Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Launch HN: Magic Patterns (YC W23) – AI Design and Prototyping for Product Teams

news.ycombinator.com

Post reply on HN
Print view View original

51–60 of 118 posts

Re: Launch HN: Magic Patterns (YC W23) – AI Design and Prototyping for Product Teams

#51

Post by jonplackett »

alexdanilowicz wrote: post #37
jonplackett wrote: post #33

Earlier quoted context omitted.

Do you not think that’ll get solved by a future generation of cleverer LLMs though? As someone pointed out in another comment, they get better results with Gemini 2.5 already. People already seem quite annoyed with Cursor based on that thread the other day with the hallucinated customer support. Interested in anyone’s opinion

> Do you not think that’ll get solved by a future generation of cleverer LLMs though? I don't. But obviously I'm biased + have spent perhaps too much time at application layer. I think there will still be a large amount of tooling + feeding of context to get the best result and I don't see a world in which we let LLMs run hog wild on our computers any time soon, especially for prototyping workflows at the enterprise…

Cool. Thanks for the insight. Good luck with it all! It seems like lots of people DO see the value in it!

Re: Launch HN: Magic Patterns (YC W23) – AI Design and Prototyping for Product Teams

#53

Post by teddarific »

airstrike wrote: post #43
teddarific wrote: post #35

Earlier quoted context omitted.

Awesome to hear — our goal is to get to a point where you can copy/paste the design, but we totally recognize that it doesn't always get you to a full high fidelity mock up with the current limitations. So the main use case we've seen a lot of success with is ideating and validating different branches, quickly. Oh that sounds very neat — definitely similar in nature! Documents and UI are both complex and can require…

FWIW I think it could get a pretty high fidelity mockup if I had spent more time with it! It was remarkably good for a single one-shot prompt which probably had 15 words or less. I'll definitely come back for more when the time is right.

Definitely — we've seen some pretty crazy high fidelity mock ups from our customers who spend hours and hours prompting away, and they genuinely blow my mind. it's honestly why we spend so much time working on the application layer (e.g. making it easier to feed context into your prompt, referencing other designs, etc) since we know that the LLM is capable of these things, it just takes much more time than most people are willing to put in.

would love to hear what you think when you give it another try!

Re: Launch HN: Magic Patterns (YC W23) – AI Design and Prototyping for Product Teams

#54

Post by alexdanilowicz »

mildlyhostileux wrote: post #46

I gave it a shot to iterate on our current UI. It did try to solve the problem we were focused on, but it also randomly removed other features that we hadn’t touched or even talked about. This makes me believe there will be a lot of back and forth that is low value just correcting small details. In UX/UI details really matter. I'd probably opt to just move pixels around my self in this case. If this were a brand new…

Our largest customers generally use us to prototype brand new UIs versus existing ones, so you're spot on.

We're certainly better at new UIs versus existing UIs. Existing UIs are tough because you need to first recreate what you have in the codebase first. This is a large reason why we built a Figma-like canvas (https://www.magicpatterns.com/docs/documentation/projects/ge...), so that you can reference existing designs to make new ones.

So, for example, you could make your navbar, button, dashboard, and then click on that and reference it to make another design.

P.S. if you're open to it, open a support chat or DM me? Would love to help you get more value and see what prompts are causing it to remove stuff. It can also be really frustrating when it removes features — what can be helpful is going back a version or also editing the code by hand. We've had code edit since day 1. Totally aligned everything needs to be a prompt to get the small details right.

Re: Launch HN: Magic Patterns (YC W23) – AI Design and Prototyping for Product Teams

#55

Post by arturmakly »

congrats on the launch! I gave it a spin.. found these issues:

1 - after selecting the Body of this page to capture as Design: https://app.visualsitemaps.com/pricing the "Render" tab > result showed be a blank box: https://share.cleanshot.com/jVGlwYND yet there was code in the "Code" tab.

after that it attempted to recreate the design, with some new additions:

- add a row of logos ( failed ) - add testimonials - add case studies - remove a row

Results >> it was 92% there: https://project-tailwind-conversion-with-lucide-icons-756.ma... had some missing images from the OG design.

Overall this is impressive for MVP.. I also like the manual click-to-select-objects for more refinement.

I was unable to find the CSS styling code however ( sorry not a React/Tailwind user ) it just showed me index.css like so: {@import 'tailwindcss/base'; @import 'tailwindcss/components'; @import 'tailwindcss/utilities';}

Re: Launch HN: Magic Patterns (YC W23) – AI Design and Prototyping for Product Teams

#56

Post by JofArnold »

I used magic patterns for a couple of months and it was one of the first no brainer AI services I've paid for outside of the main LLMs and IDEs. It did such an amazing job on quite an esoteric frontend that's very much not your normal web app. Impressive. Next time I need to design and build some more frontend code I'll be subscribing again.

Edit: to add some meat to that comment what surprised me was just how much better it was than Anthropic and OpenAI tools at that time for coming up with great looking products with minimal prompting. I also fed it other designs for inspiration and it replicated them brilliantly while incorporating my requirements. Good stuff.

Re: Launch HN: Magic Patterns (YC W23) – AI Design and Prototyping for Product Teams

#57

Post by alexdanilowicz »

arturmakly wrote: post #55

congrats on the launch! I gave it a spin.. found these issues: 1 - after selecting the Body of this page to capture as Design: https://app.visualsitemaps.com/pricing the "Render" tab > result showed be a blank box: https://share.cleanshot.com/jVGlwYND yet there was code in the "Code" tab. after that it attempted to recreate the design, with some new additions: - add a row of logos ( failed ) - add testimonials - add…

If it's helpful, the styling will be in the "jsx" file vs the css file! So when you use that "manual click-to-select-objects" feature in the chat bar, you can click on "Highlight Code" and then that will take you to the React component. In that component, you'll see a className field which will have Tailwind classes in it. That's technically what is applying the styling and boiling down to raw CSS.

Ah! Need to clean up what's happening there when using the Chrome extension to import an existing design - pushing up a fix. There's a great discussion on the extension here (https://news.ycombinator.com/item?id=42043552) btw if you're curious how it works.

When it comes to adding images, it works best if you give it an existing image URL or upload an image, which we then host for you. In this case, Sonnet 3.5 appears to have hallucinated an invalid image link. Also need to make this more robust!

Re: Launch HN: Magic Patterns (YC W23) – AI Design and Prototyping for Product Teams

#58

Post by alexdanilowicz »

JofArnold wrote: post #56

I used magic patterns for a couple of months and it was one of the first no brainer AI services I've paid for outside of the main LLMs and IDEs. It did such an amazing job on quite an esoteric frontend that's very much not your normal web app. Impressive. Next time I need to design and build some more frontend code I'll be subscribing again. Edit: to add some meat to that comment what surprised me was just how much b…

Good to hear from you and see you on here!

Re: Launch HN: Magic Patterns (YC W23) – AI Design and Prototyping for Product Teams

#59

Post by zalzal »

alexdanilowicz wrote: post #21
pelagicdev wrote: post #19

Why would you limit the tool to strictly be for React? "As per my limitations, I am designed to work specifically with React and TypeScript/JavaScript only. I cannot provide direct conversions to plain HTML/CSS or other frameworks."

Mainly to narrow the problem space: there's a lot of logic in our backend that we call "post processing" which is cleaning up the code after the AI generated it to fix hallucinations. For example, the AI often gets import statements wrong or misspells icon names. It's pretty interesting because every model seems to introduce a new set of hallucinations, so this is a problem that requires a lot of maintenance. We have…

I'm curious, if you have that philosophy (which makes a lot of sense), you must have considered building is a sort of more abstract (but extensible) UI toolkit language and library and you could code in that then compile it down to React? Or have you found the benefit of large LLMs already having detailed React training is just too high?

Re: Launch HN: Magic Patterns (YC W23) – AI Design and Prototyping for Product Teams

#60

Post by alexdanilowicz »

zalzal wrote: post #59
alexdanilowicz wrote: post #21

Earlier quoted context omitted.

Mainly to narrow the problem space: there's a lot of logic in our backend that we call "post processing" which is cleaning up the code after the AI generated it to fix hallucinations. For example, the AI often gets import statements wrong or misspells icon names. It's pretty interesting because every model seems to introduce a new set of hallucinations, so this is a problem that requires a lot of maintenance. We have…

I'm curious, if you have that philosophy (which makes a lot of sense), you must have considered building is a sort of more abstract (but extensible) UI toolkit language and library and you could code in that then compile it down to React? Or have you found the benefit of large LLMs already having detailed React training is just too high?

> have you found the benefit of large LLMs already having detailed React training is just too high?

This.

We've tried a lot. We have been around since Oct 2023. First tried fine-tuning, but it's very hard to teach an LLM something new.

At one point, our product lived in a Figma plugin (https://x.com/Teddarific/status/1729153723728011618). To do this, we had the LLM output JSON, and then converted that to Figma nodes. This is sort of what you're saying. But the big issue was it would hallucinate many things and was really only good at the examples we fed it.

Post reply on HN