Short answer: Jordan Metzner says he built it in Replit, calling Nano Banana and Veo 3.1 Fast through replicate.com. Sam Nadler says the team already uses Veo 3 for b-roll to supplement original content.
Watch this answer from Episode 18 (starts at 8:19) · Episode 18 notes and transcript
From Episode 18: TrickOrTreat!, starting at 8:19.
Sam Nadler [8:18]: Yeah. Couple thoughts. Number one, it's been really fun for our team to use. A teammate this morning told me, based on one of the videos, she it actually inspired what she was going to be for Halloween. So that's kind of fun. But, number two, I think what could be interesting here is just, like, quickly how you built this. I believe you built it using Replit, but you were telling me, yesterday that there's this, I don't know how new it is, but this planning feature on Replit that makes, builds that much easier and robust before you actually start the building process. Because I'm not sure if you used it on this exact build, but it could be fun to talk about that. And then lastly, professional applications. I mean, even in our own you know, this is more of a fun, like, Halloween themed angle, but we're already using Veo three for b roll and to supplement our actual, original content. And I think this is just another theoretical way you could take an image, you can enhance it, and then you can generate a video for it. We are gonna cover a tool that we use to publish, social media content across multiple channels, but the, theoretical acceleration of just a generic webcam video into video content that could later be published, I mean, it's so much faster with these tools. So those are the high level kind of what I'm immediately thinking.
Jordan Metzner [9:45]: Yeah. So just to kind of run through them real quick, I built this in Replit. The APIs that I'm using are Nano Banana from Google, as well as Veo 3.1 fast. I'm serving them up actually through replicate.com, as I have an account there, and I have my API keys, like, stored pretty easily already in Replit. And it's not a very complicated application. So, you simply turn on the webcam. It captures an image. We take that image. We send it to the Nano Banana API and get the new corresponding image. And then we use that image as the source image to generate the video with a little prompt about making a Halloween video. So it's been super fun to play with. You can see that it's not a perfect replica of me. It's not a perfect replica of kind of the output. A lot of the videos will say like, happy Halloween, like, blah, la. And it's not like real words, like you can kinda see the like, mumbled mumble, I guess, you would say, like mumbling of the characters.
Jordan Metzner [10:53]: And it kinda reminds me of, like, how we used to see a lot of the lettering and, like, image generation with AI kind of, like, not be real words, and most of that's kind of gotten solved to this point. So I think we'll see, like, consistent improvements. We've talked about on previous episodes how, NVIDIA is working on a video exclusive chip that should come out in 2026. And so you can just imagine, like, how much more powerful that could be to these types of video generation models. And as you know, Sora two also, could be used to do this as well. I simply went on the Google stack because I saw, the example, and I just thought it was like a fun way to build. So, yeah, I think to your point, it's definitely a cool creative way for people to get costume ideas and even just make some of their own fun social media content. And, we'll definitely try to include some of the best outcomes here. But
