Built This Week/AI ToolsAI Tools
Seedance
Seedance was the fast option in our internal AI video studio: ByteDance's model made a clip in moments, while Veo 3 could take up to ten minutes and cost about $6 per eight seconds.
By Jordan Metzner & Sam Nadler · From our episodes · Updated October 2026
Our verdict on Seedance
Seedance earned its place in our video studio on speed. When Jordan generated a clip live in June 2025, it came back in moments. Google Veo 3, which we then considered the best model, could take up to ten minutes for eight seconds.
We only showed one Seedance generation on air, and we never did a quality shootout with it. So this is a first-hand look, not an exhaustive review. Jordan's own rule applies: the only way to know which model suits your use case is to generate a bunch of samples.
What we built with Seedance
Ryz Labs Video Studio (Episode 1): an internal web app where our team picks from top AI video models, including Veo 3, Veo 2 and Seedance, writes or enhances a prompt, generates a clip, and browses a shared history with download, copy-prompt and share links.
- Why: video production was our social team's biggest bottleneck, and everyone wanted Veo 3. Instead of buying subscriptions per person, the studio uses API keys so everyone pays per use.
- How long: the first version shipped in a few days, built in Replit with models from Replicate.
- The Seedance demo: Sam suggested a dog skateboarding in Venice Beach, Jordan changed it to Venice, Italy, ran it through the "enhance with AI" button, and picked Seedance. A golden retriever skateboarding past Venetian architecture came back within moments.
How we use Seedance
- Add it through an API marketplace. On Seedance's Replicate page, Jordan copied the sample API call and the JSON schema, pasted them into the Replit agent, and the model worked "pretty much right out of the box."
- Enhance the prompt first. "Dog skateboarding in Venice, Italy" became a detailed scene with a golden retriever, warm afternoon sunlight and the dog's joyful expression. Enhance with AI was one of the first features we added, because prompt quality changes the output.
- Tell the enhancer which model you're using. Jordan's tip: give ChatGPT the target model so it writes the prompt for that model.
- Use the configuration options when you need them. The studio exposes Seedance settings such as a reference image to start from and frame rate.
- Use the fast model to explore. Since Seedance generates in moments, it's the one to try ideas on. Save the slow, expensive model for shots you know you want.
- Copy the prompt and compare models. We added a copy-prompt button so the same prompt can run on Veo 2, Veo 3 or Seedance and you can see which wins.
It's about $6 per eight second generation, and the generations are not perfect.
— Jordan Metzner, on Veo 3, Episode 1
That cost is why a fast, cheaper option mattered. Jordan said video model pricing mostly tracks generation time, and a thirty-second Veo 3 commercial could easily mean ten clips and $60 to $100 in a few minutes.
Where Seedance falls short
- Thin evidence. One clip, generated on air. We never published a side-by-side of Seedance quality against Veo.
- Not the quality leader at the time. In June 2025 we both treated Veo 3 as the best video model available. Seedance was the quick one.
- Studio bugs, not Seedance bugs. Some image models in the same studio failed and some images disappeared from storage. That was our build, not the model.
What the studio produced
Within days of launch, one of our video editors generated 20 to 30 clips in about an hour and had a capybara commercial for our offsite business ready within a day. We also released a Wes Anderson-style Ryz Labs commercial that got good reviews. We didn't say which model made which clip.
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
— Sam Nadler, Episode 1
Seedance compared
- Seedance vs Google Veo: Veo 3 was the quality pick but slow and costly; Seedance took just a few moments.
- Sora vs Google Veo: how we saw the other big video models.
Episodes featuring Seedance
- Episode 1: the Ryz Labs Video Studio and the live Seedance generation.
FAQ
What is Seedance?
ByteDance's AI video model. In June 2025 Jordan described it as ByteDance's newest model and added it to the Ryz Labs Video Studio, our internal tool for generating video with several models.
How fast is Seedance?
In our live demo it took just a few moments to generate a short clip. Jordan contrasted that with Veo 3, which sometimes took up to ten minutes for an eight-second video.
Is Seedance better than Veo 3?
Faster, not necessarily better. In June 2025 we considered Veo 3 the best video model, but it was slow and expensive. Jordan's advice was to generate samples across models and see which works for your use case.
How do you use Seedance through an API?
We call it through Replicate. Jordan copied the sample API call and JSON schema from Replicate's API page and asked the Replit agent to add the model to our studio.