Skip to content
 thumbrella.dev 

AI Platforms

Thumbrella is published as a model on several hosted AI compute platforms. If your team already provisions workloads through one of these marketplaces, you can add Thumbrella to the same flow without deploying anything yourself.

Replicate

Section titled “Replicate”
Terminal window
replicate run thumbrella/thumb \
  input=https://example.com/media.mp4

Use your existing Replicate account and API key. Replicate bills for the compute; Thumbrella does not add a separate charge in this mode.

Fal.ai

Section titled “Fal.ai”
import fal_client


result = fal_client.run(
    "thumbrella/thumb",
    arguments={"input": "https://example.com/media.jpg"},
)

When to choose this path

Section titled “When to choose this path”
  • Your project already uses Replicate, Fal.ai, or a similar platform for other workloads.
  • You prefer marketplace-based provisioning and billing over operating a binary or calling an external API.
  • You want Thumbrella to colocate with other compute-heavy tasks (e.g., diffusion models, video encoding).

Limitations

Section titled “Limitations”
  • Platform billing applies — check Replicate/Fal pricing for your expected volume.
  • Cold-start latency is higher than the hosted service or a running local binary.
  • Not a good fit for tight latency budgets or very high request rates.

Next steps

Section titled “Next steps”