studio opened and I have a video on my timeline. The video has empty spaces and cuts. What I want you to do is look at the video and edit the video for me.
Make the cuts for me. Make sure there's no repetition inside my video and also make sure there are no empty spaces. So basically, you're creating the montage for my video.
This is a one raw file which needs cuts in order to half low and sound good. So go ahead and edit this video for me on the timeline.
I'm going to put it, I'm going to put charge video right here. Let's see if this is going to work. So as you can see, it knows that I have a 25 minute recording on my timeline.
That's already insane. It's also saying I'm checking the spoken content. So it's going to listen to the audio so it knows where to make the cuts.
So as you can see over here, DaVinci Resolve started transcribing the audio and I didn't do anything. I'm not, I'm not even going to touch my mouse anymore. I'm going to keep my hands over here.
So as you can see, it analyzes the audio and it's creating a transcript. This is insane. I'm sorry.
I'm really excited. I've seen no one share this online. This really changes everything.
This literally changes video editing forever. This is just the beginning, right? Imagine 2 years from now, 4 years from now, 10 years from now.
Obviously, you're still going to need humans to make cuts and decisions if you want to make a movie or if you want to be authentic but this is going to make things so much more efficient and it's going to save editors a lot of time because honestly, creating the montage, it's boring. It's very repetitive. Over here on ChatGPT, as you can see, it says, allow ChatGPT to use Resolve.
I'm gonna click allow this conversation. No, I'm gonna click always allow because I don't want to always click allow. So from my understanding how it works it uses Resolve to create a transcript as you can see over here we have the audio with the transcript and then based on that transcript it's going to know where to make the cuts on the video because it's using the transcript combined with the audio.
It's pretty cool if you think about it. So over here ChatGPT observed that the recording has many retakes of the same lines so most of the cleanup will come from keeping one complete delivery for each point. That's exactly what I do when I have to cut and montage one video for myself.
Sometimes, I speak the same thing three times because I don't like the delivery of it and I only pick the good one. So, Chajipati should do this for me right now. Now, the reason why I keep telling you how this works and explaining all those things is because I find it incredible.
This is really interesting to understand how Chajipati is using DaVinci Resolve to understand the audio and then make the cuts. This is really, really interesting. Now for this prompt, I'm not sure if it's editing this timeline over here in real time or if it's creating another timeline for us with all the cuts ready.
We'll wait and see. So it just created a new timeline, I think, and I'm not sure what's happening. So transcribing audio again.
Let me actually check out what happened over. Oh my God, it made all the cuts. No way.
Look, it made all the cuts already.
Oh my God, wait. Right now in order to get monetized on YouTube. It keeps making cuts.
Oh my god. Let me see this. This is crazy I want to see how good the cuts are like how much empty space there is between the cuts Let me zoom in a little bit That's a lot of watch hours, oh my god, that's really good watch hours That's a lot of watch hours.
That's really good. Wow. Okay, I'm speechless.
I'm gonna stop ChatGPT now. And the reason why we didn't see the cuts in real time is because I was zoomed in on the timeline and you couldn't see the cuts. So what I'm gonna do is I'm gonna create a new project and this time I'm gonna be zoomed out so you can actually see the cuts.
And as you can see over here on the timeline, this footage also has tons of empty spaces. As you can see, I'm just looking at the camera. I'm arranging my hair.
Over here, tons of empty spaces. So I'm going to ask ChatGPT again to edit this footage for us so we can see how it edits in real time. So I'm going to ask ChatGPT to edit the footage.
Hey, ChatGPT, can you edit this footage for me? I've added a new footage. It's around five gigabytes long.
I want you to remove the pauses and make all the cuts so the video flows. So just like before, it's transcribing the audio because this is the only way it can understand where to make the cuts because the audio is synced to the footage so it exactly knows where to make the cuts. Okay, so it opened a new timeline.
ChatGPT is using your computer, okay? I have no idea what's happening. So my hands are here and I have a message telling me that ChatGPT is using my computer.
My screen recorder can't see it so I'm going to use my phone to show you what I'm seeing. my phone to show you what i'm seeing so it's saying chat gpt is using your computer and i have these blue lines around my computer oh and i think it's starting but i already see a clip on the timeline can i actually zoom in myself yeah let's see what happens okay it started adding clips and cutting them there we go it added another clip This is incredible.
Oh, it started moving my mouse. That's crazy. This is the first time I see this.
This is the first time I let ChatGPT control my computer. So it literally moved the mouse for me. I actually want to record this with my phone as well so you can see what I'm seeing.
Oh my god, it's doing cuts. So it keeps making cuts. As you can see, my mouse is literally over here.
My hands are here. And it just makes cuts for me. Look at it go.
That's so interesting.
It added another clip over here. This is incredible. So if I take the mouse myself and I zoom out, we can see all the cuts over here being made.
But this is not it. Here's where things start to get crazy. We can actually use Hixfield's MCP and then have ChatGPV create motion graphics for us and automatically add them inside DaVinci Resolve and sync them.
to our audio and visuals. Now, let me show you how you can do this. So, in order to have ChatGPT create the motion graphics for us, we need to connect the Hixfield MCP.
For that, there's going to be a link down below in the description to Hixfield's MCP. Go in the description and click on that link. Once you click on that link, it should send you to Hixfield over here.
On Hixfield, make sure you select MCP and CLI over here at the top. After you select MCP and CLI, make sure you select ChatGPT and then all you have to do is click over here where it says add Higgs Field plugin.
After you click on this button, it's going to add the Higgs Field MCP to ChatGPT. That's it. That's all you have to do.
Make sure you have a Higgs Field account and just connect that to ChatGPT. It's really that simple. Now, with Higgs Field, we can generate the motion graphics, we can generate text, we can generate 3D things, we can generate a lot of things.
So, let me show you how to do this. I'm going to open the Venture Resolve. I'm going to open ChatGPT and I'm going to prompt ChatGPT the following thing.
hey chad gpv can you use the higgs field skill to generate motion graphics for the video you just cut Basically, what I want you to do is create transparent motion graphics such as text or animations like cards that showcase whatever I'm talking about. Not only that, but you can also generate entire horizontal footage that basically showcases what I'm talking about.
And overall, I want you to keep a specific style. For example, I always want a beautiful gradient blue as the background. The text should be white and bold and beautiful.
And try to make some motion graphics that basically help visualize what I'm talking about. about in this video. That's the prompt I'm gonna be sending to ChatGPT and now ChatGPT should start creating the motion graphics using the Higgs field scale and after it creates them it's going to add them to DaVinci Resolve.
So as you can see it started editing the timeline.
So I have no idea what it's doing as you can see my mouse is over here but things are happening on the timeline. So we can see over here in ChatGPT, we can see the motion graphics playing. They look incredible.
Hicks Field is doing a really good job at generating those motion graphics. They look clean and they're all using the same blue background. Creates that consistency in our style.
Oh my God, it just started adding graphics as you can see. Okay, there we go. It added another graphic.
Okay, look at the timeline. It's adding all the graphics for me. And it keeps adding graphics.
Look at it. And it's checking everything. My hands are literally over here.
I'm doing nothing. It's editing the entire video for me. This is insane.
All right, I think it's ready. And we can already see the graphic over here. It looks amazing.
Oh my God, this is crazy. This is crazy. What am I doing?
I don't think I should even post this video. Seriously, what am I doing? Okay, I'm just going to play the video.
I'm really curious. This is crazy. So let's see the final result.
Let's continue with step number three. One of the biggest problems small creators face, what's really interesting is how synced the visuals are to the script. As you can see over here.
And this is why originality has... Originality matters. And it showcases the card with originality matters.
That's insane. Let me see what else we have here. Think to yourself that maybe people are not going to be interested.
What if people don't click? What if... Interesting.
They're talking about right now. for example if several fishing as you can see i don't like how this card over here shows on the graphics but the graphics look amazing so what we can do is make sure we send a proper prompt that will only show the graphics and this is going to look incredible fishing communities are suddenly talking about a specific fish or a specific fishing rod you can research that topic and make a video about it This looks really good.
So if there's graphics you don't like, you can just delete them and recreate them. Yeah, this is absolutely insane. Now, if you want to learn more, there's going to be a video that appears on the screen.
You might want to click on that video because inside that video, I'm going to teach you how you can monetize your YouTube channel much faster. So if that sounds interesting, you might want to watch that video.
The Hook
The bait, then the rug-pull.
A YouTuber opens mid-demo: ChatGPT is already inside his DaVinci Resolve timeline, cutting his footage while his hands stay off the mouse. The title calls it a permanent shift for YouTube. The video is the receipts.
How they asked for the click.
VERBAL ASK
10:30link
“There's going to be a link down below in the description to Higgsfield's MCP. Go in the description and click on that link... make sure you select ChatGPT and then click add Higgs Field plugin.”
Clear step-by-step click path to a referral link for the second tool in the demo, placed in the description rather than spoken as a price or ask.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
