Big Idea

The argument in one line.

A timestamped transcript turns AI image generation into an automatic b-roll pipeline: name each generated image after its timestamp, and video editing becomes drag-and-sync instead of hunt-and-place.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You make talking-head or voiceover YouTube videos and spend hours manually finding or placing b-roll to match your script.
You're comfortable installing a CLI tool and pasting commands into a terminal, even if you've never used one before.
You want a repeatable, style-flexible way to generate on-brand visuals (photoreal, sketch, etc.) for every line of a script.
You already have or can quickly get an audio narration track and just need the visual layer built around it.

SKIP IF…

You edit primarily with live-action footage or interviews rather than voiceover-driven explainer content.
You need precise creative control over composition — this produces raw AI stills, not directed shots.

TL;DR

The full version, fast.

The workflow starts with a narrated video's audio, run through TurboScribe to get a transcript with per-line timestamps. That timestamped script becomes the input to a single Claude Code prompt: generate one AI image per timestamp (via the Higgsfield CLI skill and the GPT Image model), and name each file by its exact timestamp. Claude Code then batch-generates the images in parallel — 24 in one run in the demo. Because filenames match timestamps, the images can be dragged onto a CapCut timeline and snapped directly to those points with no manual searching. From there, standard editing polish (webcam mask, background layering, blend modes, transitions) turns the raw stills into a finished video, and swapping a single line in the prompt changes the entire visual style, from photoreal to sketch.

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 – 00:49

01 · Hook + install Claude Code

Creator states the promise (edit a video in minutes, not days) and walks through downloading and installing the Claude Code desktop app.

00:49 – 04:24

02 · Install the Higgsfield CLI and skill

Opens Higgsfield's MCP & CLI page, creates an account, and runs the install / sign-in / skill-install commands in PowerShell, selecting the 'Higgsfield Generate' skill as a project-level, symlinked install.

04:24 – 05:36

03 · Verify the skill and test image generation

Confirms the skill installed by asking Claude directly, then tests it with a throwaway prompt ('create an image of Earth from space') to show it can call out to Higgsfield and return a generated image.

05:36 – 06:54

04 · Why you need the audio, not just the script

Explains that the real workflow needs the raw narration audio (not the written script) because only the audio can be run through a transcription tool to recover timestamps.

06:54 – 07:46

05 · Transcribe the audio with TurboScribe

Uploads the audio file to TurboScribe (three free transcriptions, no sponsorship) and gets back a transcript with a timestamp attached to every line.

07:46 – 08:31

06 · Why timestamps unlock automation

Makes the key point: because each script line now has a timestamp, an image can be generated and named after that exact timestamp, which is what lets everything sync automatically later.

08:31 – 09:19

07 · Build the image-generation prompt

Walks through the reusable prompt (written earlier with an LLM's help) that tells Claude Code, via the Higgsfield skill, to generate one 16:9 image per timestamp, in a specified style, and to name each file after its timestamp.

09:19 – 11:23

08 · Run the prompt with the script attached

Pastes the prompt plus the timestamped TurboScribe transcript into Claude Code and sends it, triggering the agent to start generating images against the script.

11:23 – 13:58

09 · Claude Code batch-generates and drag-syncs to CapCut

Claude Code confirms it will generate 24 images in parallel across multiple terminals; once done, each file (named by timestamp) gets dragged onto a CapCut timeline and snapped to its matching audio position.

13:58 – 15:07

10 · Layer, mask, and add polish in CapCut

Adds the creator's own webcam footage as a masked corner overlay, a paper-texture background, a multiply blend mode on the images, and quick built-in transition effects.

15:07 – 16:06

11 · Swap the entire visual style via the prompt

Re-runs the same workflow with one line of the prompt changed (requesting a Da Vinci sketch style instead of photoreal) to show the images — and the whole video's look — change completely with no new tools.

16:06 – 16:27

12 · Outro + next-video CTA

Wraps up and points viewers to a separate video about growing a YouTube channel from under 500 subscribers.

Atomic Insights

Lines worth screenshotting.

Naming AI-generated images after the exact timestamp they belong to turns image generation into an automatic b-roll placement system.
The bottleneck a transcript solves isn't the words — it's the per-line timestamps, which is why the audio (not the script) must go into the transcription tool.
A single, reusable prompt template — written once with an LLM's help — is what makes this repeatable across every future video.
Batch-generating images in parallel (24 in one run in the demo) turns what used to be hours of b-roll sourcing into a single terminal command.
Changing one style-description line in the prompt swaps the entire visual identity of a channel, from photoreal to hand-drawn sketch, with no new tooling.
The images still need manual polish (masking, layering, blend modes) after generation — this automates sourcing and placement, not final composition.

Takeaway

Timestamps turn AI images into automatic b-roll

WHAT TO LEARN

Once a script has per-line timestamps, AI image generation stops being a creative side-project and becomes a placement system — the filename tells you exactly where the visual goes.

02Install the Higgsfield CLI and skill

Connecting a generation tool to Claude Code is a one-time setup cost (CLI install, auth, skill install) that pays off on every video afterward.

04Why you need the audio, not just the script

The script alone has no timing information — only a transcription pass against the actual audio recovers the per-line timestamps this method needs.

06Why timestamps unlock automation

The moment a script has timestamps, matching visuals to audio changes from a judgment call into a lookup.

07Build the image-generation prompt

A prompt that explicitly demands 'name every file after its timestamp' is what makes the generated output machine-sortable instead of just a folder of loose images.

09Claude Code batch-generates and drag-syncs to CapCut

Parallel batch generation (24 images in one run) is what makes this faster than manual sourcing — the wait is minutes, not the hours a b-roll hunt usually costs.

11Swap the entire visual style via the prompt

Testing a style change costs one edited line and one re-run, which is cheap enough to let a channel experiment with its visual identity before committing to it.

Glossary

Terms worth knowing.

Higgsfield: An AI creative platform (and CLI/skill) for generating images and video that can be connected directly to Claude Code so prompts inside a coding agent can trigger media generation.
Claude Code skill: An installable capability that extends what a Claude Code agent can do — in this case, letting Claude call out to Higgsfield's image generation from a normal chat prompt.
TurboScribe: A transcription service that converts an audio file into text with per-line timestamps, used here to recover timing data that a plain script wouldn't have.
GPT Image model: The image-generation model selected inside the Higgsfield workflow to render each timestamped still from a text prompt.

Resources

Things they pointed at.

01:00toolClaude Code ↗

01:19toolHiggsfield ↗

06:12toolTurboScribe ↗

10:40toolCapCut ↗

Quotables

Lines you could clip.

00:00

“I just pasted my YouTube script into Claude Code, gave it one prompt and it helped me edit my entire YouTube video for me.”

states the entire promise of the video in one sentence→ TikTok hook↗ Tweet quote

07:40

“These timestamps in our script now are the reason why we can actually automate the editing of our video.”

names the exact mechanism that makes the whole workflow work→ newsletter pull-quote↗ Tweet quote

12:07

“Cloud Code says, I'll generate all the 24 images, and it's using different terminals to generate each image.”

concrete proof-of-work moment showing the automation actually running→ IG reel cold open↗ 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.

I just pasted my YouTube script into ClockCode, gave it one prompt and it helped me edit my entire YouTube video for me. So instead of spending hours or even days editing your YouTube video, you can now do it within minutes using ClockCode. So in this video, I'll show you how to set everything up, what prompt to use, and how to turn a simple YouTube script into an edited video with ClotCode.

Let's begin. So as usual, you need a Clot account. So just search for Clot AI over here on your browser, go to their website, and create an account if you don't have an account.

Now after you create an account on clot, I want you to go back to your browser and search for clot code, and then click on this link over here, which is called clot code by Entropic. Click on this, and in order to download clot code to your computer, click on this button over here which says get clot code and select desktop.

And it's going to send you to this page over here, and then just click on this button over here which says download for Windows. And it's going to download the installer, click on the installer, and install Cloud Code on your computer.

Now after you install Cloud Code, I want you to open the application. You can literally just go over here to the search bar, search for Cloud, and open the application. And this is going to be your dashboard.

You have over here the chat, the co work, and the code. I want you to select code over here to the top left. Now in order to have Claude code help us edit videos for us, we need to install the right skill, and for that, we're going to need Higgs Field.

So what I want you to do is go to Hicksfield. There's going to be a link down in the description. If you click on this link, it's going to send you over here to hicksfield.ai.

I want you to click on this link and go over here to Hicksfield. If you don't have an account, create an account by clicking over here to the top right, sign up. Click on sign up and create an account or log in with your Google account, whatever you prefer.

Now after you create an account, I want you to click over here to the top left where it says MCP and CLI. Click on this. Now over here, we have a bunch of options.

We have MCP, CLI, and scale. This is pretty much the scale that will help us create footage for our videos, and the easiest way to install it is to use the CLI over here. So click on the CLI over here, and the first thing you wanna do is copy the first command over here.

So in order to copy, you can just click on this icon over here one time, as you can see, and it's going to copy this command. Now what I want you to do is go to your search bar over here and search for terminal. Just like this.

As you can see over here, it says terminal. Or you can also search for PowerShell. They're the same thing.

Open terminal or open PowerShell, and now paste this command over here. As you can see, I've pasted this exact command from Higgs field. As you can see, it's the exact command.

And then click enter. Now once it's done installing, it should say something like this. Change one package in three seconds.

That's all it should say. If it also says NPC notice, that means there's an update. But if you don't see this notice over here, that doesn't mean you didn't install it.

You still installed it if it just says change one package in thirty three seconds. Now what I want you to do is go back to Higgs Field over here and copy the second command, which is signing in. So copy this command over here by clicking on this icon over here.

Go back to your terminal over here and paste the command. What this does is going to ask you to authorize access to Hicks Field with your account. As you can see, it literally opens a new window, and all I have to do is click over here on the button that says connect.

So click on this button over here that says connect, and now you've basically connected your terminal with Hicks Field. You can now close this window and go back to Hicks Field and copy the last command, which is the skill over here. As you can see, this one over here.

Click on this icon and copy the command. Now go back to your terminal over here and paste the command. Hit enter.

And now in order to continue, press I on your keyboard over here as you can see. Okay. To proceed, I.

So type I and then click enter. Now over here, you have a few different options, and you can select them using your arrow keys on the keyboard. As you can see, you can move through them.

But the option you wanna choose is the first one which is Higgs field generate. To select this option, just press space on your keyboard. So not enter but space.

So press space, and as you can see, you've selected this option. Now you can press enter on your keyboard. Now you have the option to select if you wanna install this skill as a project or global.

So I'm gonna select project and press enter. Then it's going to ask you for the installation method, which is Symlink or copy to all agents.

I'm gonna choose Symlink. So choose Symlink and press enter. Then it's going to ask you if you wanna proceed with the installation, so just select yes and press enter.

And there we go. We've installed one skill. It's done as you can see.

If you have this message over here, you've successfully installed the skill. Now what you can do is close this terminal over here by pressing x, go back to your Claude app, and in order to check if the skill is actually installed, you can literally prompt Claude something like, is the skill Higgs field installed?

So if Claude gives you this message over here, that means you've successfully installed the Higgs field skill, and now you're going to be able to generate footage for your YouTube videos. Now let me show you how we can actually use this skill to help us edit our YouTube videos because this skill is good at generating images.

Right? So for example, if I type over here in the chat something like create an image using Higgs field skill of earth from space.

Right? And in a few seconds, Claude code plus Higgs field is going to generate me the image.

So Claude created the image. If we click over here, we can look at the image. So over here, we have Earth as seen from space, which is super awesome.

This is a super high quality image, but this is not going to help us automate the creation of YouTube videos or social media content. Therefore, I wanna show you how we can use this skill to actually automate the process of creating content for YouTube or any social media platform. Now for this to work, what you need is to have your video ready.

So as you can see over here, I have my video ready, which I'm about to upload to my YouTube channel in the future. So if I hit play, I want you to listen to the audio. Right?

Every creator runs into the same question at some point. How often should you An action post. This video is raw.

There are no edits. I only have the video. Now what you wanna do is render the audio only.

As you can see over here, I have the audio. Every video runs into the same question at some point. Basically, what I'm trying to say is that for this to work, you first need the audio of your video, not the script.

You also need the audio. So if you have the script, you need to go ahead and record the audio because we need the audio. So now once you have the audio, what I want you to do is go back to your browser over here and search for TurboScribe.

Now I'm not sponsored by them. The only reason I'm showing them is because they offer three free generations. So go to this website, create an account, and then on the dashboard, this is what you should see.

Over here, if you click transcribe files, you can upload the audio file over here. So I'm gonna show you. I'm gonna upload the audio file over here, and then you can click over here where it says transcribe.

Click on this and wait for your audio file to be transcribed. As you can see over here, the status is preparing to transcribe this file. This usually takes around thirty seconds to one minute.

Now once your audio file is transcribed, you should see this green icon over here, this verified icon. You can literally just click on it like this, and as you can see, we have the transcribed of the audio, which is the script.

Now you might be asking yourself, why didn't I just use the script? Well, because if we transcribe the audio, we can actually get the timestamps.

As you can see over here, we have zero point zero zero seconds, then we have three seconds, then we have six seconds, then we have thirteen seconds, seventeen seconds, and these timestamps in our script now are the reason why we can actually automate the editing of our video. Actually, just let me show you. Pretty much now what you need is the prompt, and this is the prompt I'm gonna be using.

Now it's a big prompt, but don't feel intimidated by it because you can create this prompt literally with Claude or with Chad GPT or whatever. Right? This prompt was created with Chad GPT.

As you can see, the prompt says this. Here is my video script with timestamps. I want you to use Higgs field with the gbt image two model to generate one image for each time stamp.

So this over here is the most important part because now we have the time stamps and we can use Cloud Code to generate an image for each time stamp. This already automates a lot of the process. Right?

And then the prompt says every image must be horizontal in a 16 by nine format. Right? So we want the horizontal format.

And then the second most important part is this one over here. Please name every image with the exact timestamp from the script so I know exactly where to place them in the video. So, basically, each image is going to be titled based on the timestamp.

So the first image will have the title 000. The second image will have the title 0Point03. And then over here is going to explain the style of the image, and this is something you can customize, and I'm gonna show you how you can customize it using Claude.

But for now, I'm gonna use this prompt over here, so I'm gonna copy this entire prompt. Right? Actually, I'm gonna share a link down below in the description to this prompt so you can literally just copy this prompt yourself.

So copy this prompt over here just like this. Right click, copy, and now go back to ClothCode. And inside ClothCode over here where it says type for commands, paste this prompt, and at the end of the prompt, just say here is the script.

Right? And now go back to your browser to TurboScribe over here and copy the script.

Don't download it over here as a PDF because if you download it as a PDF, as you can see, you will not have the time stamps. What you want is to copy it directly from here. For this example, I'm just gonna copy one minute of my script over here until it says one minute over here.

So I'm just gonna copy this, and then I'm gonna go back to my clot code, and I'm going to paste the script over here after the prompt. And I'm going to send this prompt with the script to clot.

Right? And now clot code is going to ask me to allow access a bunch of times, and then it's going to generate each image from my video based on the script and based on the time stamps. Now once Cloud Code is done generating the images, as you can see over here, Cloud Code says, I'll generate all the 24 images, and it's using different terminals to generate each image.

You should have a folder like this one on your computer over here. And inside the folder, we're going to have each image based on each time stamp inside our script. So if we click on an image, we'll see this is the first image, second image, the third image, and you will see that the title of each image is basically the time stamp inside the script.

Now what you can do is literally open your editing program such as CapCut or Premiere Pro or whatever editing program you use. So I'm gonna show you how quick this is. I'm gonna use CapCut for this because it's free, you can download CapCut for free on your computer.

I'm gonna select all the images. I'm gonna add them over here where it says import, and then I'm gonna drag each image on the timeline. And as you can see, I'm not even using the audio.

And let me show you how fast this is to edit. So for example, the next image over here is going to appear at three seconds. What I can do is I can drag this first image until three seconds over here and then drag the next image.

Right? And then I can look at the third image, and I can see it is going to appear at six seconds. So I'm gonna drag this one until six seconds over here, Then I'm gonna drag the next image.

The next image appears at nine seconds. So I'm gonna drag this one until nine seconds. And now if we put the audio or the video, which also has the audio, so I'm gonna put my video over here.

I'm gonna drag it, uh, over here. Let me put the images on top. And now if I hit play, you'll see how each image is synced to my audio.

So whatever I say in the audio is going to be synced to the image. So every creator runs into the same question at some point. How often should you act There we go.

Synced on YouTube. Because when you look around, the answer makes no sense. Some channels upload three videos and suddenly explode.

Other channels upload Hundreds. That's how fast you can now edit videos using Cloud Code, and you can make this look really nice. For example, what you can do is basically add your video on the top just like this.

I'm gonna mask it really quick. I'm gonna position it over here to the left. I'm going to connect all those images into one compound.

I'm gonna decrease the size of them just like this over here, and then I'm gonna move them over here to the left just like this. I'm going to add a paper background below them, increase the size of the paper background, and then on the image, I'm gonna choose the blending mode to multiply.

And we can go ahead and add a few effects to make it look nicer. As you can see over here, I just added some quick effects from CapCut. You literally just drag them on the timeline.

It takes a few seconds. And if I now hit play to see the video, it's going to look like this. Every creator runs into the same question at some point.

How often should you actually post on YouTube? Because when you look around, the answer makes no sense. Some channels upload three videos and suddenly explode.

Other channels upload hundreds of videos, and, basically, what you wanna do from here is continue to add the images Cloud Code is generating over here. Now the cool thing about this is that you can change the style of the prompt, and you can generate images that look completely different.

And let me show you how that would look like. So I used ChatGPT over here to change the style of the images over here. I said I want some kind of DaVinci sketch style like these ones, and Chad GPT recreated the prompt.

So I'm gonna take the prompt over here, copy it, go back to Claude. I'm gonna paste the prompt. Here is the script.

I'm gonna take the script again, and this time I'm just gonna take the first seconds just so I can quickly generate the images. I'm gonna paste the script with the time stamps. And, again, as soon as Cloud Code starts generating the images, you're going to have a folder on your computer like this one over here.

And inside the folder, you're going to start to have each image. As we can see, we already have the first image, and it looks like this. It's in the style of Leonardo da Vinci.

Like, this paper kinda drawing with these things over here, it looks really nice. It actually looks really fun. So, basically, what I'm trying to say is that you can create any style you want for your channel.

You can actually be very creative with this. Not only that, but you also have the images saved on your Higgs field dashboard. So if you go to Higgs field over here, you'll see that on your image over here, you have all the images you've created saved over here forever.

So if you ever wanna go back to your images, you can literally click on them and download them to your computer. And right now, this is pretty much the only way and the cleanest, most legit way to automate your YouTube content. It just looks super clean and makes your presentation easier to look at.

Now that you know how to semi automate content creation on YouTube, you might wanna watch this video over here because inside this video, I teach people who are under 500 subscribers how to grow on YouTube. I talk about the algorithm. I show my analytics, and I talk how I grew this channel over here.

If that sounds interesting, you might wanna click on this video and watch this video if you wanna learn how to grow on YouTube.

The Hook

The bait, then the rug-pull.

Instead of spending hours or days cutting b-roll to a voiceover, this walkthrough pastes a timestamped transcript into Claude Code and lets an AI image-generation skill fill in every beat automatically — turning editing into drag-and-sync.

Frameworks

Named ideas worth stealing.

07:40concept

Timestamp-Named Image Generation

Transcribe the audio (not the script) to get per-line timestamps
Feed the timestamped script into one reusable AI prompt
Name every generated image after its exact timestamp
Drag files onto the timeline — filenames tell you where they snap

A method for turning AI image generation into automatic b-roll placement: the timestamp embedded in the filename is what lets visuals sync to audio without manual matching.

Steal forany voiceover-driven explainer, listicle, or faceless-channel video that currently requires manual b-roll sourcing

CTA Breakdown