Skip to content
← 回到文章列表

怎样把你的 AI 变成世界级设计师 ​

👋 Hey there, I’m Lenny. Each week, I share deeply researched product, growth, and career advice. For more: Lenny’s Jobs | Lenny’s Podcast | Lennybot | How I AI | Become an AI-Native Builder and my other favorite AI/PM courses

👋 嗨,我是 Lenny。每周我都会分享一些下足功夫调研过的产品、增长和职业建议。更多内容:Lenny’s Jobs | Lenny’s Podcast | Lennybot | How I AI | Become an AI-Native Builder,以及我喜欢的其他 AI/PM 课程

P.S. Get a full free year of Cursor, Notion, Replit, Lovable, Wispr Flow, Linear, ElevenLabs, Factory, PostHog, Granola, Brain.fm, Waking Up, and more, by becoming an Insider subscriber (while supplies last). Learn more.

P.S. 成为 Insider 订阅者,就能免费用一整年 Cursor、Notion、Replit、Lovable、Wispr Flow、Linear、ElevenLabs、Factory、PostHog、Granola、Brain.fm、Waking Up 等等(送完即止)。了解详情。


I’d always thought AI was bad at design. But after reading this mind-blowing post by Anshu Chimala, I realize I was just doing it wrong. Anshu led software engineering and design teams at Apple for 12 years, focusing on research and prototyping for future AI products. He regularly shares design tutorials and demos on X (he’s one of my favorite follows). For deeper dives into crafting distinctive experiences with AI, check out his Substack and connect with him on LinkedIn.

我一直觉得 AI 不会做设计。读完 Anshu Chimala 这篇让人大开眼界的文章,我才发现是自己用错了方法。Anshu 在 Apple 带了 12 年软件工程和设计团队,专做未来 AI 产品的研究和原型。他经常在 X 上分享设计教程和 demo(我最喜欢关注的人之一)。想更深入地了解怎么用 AI 做出与众不同的体验,可以看他的 Substack,也可以在 LinkedIn 上加他。

Let’s get into it.

我们开始吧。


A conversational calorie tracker, built in three prompts with Claude Fable 5:

一个用聊天来记卡路里的应用,用 Claude Fable 5 三条提示词做出来:

一个用聊天来记卡路里的应用

A space exploration game, built in two prompts with Claude Opus 5:

一个太空探索游戏,用 Claude Opus 5 两条提示词做出来:

一个太空探索游戏

A dynamic landing page, built in three prompts with Claude Opus 5 + GPT-5.6 Sol:

一个会动的落地页,用 Claude Opus 5 + GPT-5.6 Sol 三条提示词做出来:

一个会动的落地页

I often post AI design demos like these on X. Every time I do, someone inevitably asks, “Why does the model create all this incredible stuff for you, but when I try, I only get generic slop? It’s like you’re using a completely different model.”

我常在 X 上发这类 AI 设计 demo。每次发,总有人问:「为什么模型给你做的东西这么惊艳,我自己试就只能拿到千篇一律的平庸货(slop)?你用的好像根本是另一个模型。」

I’m not using a different model, but I am getting more out of the models I work with. Most people only see 1% of AI’s creative potential. I want to show you how to tap into the other 99%.

我用的不是别的模型,只是从手里的模型身上榨出了更多东西。大多数人只看到了 AI 创造力的 1%,我想带你看看怎么挖出剩下那 99%。

AI models are capable of amazing creativity, but that creativity gets stifled by how they’re trained. Large language models are next-token predictors: at each step, they look at a sequence of text and predict what comes next based on millions of examples. The results may be rated by humans, and those ratings fed back into the model. This teaches the model to make consistent, safe choices that fit everyone’s preferences.

AI 模型的创造力其实很惊人,只是被训练方式压住了。大语言模型是「预测下一个 token」的机器:每一步都看着一串文本,根据几百万个例子猜下一个是什么。输出可能会交给人来打分,分数再反馈回模型。这么一来,模型学会的是做稳妥、一致、谁都挑不出毛病的选择。

This makes typical LLMs great at most tasks but poor designers. To create a design, an LLM has to build it out token by token. Whenever it needs to make a design decision—what colors to use, or how to arrange elements—the model fills in the tokens it thinks are most likely to please everyone. As a result, the design usually ends up being repetitive and bland. It’s like the ultimate case of design-by-committee.

所以一般的 LLM 干大多数活都很好,偏偏当不好设计师。做一个设计,LLM 得一个 token 一个 token 地往外搭。每当要做设计决策,比如用什么颜色、元素怎么摆,模型填进去的都是它认为最可能讨所有人喜欢的那些 token。结果设计往往重复又寡淡,简直是委员会式设计(design-by-committee)的终极形态。

Great design, on the other hand, starts with feeling and aims to create an emotional response. It bends the rules and delights users with memorable, unexpected choices. Great design is exactly the opposite of what an LLM does naturally, which is to make the most predictable choice at every step.

好的设计正相反,它从感受出发,要的是在人心里激起点什么。它会打破规则,用让人记得住、意想不到的选择打动用户。LLM 的本能是每一步都选最好猜的那个,好设计跟它恰好背道而驰。

However, if we can get the model to reach beyond the most predictable choices, we can access a vast landscape of creative ideas that most people miss out on.

不过,只要能让模型越过那些最好猜的选择,我们就能摸到一大片大多数人错过的创意天地。

This is a lesson I learned from managing human designers, before I was managing AI ones. For most of my career at Apple, I led an R&D team designing exploratory future AI products. Early on, our preconceived notions about how user interfaces should work limited our creativity and kept us returning to the same old ideas. Through rigor and new processes, we learned to stop re-creating what’s comfortable and instead look to the fringes of what’s possible, to generate something new. We became experts at polishing the little details to an Apple level of quality.

这是我在管 AI 设计师之前、管人类设计师时学到的。在 Apple 的大部分时间里,我带的是一个研发团队,设计探索性的未来 AI 产品。一开始,我们对「界面该怎么做」的成见限制了创造力,老是绕回那几个旧点子。靠着严格的要求和新流程,我们学会了不再复刻舒服的老样子,而是去可能性的边缘找新东西。我们也练成了把小细节打磨到 Apple 水准的本事。

Since my time at Apple, I’ve been working on applying that same process to my work with AI. In the past couple years, AI agents have become extremely capable. They can do in hours what used to take my team weeks. And with the right guidance, they can create designs that look completely unlike anything else.

离开 Apple 之后,我一直在把同一套流程搬到和 AI 的协作里。过去这两年,AI Agent 的能力强得吓人,以前我的团队要几周的活,它们几小时就能干完。只要引导得当,它们能做出跟别的东西完全不一样的设计。

Loosely inspired by the Double Diamond design process, I’ve reimagined the design process for a team of AI agents instead of human designers:

我大致参考了双钻设计流程(Double Diamond),把设计流程重新改造了一遍,对象从人类设计师换成了一队 AI Agent:

  1. Discover new ideas beyond the average slop by exploring a variety of directions and creating bold, ambitious design briefs.发现:探索各种方向、写出大胆有野心的设计简报,找到平庸货之外的新点子。
  2. Define an individual design identity by pushing AI beyond its familiar patterns and chaining models together to fully realize the design’s potential.定义:逼 AI 跳出熟悉的套路,把几个模型串起来用,把设计的潜力完全发挥出来,确立独有的设计个性。
  3. Deliver a stunning final result by polishing away the sloppy rough edges and focusing on the key elements.交付:磨掉粗糙的毛边,把精力集中在关键元素上,交出惊艳的最终成品。

By following these stages and applying the techniques within each one, you can create an incredible design remarkably quickly—and make people ask, “Why does AI create magic for you (and not me)?”

按这几个阶段走,用上每个阶段里的技巧,你能飞快地做出惊人的设计,然后轮到别人来问你:「为什么 AI 给你变魔术,不给我变?」

Discover: Explore the space of possibilities发现:探索可能性的空间 ​

The hardest part of the design process is looking at a blank screen with infinite possibilities. The best way to tackle that moment is to start by going broad before going deep. AI is an excellent tool to explore a wide variety of potential directions.

设计里最难的,就是盯着一块空白屏幕、面对无限可能的那一刻。最好的办法是先铺开、再深挖。要探索各种各样可能的方向,AI 是绝佳的工具。

As we know, though, models tend to overrely on familiar patterns and make conservative choices. To explore the full potential design space, we want to coax a model to do the opposite: be bold, be varied, and take risks. Below are two ways to push it out of its comfort zone.

可我们也知道,模型总是过度依赖熟悉的套路,选择也偏保守。想把整个设计空间探索一遍,就得哄着模型反着来:大胆、多样、敢冒险。下面是两种把它推出舒适区的办法。

Technique 1: Use seed strings to inject variety技巧 1:用种子字符串注入多样性 ​

The idea here is to get the model to find a new source of inspiration for designs, rather than relying on the defaults it learned from training. If you’ve tried to prompt a model to design a website or app, you’ve probably already seen what that default looks like.

思路是让模型给设计找一个新的灵感来源,别老靠训练时学来的默认选项。如果你试过让模型设计网站或应用,大概已经见过那个默认长什么样了。

As a simple example, I gave four instances of Claude Code the same prompt:

举个简单的例子,我给四个 Claude Code 实例同一条提示词:

Prompt:

提示词:

Build me a landing page for my productivity app.

给我的效率应用做一个落地页。

Claude Opus 5:

Claude Opus 5:

四个实例做出的默认落地页

Almost every time, we get a purplish gradient, text on the left, graphic on the right, and the exact same structure. It looks like every AI-designed website ever.

几乎每次都是偏紫的渐变、左边文字、右边配图,结构一模一样。跟所有 AI 设计出来的网站都一个样。

We didn’t ask the model to do anything unique or varied, so it makes sense that it keeps falling back on the same patterns it knows well. But just asking for variety doesn’t work:

我们没让模型做什么独特或多变的东西,它一直退回到自己熟悉的套路也说得通。可光要求它「多样一点」并没有用:

Prompt:

提示词:

Build me a landing page for my productivity app. Give me something totally unique. Make every design decision completely at random.

给我的效率应用做一个落地页。给我一个完全独一无二的东西。每一个设计决策都完全随机地做。

Claude Opus 5:

Claude Opus 5:

要求「完全随机」后的落地页

The results are different from before, but they’re still not varied. The model always uses the same color scheme, structure, and even the same awkward pottery metaphors. It’s predicting tokens that sound random but aren’t actually random.

结果和之前不一样了,可几次之间还是差不多。模型每次都用同样的配色、同样的结构,连那个别扭的陶艺比喻都一样。它预测出来的 token 听上去随机,实际上并不随机。

The problem is that the model can’t inherently act randomly. It can only predict the most likely token. If we want variety, we have to bring it from outside the model. One technique for this is String Seed of Thought, published by Sakana AI. We make the AI generate a random string and use it as design inspiration. That way, the model is truly making different decisions each time.

问题在于模型天生就做不到随机。 它只能预测最可能的那个 token。想要多样性,就得从模型外面引进来。有个办法叫 String Seed of Thought,由 Sakana AI 发表:让 AI 生成一串随机字符串,拿它当设计灵感。这样模型每次做的决策才真的不一样。

Prompt:

提示词:

I want you to build me a landing page for my productivity app.

Follow this procedure:

  1. Generate a long, random alphanumeric string using a shell script.
  2. Define the creative direction (color scheme, layout, typography, etc.) based on the string. Look beyond the surface for subpatterns, special numbers, anything that inspires you.
  3. Use your judgment to bring this direction to life and make it look great.

Don’t reveal the string in the design. It’s only for your inspiration.

我想让你给我的效率应用做一个落地页。

按下面的步骤来:

  1. 用 shell 脚本生成一串很长的随机字母数字字符串。
  2. 根据这串字符定下创意方向(配色、布局、字体排印等等)。别只看表面,去找里面的子模式、特殊的数字,任何能给你灵感的东西。
  3. 用你自己的判断把这个方向落地,做得漂亮一点。

不要在设计里露出这串字符,它只是给你找灵感用的。

Claude Opus 5:

Claude Opus 5:

用种子字符串做出的落地页

Suddenly the outputs are much more varied! Now we’re seeing different color schemes, fonts, and new ideas. The previous designs were ones that any Claude user could get. These designs are one-of-a-kind; no two runs ever produce the same result.

一下子就多样多了!配色、字体都不一样了,也冒出了新点子。前面那些设计,随便哪个 Claude 用户都能拿到;这些是独一份的,跑两次永远不会出一样的结果。

Technique 2: Be much more ambitious with your prompts技巧 2:提示词写得野心大得多 ​

Another approach to giving a model a strong push is to get more specific and wild with your prompts. This gives the model a clear vision to base its decisions on, rather than letting it make them up on the fly. The best way to find a unique idea is by bringing your own taste into the equation. You first imagine the inspiration—a video game, an interior design trend, an art installation—and describe how you’d like that inspiration to influence the AI’s outputs. Here are some examples:

另一种狠推模型一把的办法,是把提示词写得更具体、更疯。这样模型做决策时有一个清楚的愿景可依,不用临场瞎编。找到独特点子最好的办法,是把你自己的品味带进来。先想好灵感是什么,一款电子游戏、一种室内设计潮流、一件艺术装置,再描述你希望这个灵感怎样影响 AI 的产出。举几个例子:

“Build me a landing page for my productivity app, with a bold pixel art theme and stunning graphics. Each section should feel like a still from a video game, yet somehow it should all function as a landing page.”

「给我的效率应用做一个落地页,用大胆的像素画主题和惊艳的画面。每个区块都要像电子游戏里的一帧截图,但整体还得能当落地页用。」

像素画主题的落地页

“Build me a landing page for my productivity app, set in an isometric living 3D city, where different features are somehow represented by neighborhoods or buildings.”

「给我的效率应用做一个落地页,场景是一座活生生的等距视角 3D 城市,不同的功能用不同的街区或建筑来表现。」

等距 3D 城市主题的落地页

“Build me a landing page for my productivity app, with a radically asymmetric layout, dissonant colors and typography, and uncomfortable negative space. Break all the rules but still make it look good.”

「给我的效率应用做一个落地页,布局极端不对称,颜色和字体互相冲突,留白多到让人不舒服。打破所有规则,但还是要好看。」

打破规则的不对称落地页

Of course, the hard part is coming up with original ideas to ask for. AI can help with this too, but if you simply ask it for ideas, you’ll get the same average ones everyone else gets. Here’s a system I use to find unique prompt ideas with AI:

当然,难的是想出原创的点子去提要求。这一步 AI 也能帮忙,可你要是直接找它要点子,拿到的就是别人都能拿到的那些平均货。下面是我用 AI 找独特提示词点子的一套办法:

1. Ask AI to list a bunch of ideas, intentionally lacking detail. The goal is just to inspire your imagination.1. 让 AI 列一堆点子,故意不要细节,目的只是激发你的想象 ​

I want to come up with a bold, unique design language for my product. Can you list as many ideas as you can, with short, high-level descriptions? Go broad, not deep.

我想给我的产品想一套大胆、独特的设计语言。你能尽可能多地列一些点子吗?每个配一句简短的高层描述就行。要广,不要深。

AI 列出的设计语言点子

2. Visualize your favorites and note how you react to different directions. Then ask AI to refine them.2. 把喜欢的几个在脑子里过一遍,记下自己对不同方向的反应,再让 AI 细化 ​

Industrial Control Panel:

  • I’m imagining something tactile. Clicky, satisfying buttons, nice sounds.
  • Initially I pictured something cartoony or skeuomorphic, but this feels tacky to me. Avoid that.
  • Instead, want consistent components and little touches that land this look without going overboard.
  • Gray gradients would look boring. Need more texture. Maybe we can incorporate some color, while retaining the control panel feel?

Can you sharpen this one based on my tastes?

工业控制面板:

  • 我想象的是有触感的东西。按下去咔哒一声、很解压的按钮,好听的音效。
  • 一开始我脑子里是卡通风或者拟物风,但感觉有点土。避开这个。
  • 我想要的是统一的组件和一些小细节,把这个感觉做出来,又不过火。
  • 灰色渐变会很无聊。需要更多质感。能不能加点颜色,同时保留控制面板的感觉?

你能按我的口味把这个方向再磨一磨吗?

AI 按口味细化后的方向

3. Iterate until you’re satisfied, then ask AI to write the prompt to build it.3. 反复迭代到满意,再让 AI 写出用来实现它的提示词 ​

Can you write a concise prompt that an AI agent could use to build an initial POC page with this?

你能写一条简洁的提示词,让 AI Agent 能照着它做出一个初版的 POC 页面吗?

AI 写出的实现提示词

If you just paste AI-generated ideas back into AI, it’s hard to get something unique. After all, anyone else could have done the same thing. However, when you actively steer the design direction, you end up with something only you could have created.

要是你只是把 AI 生成的点子原样贴回给 AI,很难得到什么独特的东西,毕竟谁都能这么干。但如果你主动掌舵设计方向,最后得到的就是只有你才做得出来的东西。

Don’t be afraid to try ideas that sound terrible. If you find yourself thinking, “There’s no way this will work,” you’re on the right track. Often, your agent will surprise you, and you’ll realize you were underestimating it. If not, just throw away those results and try something else. But save the prompts that don’t work, and test them again when newer models come out. That way, you’ll know you’re taking full advantage of what the latest models can do.

别怕试那些听上去很烂的点子。如果你心里在想「这不可能行得通」,那就对了。你的 Agent 常常会给你惊喜,你会发现自己低估了它。没惊喜也没关系,把结果扔掉换个点子就行。不过那些没成的提示词要存好,等新模型出来再试一遍。这样你就知道自己有没有把最新模型的本事用足。

Define: Deepen your design direction定义:把设计方向挖深 ​

So far, we’ve looked at how to explore a broad set of ideas and hopefully land on a promising initial design. No matter how we prompt, though, our initial AI-generated designs will usually still feel generic.

到这里,我们讲了怎么大范围地探索点子,争取落到一个有希望的初版设计上。可不管提示词怎么写,AI 生成的初版设计通常还是有股千篇一律的味道。

For example, look at the designs we came up with using seed strings:

比如看看我们用种子字符串做出来的那几版:

用种子字符串做出的落地页

These have promise, but they’re still relying heavily on the same stale patterns: text on the left with a CTA button below, nav bar up top, graphic on the right.

它们有潜力,但还是严重依赖那几个老套路:左边文字、下面一个 CTA 按钮,顶上导航栏,右边配图。

Our next goal is to give each design an individual personality through distinct design choices. Below are my favorite techniques to do that.

下一步的目标,是靠鲜明的设计选择,给每个设计一份自己的个性。下面是我最喜欢的几个技巧。

Technique 3: Create positive feedback loops with subagents技巧 3:用子 Agent 搭一个正反馈循环 ​

We need to iterate on our designs to improve them. But simply asking our agent to look at the design and improve it won’t work, because the agent isn’t objective: it reviews its own code, past decisions, and previous rationale. AI can’t easily zoom out, look at the big picture, and “think different.”

设计要迭代才会变好。可光让 Agent 看看设计、再改好一点,是没用的,因为它不客观:它审的是自己写的代码、自己过去的决策和当初的理由。AI 很难退后一步看全局,去「think different」。

To solve this, instead of letting the coding agent decide when the design is good enough, have it ask another agent—a “design critic.” The critic’s job is to look at screenshots of the current design and provide feedback. It doesn’t care how the current design is implemented or how much effort went into it, only if it actually hits the quality bar.

解决办法是:别让写代码的 Agent 自己判断设计够不够好,让它去问另一个 Agent,一个「设计评审」。评审的活儿是看当前设计的截图、给反馈。它不在乎设计是怎么实现的、花了多少力气,只看它到底有没有达到质量标准。

This approach has an extra benefit: we can use a big, expensive model for the critic without breaking the bank, because we’ll only use it for executive decisions. A cheap, fast model can do the grunt work, while the strong critic model provides taste.

这么做还有个额外的好处:评审可以用又大又贵的模型,也不至于把钱包掏空,因为它只负责拍板。脏活累活交给便宜快速的模型,品味由强大的评审模型来提供。

Let’s try this on our previous designs, using Claude Fable 5 as the critic:

我们拿前面那几版设计试一下,用 Claude Fable 5 当评审:

Prompt:

提示词:

I want you to improve this design. To figure out what to focus on, use a Fable 5 subagent as a design critic.

Follow this procedure at each iteration:

  • Capture a screenshot of the current design
  • Invoke the critic in a fresh context, with just the screenshot, not the code, implementation details, or earlier iterations/critiques
  • Ask it to evaluate the aesthetic that the design is going for, imagine how a top design studio would execute this aesthetic, then outline the biggest gaps
  • Lastly, it should provide a score out of 10 indicating how close the current design is to that studio-level quality bar

Provide this guidance to the critic in its prompt:

  • It should think high-level about the overall structure and composition as well as look at the fine details
  • It should watch out for patterns that feel overdone, excessive, or otherwise obviously AI-generated, and penalize them
  • It should provide tight, specific feedback, not vague prose
  • It should be bold and opinionated, not rely on what’s safe or easy

Your work is only complete when the critic independently deems it 9/10 or higher. Do not put that criterion in the critic prompt; keep it objective in its scoring. Use the same critic prompt each time.

我想让你改进这个设计。要搞清楚该集中改哪里,用一个 Fable 5 子 Agent 当设计评审。

每一轮迭代都按这个步骤来:

  • 给当前设计截一张图
  • 在全新的上下文里调用评审,只给它截图,不给代码、实现细节,也不给之前的迭代或评审意见
  • 让它判断这个设计想要的美学风格是什么,想象一家顶级设计工作室会怎样做出这种风格,然后列出最大的差距
  • 最后,让它打一个满分 10 分的分数,表示当前设计离那种工作室级别的质量标准有多近

在评审的提示词里给它这些指导:

  • 既要从高处想整体的结构和构图,也要看细节
  • 要留意那些用滥了的、过度的或者一看就是 AI 生成的套路,并且扣分
  • 反馈要紧凑、具体,不要含糊的长篇大论
  • 要大胆、有主见,不要只挑稳妥或省事的说

只有评审独立给出 9/10 或更高的分数,你的工作才算完成。不要把这条标准写进评审的提示词里,让它打分时保持客观。每次都用同一份评审提示词。

Claude Opus 5:

Claude Opus 5:

经过评审循环改进后的设计

Instead of the same cookie-cutter layout over and over, each design now has its own identity—but still maintains its original high-level aesthetic.

不再是一遍又一遍的流水线布局,现在每个设计都有了自己的个性,同时保住了原来那个大方向上的美学风格。

Notably, in each case, Fable accounted for less than 10% of output tokens. Asking Fable to redesign the page directly would have cost twice as much and taken much longer.

另外,每个例子里 Fable 只占了不到 10% 的输出 token。要是直接让 Fable 重新设计页面,成本会翻倍,时间也长得多。

The way you set these loops up matters a lot. Here are some tips:

这种循环怎么搭,影响很大。给几条建议:

  • Make sure the criteria for the critic are as clear and objective as possible.给评审的标准要尽量清楚、客观。
    • Bad: “Judge if our design looks beautiful, not AI-generated.” This is too subjective, and the results will vary wildly from run to run.差:「判断我们的设计好不好看、像不像 AI 生成的。」太主观了,每次跑出来的结果会差得离谱。
      • OK: “Review the aesthetic we’re going for, visualize how a top design studio would execute it, then judge our design’s quality against that bar.” The prompt is still mushy, but it provides a consistent framework and quality bar.还行:「看看我们想要的美学风格,想象一家顶级设计工作室会怎么做,然后拿那个标准来评判我们设计的质量。」提示词还是有点虚,但至少给了一个一致的框架和质量标准。
      • Great: “Here are 5 designs: 4 professional examples and 1 screenshot of our product. Rank them by polish and taste level.” This instruction is concrete and objective, and gives a visual baseline for judgment.很好:「这里有 5 个设计:4 个专业案例,1 张我们产品的截图。按精致程度和品味高低给它们排序。」这条指令具体又客观,还给了一个可以拿来比的视觉基准。
  • Provide example images to demonstrate the target quality bar. You can use comparable screenshots or designs you like, or even AI-generated concept art. Instruct the critic to treat these as a baseline or a moodboard, not a target. You don’t want it to copy other designs outright.给出示例图片,说明目标质量是什么水平。 可以用同类产品的截图、你喜欢的设计,甚至 AI 生成的概念图。告诉评审把这些当作基准或情绪板(moodboard),不是照着做的目标。你可不想它直接抄别人的设计。
  • Set the stopping criteria carefully. Otherwise, the critic may never consider the design good enough, and your agent will helplessly burn tokens trying to please it. Prompt it to do one or two iterations first, and see if it’s converging before adding more.停止条件要谨慎设定。 不然评审可能永远觉得设计不够好,你的 Agent 只能无奈地烧 token 去讨好它。先让它跑一两轮,看看有没有在收敛,再决定加不加轮次。
  • Choose the right model for each job. Consider bigger models for the critic role, since more parameters generally translate to better design sense and a wider distribution of ideas. Small models can be effective as the implementer, but don’t go too small. You still need a model that’s capable of executing a design direction well.每个活儿选对模型。 评审可以考虑用更大的模型,参数越多,一般设计感越好、点子的分布也越广。小模型当实现者可以很好用,但别太小,你还是需要一个能把设计方向执行到位的模型。

Technique 4: Use image generation to enrich designs技巧 4:用图像生成让设计更丰富 ​

Coding agents love to write code, but they usually don’t incorporate images. Instead, they tend to use the easy code-based alternatives: gradients, shapes, and basic patterns. Those are all strong giveaways of an AI-generated design.

编程 Agent 爱写代码,但通常不往设计里放图片。它们更喜欢用代码就能搞定的省事替代品:渐变、形状、基础图案。这些全都是 AI 设计的明显破绽。

Some agents have image tools built in, but they underutilize them. Others don’t have image tools out of the box but can easily use the OpenAI or Gemini APIs to generate images with an API key.

有些 Agent 内置了图像工具,但用得很少。还有些开箱没有图像工具,不过给个 API key,它们就能轻松调用 OpenAI 或 Gemini 的 API 来生成图片。

Let’s try this on the designs from the last step:

我们拿上一步的设计试一下:

Prompt:

提示词:

The design is pretty plain. Add more personality using image generation. Consider shaders or 3D effects in combination with images to create more interesting visuals.

For image generation, use this OpenAI API key (only use it locally, do not store it in the code or product): sk-a1b2c3d4…

Verify that your work looks right frame-by-frame in the browser.

这个设计挺平淡的。用图像生成给它加点个性。可以考虑把着色器或 3D 效果和图片结合起来,做出更有意思的画面。

图像生成用这个 OpenAI API key(只在本地用,不要存进代码或产品里):sk-a1b2c3d4…

在浏览器里逐帧确认效果是对的。

Claude Opus 5 (before and after):

Claude Opus 5(改前和改后):

加入图像生成前后对比 1

加入图像生成前后对比 2

加入图像生成前后对比 3

加入图像生成前后对比 4

Images and effects like these can quickly add a lot of personality and make a design less obviously AI-generated, since they demonstrate more than surface-level effort.

这类图片和效果能很快给设计加上不少个性,也让它没那么明显像 AI 做的,因为看得出来下的功夫不只停在表面。

Depending on your setup, there are different ways to connect your agent to image generation tools:

把 Agent 接上图像生成工具,按你的环境有几种办法:

  • If you use Codex, Antigravity, or Grok Build:如果你用 Codex、Antigravity 或 Grok Build:
    • Tell your agent to use its built-in image generation. The agent already knows how to do this but rarely does so until instructed.告诉 Agent 用它内置的图像生成。它本来就会,只是你不开口它很少主动用。
  • If you use Claude Code or another agent but also have a ChatGPT subscription:如果你用 Claude Code 或别的 Agent,但同时订了 ChatGPT:
    • Tell your agent, “Use the Codex CLI to generate images. Help me install it if it isn’t already present. Make sure it’s billing my subscription, not an API key.” This lets you use your ChatGPT subscription for image generation without extra costs.告诉 Agent:「用 Codex CLI 生成图片。还没装的话帮我装上。确保扣的是我的订阅,不是 API key。」这样就能拿 ChatGPT 订阅来生成图片,不额外花钱。
  • If you only use Claude, or any other tool:如果你只用 Claude,或者别的什么工具:
    • The simplest path is to give your agent an OpenAI or Gemini API key to generate images. I recommend creating a separate API key with a tight spend limit, just for your agent. That way, your costs are controlled even if the key gets out or the agent misuses it, and you can easily revoke the key without disrupting other work.最简单的办法是给 Agent 一个 OpenAI 或 Gemini 的 API key 来生成图片。我建议单独建一个专给 Agent 用的 API key,把消费上限设紧一点。这样就算 key 泄露了或者 Agent 乱用,花费也在控制之内,而且随时可以吊销,不影响别的工作。
      • If you find yourself pasting keys into chats frequently, put them in a file instead, and point your agent to it in your project. Tell your agent: “Create a gitignored file called .env.agents, store this API key in it, and note to yourself in AGENTS.md/CLAUDE.md that these keys are for you to use during development (but must not ship with the product).”如果你发现自己老是往对话里贴 key,那就把它们放进一个文件,在项目里告诉 Agent 去哪找。跟 Agent 说:「建一个被 gitignore 的文件,叫 .env.agents,把这个 API key 存进去,再在 AGENTS.md/CLAUDE.md 里给自己记一笔:这些 key 是开发时给你用的(但绝不能跟着产品发出去)。」

Technique 5: For more advanced motion, use video generation技巧 5:想要更高级的动效,用视频生成 ​

Video generation models are incredibly powerful these days, but most people think of them as tools for generating UGC ads or clips of Will Smith eating spaghetti. They can work wonders for everyday design work too.

如今的视频生成模型强得离谱,可大多数人只把它们当成做 UGC 广告、或者生成威尔·史密斯吃意大利面片段的工具。其实它们在日常设计工作里也能创造奇迹。

There are many video models out there, and the best ones change frequently, so I like to use an aggregator platform like fal.ai. This way, we can give our agent a single API key and let it evaluate different options and choose the best one without needing multiple integrations.

视频模型很多,最好的那几个又换得很勤,所以我喜欢用 fal.ai 这样的聚合平台。这样只用给 Agent 一个 API key,让它自己评估各种选项、挑最好的,不用一个个去接。

Here are two ways I love to use video models in my designs:

把视频模型用进设计里,我最喜欢这两种用法:

Create stunning animated graphics做出惊艳的动态图形 ​

The trick is to generate a looping clip with a solid color background, then either chroma key it out (like a green screen) or, in more complex cases, use a video matting model to remove the background. This gives you an animation that you can layer anywhere in your UI without it looking like a video.

诀窍是先生成一段纯色背景的循环片段,然后要么用色键抠掉背景(就像绿幕),要么在更复杂的情况下用视频抠像(video matting)模型去掉背景。这样得到的动画可以叠在界面里任何地方,看上去一点也不像一段视频。

For example, I took one of our previous designs and ran this prompt:

比如我拿前面的一个设计,跑了这条提示词:

Prompt:

提示词:

Can you replace the image on this page with a looping video clip that does something more interesting? Have the crystal splinter apart and slowly spin around. It should have awesome glassy effects that refract the page background and cast shadows and light around it.

To get convincing glass refraction effects, render the video of the glass over the page background colors first (so it bakes in the refraction effects), then remove the background with a video matting model.

Use this fal.ai API key: sk-a1b2c3d4…

Find appropriate recent models for video generation and background removal.

能不能把这个页面上的图片换成一段更有意思的循环视频?让水晶碎裂开,然后慢慢旋转。要有很棒的玻璃质感,能折射页面背景,在周围投下阴影和光。

为了让玻璃折射看起来可信,先把玻璃叠在页面背景色上渲染成视频(这样折射效果就烘焙进去了),再用视频抠像模型去掉背景。

用这个 fal.ai API key:sk-a1b2c3d4…

找一些合适的、比较新的模型来做视频生成和背景去除。

GPT-5.6 Sol (before and after):

GPT-5.6 Sol(改前和改后):

水晶碎裂旋转的循环视频前后对比

This is a much richer effect than you can get with code: interesting caustic reflections, glassy refraction effects, and complex physical motion.

这效果比用代码能做出来的丰富得多:有意思的焦散反光、玻璃折射,还有复杂的物理运动。

Create fluid transitions between states在状态之间做出流畅的过渡 ​

This is a really underrated use case for video models. In addition to generating video from text, many video models can interpolate between keyframe images. This lets you take two product stills and create a transition clip between them. You can play the clip when the user takes an action (like navigating to another screen of your app) or scrub through it frame-by-frame in response to a gesture (like scrolling or swiping).

这是视频模型一个被严重低估的用法。很多视频模型除了能从文本生成视频,还能在关键帧图片之间插帧。也就是说,你可以拿两张产品静态图,生成一段它们之间的过渡片段。用户做某个操作时(比如跳到应用的另一个页面)播放这段片段,或者跟着手势(比如滚动、滑动)逐帧拖动它。

Here’s a demo page showing off a scroll effect. I built it with a single prompt using GPT-5.6 Sol in Codex:

下面是一个展示滚动效果的 demo 页面,我在 Codex 里用 GPT-5.6 Sol 一条提示词做出来的:

Prompt:

提示词:

Build a demo page for a suitcase that uses a video model to create interactive transitions between a couple of screens. Each screen should show the suitcase in a different state, with vertical motion that feels appropriate for scrolling:

  • Initially, have the suitcase floating high up in the air
  • Then have it land on the floor and pop open
  • Finally, have its contents neatly land into it from the top

Generate the initial frame using your image generation skill. Then, generate a video clip that starts from that frame and animates to the next state. Use the final frame of that video to seed the next transition so that it continues seamlessly. Scrub through the transitions one by one as the user scrolls.

Use this fal.ai API key: sk-a1b2c3d4…

Use a video model with strong physics and consistency, like Seedance 2.5.

给一个行李箱做个 demo 页面,用视频模型在几屏之间做可交互的过渡。每一屏展示行李箱的一种状态,运动方向是竖直的,配合滚动的感觉:

  • 一开始,行李箱高高地飘在空中
  • 接着它落到地上,弹开
  • 最后,里面的东西从上方整整齐齐地落进去

用你的图像生成 skill 生成第一帧。然后生成一段从这一帧开始、动到下一个状态的视频片段。用这段视频的最后一帧作为下一段过渡的起点,让它无缝接上。用户滚动时,一段一段地拖动这些过渡。

用这个 fal.ai API key:sk-a1b2c3d4…

用一个物理效果和一致性都很强的视频模型,比如 Seedance 2.5。

GPT-5.6 Sol:

GPT-5.6 Sol:

随滚动播放过渡的行李箱 demo 页

The transitions between pages scrub fluidly with the user’s scrolling and are fun to play with. Design like this makes the user want to keep scrolling and reading more about your product. And it only took one prompt!

页面之间的过渡跟着用户的滚动流畅地走,玩起来很好玩。这样的设计会让用户想一直往下滚,多看看你的产品。而且只用了一条提示词!

Deliver: Polish your design into something users will love交付:把设计打磨成用户会爱上的东西 ​

Once we’ve gotten to a unique, standout design, the final step is to clean up the details and get it ready for production use. AI can build amazing, striking visuals, but your judgment will be key to making sure the design makes sense, flows well, and serves its practical purpose for your users.

拿到一个独特、出挑的设计之后,最后一步是把细节收拾干净,让它能上生产。AI 能做出惊艳、抓眼球的画面,但设计合不合理、顺不顺、能不能真正为用户所用,关键还得靠你的判断。

Technique 6: Cut out elements that don’t add value技巧 6:砍掉没有价值的元素 ​

AI loves to add more, but it rarely takes away. One of the biggest signs that a design is AI-generated is that it overexplains everything or contains elements that don’t serve any practical purpose. By contrast, a design that exercises restraint immediately looks premium and tasteful.

AI 爱加东西,很少减。一个设计是 AI 做的,最明显的标志之一就是什么都解释过头,或者塞着一些毫无实际用途的元素。反过来,一个懂得克制的设计,一眼看上去就高级、有品味。

When polishing AI designs, most of my effort goes into removing things. For example, when I was building my calorie tracking app, this was my initial design from Claude:

打磨 AI 设计时,我大部分力气都花在删东西上。比如做那个记卡路里的应用时,Claude 给我的初版设计是这样的:

Claude 给出的卡路里应用初版设计

I’d described the app’s functionality and specifically asked for a “clean, minimalist design.” The results weren’t bad, and were certainly impressive for being fully AI-generated. However, despite my asking for minimalism, a lot in the design wasn’t adding value:

我描述了应用的功能,还特意要了「干净、极简的设计」。结果不算差,作为完全由 AI 生成的东西,确实挺厉害。可尽管我要的是极简,设计里还是有很多东西没带来任何价值:

  • Pink glowy effects in the background and on the progress bar背景和进度条上粉色的发光效果
  • Random colors and highlights on text文字上随意的颜色和高亮
  • Extra labels and empty space when displaying all the foods for a day, when the images already communicate this展示一天吃的所有食物时多余的标签和空白,图片本身已经把信息说清楚了
  • Custom buttons and text fields that look worse than built-in iOS components自定义的按钮和输入框,比 iOS 自带组件还难看

I asked Claude to dial things back:

我让 Claude 收一收:

  • Simplify the layout into an image-centric grid把布局简化成以图片为中心的网格
  • Get rid of gradients, glows, and unnecessary containers去掉渐变、发光和不必要的容器
  • Aim for a truly minimalist aesthetic that feels Apple-native做到真正的极简,有 Apple 原生的感觉

This was the result:

结果是这样的:

收敛之后的卡路里应用设计

To my trained eye, the result is much better. It’s opinionated and allows the visuals to speak for themselves. It uses native iOS components, and the excessive colors and gradients are gone. The text is smaller, simpler, and tighter. This is good design.

以我练过的眼光看,这版好太多了。它有自己的主张,让画面自己说话。用的是 iOS 原生组件,过多的颜色和渐变都没了。文字更小、更简单、更紧凑。这才是好设计。

Today’s AI models would never think to make these choices on their own. Remember, AI doesn’t like to take risks, and it’s risky to strip down a design and delete code. The model needs a push from you. Look over your design and ask yourself what really needs to be there. Often, putting less on the screen communicates more, because you can hold your users’ attention without overwhelming them with clutter.

现在的 AI 模型自己永远想不到要这么选。记住,AI 不喜欢冒险,而把设计扒到只剩骨架、删掉代码,本身就有风险。模型需要你推一把。把你的设计从头看一遍,问问自己哪些东西是真正必须在那儿的。很多时候,屏幕上放得越少,传达的反而越多,因为用户的注意力被抓住了,又不会被一堆杂乱的东西淹没。

最后更新: