Y Combinator Startup Podcast
00:22:20 · Marvin Schwaibold
[00:22:20] S1: So I fed the video into into into Codex for the first time and just explained it my... The the the project and the process that I was... I wanted you to go through. Take the video, turn this into a shader. I want the shader running running in the web, and I want you to be as as as meticulous as you can, be very pixel perfect, grab all the colors, make sure that you're using the same hex codes, understand each part of the image, and then try to recreate that. And it went off and said, no problem. I got you, and it came out with this, which obviously doesn't look at all where we landed... Where we ended up now. Yeah. But then the process from here was I would screenshot this and my original my original video. Mhmm. I would open up ChatGPT, and I gave it to another agent there. I explained that agent exactly what I was doing as well. I said, why is the other agent not understanding what I'm trying to get to? And that ChatGPT then helped me kind of, like, understand my own process. Like, well, you have to explain to it where the light is coming from and what what shadows you're using, and maybe break the image actually down into, like, six parts and have it do each one of those individually and put that back together. Mhmm. And as I was kind of thinking about all of these different things, like, I was trying to come up with how could I become very clear with the... With with AI and explain to it exactly what it was actually looking at. So one of the tricks I used was I went into ChatGPT, and I had it render basically a topographical map of the underlying structure. Mhmm. AI, as I was going through this, I was... It was kinda getting closer, you could see that, you know, it was painting more pixels into into the grass blades here, and these started to look like bushels a little bit, which which felt more realistic. And then over here, we... I was like, okay. We're getting somewhere here now. This doesn't look this doesn't look terrible anymore. But I kind of figured out it doesn't understand that there's... It's not a flat layer underneath. These are little hills. Right. Basically, I fed this back into into AI, and and and it understood, oh, okay. So there's there's a... An underlying structure in a map that has different depths and heights. Mhmm. And if I can rerender the grass on top of those...
00:24:12 · Marvin Schwaibold, Aaron Epstein
On top of that infrastructure or that structure in general, it'll be much easier for for me to to create the scene, and that kinda worked really well. And, you know, in this in this prototype, your friends, you can actually bind... There's a binary toggle to turn the grass on or off, and you can actually play with the terrain quite a bit here as well. [00:24:29] S0: Right.
00:24:29 · Marvin Schwaibold
[00:24:29] S1: This was all unlocked via Josh to me at a lunch the other day where he kind of, like... We installed DialKit into this demo Mhmm. Because he looked at this and he's like, this looks pretty good, but you should look at the length of the grass blades. I was like, woah. Okay. That's true. So I asked I asked Codex, hey, how many how many grass blade variations do we have? And it said, we have about 217,000 different grass blades, but they only differ by, like, point six, point seven, point eight. So we were like, okay. Let's just build a toggle for it so that we can see it ourselves. Yeah. I don't wanna have to go back and forth with the interface and and chat about this. Expose the underlying structure. Exposing the underlying structure would also show me where AI had made mistakes. The original version of the terrain map that it had implemented was completely wrong and and and skewed. Mhmm. So I couldn't get to my outcome anyway because AI had misinterpreted something five prompts ago that I didn't know about anyway. Dialkit kind of unpacked that for me. But then, in general, you can just mess with a few of these values, and they kind of help, like, bring bring some of this... Like, they they they help visualize Mhmm. What what I what I really wanted to see. And if if if, you know, the the length here is longer, then you you can you can introduce more wind. This was something I wanted to to test and see how how crazy I could get with the Yeah. With the wind movement. You can see it it gets a little wild. But, yeah, I think this is just a small example of a little bit of, like, a a fun project just to try and, like, see how far can you get with with the tools. And these are things that I think before AI, I I I couldn't do. I just... I I might have had these ideas, but I but but I but I wasn't allowed or or able to execute on them. And I think that's just in general what's changing with a lot of the... With with the design role right now. Any... Anyway, like, it's it's it's... Designers are becoming more full stack as are engineers and PMs. It goes it goes it goes in every which way.
00:26:13 · Aaron Epstein, Marvin Schwaibold
[00:26:13] S0: Yeah. There's no limitations now. Right? And, actually, this is one of the one of the common patterns that we've been seeing a lot, creating prototypes and then creating, little widgets that allow you to control different variables in them to to be able to test it really easily and get it right. Yeah. And previously, it wouldn't have been worth anybody's time [00:26:32] S1: Yeah. [00:26:32] S0: To do that. You would just, you know, maybe play with variables and code or something until you got it, and then you'd call it good. And that would take way longer, or you wouldn't even spend the time doing it. Yeah. Now it's so easy to create new little software tools to be able to control these variables and test things and dial it in perfectly Yeah. Which I guess is where the name DialKit comes from. Yeah. So, yeah, maybe tell us more about DialKit here. So this is a product that makes it really easy to create and then tweak a lot of these variables in whatever project you're working on. Right?
00:27:01 · Marvin Schwaibold, Aaron Epstein
[00:27:01] S1: Exactly. So so yeah. That's that's that's basically exactly how it works. You can just run NPM run via via his website. There... He has a little prompt for it. Mhmm. You can just plug it into to whatever you're working on. I I have another project that I can quickly show. This is just another version of DialKit Mhmm. In this specific in the specific project where I was manipulating a a carousel. Mhmm. This is inspired by Mike Matas. He's a design lead at at Love From, and and his portfolio carousel works in this way, and I wanted to see, like, there's so much nuance in the way that carousel works and the way you can drag down an artifact. And when you let go of that artifact, the way... Basically, almost like energy ripples out from that from that specific artifact and triggers a collision effect with the other artifacts left and right. Building out this... The side panel here right now helped me basically look under the hood Mhmm. Of of my own... Of of of the own Frankenstein that I created here [00:27:59] S0: Yeah.
00:27:59 · Aaron Epstein, Marvin Schwaibold
[00:27:59] S0: Yeah. [00:27:59] S1: With 91 prompts, and it just helps you dial in some of these some of these different some of these different parameters that are really important. Like, pull pull boost means the faster I drag this down, the stronger the stronger the boost is. You wanna play around with that toggle a little bit. It can't feel too intimidating or or too or too strong. Otherwise, the otherwise, the whole system just feels off. And I know when it feels off, but the back and forth with chat just doesn't work. Yeah. You need to be able to to to always expose those kind of those kind of values. [00:28:27] S0: It seems like a lot of... All the projects that we've looked at here and a lot of what you're describing is just having... Setting an unreasonably high bar, and now we have the tools to actually be able to meet it much faster than it would have taken before. Yes. And so what is the role of the designer? And and maybe even nondesigners, how should they be thinking about design? You know, a lot of the audience that watches this are, maybe founders of early startups that that maybe feel like they're engineers but don't have a design bone in their body Yeah. Or, designers that are maybe trying to figure out, like, could I start a company, or could I actually build the product? Yeah. I'm so used to, you know, designing a Figma and passing it off to somebody maybe where where you guys have been in the past. How should they think about themselves and being able to uplevel the way that they use AI to create higher quality design?