andupotorac

This is like "all it takes is one lottery ticket that wins".

When they say skill issue, they mean it. Have you spent at least 4-5 hours on specs (mermaid, markdown)?

It is following the same approach everyone takes when posting what they built, and sharing links. Or what is it doing differently?

"So thanks for nothing Hacker News." cut out the emojis :-)

Festivals? :)

This is a use-case, but the idea is anyone can now add "Take a photo" to their site, instead of letting users upload photos. Even without the AI gen part.

Added support for Playwright, so that the LLM can get the screenshot of the page the user wants to copy. While this seems like a small thing, what is required to build it, and what the difference in the output is are quite large.

Here are some examples of different tools, compared to Same.dev that also captures screenshots of pages - and all the others.

https://x.com/andupoto/status/1992928743925690382

I'm vibe-coding a vibe-coding platform.

I'm going to speak this update because that way I can have the text written more naturally. So, let's do that. So, since October last year, this means about six months, I've been using Cursor as a non-coder. Actually, I've been coding the last time, I guess, back in the days of Flash in ActionScript. So, I used Cursor for the past six months, maybe around 12 to 15 hours a day on average, you know, some days off, more hours in other days. And I've been working together with my co-founder, who's a developer, but he's also using AI, except he's using AI less than I do. So, we've been working on several things, of which one of them is closest to being released, maybe 80% done, which is, as I said, a vibe coding platform.

So this particular vibe coding platform will actually not compete with the likes of Replit, Bold, Loveable and about 40 other platforms that are basically the same: chat + IDE + preview area. What we're doing is actually different because we are rewriting a project that we built about 10 years ago. And I guess in a way similar to how Replit and Bolt had completely pivoted because of AI. The same thing is happening with this project of ours. So the project was actually about offering interactive widgets to people for their websites, but we had to build those manually. So now we have rewritten the platform from scratch using AI. And we offer zero widgets because with AI we are able to offer an infinity of widgets. That's exciting.

So this was supposed to stay a side project and frankly I'm pretty surprised that there's so much demand and the growth is amazing for all these AI code gen platforms. And if that happens to us, to this project, I don't know exactly what we are going to do because our plan is actually to tackle a bigger project that we are also working on. And the strategy here is to do what we call compounding startups. So instead of building a larger project with many features, we actually split that in several projects and we are doing them one at a time and giving each the ability to actually generate revenue. At some point we are likely going to raise capital. We had this plan before but with AI things have completely changed. We are able to do now in a week what five developers were doing in six months.

So the other projects are Talbo, Reclona and Tradul. Talbo is the what they compound into. But I am not ready to talk too much about them right now. :)

Here's a tip: do it for non devs (vibe coders).

I can check it out, feel free to send a link to @andupoto on Twitter

Feel free to hit me up with a link if it's in english @andupoto on Twitter.

Actually there are models for doing the layering.

1. https://github.com/SketchSeg/SketchSeg-Natural-Prior

2. https://github.com/IamCreateAI/LayerAnimate

3. https://github.com/hmrishavbandy/FlipSketch

Regardless if this or other paper is what you end up implementing, I wanted to say kudos for doing the right thing - having the right approach. Building stuff that's just now possible - in AI, from research papers - and productizing it, is how you get ahead of everyone else.

Maybe replace Pixave with a better product if you're at it.

What are you doing for privacy and security of both the models, and inferences? Also, do you train for body shape as well?

We used this in our project: https://github.com/CaptainCodeman/svelte-color-select

Would love to try it out. Is it for large code bases or what specifically?

I've been vibe coding for 6 months now. You can do anything a developer can with it. It's not just for docs. :)

Does this allow installing dependencies with nodejs for example? I know CodeSandbox did this a while back, so I imagine it's not something that browsers can just do.