The argument in one line.
You can generate production-ready HTML presentations using Claude's extended thinking and Opus 4.5, then refine them locally in Cursor without knowing how to code.
Read if. Skip if.
READ IF YOU ARE…
- You create presentations regularly (monthly or more) and want a repeatable system that generates custom HTML slides faster than traditional design tools.
- A non-technical founder or content creator who wants visually animated presentations but has avoided coding because you thought it required programming knowledge.
- You're comfortable with Claude chat and want to move beyond static PowerPoint decks to deploy shareable, interactive presentations online via GitHub Pages.
SKIP IF…
- You need presentations built into Google Workspace or Microsoft 365 for real-time team collaboration — this method produces standalone HTML files only.
- You're required to use your company's locked-down PowerPoint template or corporate design system that won't transfer to HTML code generation.
- You present offline without internet access or work in environments where deploying via GitHub Pages isn't an option.
The full version, fast.
You can build polished, animated presentations entirely from natural language by treating slides as a single HTML file rather than a PowerPoint deck, which unlocks AI's strongest capability and removes the formatting ceiling of traditional tools. The method runs five steps: draft a script or structured outline, set up a reusable Claude Project with a system prompt refined through Anthropic's prompt improver, generate the deck using Opus 4.5 with extended thinking, iterate at most once or twice inside the chat before context degrades the model, then download the HTML and finish edits locally in Cursor or Claude Code. Match model strength to edit size, use Cursor's select-element feature to scope changes, and deploy through GitHub Pages for a shareable link.
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 – 00:19
01 · Cold open
Social-proof hook: 100 people asked about the slides. Promises complete breakdown going into 2026.
00:19 – 01:38
02 · The 5-step process overview
Lays out the full system: script → Claude project → Opus 4.5 → 1-2 chat edits → download HTML → local editor. Shows the numbered slide deck he built using the method.
01:38 – 03:29
03 · Why code instead of PowerPoint
Two arguments: (1) truly unique — animated diagrams, things PPT can't do; (2) code is AI's superpower right now. Acknowledges PPT loyalists, redirects them to a different video on Claude Skills.
03:29 – 03:50
04 · Step 1 — Write your script
Word-for-word or structured outline. Acts as the foundation for the AI to generate visuals from.
03:50 – 06:59
05 · Step 2 — Claude Project + system prompt
Creates a Claude Project live. Shows the system prompt — written using console.anthropic.com's prompt improver with thinking-model checkbox. Key rules: single HTML file, minimal text per slide, high-contrast dark/light, animated charts over text.
06:59 – 10:00
06 · Step 3 — Generate + iterate in chat
Pastes structured notes above a line break; dictation above it. AI thinks 2-5 min, returns complete HTML. Demonstrates one iteration using Apple Notes dictation for a batched edit list (5-6 changes at once). Cap: 1-2 turns before model degrades.
10:00 – 17:05
07 · Step 4 — Local editing in Cursor or Claude Code
Downloads HTML file. Opens in Cursor. Shows 'Select Element' feature for targeted AI context injection. Demonstrates adding Cursor + Claude Code logos to a slide via @-file references and dictation. Covers model matching: Sonnet/Composer for minor tweaks, Opus 4.5 for big restructures.
17:05 – 18:58
08 · Step 5 — Deploy via GitHub Pages
Create a public repo, upload the index.html + assets at root level, enable Pages under Settings → Pages → main branch. Live shareable URL in 2-5 minutes. Free.
18:58 – 19:47
09 · Recap
Fast recap of all 5 steps. Reiterates Opus 4.5 extended thinking as the key model. Plugs 30-day email series and coaching.
19:47 – 20:24
10 · Outro + next video CTA
Teases follow-up video on Claude Skills for pixel-perfect PowerPoint/Google Slides formatting. Direct next-video CTA.
Lines worth screenshotting.
- HTML presentations generated by Claude can include animated visualizations, animated diagrams, and interactive elements that PowerPoint and Google Slides structurally cannot produce.
- Code is AI's superpower — leaning into that for any task with a visual output, including presentations, produces higher quality and more uniqueness than tool-constrained alternatives.
- Staying inside Claude Chat for more than one or two rounds of edits degrades the model's intelligence — downloading the HTML file and switching to Cursor or Claude Code preserves output quality.
- Opus 4.5 with extended thinking consistently outperforms Gemini 3 Pro and GPT 5.2 extended thinking on presentation generation across a three-week head-to-head comparison.
- A Claude Project with a presentation system prompt allows a recurring workflow — drop in a new script and get back a consistent, high-quality presentation every time without re-explaining the format.
- The system prompt should be written by asking Claude to improve a rough initial prompt rather than crafting it from scratch — the model produces better system prompts than most humans write.
- Deploying the final HTML file to GitHub Pages takes minutes and produces a shareable public URL without any hosting infrastructure or cost.
- Keeping all creative and brand decision-making conversations inside a Claude Project with memory means every future presentation benefits from the full history of previous style decisions.
- One to two rounds of chat-based iteration is the sweet spot before switching to a code editor — beyond that, the chat context becomes a liability rather than an asset.
- The presentation-as-code approach produces files that are versioned, diffable, and maintainable in ways that PowerPoint binaries are not — which matters when formats and brand guidelines change.
- A script or structured outline is the minimum input the AI needs — word-for-word scripts produce the most accurately targeted slide content, but even bullet points work as a foundation.
- AI's superpower in presentations is not speed but uniqueness — every HTML presentation is custom-rendered code that cannot be duplicated by someone using the same template.
Steal the proof-of-demand hook.
JoeFlow + Mod Boss playbook
Dylan never says 'I'm going to teach you' — he says '100 people asked, so here's my answer.' That single reframe turns a tutorial into social proof.
- Open any tutorial with a real question you've been asked repeatedly — even 5 times is enough to say 'I keep getting asked this.'
- The 'context window degradation = dumber model' framing is sticky and technically accurate — use it in JoeFlow onboarding to explain why dictation sessions should stay focused.
- The Batch Edit via Dictation pattern (Apple Notes → paste once) is a literal JoeFlow demo moment. That's the product.
- Dylan's 'Match the model to the task' two-tier framing maps directly to JoeFlow's engine selection — steal the visual (two cards, fast/cheap vs smart/expensive).
- The single HTML file rule + GitHub Pages deploy is a repeatable formula for any tutorial format Joe makes — ship it as a template.
- The prompt improver at console.anthropic.com is a specific actionable recommendation most creators haven't touched — worth a standalone short.
Terms worth knowing.
- HTML slides
- Presentations built as web pages using HTML and CSS rather than traditional slide software, enabling custom animations, interactive elements, and easy sharing via a URL.
- Claude Project
- A persistent Claude workspace where you can store system instructions and context so they apply automatically to every new conversation, enabling a repeatable workflow without re-entering setup each time.
- system prompt
- A set of background instructions given to an AI before a conversation begins that defines its role, style, and rules — stored in a Claude Project so it applies to every session automatically.
- extended thinking
- A Claude model mode where the AI reasons through a problem step-by-step before generating its response, producing more accurate and nuanced output on complex tasks at the cost of longer processing time.
- Opus 4.5
- Anthropic's highest-capability Claude model at the time of this video, recommended for generating HTML presentations because it outperformed competing models on visual and structured output quality.
- context degradation
- The phenomenon where an AI's output quality decreases over a long conversation as its context window fills with iterative edits, making it harder to maintain coherent, high-quality responses.
- GitHub Pages
- A free static website hosting service from GitHub that lets users publish HTML files from a repository directly to a public URL with no server configuration required.
- Cursor
- An AI-enhanced code editor (based on VS Code) that helps developers read and edit code files, used here to make precise local edits to generated HTML presentations outside of the Claude chat interface.
Things they pointed at.
02:46productDylan's 30-day AI insight email series
Lines you could clip.
02:30
“Of all the things that AI can do, code is by far its superpower. I wanna lean into that superpower as much as possible for any use case that I have.”
08:30
“The more information you put into this conversation, the fuller the AI's head gets, and the fuller it gets, the dumber it gets.”
00:20
“You don't need to know a single thing about coding. That's what Claude's for.”
00:00
“How did you make those slides? I've gotten this question at least 100 times in the last two months.”
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.
story
00:00How did you make those slides? I've gotten this question at least a 100 times in the last two months. So here's my complete breakdown going into 2026.
00:07This is exactly how I'd use AI to create presentations. We're not using PowerPoint. We're using code.
00:12And before you click away, you don't need to know a single thing about coding. That's what Cloud's for. Let me show you how it's done.
00:19Okay. So to create AI slides, there's just five simple steps. The first one is writing a script.
00:24So the AI needs something to reference to create the visuals from. So this is going to be the things you wanna talk about per slide. It could be word for word or it could be just a structured outline with bullets.
00:32Either way, the AI needs something to reference so it can create visuals from it. After you've drafted your script, you're gonna likely wanna create a Cloud project because we wanna have a process that we can come back to on a recurring basis if you're creating presentations all the time. Because once you've created this once, you can just drop in a transcript in the future, and it's automatically going to give you that presentation back every single time in a high quality and consistent way.
00:51When using the Claude project, you wanna make sure you're using the high end model of Opus four dot five extended thinking inside of Claude. I've tested all the state of the art models, Gemini three Pro, GPT five dot two extended thinking, and Opus four dot five, and Opus four dot five has consistently outperformed the others over the last three weeks.
01:06This may change over time, but for now, this is the best. After you get your first version back, you're gonna wanna make some edits, and I recommend making one to two iterations of edits inside the Clawd Chat itself. After you've done that, you wanna then download the HTML file, which is what it's gonna give you back, to your own machine on your computer so you can make additional edits with a dedicated tool to editing code.
01:24We're doing that because if you stay inside the Clod Chat for too long, you're gonna degrade the intelligence of the model, and you're gonna get worse outputs. After you've gotten the final version of the presentation you're happy with, you can either present it and you're done. Or if you wanna share it with the world, I'll show you how to do that as well in a very easy way.
01:37Now that's the process. But before we get into the details of walking through each one of the steps, I wanna share with you why we're using code and not PowerPoint. So I'm sure a lot of you are gonna say, hey, I wanna use PowerPoint.
01:47I wanna use Google Slides, etcetera. You can do that, but the quality of the output is not going to be as unique and customizable as you're seeing here. And there's two primary reasons I use code with AI.
01:57The first one is to the point that I just made, is that this is truly unique. I can get outputs from AI via code, and I can have animated visualizations. I could have animated diagrams and charts.
02:07I can have all types of things inside of my presentation that you can't do within PowerPoint or Google Slides because they're constrained by the tools themselves. And the other one that's equally if not more important is that right now, of all the things that AI can do, code is by far its superpower. I'm not sure about you, but I wanna lean into that superpower as much as possible for any use case that I have.
02:25And those are the two reasons. And I know still after I've stated that, some people are gonna be like, hey, I still wanna use PowerPoint. I still wanna use Google Slides.
02:32If that's you, you can go watch this video here. So I've made a video of how you can use a specific feature inside of Claude called Claude Skills so you can get a pixel perfect match to the formatting your company has because I'm sure there's a very dedicated format that many of you have for your company for different types of documents, including presentations.
02:46So you can use this cloud skill on a recurring basis to get that match to the format that you care about. This is the thumbnail for the video and the exact title oh, you don't wanna hear that. The title of it is I fixed AI document formatting with one cloud feature.
02:59So if you wanna see that, you can go check that out, and it'll give you more details on how you can create presentations with that cloud skill if you're focused on PowerPoint and Google Slides. Quick pause in your regular programming. This video is brought to you by me.
03:10So two quick things. Below is a thirty day AI insight series completely free. You'll get 30 insights in your inbox so you can apply AI to your business and your work.
03:18The second thing is if you'd like to work with me, below are a series of offerings to see if there's good fit between the two of us, such as an AI community for private business owners and leaders, or if you wanna work one on one via coaching. That being said, let's get back into the video. Now if you're still here, you're probably interested in taking the path that I've shared with you, is the code route.
03:34Because remember, you don't have to know code. You just need to know explain what you want. And if that's you, let's continue.
03:39So the first step, like I said, is writing a script. We've already talked this through, so you want it to be word for word for the script, or it needs to be a structure that the AI can reference. This acts as the foundation for the AI to then build the visuals.
03:49Once you've written your script, we wanna then create the Cloud project so we can do this on a reoccurring basis. And you may ask yourself, what is a Cloud project? I'll show you quickly.
03:57So if I go to Cloud here and I open this up, on the left hand side, you can see the menu. And right here, we have some folders that says projects.
04:03When I select this, it's gonna take me to all the projects I've created, and you can create new projects here. So if I do new project, it's gonna give me the option to name it. So let's name it let's do it now slides creator v two.
04:16I'll hit enter. It's gonna create the project. Now it pulls us into this screen.
04:19So this is the screen for the projects themselves. On the left hand side, you can see the title. So this is the title I just gave it.
04:25This is the chat where you'd paste in your transcript or not your transcript, but the note you've drafted. And then on the right hand side, we have a few things that are important, most of which is the instructions. So the instructions is gonna be where the system prompt sits, and this is what the AI is going to look at before it does anything else.
04:38And to add these, you just do the plus sign here and put in your system prompt. So what I'm gonna do for you is I'm gonna share with the share with you the exact system prompt that I use on a reoccurring basis so you can start with that, and you can make edits to it and tweaks to it based off of your interests. And to show you what that system prompt looks like, I'm gonna go back to our slide deck here.
04:53Go to the next slide. So here is our assistant prompt that sits behind the scenes for the one project that I've created for myself and Claude. Instead of walking you through all the detail here, I'll I'll highlight a few things that I think are interesting.
05:03One of the most important things I wanna call out here is I didn't write the system prompt from scratch by myself. What I did is I wrote some base points, and then I ran it through a prompt improver. So there's prompt improvers both inside of OpenAI and Anthropic.
05:14You paste in your base prompt. It then automatically injects all the best practices for prompt engineering their models and gives you back something that's very effective for getting that tax done. And to get access to that for Claude specifically, you wanna go to console.anthropic.com.
05:28It's gonna take you to a page like this. You may need to sign in. Once you've done that, you do generate prompt, and you wanna check this box down here because this enables for thinking models because we're gonna use the thinking model.
05:39You would then paste in your base prompt here, so the basic thing you've drafted. Once you've drafted that, you'll then push generate. It's then, like I said, gonna put all the best practices in it, and it's gonna give you something high quality that looks something like this.
05:49Now let's talk through some important points that I've added here. So the one thing that I wanna call out first is that this is an HTML presentation. Importantly, I've emphasized this multiple times throughout the prompt that this needs to be a single HTML file.
06:01So the reason we're doing this is if you don't, sometimes the AI will write a series of files that represents a presentation, and that's too complex and convoluted to move around and play with on your local machine if you're not technical. So you want it to sit all in one single file. After that, some preferences I have for my presentations are minimal text per slide, and also I wanna have maximum visual impact per slide.
06:19That's the beginning portion. There's a lot of requirements that I've given in the prompt here around the content, the visuals, the technicalities, etcetera.
06:26Some things we'll call out here is that for me, I prefer to have charts, diagrams, and things like that over text, so I prioritize that in this bullet here. I also prefer, as you can see from visuals here, to have high contrast colors. So I wanna have a dark background and a light foreground highlighting certain aspects of the presentation so you focus your attention there.
06:44And then I have some stuff around navigational structure, etcetera. One thing that's interesting here is that when I used the prompt improver over here, this one that I showed you, it gave us back a scratch pad. So this is basically a tool that the AI uses internally in its head to think through different ideas before it then gives you your output.
06:58And that's the overall system prompt. And now that we've walked through this, let's actually see it in action. So what I'm gonna do is I'm gonna go back to my projects.
07:03I'm gonna open up the actual project that I use, the slides creator, and I'm gonna go to a previous conversation that I had earlier. I'm gonna scroll to the very top to show you how I created this exact slide deck you're looking at. So if I expand this open, this is the first message I sent to the AI.
07:16I took basically right below this line break. Let me zoom in a bit. Right below that oh, not zoomed out.
07:21That's in. Right below the line break are all the notes that I drafted for this video. So if I go to this Google document, you can see there's a bunch of text in here of notes that I've drafted.
07:29And you don't necessarily need to know what's on the text. You just need to know the structure. So at the very top, I gave some kind of pre context of the presentation around introducing the idea and the problem.
07:37I then wrote out the process. So this is what I do. Take that step.
07:41Take that step. I then paste it in the system prompt that we just walked through. And then I walk through some more points.
07:45I then do this, and I then do that, etcetera. I'm just basically writing out the steps. This isn't super detailed.
07:49I then come back here, and I then paste all of that below that line break that I highlighted there. And above this line break is dictation, so basically me speaking out some ideas. And this step is completely optional.
07:58You don't necessarily have to do it because the system prompt has this information inside of it. I just do it for my own kind of personal sanity to ensure that the AI does what I want when it comes to these certain elements, but again, it's not always necessary. A few things that I did remind the AI of is that I prefer my slides to be minimalistic when it comes to text.
08:15I then talked about the importance of having it in high contrast. I then called out that I like to have single accent colors per slide so I can have a focal point to focus the viewer's attention. I then reinstated that it's important to have a modern minimalistic look to the design as well.
08:27So that's the first message we send the AI. Then we just hit enter. After we've done that, if I scroll down here, the AI thinks for about two to five minutes, and it writes up all the encode for us.
08:36So you can see it's thinking in the scratch pad. If If I scroll down here, you'll see that we have a response. And this response is the HTML file.
08:44When I expand this out, you'll see that the slide deck it gave us back is actually pretty good. And it's almost done, basically. There's a few minor tweaks I wanted to make, but it's really, really good.
08:53And once we got that first response back, like I mentioned earlier, we can have one to two iterations inside of this conversation, but no more because we don't want to degrade the intelligence of the AI. Because the more information you put into this conversation, the fuller the AI's heads gets, and the fuller it gets, the dumber it gets.
09:07So in this case, I only did one iteration. So if I scroll down here, this again is dictation. And the way that I usually do this is I'll have the AI make multiple changes.
09:14So instead of doing one change at a time, I'll give it maybe five to six changes. And the way that I do this is I use an Apple note. So I'm gonna show you this.
09:21Pull this over here. So this is basically an Apple note. It's just a like a notepad for Apple.
09:26And what I do is I use dictation. So what I'll say is I'll open up the thing, and I'll say, hey. I want you to delete slides one and two.
09:30And it it it got some of the other stuff. But you can see over here, it said delete slides one and two. That's And what I'll do is I'll walk through the presentation going through slide by slide.
09:38And as I go through slide by slide, I'll then look at something, okay, I don't like that. I'll then open up the Apple Note. I'll then make the dictation.
09:44I'll go back and forth. And then once I've made all the changes and I've listed it all out like this, I'll then paste that in here. Once I paste it in there, I hit enter and let the AI make the changes.
09:52So if I scroll down here, you'll see that it thought for 10 steps. It took about two to three minutes, and then it gave me back a new version of the slide deck that had those adjustments to it. And that's how we're gonna make the changes inside the Claw Chat.
10:03But after we've done that and we've done one to two iterations, it's important that you take it outside of this and put it on your machine to make any future additional edits that you wanna make. And the way we can get the file is very easily. You just do the drop down here.
10:14Once you've selected the drop down, you do download. That's gonna download the HTML file onto your machine. So if I actually so you can see this download came here.
10:21If I click that once, it's gonna open that presentation in my browser, and it's gonna show me the exact slide deck that we saw inside of Claude. Now, what tools are we gonna use to actually make these edits? So there's two options.
10:31Well, there's actually like 30 options, but I'm distilling it down to two that I think are the good places to start. So I think most people should start with Cursor because this is very user friendly, especially for people that are nontechnical. And then if you're more adventurous and you wanna try something that's a bit more technical, you could try Cloud Code, both of which are great tools that all you have to do is speak to, and then you can have it make the edits to the document for you.
10:51So to download either of these, let me actually show you both websites. So first we have here is Cursor. So if you just go to cursor.com, it's gonna take you to this website.
10:58Depending on the operating system you're using, so if you're using Windows or Mac, it's gonna then show you which to download. So you can download that for your system. And once you've downloaded it, I'd recommend paying for the beginner plan.
11:07So the beginner plan is $20 a month. Reason being is you get access to higher end models for an extended period of time. And that could be useful if you're trying to make multiple presentations and also if you're trying to make more complex adjustments to the slides themselves.
11:20But if you don't wanna pay for it, that's fine. You can also use the free version as well. Just note that you're gonna get limited access to higher end models.
11:25So that's Cursor. And then now here's Cloud Code. So this is their documentation page.
11:29If you scroll down here, you'll see that you these are the ways you would download it for your system. So, again, this is more technical, so you'd have to use a terminal to do this. But all you have to do is copy and paste out one of these little commands depending on the operating system you're using.
11:40And those are the two tools that we would use locally. So once we've downloaded that, and I'm assuming you have it downloaded and ready to go, that's when we're gonna start making changes locally. And before we get into making additional edits, I do wanna call out matching the model to the task.
11:52So if you're making a very minor update to the presentation itself, so changing font sizes, colors, rounding corners, etcetera, I'd recommend using the composer model from Cursor because this is super fast and also super cheap. But if you wanna make a really big change, so you want to restructure a diagram or completely rethink a way that a visual is shown, I'd recommend defaulting to a higher end model such as Opus 4.5.
12:13And if there are multiple changes you wanna make like I did previously, if you wanna make five, six, 10 changes, you wanna then use a higher end model again so you can ensure that there's a higher chance that it's going to achieve that in the first try. So just ensure that you're matching the tasks of the model. Now let's assume that you downloaded the application.
12:27If you've downloaded Cursor, you're gonna wanna open that up. Once you've opened it up, you're gonna then open up this specific folder you've created on your desktop because you wanna create a folder on your desktop with that file inside of it. So you can name it whatever you want, grab that HTML file that you download from Claude, and put that in that folder.
12:41For me, you can see that I've already done this. So this folder right here called Claude Prezos, that's the folder that has my file in it. So when I open this up, it's going to show the file here.
12:50So it says index. Html. And let me quickly walk you through what you're looking at here inside of Cursor.
12:54So I'm not gonna walk you through all the details, but just enough so you can make changes to your presentation. On the left hand side, we have kind of like the File Explorer. So this is where all the files sit.
13:02So think of, like, Finder on a Mac or File Explorer on a Windows. This is where your images and files and all that stuff sit within that folder. So this folder right here is the one that I mentioned to you that we opened.
13:10On the right hand side is where you're gonna spend almost all of your time, which is talking to the AI. You can think of this kind of like a Chattypeteer or a Claude situation where you're just talking back and forth with an AI inside of a chat window. And in the center is where all of the text and or code sits.
13:23Now on the right hand side, let me expand this out a bit and show you the different models you can use. So if you select this button here where it says Composer one, there's a bunch of different options here. When you initially sign into Cursor and use it, it'll probably be switched to auto, so you're not gonna see any options here.
13:37To turn that off, you'd switch the toggle off, you'd get all the options for the models. And before we start making any changes, there's one thing that I recommend doing is making sure the presentation is shown inside of Cursor. And the way that we're gonna do that is using a browser and Cursor.
13:48So there's a little button here that looks like a world or a globe. I'm gonna turn this off because that's what it's gonna be set up for you when you initially sign on. And to turn this on, you would click it, and you say Browser tab.
13:57That's gonna then open this specific screen here. Now what we wanna do is we wanna get an error so we can have AI fix it. And the way we're gonna get the error is getting the URL to this file.
14:06And the way we can do this is let's go to our file and double click it. So I'm gonna go to my finder. You'll go to that specific file, and we have the file here.
14:14So I'm gonna double click this, and it's gonna open a browser. And then in my case, it's Firefox. Now up here, we have the file path in the URL.
14:21I'm gonna copy that. I'm gonna go back to cursor. I'm gonna paste it into this URL.
14:24I'm gonna hit enter. We're gonna get an error. I'm gonna do two things.
14:27I'm gonna show details, and I'm gonna take a screenshot. So I took the screenshot. I'm gonna then drag and drop that into the chat window.
14:34I'm also gonna copy these details here. I'm gonna paste that into the chat window as well because I wanna have the AI to have as much context as possible as to why this is airing out. And then I'm gonna have it fix it.
14:45Can you please ensure that I can render this presentation inside of the cursor browser? Right now, I'm getting an error. I need you to fix it.
14:50And that's it. I just talk to the AI and have it fix it for me. And there you go.
14:53It took, like, ten seconds. So the presentation is now inside of the browser sorry. It's inside of cursor.
14:59And the reason that's important is there's a very specific feature I want you to use when making changes. So at the top right hand corner, there's this little mouse here inside of a box. It's called select element.
15:08When I click it, it turns blue, and I can then see all these boxes are showing up when I hover my mouse over it. Now what I can do is if there's a very specific part of the presentation I want the AI to change, say I wanted to change, for instance, this box for whatever reason, when I select that, it automatically does a few things.
15:23It opens this window on the right hand side, which we don't need, so we can close that out by clicking that button. But also, it adds this little blue box here in the chat. So it already filled in something for me.
15:32And that something is basically telling the AI this specific section in the code needs to change. So it's focusing the AI's attention on exactly what needs to change inside the presentation itself, which increases the chances of it achieving the task in the first try. So that's one feature I'd recommend using when using this.
15:47And that's actually make some changes so you can see what it looks like. So I'm gonna delete that, and there's a specific slide here that I wanna add some images to. So I think it's slide eight.
15:57I would like to add the cursor logo here and the Cloud Code logo here. And you can see I have two images in this folder already. If you wanna embed any images in your presentation, you need to make sure the images are inside of the shared folder where that file is.
16:10So it has to be right next to the index index dot HTML file. And so what I would do here is I would at these files.
16:16So I'm I'm doing at symbol on my keyboard, and then I'm typing in the image itself. So by doing that, I'm putting that into the AI's context so it knows what images I'm referring to. And then I'm gonna do dictation.
16:26I would like you to add both of these images into slide eight so it's rendered inside the presentation, and make sure they're placed in the correct boxes and are sized proportionately. Then I hit enter. And that's all you have to do.
16:37And that's the beautiful part about this. I don't have to know how to write code. I don't have to know anything about code.
16:41I just need to know how to use the tools effectively and speak out exactly what I want the AI to do for me. You'll see that on slide eight, those new images are now there. So I have the cursor logo inside this box, and I have the Cloud logo inside this box.
16:52And this is the exact process I'd recommend you go through to make additional adjustments to your presentation. Now let's assume that you're done making your changes and you're ready to move on to sharing this with the world. Either you can just present it as is by double clicking on that index dot HTML file and showing your screen, Or if you wanna share this with other people so they can get access to view it, we can do that simply through GitHub.
17:10So how are we gonna do this? It's very, very easy. So all you have to do is go to GitHub, and you have to create an account.
17:15So if you once you've created an account on GitHub, you wanna then click your little circle here in the top right hand corner, select repositories. You'll then wanna create a new repository. So if I go here and select new, this green button, we're gonna name it whatever.
17:27So in this case, we'll name it for my situation, we'll name it Claude Prezos v two.
17:34And then I'm gonna make this public so we wanna be able to share with others so they can see it. I'm gonna say create repository. Once I've created the repository, we need to put stuff inside of it because the repository is basically just a folder.
17:44And the way we can do that is selecting this blue link right here. So upload existing files. So I'm gonna select that.
17:49I'm then gonna go to the choose your files, and I'm gonna select the exact files that we have in this folder. So we have the CloudPresos folder that I showed you. I'm just gonna select these three files from here, and I'm going to then upload those into the repository.
18:02I'm gonna select commit changes, this green button here, and it's gonna push all that up into the repository for me. There's really one important thing here I wanna call out is the placement of this matters. So these files need to be at the root of your repository, meaning there are no other folders in here, just these files.
18:17And also, the name of this file needs to be index dot HTML for you then to use the feature I'm going to show you in a second. So make sure you have those things set up.
18:25Now to deploy this, you would go to settings at the center right here. After you go to settings, you go down to pages on the left hand side. And then we're gonna change this dropdown from none under branch to main, and we're gonna say save.
18:37Now this is gonna take probably between two and five minutes. And then after two and five minutes, just refresh the page, and you'll see up here at the top, there will be a URL specifically for your presentation. And there you go.
18:47It took me probably three minutes. I then have this specific URL dedicated to my presentation. So when I select this, it's going to take me to that URL that I can share with anybody, and they'll then have access to the presentation they can go through themselves.
18:58Now as a quick recap, the overall process, remember, we wanna start with writing our script because that's gonna be the foundation of the visuals the AI creates for us. So it can either be word for word or just bullets and sub bullets for a framework and a logical flow. After that, we wanna create a cloud project so we can use this on a recurring basis and always get high quality, consistent presentations.
19:17And then when using the cloud project, we wanna use the highest end model we have access to us. And right now, OPUS four dot five with extended thinking turned on is the key model to get high quality presentations out. And And after you get your first version back from the AI, you may wanna make one to two iterations of changes inside the Cloud Chat itself.
19:32But after that, if you wanna make any additional changes, I recommend downloading the HTML file onto your computer and using a dedicated tool to that task, such as Cursor or Cloud Code. For most people, you're gonna wanna use Cursor. And then if you wanna deploy it, just use GitHub Pages.
19:45It's very straightforward and it's easy to do. And that's it. So if you enjoyed this, please reshare it with your friends.
19:51And as a reminder, two things. First off, Blow is a thirty day AI insight series. Completely free, you'll get 30 insights in your inbox so you can apply AI to your business and your work.
19:59The second thing is if you'd like to work with me, Blow is a series of offerings to see if there's a good fit between the two of us. And that's slides. But what about all the other documents you have to make it work, such as SOPs, proposals, contracts?
20:09They all have to match your company's format perfectly. I found a way to make Claude reverse engineer my design, so now every time I get a document that comes out, it's pixel perfect. I'll show you exactly how to do it right here.
20:20Go ahead. Click that link. I'll see you next time, Internet.
The Hook
The bait, then the rug-pull.
Dylan Davis opens with social proof disguised as a question — 100 people asked, so here's the answer. By framing the entire video as a reply to audience demand rather than a self-initiated tutorial, he establishes credibility before a single tool is named. The title does the rest: it's an abandonment story, and abandonment stories convert.
Named ideas worth stealing.
00:19list
5 Steps to AI Slides
- Write Script
- Claude Project
- Opus 4.5 (extended thinking)
- Quick Edits in Chat (1-2 max)
- Code Editor (Cursor / Claude Code)
End-to-end repeatable pipeline for generating HTML presentations with Claude on a recurring basis.
Steal forAny JoeFlow / Mod Boss tutorial showing Claude-assisted workflow — same numbered-step structure works for dictation pipelines
08:30concept
Context Window Degradation Rule
The fuller the AI's context window, the dumber it gets. Cap chat iterations at 1-2, then export and work locally.
Steal forAny tutorial that warns against over-relying on a single chat session — sticky mental model for non-technical audiences
12:30model
Match the Model to the Task
- Sonnet 4.5 / Composer — minor tweaks (font, colors, rounding) — fast and cheap
- Opus 4.5 — big restructures, diagram overhauls, 5+ simultaneous changes
Two-tier model selection: cheap/fast for polish, premium for heavy lifting.
Steal forJoeFlow settings or Mod Boss onboarding — exact framing works for explaining model tiers to non-technical buyers
08:30concept
Batch Edit via Dictation
Open Apple Notes, dictate 5-6 changes while reviewing the presentation slide by slide, then paste the full list as one message. Avoids multiple round-trips.
Steal forJoeFlow demo — this IS the JoeFlow use case. Literal product placement opportunity.
How they asked for the click.
VERBAL ASK
19:10newsletter
“Below is a thirty day AI insight series completely free. You'll get 30 insights in your inbox so you can apply AI to your business and your work.”
Double CTA: free email series + paid coaching/community. Delivered twice — once mid-video, once in recap. Low pressure, clearly free first.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
