Big Idea

The argument in one line.

A single master prompt loaded into Claude Code is enough to replicate any YouTube channel style end-to-end -- branding, scripting, visual generation, and thumbnails -- without writing a line of code.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You want to build a faceless documentary or AI-visual YouTube channel and have no idea where to start.
You have tried juggling separate tools for scripting, design, and animation but cannot get a consistent output.
You are comfortable following a step-by-step screen-capture tutorial and want a repeatable system.
You want to understand what Claude Code can actually orchestrate end-to-end in a content workflow.

SKIP IF…

You already run a production pipeline and want API-level automation -- this stops at manual prompt-to-tool handoffs.
You want final-cut video editing instruction; this covers prompt generation, not post-production assembly.

TL;DR

The full version, fast.

A master prompt loaded into Claude Code turns it into a 10-stage content engine. Pick a channel to model, feed in transcripts and screenshot references, and Claude generates: channel name and branding prompts, a niche-matched script with runtime tied to word count (200 words = 1 minute), 167 scene-level image prompts for a 10-minute video, per-scene animation prompts, and thumbnail concepts. External tools handle actual media generation; Claude handles the strategic orchestration. No coding required.

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 – 01:10

01 · Cold open: the claim

Promises a full A-to-Z pipeline for cloning any YouTube channel with Claude Code

01:10 – 02:30

02 · Setup: Claude Desktop and master prompt

Download Claude, switch to Claude Code workspace, paste the master prompt from the description

02:30 – 05:08

03 · Stage 1: Channel to clone and niche description

Pick a model channel (Fern, Blackfiles), describe the niche, add screenshots for visual reference

05:08 – 06:28

04 · Stage 2: Branding brief

Claude outputs channel name variants, description, logo prompt, banner prompt

06:28 – 08:00

05 · Disclaimer and transcript collection

Educational disclaimer; how to pull YouTube transcripts from rival channels and save as PDF

08:00 – 09:30

06 · Stage 3: Upload transcripts and get video ideas

Upload PDF of model transcripts; Claude generates 15 niche-matched video idea titles

09:30 – 10:56

07 · Stage 5-6: Style DNA and script generation

Claude maps the niche style DNA; writer requests a script with a target word count (200 words = 1 min)

10:56 – 13:30

08 · Stage 7: Visual style and 167 image prompts

Upload 5 screenshot references; Claude studies the visual style and generates a prompt per scene

13:30 – 16:00

09 · Stage 8: Image generation on Google Flow

Copy prompts into Google Flow / Nano Banana 2; use reference locking to keep character consistent

16:00 – 18:10

10 · Stage 9: Video and animation prompts (optional)

Claude generates per-scene animation prompts; use VOT model in Flow to animate each image

18:10 – 21:21

11 · Stage 10: Thumbnails

Upload 5 model thumbnails; Claude reverse-engineers style and generates thumbnail concepts and full prompts

21:21 – 25:38

12 · Export and next video teaser

Claude packages everything into a Google Doc; teases voiceover and API automation in the next video

Atomic Insights

Lines worth screenshotting.

A single master prompt can orchestrate 10 distinct production stages inside Claude Code without any additional code.
The 200-words-per-minute rule gives you a direct lever on video runtime before you write a single line.
Uploading competing channel transcripts as a PDF teaches Claude the niche pacing and cadence faster than written descriptions alone.
Feeding 5 screenshot references before requesting image prompts is what makes Claude visual outputs match a specific channel aesthetic.
Reference locking -- using a generated image as the start-frame for the next generation -- is the only reliable way to keep a faceless character consistent across 167 scenes.
Claude Code does not cap prompt length the way the chat interface does, which matters when you need 167 detailed scene prompts in one output.
Google Flow is the strongest free image-generation platform for this workflow because it supports reference images; tools without that feature produce inconsistent results.
The thumbnail stage is where Claude reverse-engineering is most impressive: it reads layout, typography, and lighting from 5 sample images and generates concepts that could go live without editing.
Describing a channel by niche and aesthetic produces better style outputs than listing a channel name alone.
The whole pipeline costs near zero if you stay on Google Flow free tier; the only paid step is the VOT video model at 20 credits per clip.
Stacking multiple reference channels from the same niche sharpens the style model; one channel produces decent outputs, three produces consistent ones.
The export step packages every generated asset into a single plain-text document for reuse across projects.

Takeaway

Claude Code as a production orchestrator, not a writing tool.

WHAT TO LEARN

The most transferable insight here is the pattern: a sequenced master prompt that advances on a trigger word turns a general-purpose AI into a domain-specific multi-stage pipeline.

A master prompt that advances on a trigger word lets you chain 10 or more distinct tasks without re-explaining context each time -- a pattern that applies to any multi-step workflow, not just video.
Feeding model transcripts before requesting new scripts is more reliable than describing a style in words; the AI infers tone, pacing, and structure directly from examples.
The 200-words-per-minute formula gives you runtime control before you start writing; spec the word count, not the topic, when length matters.
Reference locking -- using a generated output as the reference for the next generation -- is the practical solution to visual consistency in any multi-scene AI image project.
Claude Code willingness to generate very long outputs in one pass -- 167 detailed prompts or 1000-word image descriptions -- is a meaningful capability gap over chat-interface models for production workflows.
The bottleneck in this pipeline is not AI capability but tool handoffs: every stage requires a manual copy-paste between Claude and an external generator, which is where API integration would eliminate friction.

Glossary

Terms worth knowing.

Master prompt: A single long system-level prompt that programs Claude Code to act as a multi-stage content production engine, advancing through numbered stages on the word 'next'.
Style DNA: The model internal representation of a channel visual and editorial fingerprint: tone, pacing, format, and typical hook structures, derived from transcript and screenshot references.
Nano Banana: An image-generation model available inside Google Flow, known for photorealistic 3D-rendered characters; the preferred model for the faceless documentary aesthetic shown in the video.
Reference locking: Attaching a previously generated image as a visual reference when requesting the next image, forcing the model to maintain character and style consistency across scenes.
VOT model: Video-over-time model inside Google Flow that animates a still image using a text prompt; costs 20 credits per clip and produces short cinematic motion sequences.
Faceless channel: A YouTube channel that produces content without showing a real human presenter, relying instead on AI-generated visuals, voiceovers, and automated scripting.
YouTube automation: A content strategy that uses AI tools to handle scripting, visuals, voiceover, and editing so a channel can publish without manual creative labor for each video.

Resources

Things they pointed at.

01:30toolClaude Code / Claude Desktop ↗

13:30toolGoogle Flow ↗

12:36toolHiggs Field ↗

16:10toolElevenLabs ↗

20:10toolCapCut ↗

02:32channelFern ↗

03:58channelBlackfiles ↗

Quotables

Lines you could clip.

06:20

“Roughly every 200 words gives you about a one-minute video.”

standalone formula, no setup needed→ TikTok hook↗ Tweet quote

20:50

“This is basically ready to go. You could literally upload this straight to YouTube as your thumbnail.”

visual proof moment, punchline delivery→ 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.

00:00Now with the release of Claude Code and Opus 4.7, it is literally possible to clone and model almost any YouTube channel, whether it's documentary, two d, history, Pixar, fictional, true crime, or pretty much any other niche.

00:16And after hours of research, I've found the perfect pipeline you can use to replicate and model any channel in any niche. So in this video, I'm going to walk you through the entire process step by step.

00:29I'll show you what you need, how to get it, and pretty much everything you need to automate the full process from start to finish. Okay. So the very first thing you're gonna need is a channel you want to model.

00:39For me, I found this channel called Fern. They mostly use these mannequin style silhouettes as visuals, and then they build documentary style stories around them.

00:48Now, the next thing you're gonna do is head over to your browser and search for Claude desktop. When you see the option that says Claude download, go ahead and click on that. Now from here, you just need to pick your operating system.

01:00I'm on Windows, so I'll go with that. But if you're using a Mac, just click on the Mac option and download the macOS version. Alright.

01:07So once you logged in to Claude, what you wanna do is click on those three lines at the top left corner of your screen. That opens up the sidebar, and you'll see a few different workspace options there. You'll notice one called code.

01:17That's the one that gives you access to Claude code. Then there's chat, which is just the normal Claude chat interface. There's also a co work option, but we're not really gonna touch that for this.

01:26So just switch over to Claude Code. Alright. So once you've got Claude set up and you're inside Claude Code, the next thing you're gonna need is the master prompt.

01:35You'll find the link for that in the description. Go ahead and open it, press control plus a to highlight everything, then right click and copy the whole thing. Don't miss anything.

01:46It's usually pretty long. Then jump back into Claude code, paste it in, and send it. Now what this prompt does is basically turn Claude into a content engine.

01:56It helps it understand how to take a channel and sort of replicate that style for you. Alright. So once you send the prompt, Claude is gonna ask you a simple question.

02:05What channel do you want to replicate? Now here, you can just type the channel name, and if you've got the link, even better, you can drop the URL in there too. But from what I've seen, the best results come when you actually describe the channel a bit.

02:18So for me, I'm gonna say something like, a three d documentary channel that makes videos on real life events and historical content. That way, Claude really understands what you're going for. And one more thing, you can also add screenshots from similar channels in that niche.

02:33So if I go back to my browser, I can just grab a few screenshots of the kind of content I'm trying to model and use those as extra reference. Now here's the thing. The more channels you can stack into a specific niche, the better your results are gonna be.

02:48For example, there's this channel called Black Files. It's pretty similar to Fern, but they lean heavily into AI. Their voiceovers, scripting, pretty much everything is automated.

02:58So what that means is the more references you give Claude, the clearer the picture becomes. All you've gotta do is drop in those screenshots, maybe add a short description of the niche you're trying to replicate, and then send everything into Claude code.

03:12Once you do that, Claude takes it, processes everything, and figures out exactly what you're going for. Then it starts giving you outputs, things like channel name ideas, channel descriptions, and even logo and banner prompts you can use to actually brand your channel.

03:26So for me, it came up with names like Red Files, Ghost Files, and Black Vault. And not gonna lie, those actually sound really good. They fit perfectly with the kind of channels in this niche.

03:37I personally like Red Files and Black Vault the most. If I was starting a channel right now, I'd probably pick one of those. It also gave me a really catchy channel description that I can just paste straight into my channel setup.

03:50And on top of that, it even gave me prompts for generating a logo and a banner, which is honestly pretty cool. So just from this step alone, we've pretty much covered the whole branding side of things. Now what I'm gonna do next is just type next.

04:05Once you do that, Claude moves on and asks you for transcripts. Okay. So transcripts are basically just the script, the exact words people are saying in videos from the niche you're trying to replicate.

04:16So if you're trying to build a system where you don't have to write your own scripts, what you really need are model scripts. Now to get those, just go find channels that are already posting the kind of content you like.

04:29Once you find one, click on one of their most viral videos. Watch it for a bit, get a feel for how it's put together. Then scroll down to the description, click on more, and you'll see an option that says show transcript.

04:41Go ahead and click that. Once you do, YouTube will pull up the entire transcript of the video from start to finish so you can see exactly what was said in that video. Alright.

04:52So once you've got the transcript open, just copy everything, literally the whole thing, and save it somewhere safe.

05:00Now from here, go find another video. Try to pick one that's doing really well, like a popular one with strong views. Do the same thing again.

05:09Open it, grab the transcript, copy it, and paste it into a Google Doc or wherever you're keeping your files. And then just keep repeating that process. Do it again and again until you've got maybe two, three, or even five transcripts saved up.

05:25Honestly, the more you collect, the better. Alright.

05:27Real quick disclaimer. This is strictly for educational purposes. I'm not teaching or enabling anything illegal here.

05:34I'm just showing you how to create channels in any niche you want to model using Claude code. So once you've got your transcripts and you've saved them into a PDF like I did, here's what you're gonna do next. Upload that PDF into Claude code.

05:49Once it's uploaded, just type something like transcript attached and send it. Now Claude might ask you for a couple permissions. Just go ahead and allow them.

05:58It needs that access so it can actually read through the transcripts. And here's where it gets interesting. Once it has those model scripts, it starts understanding the pacing, the format, the tone, the cadence, basically, how those scripts are built.

06:12And from there, it can start generating new scripts that follow that same pattern. Alright. So once Claude Code finishes reading your transcripts, it's gonna ask if you want ideas in that niche.

06:24Just say yes, you want ideas, and send it. Now once you do that, Claude Code starts generating a bunch of video ideas for you, usually around 15 of them.

06:34And if you actually read through the titles, you'll notice they match the exact style you're going for. You'll see stuff like The Teen Who Hacked NASA From His Bedroom or the spy who sold the West to the KGB for twenty years. Just that same kind of vibe we're trying to replicate.

06:52So from here, it's pretty simple. Just go ahead and pick one idea you want to work with. Alright.

06:57So for me, I'm picking idea number one, the teen who hacked NASA from his bedroom. I'm just gonna type one into the prompt box and send it. Now once you do that, Cloud Code gives you a full breakdown of how your video is gonna look.

07:11It covers everything. The niche, the target audience, the hook style, the script flow, the tone. Basically, a full analysis of the kind of content you're about to make.

07:21After that, it's gonna ask if you want a script. For me, yeah. I want the script.

07:26I've already looked through the overview it gave me, and everything checks out, so I'm good to go. One small thing I like to do here. When I ask for the script, I include a word count in brackets.

07:37So I'll say something like 2,000 words. That part actually matters because the length of the script directly affects how long your video is gonna be. Alright.

07:48So here's a quick way to think about length. Roughly every 200 words gives you about a one minute video. So if you go for something like 2,000 words, you're looking at around a ten minute video.

08:00From there, it's just basic math. You can scale it however you want depending on how long you want your video to be. So if you're aiming for, let's say, a twenty minute video, you'd want around 4,000 to maybe 4,500 words.

08:15Now once you send the request, Cloud Code starts writing the script for you. And as you can see, it's actually following the same tone and cadence as the model scripts we gave it.

08:26It's structured properly, includes time stamps, everything looks really on point. And honestly, one thing I really like about Cloud Code is that it doesn't limit you the same way other models do.

08:37You can push it up to, like, 5,000 words if you want, which is a huge win if you're trying to make longer videos. Not gonna lie, this script actually looks really good. It genuinely feels like something you'd see on a proper model channel.

08:48So, yeah, from here, just copy the whole script and paste it into a Google Doc or somewhere safe so you don't lose it. Alright. So from here, just type next to move on to the next part of the pipeline.

09:01Once you do that, you'll land on stage seven. And this is where you add your sample visuals. Now these are basically anything that shows the kind of style you want.

09:09Could be images, silhouettes, even short video clips. Just something that represents the look you're going for.

09:16And the easiest way to get these? Just go back to the model channels you're trying to replicate and take a few screenshots. That's usually enough to give Claude Code a solid idea of what your visuals should look like.

09:28Alright. So at this point, just grab a few screenshots from videos, like three, four, maybe five, and save them on your computer. The key thing here is this.

09:37The more examples you give Claude Code, the better it understands what you're going for. So don't stress it. Just pick some solid screenshots that match the style.

09:46Once you've got them, go ahead and upload them into Claude Code. For me, I already have visuals I like. They're these glossy mannequin style shots with mannequins wearing police uniforms, kind of similar to the channel I'm trying to replicate.

09:59So I'm just gonna upload those. Quick heads up though, ClaudeCode only lets you upload up to five images, so keep that in mind. Alright.

10:07I've added my five images. Now I'm just gonna type images attached and send it.

10:14Once you do that, Claude code basically studies those visuals and understands the style you want. It learns how to prompt image models to recreate that same look. Now if you're a bit more advanced with coding, you can connect something like a nano banana API to automate all your image generation, which is pretty powerful.

10:31Claude would just handle everything for you. But if you're a beginner, don't worry about that at all. Just use any external tool to generate your images and keep it simple.

10:40So, yeah. Once Claude understands your visual style, all you've gotta do next is say, I need prompts for my script. Alright.

10:46So once you ask Claude code for prompts for your script, here's where it gets really cool. What it does is go through your entire script and figure out how many scenes are actually in there. So in our case, we had a 2,000 word script, and Claude analyzed it and came back with about 167 scenes for the full ten minute video.

11:07And honestly, that's huge. Because now you can create visuals that match every single part of your script instead of just guessing. Another thing I really like about this is that Claude code doesn't really hold back on output.

11:21It can generate super detailed prompts, like 500 to even 1,000 words, just for your visuals. So if you're thinking bigger, like making longer videos with hundreds of beats, maybe 700 or 800 scenes for something close to an hour, Claude can actually handle that pretty easily.

11:40And that's something you don't usually get with other models, since they tend to have limits on how much they can generate in one go. Alright. So right now, Claude code is basically doing its thing.

11:50Just running in the background and writing out every single image prompt you'll need to match your script. Now, quick note for the more technical people. If you've got an API key for something like nano banana or wave speed, you can actually plug that into Claude code and have it generate all your images automatically.

12:09Like, it'll go through every prompt, create each image, and even store all of them on your computer.

12:15So in our case, that's like 167 images done for you. But, yeah, that part is a bit more advanced.

12:23And not gonna lie, it usually costs money since you need to pay for access to those APIs. And just so you know, an API is basically how different software tools talk to each other. We'll get into all of that in another video.

12:36For now, don't worry about it. Just let Clogcode run and finish generating all the prompts you need for your images. Alright.

12:44So once you've got all your prompts, the next thing you're gonna need is an image generator. Now if you're on a paid plan or you don't mind spending a bit, you can use tools like Higgs Field or Open Art.

12:55Those give you access to models like Nano Banana, Sea Dream, or Flux Pro, and those are really good for getting high quality, realistic images. But if you're going the free route, no problem.

13:07You can use stuff like image effects, Wisk, or Google Flow. Personally, I'd say go with Google Flow or Wisk. And here's why.

13:16They let you upload a reference image before generating anything. That part is huge because once you give it a reference, the model really understands the style you're going for and keeps your visuals consistent. I'll break that down properly in a second.

13:29Alright. So Claude just finished giving me all the prompts and wow, it's a lot. It gave me about 167 prompts, which is honestly pretty crazy.

13:38Now to actually generate your images, you're gonna need an external platform like we talked about earlier. So what I'm gonna do is copy my first image prompt. Just copy it.

13:47And then head straight to my browser. From here, you can use platforms like Flow, WISC, or ImageFX if you're sticking to free tools.

13:56But quick heads up, most of these platforms have limits. And from what I've seen, some of them might even get discontinued by April 30. So, yeah, just keep that in mind.

14:05That said, Flow is still a really solid option. I think the subscription is around $2, and it unlocks a lot more.

14:13Plus, you usually get about a 150 credits for Nano Banana, which is pretty useful for generating your images. So I'm just gonna create a new project in Flow. Once you're in, go over to the model section.

14:23Here, you can choose what you wanna use. Nano Banana, Nano Banana Pro, or even Imagen four. For me, I'm gonna go with Nano Banana two.

14:32Then I'll just paste in my prompt and we're good to go. Alright. So once I've pasted the prompt, I'm just gonna send it.

14:38Give it a couple seconds and you'll get an output that looks like this. Now, this one is more like a background image since it's showing the intro part of the script. And, honestly, it looks really good.

14:49So I'm just gonna download it. Next, I'll head back to Claude Code and look for a scene that actually features a person. I'm gonna grab the second prompt, highlight just that section, and copy it.

14:59Quick heads up here. Don't right click and hit copy message, because that's gonna copy everything. What you wanna do is select only the part you need, and then press control plus c on your keyboard.

15:11Once that's copied, go back to Flow, paste it in, and then just send it to generate your next image. Alright. So this is what the next output looks like.

15:19And as you can see, it pretty much matches the style of the model channel. It looks really good. Not gonna lie.

15:25It's kinda crazy that this came from just one prompt. Even the small details like the posters on the wall look super realistic. Overall, the quality is really solid.

15:34Now, one thing I really like about Flow is this. If you start getting inconsistent results, you can just use a model reference.

15:41So, for example, I can click on this image as my reference and then change the prompt to something like, have the character sitting in a movie theater, watching a movie with popcorn in his hand.

15:51Once I send that, Flow keeps the same character but changes the setting. And that's actually really powerful because now you're not starting from scratch every time. And honestly, with the way Claude writes these prompts, this whole process becomes really easy.

16:04Anytime you see a prompt that involves your main character, like the mannequin, you can just reuse that reference from one of your generated images. That way, you keep the same character across different scenes, and your results stay consistent without you doing anything complicated. So, yeah, that's pretty much how you go about generating your images.

16:21Alright. So you can also use something like image effects. Let me show you real quick.

16:26If I go back, copy one of my prompts, and paste it into image effects, you'll see the output actually looks really good. Image effects is honestly pretty cool. The results come out clean and impressive.

16:37But there is a downside. It might not even be available after April 30. So, yeah, keep that in mind.

16:44Also, one big limitation is that you can't upload a reference image, and that matters a lot. Because without a reference, it's harder to keep your visuals consistent. That's why I always lean toward Google Flow.

16:57Whisk is there too, but it's not really as strong in my opinion. Google Flow just wins overall, especially when it comes to keeping things consistent.

17:05So, yeah, that's pretty much how you generate your images once Claude gives you the prompts. Now, from here, just head back to Claude, type next step, and send it to move forward. Now, this takes you to step nine, which is the video prompts.

17:18Basically, these are the animation prompts you use with your images to turn them into actual videos. Now, quick thing to keep in mind. This step isn't always needed.

17:27It really depends on the kind of content you're making. Some niches don't even require animation, so you can skip this part if it doesn't apply to you.

17:35You'll also see that Claude gives you an option here. It'll ask if you want video prompts for your niche. So you can just say yes if you want them or no if you don't.

17:45For me, I'm gonna need video prompts, so I'm just gonna say yes. But, yeah, just keep in mind, not every niche needs this, so you can always skip it if it doesn't apply to what you're doing. Now once you type yes, Claude basically goes back through your script and all the image prompts it gave you and then starts generating animation prompts for every single scene.

18:05These are what you'll use to turn your images into videos. Now if you're still using Google Flow, here's what you do. Go to the nano banana section and switch from the image mode to the video model.

18:16Once you switch that, the whole interface changes, and now you've got a video setup to work with. From there, you can start generating videos using the v o t model. And honestly, this is such a cool feature.

18:29This is why I really like Google Flow. Everything's just in one place, and it makes the whole process way easier. Now quick heads up.

18:37Even though Nano Banana is free on Google Flow, the VOT video model isn't. It costs about 20 credits each time you generate a video, so just keep an eye on that. If you're on a paid plan, you've got more options too.

18:51You can use models like Cdance or whatever's available on platforms like Higgs Field or Open Art. So it really depends on what you wanna work with. From here, it's pretty simple.

19:01Once you've got your image, just upload it back into Google Flow, copy the video prompt Claude gave you, paste it in, and you're good to go. So just copy the prompt Claude code gives you, head over to Google Flow, paste it in, and send it. Now if we jump back into Claude code, you'll see it gave us about 167 animation prompts.

19:20And that's because it remembers we had around 167 scenes earlier, which is honestly really cool.

19:26It means every single image you created can now be turned into a video if you want. So everything lines up perfectly. You don't have to guess anything.

19:34It's all mapped out for you. Now if you're more on the developer side, you can actually plug in something like a VOD API and automate the whole thing. If you've got APIs for certain platforms, you can connect them to Claude code, and it'll handle everything, from generating images to turning them into videos.

19:50But, yeah, that part gets a bit technical. If you're just starting out, don't worry about it. Just stick with external tools like Google Flow for now, because adding APIs can make things way more complicated than they need to be.

20:03So, yeah, just keep it simple and follow the process we've been using. Alright. So what Claude code is basically doing for you as a beginner is giving you every single detailed prompt you need.

20:14From there, you just plug those into different tools: Eleven Labs for your voiceover, Nano Banana for your images, and VOD to turn those images into videos.

20:24Once everything's ready, you can bring it all into CapCut, edit it together, and you'll get a really solid result. Now after sending one of the prompts to v o d, this is the kind of result we get.

20:34And, honestly, it looks really good, especially for an intro. Let's try another one real quick. I'm going to animate a different image, the one where he's standing in front of the screen.

20:43So I'll just drag that image into the start frame in Google Flow, then head back to Claude code, copy the prompt for that specific scene, come back, paste it in, and send it. Alright.

20:54So after a couple minutes, we get an output that looks like this. And honestly, it looks really good. You've got the camera zooming into the computer while text is being typed out, then it pans over to the mannequin's face.

21:05It matches the style of those model channels really nicely. So, yeah, this is basically how you turn your images into videos and handle your visuals using these external tools. Now once you're done with all that, just head back to Claude and type next in your prompt box.

21:19At this point, we've already covered the script, the images, and the video animation. So what comes next is where everything starts coming together. Alright.

21:27So once you type next, Claude code takes you to stage 10, which is the thumbnail part. Now here, you're gonna need to upload some sample thumbnails. Basically, examples of the style you want your thumbnails to look like.

21:39The easiest way to do this is just go to the model channels you're studying, find some of their best thumbnails, and take a few screenshots. If you already have some saved on your computer, that works too. But most people just grab them directly from those channels.

21:51For me, I've already got mine ready, so I'm just gonna upload These are the kinds of thumbnails I wanna replicate.

22:05Quick heads up. Claude code only lets you upload up to five images here, so make sure you pick your best ones. Alright.

22:12So I'm just gonna pick these five images, click open, then type thumbnails attached, and send it. Now once you do that, Claude Code basically studies those thumbnails and figures out the style you're going for.

22:25It understands everything. The text, the silhouettes, the lighting, the pacing, even how the text is placed and how the background is set up.

22:33It pretty much reverse engineers the whole thing. After that, it gives you a bunch of thumbnail ideas you can use for your video. And the cool part is, these are all concepts that actually fit your niche and would work well.

22:46So from here, it's simple. Just pick the one you like. Once you've chosen, you can ask for the full prompt.

22:52For me, I'm just gonna say, give me the full text to image prompt for concept three, and send it. Alright. So once you've picked your concept, Claude code gives you the image prompt you need to create that thumbnail.

23:03Now, if you want, you can ask it to generate prompts for all the concepts. For me, I only wanted concept three, so that's what I went with. Once you've got your prompt, just highlight everything, copy it with control plus c, and head over to something like Google Flow.

23:18Then switch to the image model and paste it in. Now here's a little trick to get better results. Attach a model reference.

23:25This could be one of the thumbnails you grabbed earlier or any image that matches the style you're going for. It helps the model understand exactly what you want so your output comes out way closer. So what I'm doing right now is uploading a model thumbnail I want nano banana to follow.

23:40Then I'll paste in my prompt, select that reference, choose how many outputs I want, I'll go with two, set the aspect ratio, and then send it. Alright.

23:49So after a couple seconds, this is the kind of output we get, and, yeah, it looks really good. It's cinematic, the style matches, the text is on point, everything just fits together nicely. Not gonna lie, this is basically ready to go.

24:01You don't even need to edit it. You could literally upload this straight to YouTube as your thumbnail. So, yeah, I'm just gonna download it.

24:08Now, quick tip. You can generate as many thumbnails as you want. If you don't like what Claude gives you, just ask for another concept or tell it to tweak something.

24:16You're not stuck with one result. A lot of people forget this, but it works like a chatbot. If something's off, just tell it and it'll adjust.

24:24So, yeah, at this point, you've seen everything. How to get good scripts, generate images, animate them, and now create thumbnails. And, honestly, this is really powerful.

24:35It makes the whole process way more accessible, as long as you're actually willing to put in the time and follow through. Alright.

24:42So once you're done with everything, you can go back to Claude code and just type next again. Now what this does is pretty cool. Claude will ask if you want to save everything you've done into a Google Doc.

24:52That includes your branding brief, your style DNA, your full script, which is about 1,900 words, plus all 167 image and video prompts, and even your thumbnail concepts.

25:04This is really useful if you're working on multiple projects and wanna keep things organized. For me, I don't really need it right now.

25:12But if you do, just say yes. Once you do that, Claude will create a file on your computer with all of that saved in plain text, which is super convenient. And, yeah, that's pretty much it.

25:22In the next video, we're gonna take this even further. I'll show you how to create full videos from start to finish with voice overs and more advanced animations all using Claude. We'll also bring in API so Claude can basically run the whole system for you.

25:36Alright. That's it for this one. Have a good

The Hook

The bait, then the rug-pull.

A single master prompt dropped into Claude Code and the whole problem of building a faceless channel dissolves: branding, scripting, 167 scene-by-scene image prompts, animation directions, thumbnails -- all sequenced by the model, no editor required.

Frameworks

Named ideas worth stealing.

01:30list

Sequential workflow driven by a single master prompt; each stage is triggered by typing 'next' in Claude Code.

Steal forAny repeatable content production system

06:20concept

200 Words Per Minute Runtime Formula

Target word count in your script request directly controls runtime: 2000 words = 10 min, 4000-4500 words = 20 min.

Steal forSpec any AI-generated script to hit a target video length before writing

13:50concept

Reference Locking

Use a generated image as a visual reference when requesting the next image to keep character identity consistent across all scenes without extra prompting.

Steal forAny multi-scene visual project where character consistency matters

CTA Breakdown