Short answer: Jordan Metzner picks a model such as Seedance, rewrites the prompt with an enhance-with-AI button powered by ChatGPT, and notes Veo 3 can take ten minutes for eight seconds while Seedance takes moments.
Watch this answer from Episode 1 (starts at 8:37) · Episode 1 notes and transcript
From Episode 1: How We Built an AI Video Studio in 3 Days, starting at 8:37.
Jordan Metzner [8:36]: Okay. Cool. So, yeah, this is the tool. We call it Ryz Labs's video studio. We just added images, so maybe now we need a new name for it. What I've done is I've gone and find kind of the top best video generation models on the market today. In the case, that's these, six right here. They're all a little bit different. They're all specialized in kind of different output qualities, and kind of the only way to really know which one's the best for your use case is to just generate a bunch of samples and see which one works the best. Some of the pricing I've included here as well. So, they're all over the place. This is, 30¢ per second. We know this, Veo three is six dollars per eight seconds. This one has a token quantization pricing. So they're they're all a little bit different, but, essentially, the way the tool works is you pick a model. In this case, we can pick something like Seedance. We have all the configurations here, so you can start with a reference image. You can change your frame rate. For now, let's just leave kind of all of this just as it is. And you can start with, like, a dog skateboarding
Sam Nadler [9:46]: In Venice Beach.
Jordan Metzner [9:51]: Oh, cool. Well, why don't we go to Venice, Italy?
Sam Nadler [9:54]: Oh, nice.
Jordan Metzner [9:55]: Okay. Cool. So, dog skateboarding in Venice, Italy. And then I built this, like, enhance with AI, which basically, I highly recommend using something like a ChatGPT to enhance your prompts. Inside the tool, I just built that. But basically, what you do is you take this and you kinda enhance it with ChatGPT by telling the model that you're working with. You tell ChatGPT the model that you're working with, and you can get kind of a more enhanced prompt and, here it is, let's capture the dog's joyful presentation, as he's on the iconic Venetian architecture and gondolas. So anyway, wrote a really nice prompt here.
Sam Nadler [10:32]: But wait really quick, the enhance with AI was one of those kind of second features we built after initially using the tool and seeing kind of the video quality and then, understanding that obviously the prompt to is so important in creating the types of videos you want. So that was I believe one of our first features that we built, right as soon as we launched the initial product.
Jordan Metzner [10:56]: Yeah. I mean, I wrote like have a dog skateboarding in Venice, Italy and, here it says a golden retriever skillfully skateboarding through the vibrant streets of Venice, Italy, capture the dog's joyful expression under the warm afternoon sunlight, you know. So it's just so much more robust in helping the model generate a high quality output versus, something as simple as the initial prompt. Both will work, but quality of output will definitely have an impact. Another thing to note is all these models kind of have different times to generate. So, the price for the most part is really relative to how long it takes to actually make the video. In the case of something like Veo three, sometimes takes up to ten minutes to generate like an eight second video and hence, why the price is so expensive. In the case of Seedance, which is, ByteDance, their newest model, might maybe take just a few moments, to make it. So what I do here is after I generate it'll say it's in progress, know, and then, we built a small, video history page here and here you can see, the video is being generating. So I can just keep hitting refresh.
Jordan Metzner [12:06]: It's it's processing and hopefully here we go. Awesome. Okay. Cool. So, if you see here, there's the dog skateboarding in Venice, Italy. Cool. And, I can make this mini bigger. I don't know if you guys can see it bigger.
Sam Nadler [12:24]: Yep. Cool.
Jordan Metzner [12:29]: Alright. I mean, pretty good. So, what else can I do with that? So obviously, like, I can download that. I have a copy button here in case I wanna use that prompt again. And then I also have like a share link as well, so if I wanna share that with someone or I could I can throw it in the trash. So, if we just use this as an example here, I'll just take the same prompt, I can go back to my model, I'll pick something here like veo two which is, Google's other model, the previous model to veo three, I'll paste the same prompt in and we can generate a video. And yeah, it'll take a few moments but we'll see the quality of the veo two video versus the veo three video and there you can kind of compare and say like, is it worth it? I would do a veo three but it's gonna take like ten minutes right now so it's probably just not best to do it, while we're doing it here.
Sam Nadler [13:16]: And that copy prompt feature is also one of those like quick iterations we made after we started using the project product and we wanted to, test the prompts on different platforms to see which one worked best or we wanted to make a small tweak because it didn't generate exactly what we want. And instead of trying to like recreate the prompt or you know, save it somewhere else, we created this copy prompt feature which has been really really useful.
Jordan Metzner [13:43]: Yeah. I think I mean, it's easy to see where our road map is going where, you write one prompt and then we can generate multiple videos across multiple models. So we can go horizontally, so, produce the same video across all different models at the same time, as well as go vertically across each model. So make multiple generations using the same model with the same prompt just to get different outputs as the model will not always produce the same output with the same prompt. And so you can get to a point where, with one single prompt, we'll generate maybe 25 videos or something like that. And so your ability to bring all these assets together, is super feasible. And again, it's just simply a limitation of cost. So, here's using Google Veo 2. Here's the dog skateboarding in Venice. This one looks, in some cases a little bit more realistic in the sense that it looks a little bit more like Venice, golden retriever.
Jordan Metzner [14:38]: But you can see the video quality a little low on the pixels, etcetera. Cool. So so that's how it works.
Sam Nadler [14:47]: What a great tool. And I wanna get into like how our team has specifically used this tool in the past few days. I mean, we've quickly made, two social campaigns and TV campaigns that I think are worth talking about. But before we do that, I want you to walk me through kind of the major building blocks of getting this out the door in just a few days. What tools did you use? What were some of the roadblocks you encountered? And yeah.
Jordan Metzner [15:16]: Okay. Yeah. Happy to do that. I think one thing I should show you, before we do that, right now would be a great time to show kind of the capybara commercial.
Narrator [15:25]: Capybaras are the most unique and relaxed creatures in the animal kingdom, yet the trials of modern life sometimes push their limits. But nature, in its infinite wisdom, understands that even capybaras require a chance to get off-site and restore their tranquil harmony. Offsiteio handles the stress of planning great company retreats, saving teams money. We handle the details, you just chill. Offsiteio, your off-site event planner.
Jordan Metzner [15:56]: I think that's pretty funny. We also have, obviously, the Wes Anderson one we did for Ryz Labs.
Sam Nadler [16:00]: This is Kyle, our DevOps guy. He lives in a tree now. Something about latency and forest spirits.
Jordan Metzner [16:07]: This is Jess, our designer. She's never used Figma, but she does love clip art.
