Skip to main content

Dry runs

A dry run goes through a blueprint's stages like a real run, but every AI model is replaced by Reelcraft's free test provider. Nothing is charged. Use it to check that a blueprint's stages are connected correctly before you pay for a real run. It has limits: see What a dry run can't test.

Start a dry run​

  1. Open the blueprint's canvas and select Run in the top bar.
  2. In the dialog, select Dry run. The title changes to Dry run. Fill in the inputs, as for a real run. Uploads work the same way.
  3. The Budget cap (USD) changes to $1. Keep it small.
  4. Select Start dry run.

Like a real run, a dry run needs a saved, runnable version.

What is different​

  • Text, image, video and speech stages, and the quality control judge, use the test provider. They return placeholder results that look right but aren't real: a plain picture, a short test clip, made-up text.
  • Human stages still pause for you: Human Input, Human Timeline Edit and approvals work exactly as in a real run, because you're the one acting.
  • Local stages run for real: Concatenate Video, Render Timeline and Export Subtitles do their actual job on the placeholder media, and cost nothing either way.
  • Analyze Media keeps its own model. With Probe, nothing is paid. With Transcribe align, it still calls Deepgram. See Analyze Media.

What a dry run can't test​

The placeholder results are very simple, so some things fail in a dry run that would work in a real one:

  • Structured data. The test provider can only return plain text. A Generate Text stage whose output is data fails its schema check ($: must be object) and the dry run stops there. This includes any blueprint that plans its scenes as data.
  • Media checks. A placeholder image is a single pixel, a placeholder speech file is silence with no length, and placeholder video is a short test clip. Checks that read a file's details, such as wpm, duration_range and media_format, fail on them, for example wpm: artifact has no usable probe duration.
  • Content checks and quality control. Checks and judging that look at the content behave differently on made-up content.

A dry run is best for checking that stages are connected and that the blueprint reaches its end, for blueprints built from text, image, video and speech stages. To try a stage that a dry run can't handle, use the canvas Run tab with a real model, or temporarily turn off the check that fails.

Finding dry runs afterwards​

Dry runs are hidden from the Runs list so they don't crowd out your real work. Turn on Show test runs to see them. They're marked with a dry run badge.

Dry runs on the canvas​

The Run tab on the canvas starts real runs. To try a blueprint for free, select Run in the top bar and choose Dry run.