How a run works
A run is one execution of a template. This page explains what happens between clicking Run and getting an export back, so the run view makes sense while you watch it.
Templates, steps, and runs
- A template is your recipe: an ordered list of steps plus project settings (resolution, frame rate, timeline levels).
- A step runs one processor, the unit of work. Each processor does one job: transcribe audio, choose cuts, place b-roll, render captions.
- A run executes those steps in order against your uploaded files.
Every run belongs to a run group, which is where reusable work is kept. See Run groups and variants.
What happens during a run
- Your template is checked. File references, step references, and required parameters are validated. If a processor needs a service key you have not set, the run stops before doing any work and asks you for it.
- Steps run in order. Each step reads its inputs (uploaded files, or the output of an earlier step), does its work, and passes its result on. The run view shows each step as pending, running, done, reused, or failed.
- Steps you marked for review pause the run. You approve, edit, reject, or regenerate before the run continues. See Review gates.
- Results are assembled into a timeline. Every clip a step placed is collected onto its timeline level, and adjustments such as reframing or background removal are applied to the clips they target.
- The edit is finished and made available for export. You can preview it, nudge clip timing, and export.
What a step can produce
Not every step puts clips on the timeline. Steps fall into a few kinds, and the processor reference says which kind each one is:
| A step can produce | What that means | Examples |
|---|---|---|
| Clips | Places media on a timeline level | broll_place, cut, title_card |
| Analysis | Data for later steps, such as a transcript or footage notes | generate_transcript, analyze_footage |
| A script | Narration text for later steps to speak or display | write_script |
| A plan | Which kind of content belongs in which time range | video_director |
| Adjustments | Changes to clips other steps already placed | reframe, speaker_center |
| Replacements | Swaps a clip's media for a processed version | chroma_key, noise_reduce |
This is why order matters: a step that adjusts clips has to come after the step that placed them, and a step that needs a transcript has to come after the step that made one.
Wiring steps together
Steps are connected by named inputs. An input is fed by either a file you uploaded or an earlier step's output:
- id: ai_cut
processor: ai_agent
input:
- id: transcript
step: transcribe # the output of the "transcribe" step
- id: speaker
file: speaker-footage # a file from your workspace
Only steps earlier in the pipeline can be referenced. Each processor's page lists the inputs it expects.
Exports
When a run finishes you can export:
- FCPXML, a Final Cut Pro bundle you can open and finish in FCP
- SRT subtitles
- MP4, a rendered video produced on demand
Next
- Timeline and levels, how tracks stack and combine
- Run groups and variants, where reusable work lives
- Reusing work between runs, how to avoid paying twice
- Review gates, pausing a run for your approval