Built This Week/Builds

Builds

Ryz Labs Video Studio

Our team's AI video studio puts Veo 3, Seedance and other models behind one Replit app on pay-per-use API keys. The first version took a few days; a capybara ad took one day.

Ryz Labs Video Studio is the internal tool Jordan built so everyone on our team can generate AI video and images from the top models in one place. Pick a model (Veo 3, Veo 2, Seedance and others), enhance your prompt with AI, generate, and browse a shared history. Jordan built it in Replit on Replicate, and the first version shipped in a few days. It was the subject of our first episode, Episode 1 (June 27, 2025).

Why we built it

We make a lot of social video across several brands, and production time was the biggest bottleneck. When Veo 3 launched, our social team asked for access right away. Rather than buy per-seat subscriptions to many video tools, we did what we'd already done for chat: use API keys as a consumption layer so everyone gets every model and we pay only for what's generated.

Why not just give everyone Veo 3? Cost. At 06:42 Jordan explained:

It's about $6 per eight second generation, and the generations are not perfect.

— Jordan Metzner, Episode 1

A 30-second ad needs at least four clips, and with retakes you can spend $60 to $100 in minutes. Cheap next to a film shoot, but it adds up fast for a startup.

The stack

  • Replit: the app, plus its built-in database, storage, environment keys and one-click redeploy.
  • Replicate: API access to every model.
  • ChatGPT: improved the first prompt for Replit, and powers the in-app "enhance with AI" button.
  • Video models: Google Veo 3 and 2, Seedance and others, six in all.
  • Image models: FLUX Pro, MiniMax Image 01 and two more.

How we built it, step by step

  1. Sharpen the first prompt. Jordan most likely asked ChatGPT (probably GPT-4o) to improve something like: a web admin panel with a left-hand sidebar that uses different models to generate video.
  2. Build in Replit. At 20:04 he explained when he picks it:
But when I wanna build something that I need like an API and I need to make some some functional calls, I'll go to Replit.

— Jordan Metzner, Episode 1

  1. Add models from Replicate. For each model, copy the sample API call and JSON schema from its Replicate API page, paste them into the Replit agent and ask it to add the model. It usually works out of the box.
  2. Add "enhance with AI." ChatGPT rewrites your prompt for the chosen model. It was one of the first features added after launch, because prompt quality made such a difference.
  3. Add history and reuse. A video history page, download, share link and a copy-prompt button for testing the same prompt across models.
  4. Share one account. Everyone sees what colleagues generated.
  5. Add images. A newer page with the top four image models.

How it turned out

Within days of launch, we released a Wes Anderson-style Ryz Labs commercial that got good reviews. Then one of our video editors made a nature-documentary-style capybara ad for our offsite business, Offsiteio. Sam at 28:13:

Had it done within, you know, an hour maybe, stitched them together, layered on the voice over, presented it to us, few small edits, and within a day we had a commercial that previously would have cost, you know, thousands of dollars or just been too time consuming or costly for us to even pursue.

— Sam Nadler, Episode 1

He estimated the editor generated 20 to 30 clips. Not everything worked: some image models failed in the demo, and some images disappeared from storage.

What we'd do differently

  • Enhance prompts from day one.
  • Watch generation time. Price tracks it: Veo 3 could take ten minutes, Seedance moments.
  • Fan out. The roadmap: one prompt sent across many models and many runs, maybe 25 videos at once, limited only by cost.

See also Seedance vs Google Veo and our later Sora 2 API video generator.

FAQ

How do you build an AI video generator?

Jordan built a web app in Replit and added each model through Replicate: copy the model's API call sample and JSON schema from Replicate's API page, paste them into the Replit agent and ask it to add the model. He added AI prompt enhancement, a shared video history, download, copy-prompt and share links.

How much does Veo 3 cost per video?

On the show in June 2025, Jordan said Veo 3 was about $6 per eight-second generation and could take up to ten minutes. A 30-second ad with retakes could cost $60 to $100 in minutes.

Why build your own AI video tool instead of buying subscriptions?

Per-seat subscriptions to many video tools add up across a team. Using API keys as a consumption layer gives everyone every model and you pay only for what is generated.

Why use Replit for an internal AI tool?

Jordan goes to Replit when a tool needs a back end, database, storage, environment keys or a bit of auth. For front-end-only demos he uses tools like Bolt, Lovable or Cursor.