Split into Chunks

The Split into Chunks node cuts a long video or audio file into consecutive chunks. Every chunk is between a minimum and a maximum length that you set, and no chunk overlaps the next. With Cut by: Shots, the node instead cuts a video into one part per shot. Connect an output in Each mode, and the next node runs once for every chunk, so you can transcribe, dub or caption a long recording piece by piece.

When to use it

• Split a long podcast or interview into segments before you transcribe them with Transcribe.
• Cut a long video into short clips for social posts, and process every clip with the same chain of nodes.
• Cut an ad or a music video at its shots, then restyle or upscale each shot on an Each connection.
• Break a recording into parts for Dubbing or Voice Changer, then join the results again with Combine Videos.

Quick start

Inputs

Connect at least one input. With both inputs connected, the node splits each one on its own. A file can be up to 3 hours long. A longer file fails before any work is done, so trim it first.

Outputs

What a connection passes on

Connections drawn before Each became the default keep passing one chunk. Workflow JSON written through the API or MCP passes one chunk when an edge has no outputMode. Set "outputMode": "each" on the edge's data to run the next node once for every chunk. See Lists and batching.

To keep the chunks as rows, connect a chunk output to the + input of a List. The List adds a video or audio column with one row per chunk. Nodes after the List then read the chunks like any other list.

Settings

How the chunks are cut

• As long as the maximum allows. Chunks are cut in order, each as long as the maximum. A video chunk is cut on whole frames and rounded down, so it is never longer than the maximum. At 29.97 fps, a 10-second maximum gives chunks of 299 frames, or 9.98 seconds.
• The last chunks are evened out. If the leftover at the end is shorter than the minimum, it is evened out with the chunks before it. Every chunk then stays between the minimum and the maximum.
• The sound never runs past the picture. A video chunk's sound ends at most about 21 ms before its picture does.
• Audio files are cut exactly too. MP3 and WAV chunks are cut on exact samples. AAC chunks are cut on whole audio frames, and the last few milliseconds of the file, under 21 ms, are left out.
• At most 200 chunks. A split that would make more fails before any work is done. The message names the shortest maximum length that fits.
• A file that cannot be split fails with a message. This happens with a file shorter than the minimum, or a file that cannot be cut into chunks between the minimum and the maximum. For example, an 11-second file with a 6-second minimum and a 10-second maximum fails.

For example, a 60-second phone video at 29.97 fps has 1,798 frames. With a maximum of 10 seconds and a minimum of 4 seconds, it makes 7 chunks. The first five have 299 frames (9.98 seconds), and the last two have 152 and 151 frames (5.07 and 5.04 seconds). Without the minimum, the last chunk would be 4 frames long.

Cutting at the shots

With Cut by: Shots, a video is cut into one part per shot. The cuts are measured from the video itself: every decoded frame is scored, and each hard cut starts a new shot. Dissolves and fades are not cuts. A flash held for two or three frames before the picture comes back is not a cut either. A one-frame flash is a cut.

• Each part is cut on whole frames and keeps the video's own sound, which never runs past the picture. There is no minimum length, so a one-frame shot is a one-frame part.
• Shots mode cuts a video only. The Audio input is not used. With an audio file connected, the run fails with a message: disconnect the audio, or cut by part length.
• A video can be up to 30 minutes long, with at most 200 shots. A longer video, or one with more shots, fails with a message. Trim it first.
• The Output Chunk menu shows the start and end of each part. Through the API, the job's output also lists each shot: shotCount and shots: [{ index, startSec, endSec, durationSec }], in seconds to the millisecond.
• A shot's start is placed on the nearest frame. It is exact for a constant-frame-rate video, and within half a frame for a variable-frame-rate video.

Download the chunks

After a run, the settings panel lists the Audio Chunks and, for a video, the Video Chunks. Every chunk stays in these lists, whatever you choose under Output Chunk. Use Select All or Deselect All to choose chunks, then click Download to save them.

Credits

Split into Chunks costs 22 credits per run, whatever the number of chunks. Cutting at the shots costs the same. Each node after it on an Each connection is charged once for every chunk it runs on.

Before the run, the estimate counts the chunks from the input's recorded length divided by the maximum. It can count one chunk more than the run makes, never fewer. When the length is not known, for example for a video URL, the estimate assumes a 10-minute file. When cutting at the shots, the shots are only known once the video is measured, so the estimate assumes one shot every 4 seconds of the input, at most 200. After the run, the chunks are counted exactly.

Tips

• Choose the format for the job. WAV keeps lossless quality. AAC gives smaller files, for example for mobile delivery.
• Process only some chunks. Set a range or a list on the Each connection, for example chunks 1 to 5.
• Pick the chunk that moves on. Output Chunk only changes which chunk a Selected connection passes on. All chunks stay available for download.
• Cut one part only. To take a single section of a video instead of equal pieces, use Trim Video.