2026-03-27 | Chase AI | 7 Levels of Building Elite Websites with Claude Code

掌握Claude Code前端开发核心秘籍,七个维度带你从小白进阶高端网站构建

视频 科技 商业
Claude Code 前端开发 AI辅助编程 UI/UX设计 生成式AI 提示词工程 SaaS落地页 交互设计 视觉叙事 网站构建策略 +6

媒体详情

上传日期
2026-04-07 09:50
处理状态
已完成
转录状态
已完成
LLM 提供商/模型
cursorhub/gpt-5.4 (reasoningEffort=high)
音频播放器

转录

下载为TXT
speaker 1 [00:00:00-00:00:57]: Claude code sucks at front end design, and it's your fault. But this is a problem we can solve with a simple roadmap. You just got to know where you're going. And in this video, I'm going to break down the seven levels to Claude code front end design and explain what you can expect at each level, the traps to avoid, and what you need to do to progress to the next one. And by the end, you will have a simple repeatable process. To create frontend designs with cloud code that aren't yet another generic boring template. Now our case study for today's video is going to be our fake SaaS web app known as Argus. Argus is going to be a social media intelligence application, so essentially a place that content creators or really whoever can go to and find trending topics in their niche. And it is through Argus we are going to go through the entire seven level. Process. So let's begin with level one, and sadly, a lot of people are actually stuck here. This is just you and a prompt.
speaker 1 [00:00:57-00:01:57]: You're gonna have claude code open, and you're just gonna say something like, hey, let's create a landing page for argus, my social media web app. That acts as an intelligence app for content creators to find and identify trending topics in their niche. Now, even if i do this, and i put it into plan mode, what we are going to get is something very, very generic. Why? Because we've really given it no design direction, and we're kind of relying on claude code to fill in the gaps of like. What looks good? You hear this often, like ai has no taste. True, but actually most people have no taste, and because of that problem, they can't articulate it. To cloud code, right? If you don't know what right looks like, how is cloud code supposed to figure it out? So what it is here at the most base level that we need to begin to master a few skills, and those three skills are writing descriptive prompts, specifying frameworks, and understanding what these frameworks even are. And then beginning to build a basic design vocabulary. Now, by simply using plan mode in this process, you're gonna begin to build some of these skills, because it's gonna ask you questions like, what tech stack do you wanna use for this landing page?
speaker 1 [00:01:58-00:02:55]: Right, next js, plain html, astro. You're gonna have no clue what any of these mean, but by simply typing something, asking what are these frameworks, right? You're gonna begin to build some level of competence in this space when it comes to frameworks and things like this, you just have to ask questions. Play mode is also gonna force you to think about design decisions and the actual goal of your website, because it's not enough to say, hey, i just want a landing page for argus. What is the point of this, right? What is the actual outcome you are shooting for? In this case, we want both a waitlist sign up and a product showcase, and then it's going to explicitly ask us questions about style. Now right away, you know these sorts of visual styles, it's going to give you again are kind of generic dark and techy clean and minimal right? Like this isn't specific enough to really get the outcomes we want yet. This is where everyone stays at, and this is why everything looks so generic. So for us, we'll go ahead and we'll do clean and minimal. And then we'll just do the hero, the features and a call to action and submit this guy.
speaker 1 [00:02:55-00:03:55]: Now, you're probably thinking, hey, if this is all super basic, how am i actually going to write descriptive prompts and build that basic design vocabulary? Well, frankly, it's going to just take some time and it's going to take experience. The more you are exposed to again, what right looks like creative front-end designs, the more you're going to pick up on these things and we'll talk about that more in future levels. So let's take a look at what it built us, and holy AI slob! This is hideous, right? This is as generic as it gets. We even have the purple gradients, which AI absolutely loves for some reason. So what happened? What went wrong? Even with plan mode, this is what we got. Well, the truth is, like you already know, cloud code and AI in general isn't the best at front-end design, and if you give it sort of non-descriptive. Boring prompts. This is the sort of output you're going to get right? And yet this is what you tend to see all over the place with maybe a few differences. So what can we do? What can we do to get out of level one and get out of design? Hell. But before we do that, a quick word from everybody's favorite sponsor me.
speaker 1 [00:03:55-00:04:55]: So i just released the cloud code masterclass not too long ago and it is the number one place to go from zero to ai dev, especially if you don't come from a technical background, i update this every week and since it's released, we've already added over an hour of new content. and updated 20 plus modules. so if you want to get your hands on it, you can find it in chase ai plus. there's a link to that in the pinned comment. but if you're brand new to ai, that's just a bridge too far. make sure to check out the free chase ai community. there's a link to that in the description. well, to move on to level two, we need to give claude a design education, and this is where we begin to bring in skills. this is where we start bringing outside specialized prompts that tell claude code how to design better. now, these skills we rely on, like the front-end design skill in the ui ux pro max skill, are going to do a lot of the heavy lifting for us. they are going to inject cloud code with these prompts that say hey, if you're creating a web page in this industry, here's the sort of things you could be looking for. here's a list of all like the ai slop design techniques avoid these.
speaker 1 [00:04:56-00:05:55]: so essentially it's going to supercharge our prompts even if we show up with something that's rather poor. and here's a look at that uiux pro max skill. this is an open source repo on github. it's got 52000 stars. we can go inside of here in the skills folder and actually see how everything is set up if we want to right. it's all just text prompts. And it's a bunch of text prompts that are acting like a checklist for cloud code to again get it out of this generic stuff so we don't get stuck with this. Now to install this particular skill and all other skills like it, it's very simple. If we go inside the GitHub, we have instructions laid out for us right here. We're just going to do forward slash, plug in marketplace, add the skill, and then install the skill. So we just run this inside of our cloud code terminal. And if that is a bit too confusing, understand you can always just copy paste the URL of this, dump it into cloud code and say, hey, I would like to install the skill. So let's put this skill to the test. I've loaded the skill with a forward slash. You can also use natural language and say something like, hey, cloud code, use the UIUX. And we're going to say, I want you to recreate the landing page using the skill, asking any questions you need.
speaker 1 [00:05:55-00:06:54]: Now similar to standard cloud code plan mode, it's going to ask us some questions about the style. We'll do what it recommends this time, right? Going for some typography, sure sounds great. And then a CTA color, let's do orange. We want to get away from the purple at all costs. So cloud code finished the redesign, so let's take a look. So here's the landing page using the UIUX Pro Max Skill, and right away, this is a huge improvement. Right, we actually have like a proper background and have this little square thing. The buttons themselves have like a subtle glow and they even change color when you mouse over them. We have a graphic and the colors change from section to section. So this is way, way better, even though it's still... It absolutely loves the purple. That being said, this is clearly made by AI, right? This isn't something that you would see and be blown away with what was created. This still looks like an AI template, looks like a designed. Ai template, but this is still an ai template, and at its core, it's the same exact website you kind of saw before, which is a little bit of window dressing. So, how can we actually improve this? How can we move out of level two into level three?
speaker 1 [00:06:55-00:07:56]: Because these skills are great, they are bringing some ideas of color theory and typography spacing and just general layout design. But at its core, right, this is still generic. And the truth is, when it comes to the skills to master, the biggest one is like, how do we evaluate this output with a designer's eye? And then turn those critiques into prompts. This is still a prompting. Problem, but the issue with prompting, and this is where we get into the trap here, is like there is a ceiling to how descriptive we can be, especially when we first start, right? We lack the knowledge to actually articulate the problems. Right, i'm trying to explain a visual medium, and a visual problem with text. Inherently, there's some loss there. So what can we actually do? Well, we need to start showing. Cloud code, the problems, we need to start showing cloud code, actual examples of what we're trying to push for. Right, so to unlock level three, right? That's where we start moving away from just text prompts and we start moving into the visual medium, right? We start showing it, not just telling. And so level three is where you become the visual director.
speaker 1 [00:07:56-00:08:55]: This is where we begin to give Claude code concrete examples of what we are shooting for, right? That way we aren't just relying on text and skills. We're now going to use text plus skills plus visual examples. Now, this is a huge unlock for two reasons. One, Claude code's obviously going to be able to better mimic an image that you show it versus your description of the image. But secondly, because we are using images, that implies that we're looking at other websites and we are now going to be able to see again what right looks like. What are actual pros in the space doing and how can we mimic them? Now, there are a number of places where you can find inspiration for your front-end designs and i'll list four of them. The first one is awards, that's awards with three w's. This is a place where people submit their front-end work and they get graded on it. So it's very much has a creative bent to it. Secondly is godly dot website. It's an infinite scrolling website where you can just again, see a bunch of different designs. Number three is pinterest, which surprisingly has some pretty cool stuff. So here, i'm just looking for sas landing page. When i find one, i want, i can also see a bunch of similar ones as well.
speaker 1 [00:08:56-00:09:55]: And then lastly, we have dribble, that's dribble with three bees here. I just put sas landing page and again, i'm seeing a bunch of different stuff. That looks pretty cool. Let's say and laid it on a website like open hands that I like. I like sort of the you know scrolling thing they have on the top with the proven workflows. I kind of like the different color scheme. It's something unusual right away. They have the social proof and just in general, I'm liking the style right? For example, so again, what I'm going to do very very simple. I'm just going to take a bunch of screenshots of this web page, dump it into cloud code and say hey, this is the visual style I want to move towards. So inside of cloud code, I said actually I want our website to match the style of open hands. I gave it the URL, and then I said here's some screenshots for reference, and I dropped all the screenshots in there. And so now it has actual visual references to go off of. It'd be kind of difficult for me at least to describe in words what I was looking for here, even with the help of something like the UIUX Pro Skill. And it's through this process we're going to begin to curate really good visual references.
speaker 1 [00:09:55-00:10:55]: There's so many of them out there, and once you begin to see them more and more and more stuff like this is just a complete and total turnoff. And it's like why would I ever do anything like this when all this cool stuff is available? And as you see more of those references, even though we're just giving cloud code screenshots, this is going to help you communicate what you do specifically like because you're going to see the same things over and over again as you look through all these references. And what you also are going to want to start doing is combining references from multiple sites. Even though we were using open hands here, there's probably some cool stuff I found on Pinterest or Godly websites. I'm like, let me pull this from here and let me pull this from there. Right? I want to take the best of all these different sites and combine it. However, it's very easy to get stuck in this particular trap on this level. And I would say most people kind of stop here at level three where they understand, yeah, I got to use skills. Cloud code sucks off the bat, but with skills and with screenshots, I can get pretty far. The truth is there is a ceiling here because we're going to get kind of stuck at the vibe gap, like you're going to see with what it comes out with. It's going to be like kind of close, right?
speaker 1 [00:10:55-00:11:56]: But it's not going to be perfect because again, there's still that effect that like lost the effect of me telling cloud code, which is very much right. It does everything in code, showing it a visual image and saying hey translate the visual to code. And that's kind of the problem with a lot of front end design. It's like it's a visual medium, yet I'm trying to type it out to text. At the same time, if i throw it visual screenshots, there's still that translation piece. Like, how does it actually do it properly? Well, the truth is, it has a limit and we can see that limit in action right here because this is what it created with all those screenshots. Here's the original. Here's ours. And like i've said in the past, like this is where people get stuck because if we scroll through this, like, you know, if you squint your eyes, you can kind of see what it was going for and maybe got like 50% of the way there and there's some things we do like. But what people do now is they're just gonna go here, take another screenshot of the hero, ok, let's work on the hero piece over and over and over again, a bunch of prompting, all right, go back, screenshot, go back, and it just takes forever, and you never really nail it, right?
speaker 1 [00:11:56-00:12:56]: Because it's that issue with the translation from screenshots to code. And so to move out of level three, and start moving on to level four, we need to not only look at the surface of these websites we like, and grab inspiration from them. We need to learn. How to actually clone them properly and actually get the underlying code that is powering something like open hands, and this goes beyond simply just like 0 we just want to clone it and we are just like copying them. No, the process of cloning it and then figuring out how to actually copy these things comes with some huge benefits for us, not just do the templating piece which is great, but you are going to then get exposed to the actual like. Code that is happening beneath the surface, and so to move from level three to level four, we now have to go beyond simply the surface level of looking at these websites for inspiration with the screenshots. We now need to learn how to go below their surface, look behind the curtain, and see how these websites are built in terms of their actual coding components: their HTML, their CSS, and their JavaScript.
speaker 1 [00:12:56-00:13:56]: And so level four. Is the cloner. This is where we learn by stealing from the pros. Now, let me be very clear. This level is not about, oh, we're just gonna go plagiarize websites and copy and call it our own. That's not the case here. What is the case is we are going to find websites that we like, and we are going to deconstruct them, and then use them as a template that we build upon. And in the process of doing so, we are going to learn a ton, just like any skill in any domain. How do you get better at it? You see what the pros do, you try to copy them, you learn a ton by trying to emulate them, and over time, you give it your own spin, and you create something original. That's what this level is all about. And so to really get the most out of this, we kind of need to begin to understand what's going on with front-end design. Front-end design, to make it kind of simplified, has like three parts. It has the html, think of that is like the bones of the website, right? How it's actually structured. We have the css, that is like the styling, right? Of the website, think of it as the clothes on the website, and then we have javascript.
speaker 1 [00:13:56-00:14:56]: Think of javascript as like the muscles, right? It's what actually allows things to happen when you click a button or do whatever. And it is by looking at these three parts of a website that we are a gonna be able to better clone or template a website, but b understand how they work, right? That whole idea of deconstructing the website and learning how they do things. So how do we do this? How can we deconstruct, for example, open hands? Clearly, screenshots don't work because this is not the same thing. Well, it's relatively simple. So to grab the HTML, we're just going to go to the website and then do control U. What you see here is the entirety of the HTML. So what we're going to do is we are going to copy this, right? We're going to copy this whole thing, and then we're going to paste it into cloud code. Now this will usually be rather large. In this case, it is 1152 lines of HTML. Now at the very, very bottom of this, you can see that there are also links to the CSS files as well as the JavaScript.
speaker 1 [00:14:56-00:15:56]: And that's what I'm highlighting down here. now, this is going to be at the bottom of essentially all of these html files. and i'll kind of zoom out. it's kind of hard to see. but at the bottom right, we have the javascript as well as the css. so, these are actual files that also include a ton of code. so, what we are going to do is after we give it the html, we're also going to tell cloud code, hey, take a look at the javascript and take a look at the css to really understand what's happening. so, i wrote, take a look at the css and js files as well. now, you can stop here and it will use webfetch on its own to take a look at those. the problem is the way webfetch works. Is cloud code does web fetch? It calls a smaller model to do it for it, right? So this instance of cloud code doesn't actually go to that web page. It uses a smaller model. The smaller model takes a look at summarizes it and gives it back to cloud code. The issue with that is the css and js files are pretty large and they contain again like a lot of the very specific ways this website was put together. So we want all of it, right?
speaker 1 [00:15:56-00:16:59]: I don't want to summary. Of these files. So what i have done is i have created a skill that allows cloud code to do this better. It's essentially an enhanced prompt. So it grabs as much as the css and the javascript as it needs. So i just say use the site tear down skill. Now, i will put this site tear down skill inside the free school community and link to that is in the description. So you just have to load that up and you'll be able to do this as well. And so in total, what did i say? I said, here's the html for open hands. Take a look at the CSS and JS files as well, use the site tear down skill and use this info to better clone the original OpenHands site. Now how is this going to educate us? Well, first things first, you probably have no idea like how is able to do things like this background, right? You probably don't really understand or even know what sort of vocabulary you should have used to have it better mimic this, right? When we just gave it screenshots, we got this like you know kind of sort of close thing. but by having claude code actually look at what's going on under the hood, we can now rely on claude code to teach us these things.
speaker 1 [00:17:00-00:18:04]: because now that claude code actually understands at a coding level what's going on, we can now have an intelligent conversation with claude code and say, hey, how did they do this? what does that mean? what is a gsap? like, how did they do the scrolling animations? how did they do this? how did they do that? and over time, after you clone more and more of these websites and use them as templates for your own thing, because again, you're always going to put your original spin on it, you're just using this as like a baseline foundation. You will become better at front-end design. Just like you do with any other skill and this is the process by which you are not only exposed to but begin to bring in pro techniques into your builds. But to move on to the additional levels, like i said, you can't just be, you know, a monkey at the keyboard with cloud code. Just like with everything else we talked about, can't just say to accept except next next next because how do you differentiate yourself? So what skills do we need to master in this process? We need to learn to read and understand the source code. Cloud code helps us there. We need to identify which techniques equal each effect, because at a certain point we kind of have to be able to see these websites and kind of recognize right away like 0 this is what is happening.
speaker 1 [00:18:04-00:19:04]: And lastly, we need to learn how to adapt these clone patterns to our own designs. Right? We are not plagiarizing, we are getting inspiration. And if you do not master these skills, you are going to get stuck in this trap, right? The clone ceiling. You are just going to copy, but you can not create. You are not going to understand the why, and so you are going to hit a hard limit. And I at the end of the day could replace you with any dude off the street and just tell him to do steps one, two and three, right? That won't get you as far as you want to go. So let's see what Cloud Code was able to build when we gave it all this additional coding context. And here is the updated version. And this one obviously looks a lot closer to the original, especially now that the background is actually looking. So if we compare these two, now obviously it isn't one for one just yet. This was essentially its first pass. But much better, much closer to what we had. And if we want to continue to iterate at this point and bring it closer to whatever we like here, well, that's much easier, right? Because now it's looking essentially at the same cookbook open open hands is looking at and we can say, okay, let me change this. Let me change that.
speaker 1 [00:19:04-00:20:03]: I see how they did it. Here's how we can emulate it, right? And again, this is where we can begin that education process. For example, like this cool background, how was that working? And also, How can we get the same thing open hands has where the mouse scrolls through and you know sort of eliminate some of it? And all you have to do is ask cloud code, how's the background working? How can we match that effect? And every time you do this, you're just adding another tool to your toolbox. And so cloud code explains how this effect works, and better yet, it now does the proper thing on our own background. Right? Pretty sweet. But for obvious reasons, we don't want to sit in level four. We don't want to just be a cloner, right? We want to put our own spin on it. We want to begin to make this ours. How do we do that? How do we begin to move on to level five and put our own twist on things? Well, this is where we begin to have fun with components and also our own custom asset designs. This is where we bring in our own stuff, and we're no longer just copying exactly what some other website is doing. Now, one of the easiest ways to begin doing this is finding quality components on websites like 21st Dev.
speaker 1 [00:20:03-00:21:04]: So in many ways, this is similar to how we found inspiration for websites and bringing in those parts, but at like a micro level on a component level. So on a place like 21st Dev. For example, I can look at things like buttons over here on the left. And what I'm going to get is like a bunch of different essentially button designs, and if I find a button design I like, let's say you know something like this, well then I can just copy this prompt, bring it in a cloud code and say hey, let's integrate this button design. And this applies to essentially any sort of button, right? We have stuff for carousels, we have stuff for scroll areas, maps, navigation menus, images, hooks. Everything and it isn't just limited 21st dev, we have other websites like codepen, right? Codepen has a lot of cool designs that we are able to bring in as well. Monet is another really cool one. In fact, there's a bunch of places where you can find these sort of cool components and begin integrating them into your web page. But what if we want something that's a bit more custom? I don't want to just copy a bunch of random components I found somewhere else.
speaker 1 [00:21:05-00:22:04]: How do I really make it original? Well, you have two options. The first is you still use these components from 21st. dev, but we just prompt cloud code to change it, right? It's just code. This prompt gives us the exact code, and then we can just tweak it to our heart's desire. Or we can literally create it ourselves. I mean, if we head over to Pinterest and we see cool landing pages like this that are very like almost art forward, well, there's nothing saying we can't use AI to create some sort of image like this. And let it be sort of the showstopper for when people go on our page. And these don't just have to be still images. I mean, we totally have the ability to add videos to our background, to have it tied to certain scrolling animations, to again take the ideas that we discovered in step four and make it our own. So let's actually do that. Let's quickly redesign our web page, get away from this open hands clone, and actually design some like cool piece of AIR and let that. Be what you first see when you come onto our site. And so, you know, the drill by now, first we need to find some sort of inspiration.
speaker 1 [00:22:04-00:23:03]: And after scrolling through pinterest for a little bit, i'm seeing a lot of sites like this, where we have some sort of almost cartoonish slash stylized image in the background for the hero section, and then we'll have some of our text on the left. Now, this is where you can start to bring in some of your visual storytelling, right? What is the app we're creating? The app is argus, right? And we need to come up with some sort of tagline that ideally also kind of ties into the image itself, like what are we trying to get here with Argus? Well, Argus, you know, mythological figures 10000 eyes. We're talking about social media, we're talking about finding things before the people do. So the tagline I came up with was 'see what's next'. Simple, it's tight. And you know who helped me come up with that? Obviously, it was Claude Code. And then what I did with Claude Code is I pretty much said like hey. You know, we talk about like visual storytelling here. What are some ideas we could come up with imagery, right? Because i don't want to at this stage, we're doing our own thing, right? I can rip the, you know, general concepts of what you see here image wise in terms of how they set things up, but the image needs to be my own.
speaker 1 [00:23:03-00:24:02]: And again, having some sort of visual storytelling that ties into what your application actually does, we'll go very, very far. So what i ended up doing is then creating an image. Based off of some ideas, cloud code gave me and i actually leaned on mid-journey for this one. Now, you can use any sort of image generator you want, whether that's nano banana pro, c-dream doesn't really matter. I really like mid-journey for these kind of like almost concept art type pieces because mid-journey, especially version 7, i really haven't tested out version 8 very much like you can give it kind of a. Weird prompt, like the prompt I gave it was I need a background image that will be the landing page for a website called Argus. The tagline is See What's Next. And so just with like a really poor prompt like that, that's very vague. I got a bunch of cool different stuff that I was able to choose from, and I saw this and I was like, This looks pretty sick right? And I can only... I can already imagine right now like having sort of the information over here on the left. We can have maybe some of the stuff up top, like I can play with this. Like this is.
speaker 1 [00:24:03-00:25:04]: I think very cool, maybe we even do some sort of animation at some point. So once I have my image, it's pretty easy from there. I'm just going to download this image, give it to cloud code and say hey, I want to reimagine the hero page using this image as the background. And so inside of cloud code, I said I actually want to change the front end entirely. I want to use the following image as the background for the hero. Let's put the info about the app argus on the left hand side, leave some negative space on the right to show off the image and make the tagline see what's next. So we'll see what it does at this point, and again, this is kind of the level where you need to start letting your creative juices kind of flow, because up until this point, you've been exposed to a lot of different front-end designs that are different than just the super generic stuff you see in every single SaaS template. Having some cool art, whatever again, visual storytelling is what we're going for here. Is really going to level up your game. And so here's what our updated hero section looks like with that brand new background image, and I made this a little bit smaller so you can see it in its entirety. And clarkson went ahead and also, you know, updated all of the colors.
speaker 1 [00:25:04-00:26:03]: Now, again, that was just the very first pass. All we've done is added this background, but right away, you start to see it feels a little bit. I don't know, more original, more creative stands apart, right? And now we begin to add more and more of our player to it and to start that can be as simple as just upscaling this image. You can do that inside of mid-journey because as we see here, the image feels a little stretched out and not as. You know, detailed as you would like or we can do something like add motion. We don't have to have a still background. We can actually put a video back there as well. Now, to turn it into video is really easy. You're just gonna head to somewhere like cleaning 3.0 or v03.1 just anywhere where you can do a starting frame. Ideally, you can do the end frame as well. Like you see over here on the left that way, it doesn't do some like crazy motion because. You have to remember, like ideally it would be a complete loop. Sometimes it can be hard to get AI images to loop correctly. So the second best option is just to create a rather long video about 15 seconds, and you want the motion to be subtle. So kind of what you see here, right?
speaker 1 [00:26:03-00:27:02]: If I had this as a background, even with the birds completely still, which is totally fine, it's almost like a parallax type thing you got going on. And since it's the hero page and this is lasting 15 seconds. The chances that someone's going to sit there for 30 seconds and see the whole thing through are unlikely, so if there is a hitch at the very end, you're still going to be okay. But luckily for this one, since we had the motion so subtle, you almost can't even tell when it switches from second 15 to second one. So we're going to use this one. And you're probably also asking what about performance stuff? Well, in terms of performance on desktop, they'll be just fine with the video, but we will tell Claude Code: Hey, if it's a mobile. You know, user that comes on your mobile web page, it's not gonna load the video for them, it will just load that still image. So, i gave it the new up-scaled image, and said use that for the mobile, and then i give it a link. To the video and said, hey, replace that in the hero. And here's a look at the page now with the motion, right? I think it looks pretty good. The other thing is, it's subtle, right?
speaker 1 [00:27:02-00:27:56]: You don't want to do crazy motion and have like a full-on video game going on in the background unless you know what you're doing. But something like this with like the subtle sort of cloud movement, you have the water moving at the bottom, right? Again, this gives it... This gives your web page some character, and giving your web page some level of original character is what Level Five is all about. Like I keep saying, the first four levels we're learning the fundamentals, we're seeing what pros do here. In Level Five, we're actually customizing it. So just like you did it with the hero. That's how you need to begin to approach every other section, right? Like these cards, these cards are hideous. The background here is kind of boring. It's the same exact thing every time. So what am I going to do? I'm going to find inspiration. I'll go to 21st Dev, I will go to Awards, I'll bring it into Cloud Code, and then I'm going to iterate, iterate, iterate until I get it to a place I actually like. Now it's at this stage as you're doing all these visual edits that you're going to feel a little bit constrained sitting inside of Cloud Code. Again, there's always that sort.
speaker 1 [00:27:56-00:28:57]: Between being inside of a terminal and using text and trying to mess with things in a visual medium, so in step six we're going to start moving outside of cloud code a little bit because it's here in level six where we start bringing in some outside tools that let us get even more specific with how these sort of visual things are coming together. And there are a lot of tools at our disposal. We got things like paper dot design, we got stitch, we got figma. Right, we got pencil. dev. This is where you now have the ability and the skills to bring in these outside programs to just tinker a little bit more. And a great one to try out because it's very AI forward and simple to use is stitched. It's also free, right? This is from Google and essentially gives us a visual canvas to mess with front end design type stuff. So what I did here is we're not starting anything from scratch. We're doing a redesign. So I gave it screenshots of our website, both the hero page and the rest of it. And I said. here's a look at the current web page setup. i really like the hero. i like sort of the aesthetic there, and i like the colors. but i want to update essentially the entire bottom half of our website, right?
speaker 1 [00:28:57-00:29:59]: just feels very flat, right? i want to sort of bring in sort of the imagery and colors that we have here up top. like, let's see what you can come up with. so let's take a look at what it generated for us for a redesign. and here's what we got. now, the cool thing about stitch is i can have it do redesigns of the redesigns. so if i have the selected, you can see it gets selected down below, and i can just prompt it like any chatbot. I can also right click on it and go to variants, and again change the layout, color scheme, images, etc. Now since this was a redesign in stitch, and I wasn't just building something from the ground up, this is just an image it created, but this is something I can totally right click copy, and then now bring that back to cloud code and say hey, what do you think about this? Right again, these sort of visual editor tools are just great for sort of the ideation process. and again, there's a whole bunch of them. pencil. dev is another one you're going to see all over the place. this works really well if you're inside of like cursor or vs code, because you can actually have the canvas up and edit things in real time. so there's a bunch of tools in the space and more coming onto the market like every week.
speaker 1 [00:29:59-00:30:57]: and so i dropped that image back in the cloud code and pretty much said, hey, what do you think about this? what do you think of this sort of glass morphism effect we got going on with the image in the background? let's try that out. and that's pretty much what you're going to do from now until the website is where you want it, because this is the stage in front of design where now you're just tinkering. You're going to tinker, tinker, tinker, tinker. By now you've built the skills to see like what you're kind of moving for, but this is just an extremely iterative process. There's no real shortcut here, but again, you have all the tools at your disposal now, both to find inspiration and to actually play with. So after about 20 minutes of tinkering on my own, this is what I got with Argus. So first of all, you will notice a few small changes that give it some weight, give it a more premium feel. First of all, is the loading section right? When I refresh this, what do you see? You notice sort of like a hesitation before all the text loads and it gives it some weight, right? It makes it feel heavier. You also notice I changed the fonts. Fonts are huge, typography is huge.
speaker 1 [00:30:57-00:31:58]: Where can you go to see different fonts and sort of like you know get some inspiration? Well, great place is Google Fonts. They have a bajillion fonts, they're all for free. They're broken down by different type and cloud coaching use all of them. So definitely check out Google Fonts. If you're like, i don't really know which one i should use. We added a scroll section down here. We added this sort of ticker at the bottom that also doubles as sort of a border between right our video, and then our image background down here. Before we had this border going from a video. Hero section to the same image as the background was very jarring, but this gives it a nice sort of natural border. You will also notice the changes that we brought in from stitch. We now sort of have this glass morphism effect that we still need to edit a little bit. You will notice it can get a little cut off when it bounces, but it again kind of shows off that background artwork that gives our thing some character. And then you can also see some subtle things, right? If I refresh right here, you will notice the counter just like zooms up from zero to 10 million.
speaker 1 [00:31:59-00:32:59]: I mean, these are these nice little like on the margin things that show you actually cared about the website and at some level of like professional pride. You'll even notice sort of like this lighting zoom go over like AI tools and content strategy. And again, it's just these little things. It's these little small things that truth be told most people. Won't even actually notice at all yet when you combine all of them, which you eventually get is something that looks coherent and something that you cared about and created whether or not you used AI as the tool. The point is it was crafted right? It's got just like all those little things. I mean like even like the scroll bar up top right, you see how that progresses again. It's just all these little little things. And so, how did i prompt this? Well, i mean, frankly, it was just a combination of everything we talked about up till now. Like, i've seen the kind of things i like, i take screenshots of those and i bring them into cloud code, but even more. So like, i'm no expert in web design. So what i did on certain things, like i knew right away, like, hey, like i, i like the glass morphism cards, we saw in stitch, right? Because remember stitch kind of gave us this idea first.
speaker 1 [00:33:00-00:33:57]: And when cloud code originally made them, they were just kind of flat, right? They didn't have any of this like pop off the page. So i said, hey, like. Let's give them some weight. Let's have them pop off the page. But what I also told Claude, what I also told Claude Code was, hey, what I want you to do is I want you to do a web search for the best web design practices when it comes to doing these sorts of things. And on top of figuring out, hey, how do we want to like make these cards sort of pop and give them some weight? Come up with some other things that can make our website look a bit more premium yet in a subtle fashion. It came up with a bunch of different ideas. I had it run through all the ideas, some i discarded, some i kept, but that's sort of the general, you know, cadence of doing this, right? You don't have to rely completely on yourself, claude can go out there on the web and again, see best practices. But remember, you are always in control, you are in the driver's seat, and don't be afraid to kind of have fun with this. Like i said, all these little things, this is where you again give it character, and this is where you make it yours. Right?
speaker 1 [00:33:57-00:34:57]: The more we can make it yours and reflection of your creative spirit in this space, the better it's going to look. Right, truthfully, because if it is a reflection of you in which you want on the screen and what you're trying to portray, it's hard to really call that ai slop. It's hard to call that an ai template because it's yours. And so while this level is on the surface about hey using these outside tools like stitch, i think really this level six is pretty much everything i just said like. Using these outside tools like stitch to iterate, but ultimately at this level, this is where you're finally able to sort of, I think, creatively express yourself in the front end design space and using cloud code as the tool versus in the first couple levels we were just along for the ride. That leaves us with the final layer, layer seven, right? And what is going on there? What is this stuff about three D? Well, spoiler alert. this is like a place where none of us are probably gonna be playing it at all, right? when we talk about the frontier and the architect and 3d stuff, this is where like the big leagues play.
speaker 1 [00:34:57-00:35:58]: this isn't really something we are gonna be able to do, i think, as of today, march, you know, late march 2026. i don't know if this is really in our wheelhouse yet, because this is when we start talking about things like custom webgl and shaders and 3d experiences. this is when we're talking about like websites that pretty much look like a video game. i mean, take for example, this website right here. do you think this is something where we can just do control you control a you know take a look at the h t m l and the c s s and see what's going on? No, no, this is all, this is like art, right? This is a team of designers that spend a long long time putting this all together, and it is absolutely gorgeous. Now, in theory, could ai do this? Maybe, but i will be honest, this is like, this is so far beyond you or me. And i frankly just put this in here to kind of show you what's possible, cause when you see some websites like this, and if you want to see more of them. Head over to awards, which I talked about earlier and go to like their sites of the day and sites of the month.
speaker 1 [00:35:58-00:36:58]: They all like have the same kind of thing going on where it's like this is all custom type stuff. And maybe this is something AI will allow us to play in in a few years' time, but like it's really cool. Like the whole website design space is just so interesting. This stuff is like straight up art. And like I said, when you first get started in the space, you see the same SaaS stuff over and over and over again. But when you see stuff like this, your mind is just kind of blown. And so this is very much level seven. If you know how to do stuff like this with AI, like hey, you should be the one on YouTube making videos, because I would love to learn how, but just. I purely made this level to honestly show off this igloo website because it's so cool, so definitely check it out and like just kind of have your mind blown and let it kind of think about what you could do in the future. And so this is where I'm going to leave you at the end of our seven levels of front end design with cloud code. What I hope you got out of this more than anything is sort of the. Idea of the progression of the skills to get to something like this, because it's something that's totally doable.
speaker 1 [00:36:58-00:38:02]: And it's the idea that first we need to see stuff we like, then we need to learn how to deconstruct it. That whole idea of cloning websites, I think is super, super important, right? Because you're going to learn so much by essentially having Claude code walk you through how someone else created a website you loved. And through that process, you're going to get educated on all these different techniques and effects of that you didn't even know were possible. And through that process, you can then go to your own website and begin to apply those in your creative style, right? That's what it's really all about. What it's not about is like, hey, there's this one super special skill and you use a special skill and cloud code and wow, look at the website like. truthfully told you if it was that easy, everyone's website would look good. everyone's website does not look good. So clearly it's not that easy, right? There's a little bit more to it. I mean, I think the whole idea of taste and AI not having taste, I think that's kind of true, but I think more accurately is the idea that we have a very tough time articulating our taste, and we have a tough time articulating our taste because we don't even know the correct words to use.
speaker 1 [00:38:03-00:38:46]: Because we aren't web designers, we don't know the vocabulary, we don't know the nomenclature, and that's actually really common thing with sort of like all the AI stuff in general, not just web design, and that goes for like any sort of coding stuff too. Like if you don't come from the space, you just don't know how to speak the language literally, and so that creates sort of a translation, you know disadvantage between you and Claude Code, and the end result becomes something sloppy and generic. But hopefully kind of going through this roadmap, you see a path forward to alleviating some of those things over time. So hope you got something from it. I had fun doing this. Let me know in the comments what you thought. As always, make sure to check out Chase AI Plus if you want to get your hands on the Cloud Code Masterclass. We'd love to have you there. And I'll see you around.

最新摘要

生成于 2026-04-07 09:57

核心概览

这期内容围绕一个核心问题展开:为什么用 Claude Code 做出来的前端页面常常一眼就是“AI 模板”? 结论并不只是“模型审美不够”,更关键的是:输入过于模糊、缺少高质量视觉参照、不会把偏好翻译成设计要求,也不了解网页效果背后的实现方式。

为了解决这个问题,内容给出了一条清晰的 7 级进阶路线。这条路线不是去寻找某个“万能提示词”,而是逐步建立一套可复用的方法:

先看优秀作品,接着学会描述与判断,再用视觉参考缩小偏差,然后拆解源码理解实现,最后加入自己的组件、资产、动效与细节,逐步把页面从“通用模板”做成“有品牌感的作品”。

案例使用的是虚构 SaaS 产品 Argus:一个帮助内容创作者寻找垂直领域热门话题的社媒情报应用。围绕这个产品,页面目标被明确为两件事:

  • 展示产品能力
  • 引导用户加入 waitlist

Argus 的页面演进过程也贯穿了全部七个层级:从最开始极其普通的紫色渐变 SaaS 落地页,逐步发展成带有自定义主视觉、视频背景、玻璃拟态卡片、滚动区、数字动效和更统一品牌语言的页面。


七个层级的完整进阶路线

第一级:只有提示词

第一级的典型做法,是直接在 Claude Code 里输入类似:

  • 给 Argus 做一个 landing page
  • 它是一个帮助创作者寻找趋势话题的社媒应用

即使用了 plan mode,产出通常还是很普通。原因很直接:只描述了产品,没有给出足够具体的设计方向,Claude Code 只能用最安全、最常见的默认方案来补全。

这一层需要开始补足三项最基础的能力:

  • 写更具描述性的提示词
  • 明确技术框架
  • 建立最基础的设计词汇

plan mode 的价值,在于它会反过来逼用户思考几个问题:

  • 想用什么技术栈:Next.js、Astro、纯 HTML 等分别是什么
  • 这个页面的目标到底是什么
  • 风格偏好是什么
  • 需要哪些区块:Hero、Features、CTA 等

在 Argus 案例里,页面目标被进一步明确为:

  • 产品展示
  • waitlist 收集

风格则先选了最常见的 “clean and minimal”,区块只保留 Hero、功能区和 CTA。结果也符合预期:页面虽然“能用”,但非常模板化,几乎就是典型 AI SaaS 页面:紫色渐变、标准卡片、标准按钮、没有明确个性。

这一层最常见的误区是:

  • 以为自己已经把需求说清楚了
  • 实际上只是把“产品介绍”说清楚了,并没有把“设计方向”说清楚

只靠一句提示词,Claude Code 解决的是“生成页面”问题,不是“做出好设计”问题。


第二级:给 Claude Code 补一套“设计教育”

从第二级开始,不再只依赖裸提示词,而是引入外部 skills,例如视频里展示的:

  • frontend design skill
  • UIUX Pro Max

这类 skill 的本质不是魔法,而是一套结构化提示词和设计检查清单。它会把一些设计约束和经验提前注入 Claude Code,例如:

  • 某类行业页面通常应该关注哪些布局与视觉重点
  • 哪些“AI 味很重”的设计套路应尽量避免
  • 如何更好地处理排版、色彩、间距、层次和 CTA

视频里提到的 UIUX Pro Max 是一个 GitHub 开源仓库,约有 5.2 万 stars。它本质上就是文本规则集合,可以通过 Claude Code 的 marketplace 安装;如果不熟悉安装流程,也可以直接把仓库链接丢给 Claude Code,让它协助安装。

在 Argus 案例中,使用 skill 重新设计后,页面马上比第一级好了不少:

  • 背景更完整
  • 按钮有轻微发光与 hover 变化
  • 区块之间的配色有了变化
  • 视觉层次明显提升

但第二级仍然有清晰上限:它只是把“粗糙模板”升级成了“设计过的 AI 模板”。

也就是说:

  • 比原始版本顺眼很多
  • 但依然能看出是通用套路
  • 还没有形成真正的品牌辨识度

这一层最关键的能力,是开始用“设计眼光”去评估结果,再把评估转成下一轮提示。难点在于:视觉问题很难只靠文字精确表达。 这就引出了第三级。


第三级:从“告诉它”转向“给它看”——成为视觉导演

第三级的核心变化是:不再只靠文本描述,而是给 Claude Code 直接看视觉参考。

这一步之所以是巨大跃迁,有两个原因:

  • Claude Code 模仿图像,通常比理解抽象描述更接近目标
  • 寻找参考图本身,就是在训练“什么叫好设计”的判断力

内容里给出了几个明确的灵感来源:

  • Awwwards
  • Godly.website
  • Pinterest
  • Dribbble

这些站点的作用不是照抄,而是帮助建立视觉参照系。比如在 Pinterest 或 Dribbble 上搜索 “SaaS landing page”,很容易看到不同风格的排版、配色、Hero 构图和组件设计;一旦看到喜欢的方向,就可以截图保存。

Argus 案例在这一层选中了 OpenHands 作为主要参考对象。看中的点包括:

  • 顶部滚动内容的处理方式
  • 比常见 SaaS 页面更不寻常的配色
  • 社会证明的组织方式
  • 整体视觉气质

具体操作非常直接:

  • 把目标网站 URL 给 Claude Code
  • 再把多张截图一并发进去
  • 明确说出希望页面向该视觉方向靠拢

这种做法比“我想要高级一点、科技一点、简洁一点”有效得多,因为后者依然过于抽象。

第三级还能进一步升级的一点是:混合多个参考源。

不是只盯着一个网站,而是:

  • 从 Awwwards 拿整体气质
  • 从 Pinterest 拿 Hero 构图
  • 从 Godly.website 拿某段滚动效果
  • 从 Dribbble 拿局部卡片或排版细节

这样更容易拼出自己的方向。

但这一级有一个非常典型的瓶颈,可以理解为:

“风格接近、实现不到位”的落差。

也就是:

  • 看起来有点像
  • 但始终只像了五六成
  • 反复截图、反复提示、反复迭代,时间成本很高
  • 很难真正把截图里的效果准确翻译成代码实现

这说明仅靠“看图仿形”还是不够,需要进入下一层:不仅看表面,还要看网站底层是怎么做出来的。


第四级:拆站者——从专业网站里学实现方式

第四级不再只是看截图,而是开始拆解优秀网站的源码结构。重点不在抄袭,而在于:

把优秀作品当作样板来解构、学习,再改造成自己的版本。

内容把前端简化为三个部分来理解:

  • HTML:网页结构,相当于骨架
  • CSS:网页样式,相当于外观与服装
  • JavaScript:交互与动态,相当于动作与反应

进入这一级之后,学习目标发生变化:

  • 不只是“这个页面看起来为什么高级”
  • 还要知道“这个效果是怎么实现出来的”

具体方法包括:

  1. 打开目标网站
  2. Control + U 查看源代码
  3. 复制 HTML
  4. 再进一步查看页面关联的 CSS 和 JavaScript 文件

Argus 案例里,OpenHands 的 HTML 源码大约有 1152 行。内容特别指出了一个常见问题:

  • 如果让 Claude Code 自己用 webfetch 抓取网站
  • 它常常会借助一个更小的模型去做摘要
  • 对于很长的 CSS 和 JS 文件,这种摘要会丢掉很多关键细节

而很多真正决定页面质感的东西,恰恰藏在这些细节里,比如:

  • 背景效果是怎么做的
  • 滚动动画怎么触发
  • 某些动态交互用了什么方式
  • 页面层次和过渡靠什么实现

为了解决这一点,内容里使用了一个自定义的 site teardown skill,目的是让 Claude Code 尽量完整地理解 HTML、CSS、JS,而不是只看摘要。

这一级的价值非常大,因为它把 Claude Code 的角色从“生成器”变成了“讲解器”和“拆解助手”。进入这个阶段后,可以开始问它:

  • 这个背景效果是怎么实现的
  • 这里为什么会有这种滚动动画
  • GSAP 在这里起了什么作用
  • 某种视觉效果对应的是哪种代码模式

于是,学习过程就变成了:

  • 先挑一个喜欢的网站
  • 拆它
  • 让 Claude Code 解释实现逻辑
  • 再把其中的技巧迁移到自己的页面里

Argus 在这一层重新生成后,页面与 OpenHands 的接近程度明显提高,尤其是背景和结构层面,不再只是“神似”,而是开始能在实现层面真正靠近原站。

这一层的核心风险是 “克隆上限”

  • 只会跟着复制
  • 但不理解为什么这样有效
  • 最后做不出自己的东西

所以第一级到第四级,更像是在打基础:学输入、学判断、学参考、学拆解。真正开始形成个人表达,要到第五级。


第五级:组件与自定义视觉资产——开始做“自己的页面”

第五级的重点,是从“像别人”转向“有自己”。这一级不再满足于继续克隆整站,而是开始通过两条路径加入原创表达:

  • 引入成熟组件后再改造
  • 制作自己的视觉资产与主叙事

1)从组件层借鉴,再让 Claude Code 重写和二次设计

这一层不只是做主视觉,也包括从成熟组件来源抽取局部交互,再让 Claude Code 按需重写。

内容里提到的组件来源包括:

  • 21st.dev
  • CodePen
  • Monet

例如在 21st.dev 上,可以按组件类型查找:

  • 按钮
  • 轮播
  • 导航菜单
  • 滚动区域
  • 地图
  • 图片相关交互
  • hooks

如果看到某个按钮、轮播或卡片效果喜欢,可以直接拿它提供的 prompt 或代码,再交给 Claude Code 集成到当前项目里。这里的关键不是“拿来就用”,而是:

  • 先引入一个高质量局部
  • 再让 Claude Code 调整样式、结构、配色和交互
  • 让它适配自己页面的整体语言

这意味着第五级已经从“整站模仿”进一步细化到“部件级重构”。

2)制作自己的主视觉,让页面拥有视觉叙事

真正能把页面从模板感里拉出来的,是和产品含义绑定的视觉叙事

Argus 案例在这一步先从产品本身出发思考:

  • Argus 是社媒情报应用
  • 名称又让人联想到“能看见更多”的意象
  • 页面不应该只是一组功能卡片,还应该有一个能承接产品想象的视觉主角

随后,为页面想出一句更适合作为品牌表达的口号:

  • See What’s Next

这句话在语义上更接近:

  • 看见接下来会发生什么
  • 预见下一步

它不是把含义局限在“趋势”二字上,而是更宽泛地表达“提前看见将要出现的东西”。

接下来,围绕这个口号与产品定位,让 Claude Code 帮忙发散主视觉想法,再使用图像生成工具制作真正的画面。内容里实际使用的是 Midjourney,并提到这类“概念感、艺术感”的图像它生成得尤其好。

生成图片后,Argus 的 Hero 区重做为:

  • 左侧放产品信息与口号
  • 右侧留出负空间展示主视觉图

这样一来,页面的第一屏不再只是“标题 + 副标题 + 按钮”,而是开始有了更明确的品牌氛围和故事入口。

3)从静态图升级到轻量动态背景

主视觉做好之后,还继续往前推进了一步:把静态背景图做成轻微循环的视频。

这一步使用了支持起始帧视频生成的工具,并强调了几个原则:

  • 背景动效要细微
  • 不要喧宾夺主
  • 最好做成相对自然的长循环
  • 即便循环点不完美,只要整体足够克制,也能接受

Argus 最后使用的是一个约 15 秒 的轻微循环背景视频,效果主要体现在:

  • 云层轻微移动
  • 水面有细小动态
  • 整体更像轻微的呼吸感,而不是强烈动画

同时还做了一个明确的性能决策:

  • 桌面端加载视频背景
  • 移动端只加载静态图

这样既保住了桌面的视觉表现,也避免了移动端负担过重。

第五级的核心,不是“多加点特效”,而是:

开始把页面做成“这个产品的页面”,而不是“某类 SaaS 常见页面”。


第六级:借助外部可视化工具做精修与风格统一

到了第六级,单纯在终端里改视觉会越来越别扭,因为这时候处理的已经不是“有没有页面”,而是:

  • 版式是否平衡
  • 卡片是否有重量感
  • 层次是不是自然
  • 细节够不够统一
  • 是否有“被认真做过”的质感

因此,第六级开始引入外部可视化工具,内容里提到的包括:

  • paper.design
  • Stitch
  • Figma
  • Pencil.dev

其中重点演示的是 Stitch。它的用法不是从零写代码,而是先把当前页面截图丢进去,让它做可视化重设计。比如在 Argus 的例子里:

  • Hero 区整体满意
  • 下半部分太平、太扁
  • 希望把上半部分已经建立起来的色彩和视觉语言延续到下半部分

Stitch 生成的新方案虽然本质上只是图像,不是可直接上线的代码,但它在第六级非常有用,因为它能快速完成这些事情:

  • 帮助探索不同排版
  • 快速尝试颜色与层次变化
  • 给出新的局部视觉思路
  • 形成可以再交还给 Claude Code 的“明确参考”

之后再把 Stitch 的结果截图回传给 Claude Code,让它继续实现、改写、微调。

这一层最重要的特点是:页面进入“打磨期”而不是“搭建期”。

Argus 经过一段时间精修后,出现了很多不会单独特别显眼、但叠加后很能提升质感的细节,例如:

  • 页面刷新时,首屏文字有轻微延迟加载,让首屏更有分量
  • 更换字体,提升整体气质
  • Google Fonts 获取更多字体选择
  • 增加滚动区与底部 ticker,让 Hero 视频区域与下半部分背景衔接更自然
  • 引入玻璃拟态卡片,让背景艺术资产能透出来,同时增加层次
  • 数字统计从 0 动画增长到目标值
  • 某些标签区域加入轻微扫光
  • 滚动条顶部的进度反馈更细腻

这些细节很多用户未必会逐项注意,但它们叠加起来,会形成一种明确感受:

页面不是“被生成出来”的,而是“被雕琢出来”的。

这一层还有一种很实用的工作方式:在已经有初步方向后,不必只靠自己硬想,可以让 Claude Code 去网上搜索某类设计的最佳实践,比如:

  • 怎样让卡片更有重量感
  • 怎样让玻璃拟态更高级而不过度
  • 有哪些微交互能让页面更 premium、但又不过分张扬

然后由人来筛选:

  • 哪些保留
  • 哪些舍弃
  • 哪些继续微调

第六级的本质,是人开始真正进入驾驶位,Claude Code 变成执行与协助创意的工具。


第七级:前沿层——3D、WebGL、Shader 与沉浸式网站

第七级对应的是更前沿的网页体验,包括:

  • 3D 场景
  • WebGL
  • Shader
  • 更像互动艺术或游戏体验的网页

这一层已经不是普通 landing page 的升级版,而是接近大型创意网站、品牌艺术站或实验性交互作品。

内容里明确指出,就 2026 年 3 月下旬 这个时间点而言,这还不是大多数人能稳定用 Claude Code 驾驭的层级。原因在于:

  • 这类网站的复杂度远超一般 HTML/CSS/JS 页面
  • 很多效果不是简单看源码就能学会
  • 背后往往是设计团队、开发团队和长期打磨的结果

第七级被放进路线里,更多是为了说明前端设计空间的上限:

  • 网页不只是 SaaS 模板
  • 也可以接近互动艺术
  • AI 未来也许能帮助普通人更多地进入这个领域
  • 但在当前阶段,它更像是“未来方向”,而不是“眼下的通用做法”

Argus 案例的完整演进链路

围绕 Argus,这期内容实际展示了一条非常完整的页面升级路径:

  1. 模糊提示词起步

    • 先得到一个极其普通的 SaaS 落地页
    • 暴露出“产品说清楚了,设计没说清楚”的问题
  2. 引入设计 skill

    • 用 UIUX Pro Max 等 skill 提升基础排版、色彩和层次
    • 页面变好,但仍然带明显模板感
  3. 加入视觉参考

    • 从 Awwwards、Godly.website、Pinterest、Dribbble 等处搜集灵感
    • 把 OpenHands 之类目标网站的截图交给 Claude Code
    • 页面开始在风格上接近参考方向
  4. 拆解源码

    • 不再只看截图,而是复制 HTML、分析 CSS 和 JS
    • 借助 site teardown skill 理解底层实现
    • 真正接近目标网站的效果做法
  5. 摆脱纯克隆

    • 从 21st.dev、CodePen、Monet 等处找局部组件灵感
    • 同时开始制作自己的主视觉
    • 用 Midjourney 生成与 Argus 品牌相关的 Hero 图
    • 确立 See What’s Next 作为页面口号
  6. 加入轻量动态

    • 将 Hero 主视觉升级为轻微循环视频
    • 桌面端用视频,移动端用静态图
    • 页面更有呼吸感与记忆点
  7. 外部工具协同精修

    • 用 Stitch 等工具探索下半部分重设计
    • 再交回 Claude Code 实现
    • 增加字体、ticker、玻璃拟态、计数动画、扫光和加载细节
    • 最终形成更完整、更统一的页面语言

关键方法与可复用操作链

这期内容最有价值的部分,不在于某个单独工具,而在于它把“前端设计怎么从 AI 模板走向可控作品”拆成了一条可执行路线。

1. 先补输入质量,而不是先怪模型

如果一开始只说:

  • 做个 landing page
  • 要科技感
  • 要简洁

那么输出大概率仍然是模板化结果。要逐步补上的,是:

  • 页面目标
  • 技术框架
  • 区块设计
  • 风格参照
  • 设计词汇

2. 先学会看,再学会说

很多人不是没有偏好,而是:

  • 没看过足够多的高质量参考
  • 不知道自己喜欢什么
  • 即使喜欢,也说不出是哪里好

所以前期大量看站、截图、对比,本身就是训练的一部分。

3. 视觉参考比抽象形容更有效

与其说“高级一点”“未来感一点”,不如直接给:

  • 网站截图
  • URL
  • 局部组件参考
  • 具体想保留和想舍弃的部分

给它看,比跟它解释,往往更有效。

4. 截图只能解决一半问题,想真正逼近效果必须拆实现

如果只给截图,很容易陷入:

  • 看着像
  • 但做不到位
  • 来回提示很多轮仍然差一点

这时候就需要查看 HTML、CSS、JS,理解效果是怎么做出来的。“拆站学习”是从模仿走向真正理解的关键步骤。

5. 只会克隆不够,必须加入自己的叙事和资产

真正能摆脱模板感的,不只是“还原得更像”,而是:

  • 有自己的 Hero 构图
  • 有自己的品牌口号
  • 有自己的视觉资产
  • 有自己的局部组件风格
  • 有与产品含义相连的叙事

6. 高级感往往来自一堆小细节的叠加

不是某一个大特效决定页面高级,而是很多小地方共同作用,例如:

  • 字体
  • 间距
  • hover
  • 加载节奏
  • 卡片重量感
  • 分区过渡
  • 轻微动画
  • 颜色统一性

质感通常是“多处小细节都不差”的结果。


关键工具与数据点

  • 7 个层级:整套 Claude Code 前端设计进阶框架
  • Argus:案例产品,一个帮助用户寻找社媒热门话题的情报应用
  • UIUX Pro Max:GitHub 开源 skill,约 5.2 万 stars
  • OpenHands:第三级与第四级的重要参考网站
  • 1152 行 HTML:案例里拆解参考站时复制的页面源码规模
  • 15 秒视频背景:Hero 背景的轻量循环动效长度
  • 约 20 分钟精修:第六级中,用来迭代页面细节的示例时长

常用灵感与工具来源包括:

  • Awwwards
  • Godly.website
  • Pinterest
  • Dribbble
  • 21st.dev
  • CodePen
  • Monet
  • Stitch
  • Figma
  • Pencil.dev
  • Google Fonts
  • Midjourney

结论回顾

这期内容最终落到一个非常明确的判断上:

Claude Code 做不好前端,不只是因为 AI 本身有限,更因为多数人缺少视觉参照、设计词汇和对实现方式的理解。

因此,更有效的路线不是找“神级提示词”,而是按顺序完成这几件事:

  • 看大量优秀作品,建立审美参照
  • 用截图和组件参考提高表达精度
  • 拆解 HTML、CSS、JS,理解效果对应的实现方式
  • 把学到的技巧迁移到自己的产品页面
  • 加入自己的视觉资产、叙事和细节打磨

用一句话概括,这套路线的目标是:

把 Claude Code 从“自动吐模板的生成器”,逐步变成“帮助完成个人设计表达的协作工具”。

相关推荐 (...)