> ## Documentation Index
> Fetch the complete documentation index at: https://docs.overlap.ai/llms.txt
> Use this file to discover all available pages before exploring further.

> Find the best moments in an input video and pass clip outputs to the rest of a workflow.

# Findclips

> Find Clips scans incoming video and creates clip outputs from your clipping strategy, duration, prompt, keyword, and optional model guidance.

The **Find Clips** node is for selecting moments from the source video. It does not apply visual edits such as subtitles, reframing, b-roll, watermarks, or outros. Add those editing nodes after Find Clips when you want to change how the clips look.

<Info>
  ### Schema

  * **Input**: Video or clip output from a trigger or earlier node
  * **Output**: Clips selected from the input video
</Info>

## Add the node

Open **Workflows** from the left sidebar and click **New** to open the workflow builder.

Add a trigger first, such as **Manual Trigger**, **New YouTube Video**, **RSS Feed**, or another source node. In the **Editing** stage, choose **Find Clips**. Overlap adds the action node to the canvas and connects it to the previous video output.

Click the **Find Clips** node on the canvas whenever you need to reopen its settings on the right side.

## Choose a clipping strategy

Choose one of the visual **Clipping Strategy** cards at the top of the sidebar:

* **Highlights** finds the strongest standalone moments in the source.
* **Montages** are themed clips that combine related moments with discontinuous cuts.
* **Even Clips** splits the complete source into continuous clips near one target length.

Selecting a card expands the settings that apply to that strategy directly
below it. Highlights includes a model choice and a minimum/maximum duration.
The Montages strategy includes a minimum/maximum duration. Even Clips includes one target
duration.

The saved workflow contract still treats Even Clips as the uniform strategy;
the clearer label only changes how the option appears in the sidebar.

### Videos with no speech

Even Clips works on videos with nothing spoken, such as ASMR, b-roll, or
music-free screen recordings. With no sentences to align to, the source is cut
into exact target-length clips from the start, and each clip is titled after
the source video (`<source title> (Part 1 of 12)`). Subtitles on these clips
are empty.

Highlights and Montages on the default model choose clips from what is said, so
they stop with "This video has no speech" on such a source. Use Even Clips, or
the Multimodal model for Highlights, which selects from frames.

## Choose a Highlights model

When **Highlights** is selected, use **Model** to choose how Overlap should
evaluate the source video:

* **Conversational**: for spoken conversations, podcasts, interviews, and similar talk-driven videos
* **Multimodal**: for sports or other videos where non-verbal context is important

The Montages and Even Clips strategies choose their required clipping model automatically, so
they do not show a separate model setting.

## Configure duration

For **Highlights** and **Montages**, use **Minimum** and **Maximum** under
**Duration** to set the target range for generated clips.

The current defaults are:

* **Minimum**: `30 s`
* **Maximum**: `120 s`

Clips can be between `10` and `600` seconds, and the minimum must stay below the maximum.

For **Even Clips**, set one **Target clip length** from `10` to `600` seconds.
The separate minimum and maximum controls are hidden because they do not apply.
Find Clips partitions the complete source at approximately that length, using
nearby transcript boundaries when possible, without a fixed maximum number of
clips.

## Guide clip selection

For organizations with transcript reuse enabled, Find Clips preserves one
source transcript using your configured transcription provider. Selection,
General Editor cuts, and subtitles share its speech timing. Automatic cuts use
sentence starts and sentence ends; timestamp ranges you pin remain exact.
Automatic boundary reviews use sentence starts for openings and sentence ends
for endings, including when speech is separated by a pause.

Joined montage outputs reuse the selected words in playback order. Caption
corrections create an editable working version while preserving the original
source transcript. This applies to new runs; existing clips are unchanged.

Use **Prompt** to describe what makes a good clip for this workflow. You can type directly into the editor or click **Select prompt** to choose a saved prompt.

Good prompts usually name the audience, topic, or type of moment you want. For example:

* Find clips where the guest shares practical advice for founders.
* Surface energetic moments with clear reactions or surprising claims.
* Focus on segments about AI trends and avoid housekeeping or intro chatter.

Use **Keywords** when certain terms, names, products, or topics should influence what the node looks for. The node starts with no custom keywords.

For montage-style prompts, Find Clips can combine separated moments in the
playback order that best explains the chosen theme. That order does not have to
match source chronology: a later moment may become the hook when the resulting
clip remains coherent.

By default, Overlap creates one new video in the chosen playback order while
keeping each exact selected moment editable in Studio. The video also includes
about five seconds of surrounding source context on each side of every moment.
Those five-second targets snap to the nearest transcript sentence start and
sentence end, so extending a cut does not begin or end halfway through a
sentence.

The optional `flattened` mode joins only the exact selected moments and removes
the editable segment contract. The optional `segments` mode retains the long
source and absolute source-time ranges instead of creating a standalone video.

## Clip titles

After selecting each clip's source ranges, Find Clips generates a concise title
from the exact retained transcript. This title step applies to every clipping
strategy, including Even Clips; it does not change the selected
timestamps.

The generated title is clip metadata, not text burned into the video. Add a
[Style Video](/nodes/style-video) node afterward when you want visible title
text in the finished video.

## Runtime configuration

When you manually launch a workflow that includes Find Clips, open **Runtime Configuration** to add one-run guidance without changing the saved workflow.

* **Prompt adjustment** adds temporary instructions to the saved Find Clips prompt.
* **Timestamp ranges** accepts source-video start and end times. Use **Guaranteed clips** to include exact ranges as clips, or **Search within** to find individual clips inside those ranges.

## Build the rest of the workflow

After Find Clips, continue with editing nodes such as [Style Video](/nodes/style-video), [Convert to Vertical](/nodes/reframe), or [Audiogram](/nodes/audiogram). Finish with an export node so the workflow has a destination for the clips it creates.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.