Apply EDL

The Apply EDL node renders an edit decision list (EDL) into one finished video or audio file. An EDL is a small JSON description of an edit: the media it draws from, and an ordered list of segments that say which part of which source plays next. An Edit Plan node, an AI agent or you decide what the edit is, and Apply EDL turns that decision into picture and sound.

When you wire a transcript into the node, it also outputs that transcript remapped through the cut. Words in removed parts are dropped, and a word that crosses a cut is shortened to the kept part, so captions built from it stay in sync with the finished edit. Apply EDL runs no AI model.

When to use it

• Render a tightened version of a long recording, planned from its transcript.
• Assemble a set of clip ranges into one deliverable from a single edit description.
• Produce an audio-only cut, for example a podcast episode, on the same timeline as the video version.
• Output a transcript that matches the cut, for Add Captions.

Quick start

Inputs

Outputs

Settings

What an EDL contains

An EDL has two parts:

• Sources are the media the edit draws from. Each source has a URL, and it can have a role.
• Segments are the pieces of the edit, in playing order. Each segment names a source and a time window, in milliseconds, on the edit's master clock.

The joins between segments are hard cuts or crossfades. A video render needs a picture source on every segment. The MCP tools page describes the apply_edl tool, which takes the same EDL.

How the edit is rendered

• Segment order is the timeline. The node never sorts segments. An edit may go back to earlier parts of a source, for example to switch between cameras or to reorder clips.
• Cuts and crossfades. At a cut, two segments meet end to end. At a crossfade, they overlap, so the finished cut is shorter than the sum of the segments by the total crossfade time.
• Master audio. When a source has the role master-audio, every segment takes its sound from that source, so a camera switch never changes the sound. Otherwise each segment uses its own sound. A role must be master-audio, camera, wide or screen. An unknown role, for example a misspelled one, is refused.
• Checks before the render. The EDL is checked before any rendering starts. A missing or unreachable source, or a video edit with a segment that has no picture, is reported at once, and the message names the segment or source.
• Segments must fit their sources. After the sources are downloaded, each segment is checked against the real end of the track it reads. A segment that runs past its media fails the run at once, names the segment, the source and the track, and refunds the credits. A reach of up to 1 second past the end counts as rounding: the last frame holds and the sound continues as silence, so the cut keeps its exact length.
• Long edits have no gaps. A long edit is rendered in pieces that are joined frame-exact and sample-exact, so there is no gap or click at any join.

What Apply EDL refuses

Apply EDL renders one source per segment, full frame, with cuts or crossfades. It refuses the rest of what an EDL can describe, instead of dropping it silently:

• Layouts with several sources. A layout mode other than single, a layout with more than one slot, or a slot that names a different source than its segment is refused.
• Layout effects. A layout emphasis other than none, or a layout transition other than cut, such as pan or zoom, is refused.
• Crops. A region crop on a segment, a slot or a source is refused.
• Early segments. A segment that starts before the start of its source is refused.

The 180-minute limit

One render produces at most 180 minutes of output, measured on the finished cut after crossfades. A longer edit is refused before anything is charged, and the message names the edit's length and the limit. Split a longer programme into parts of at most 180 minutes and render each part on its own.

Long renders in a workflow

A final-quality render of a long episode can take hours. Inside a workflow, Apply EDL gets a time budget sized from its own edit, instead of the standard 90 minutes per node. The workflow's overall limit, normally 120 minutes, grows by the same amount. If the service restarts during a render, the resumed run reconnects to the render that is still in progress. Cancelling the run stops the render at its next piece.

Credits

Apply EDL costs 10 credits per started minute of rendered output, with a minimum of 1 minute. The length is measured on the finished cut, after crossfades shorten it.

The estimate before you run

The cost on the node, on the Run button and in the run confirmation is an estimate on the high side. You are always charged for the length of the cut that is actually rendered, computed from its EDL.

• The estimate is exact when the cut already exists: when the EDL is set on the node itself, or when you run only this node, which renders the plan already on the canvas.
• An Edit Plan that plans again in the same run is estimated from the recording, because the new cut does not exist yet. Tighten is priced at the full length of the episode. Clips is priced at twice the target clip length for each clip, times the number of clips. Without a target length, 90 seconds is assumed, and without a clip count, 8 clips. Nodes fed from the render with an Each connection, such as Add Captions, are counted once per clip.
• An EDL from any other node that runs again is priced at the 180-minute ceiling.

The balance check before a run compares your balance with this estimate. See Edit Plan for how the plan itself is estimated.

Tips

• Keep hard cuts for speech. Leave Default crossfade (ms) at 0 for talking-head and podcast edits. The EDL can still ask for a crossfade at any single join.
• Review with a proxy. Render with Proxy (720p review) while you check the edit, then switch to Final (full quality) for delivery.
• Use word-level transcripts for captions. Add Captions needs the transcript's words. Run Transcribe on ElevenLabs STT or Incredibly Fast Whisper, which return word timestamps. A Whisper transcript has phrases only, and Add Captions refuses it before the run, with nothing billed.