Stop overthinking your landing page design.
I just tested Gemini 3.0. The UI capabilities are frighteningly good.
I went from blank screen to a slick landing page in just few minutes. It's fully mobile responsive.
Full prompt below 👇🏻
Checkout this demo:
@Shpigford I think fable 5 medium is the sweet spot for me. I still let it use opus and sonnet through sub-agents when needed, which helps keep the overall cost under control.
@diegocabezas01 I think it's fancy version of /goal with everything like chrome extension, plugins, tests enabled. Good for those who have almost unlimited tokens.
@maubaron Great experiment, but why do people subscribe to your channel if the hooks are just clickbait?
Does this work for app marketing only or other niches too?
@rohanpaul_ai This is why "add a chatbot" feels too small.
agents don't want to click around your UI. They need clean actions like create invoice, approve refund, fetch status, attach file, etc.
A lot of saas will have to become agent-readable, not just user-friendly.
@neural_avb terminal output is such an underrated token leak.
agents really don't need 400 lines of npm noise to move to next step. This should be default behavior in coding agents.
@zarazhangrui exactly, don’t just end with "just say the word." finish the obvious next step, show the diff, and only interrupt me for decisions where you need more context or expert's opinion🙈
Stop asking AI to "design a clean UI"
That prompt is too vague.
Give it a design system, a product context, and a clear creative direction instead.
A simple workflow to design better interfaces faster:
1. Open getdesign. md
2. Pick a design doc/style direction you like
3. Give it to chatGPT (along with given prompt)
4. Generate a first UI concept
5. Iterate the image until the layout feels strong
6. Use claude Code or codex to convert the final image into code
The important part:
Don’t copy the design doc exactly.
Use it as a structure, taste reference, and quality then create your own version.
Here’s the prompt I use:
<prompt>
You are a world-class senior product designer, UI/UX designer, and front-end design systems architect.
I’m going to provide you with a design reference document.
Your job is to study its structure, visual principles, layout patterns, spacing, typography, hierarchy, interaction ideas, and overall design taste.
Then create a brand-new interface concept for the product/page I describe below.
Product/Page:
[Describe what you are designing]
Audience:
[Describe who will use it]
Goal:
[Describe the main business/user goal of this page]
Style direction:
Use the provided design document as inspiration only.
Do not use the exact same layout, copy, brand elements, names, illustrations, colors, or components.
Build an original design that feels premium, modern, minimal, polished, and production-ready.
Requirements:
- Strong visual hierarchy
- Clean spacing
- Premium typography
- Beautiful but usable layout
- Clear primary CTA
- Realistic SaaS/product UI
- No clutter
- No generic dashboard look
- Use realistic content, not lorem ipsum
- Make it feel like something a top-tier startup would actually ship
Output:
Create a detailed visual UI concept as an image.
Before generating the image, briefly explain:
1. The design direction
2. The layout structure
3. The key creative idea
4. Why this interface will feel high quality
Then generate the UI.
</prompt>
This turns design docs into a creative starting point.
It's all about adding taste into your workflow.
Stop asking AI to "design a clean UI"
That prompt is too vague.
Give it a design system, a product context, and a clear creative direction instead.
A simple workflow to design better interfaces faster:
1. Open getdesign. md
2. Pick a design doc/style direction you like
3. Give it to chatGPT (along with given prompt)
4. Generate a first UI concept
5. Iterate the image until the layout feels strong
6. Use claude Code or codex to convert the final image into code
The important part:
Don’t copy the design doc exactly.
Use it as a structure, taste reference, and quality then create your own version.
Here’s the prompt I use:
<prompt>
You are a world-class senior product designer, UI/UX designer, and front-end design systems architect.
I’m going to provide you with a design reference document.
Your job is to study its structure, visual principles, layout patterns, spacing, typography, hierarchy, interaction ideas, and overall design taste.
Then create a brand-new interface concept for the product/page I describe below.
Product/Page:
[Describe what you are designing]
Audience:
[Describe who will use it]
Goal:
[Describe the main business/user goal of this page]
Style direction:
Use the provided design document as inspiration only.
Do not use the exact same layout, copy, brand elements, names, illustrations, colors, or components.
Build an original design that feels premium, modern, minimal, polished, and production-ready.
Requirements:
- Strong visual hierarchy
- Clean spacing
- Premium typography
- Beautiful but usable layout
- Clear primary CTA
- Realistic SaaS/product UI
- No clutter
- No generic dashboard look
- Use realistic content, not lorem ipsum
- Make it feel like something a top-tier startup would actually ship
Output:
Create a detailed visual UI concept as an image.
Before generating the image, briefly explain:
1. The design direction
2. The layout structure
3. The key creative idea
4. Why this interface will feel high quality
Then generate the UI.
</prompt>
This turns design docs into a creative starting point.
It's all about adding taste into your workflow.
@koltregaskes If I'm right, if you start or resume a session after a while, you won't get cached input token pricing. Instead, everything will be counted as input/non-cached, so you'll pay more for all input tokens or could be cashing storage if TTL is longer.
@elvissun personally I have not used it too much but hearing people are using codex for task allocation and v4 pro for it's execution, it's quite cost efficient workflow.
@elvissun Okay understood. I got little confused cause many people are using codex along with deepseek v4 pro model from openRouter in their development workflows.