---
title: "Long Video Slicing: Extract Full Subtitles and Cut Final Clips in 6 Minutes"
description: "A 17-minute video, 6 minutes to obtain 578 timestamped subtitles, one SRT file, and 2 final clips with sound."
canonical: https://taloswork.ai/en/blog/long-video-clipping
language: en
last_updated: 2026-10-28
date_published: 2026-10-28
category: "自媒体"
---

# Long Video Slicing: Extract Full Subtitles and Cut Final Clips in 6 Minutes

# Long Video Clipping: Extract Full Subtitles and Slice Finished Clips in 6 Minutes

## Project Background

Breaking a long video into short, independently distributable clips is one of the most common content repurposing workflows for self-media creators. For a 17-minute video, manually transcribing the full subtitles, locating segments that cover a specific topic, and then cutting finished clips based on time points typically takes more than an hour (estimated).

Using Talos this time, it took about 6 minutes to produce 578 timestamped subtitle entries for the full transcription, one SRT file, and 2 finished clips with both visuals and audio intact.

## Challenges

The difficulties centered on three areas. First, locating segments about content planning and how to decide on titles within 17 minutes of footage — relying on dragging the progress bar to audition repeatedly made it easy to lose your bearings. Second, timestamps in the transcription needed to be precise to the second, and estimated time points would not match up in editing software. Third, when slicing by time points, the starting point could be affected by video keyframes, potentially leading to misaligned cuts or files with visuals but no sound.

The goal was to produce a material package ready for editing: a full second-level transcription, a segment index table, and finished clips with audio.

## Method

### Step 1: Clearly State All Requirements at Once

The core of the prompt was to communicate the materials, deliverables, time precision, and delivery cadence clearly in one go. The full prompt used in the first round:

```text
I need to cut a few usable segments from a long video.

Video link: https://www.bilibili.com/video/BV13QzbYuEWr/

This is "150 Videos in a Year? A Share of the 2024 Workflow" from the channel "Yingshi Jufeng." It is 17 minutes long and discusses their content creation workflow.
The content I'm looking for: segments about content planning and how titles are decided.

Please:
1. Open the video, fully extract the subtitles, write them into a transcription file with timestamps, and save it to the current session directory.
2. From the transcription, find all segments related to content planning and titles, and list them in a table: start time, end time, what is being discussed, and one quote from that portion.
3. Do not cut yet in this round; just show me the table.

Requirements:
- Timestamps must be precise to the second, no estimates.
- If no relevant content is found, say so directly without padding.
```

This prompt contained four key elements:

- **Material description was thorough**: provided the video link, program name, duration, and content theme, eliminating a round of follow-up questions.
- **Deliverables and storage location were clear**: subtitles had to be saved as a transcription file with timestamps, with the current session directory specified as the storage location.
- **Time precision and extraction method were specified**: required timestamps precise to the second, no estimates, and results organized by theme into a table.
- **Delivery cadence was explicit**: no cutting in this round; first provide the table for confirmation, then proceed to the next step.

### Step 2: Extract Subtitles and Save as Transcription

After receiving the prompt, Talos executed six stages:

1. **Opened the video page**: read the platform's subtitle track and confirmed the total number of subtitle entries and coverage range.
2. **Expanded subtitles one by one**: extracted the start time of each subtitle entry in minute segments, without interpolation or estimation.
3. **Saved as a transcription**: wrote 578 subtitle entries in chronological order into a file with timestamps, retaining auto-subtitle recognition errors and flagging them separately.
4. **Generated SRT**: used the same subtitle data to provide start and end times for each subtitle.
5. **Verified the timeline**: counted entries individually to confirm that times increased monotonically with no duplicates.
6. **Organized a segment index**: read through the transcription and compiled 11 segments related to content planning and titles into a table.

The subtitle extraction and segment location phase took about 3 minutes, producing three files: the transcription, SRT, and segment index table.

### Step 3: Slice by Time Points

After confirming the segment index table, only two time points were reported in the second round:

```text
Cut out these selected segments for me: 3:46-4:10; 4:38-4:43

Requirements:
1. Save each segment as a separate video file, named in the format Clip_SequenceNumber_StartTime.mp4, in the current session directory.
2. Include both visuals and audio; do not cut visuals only.
3. After cutting, give me a list: file name, duration, and one or two original lines corresponding to each segment.
```

The slicing phase took about 3 minutes, results for the two clips:

| File | Duration | Content |
| --- | --- | --- |
| Clip_1_3m46s.mp4 | 24.0 seconds | 3:46-4:10, discussing why some people make the same topic interesting and how to package content as a "shell" |
| Clip_2_4m38s.mp4 | 5.2 seconds | 4:38-4:43, playback data for the moiré pattern video |

Both files were 1920×1080, H.264 video with AAC stereo 48kHz, with both visuals and audio intact.

Two deviations during the process were handled separately. For the first segment, after keyframe alignment, the starting point landed at 3:45, one second earlier than requested; Talos used ffmpeg to precisely trim it back to 3:46. For the second segment, a 5-second cut starting at 4:38 could not be achieved by keyframe-based slicing; instead, it was recorded in real time from the player, with start and end points accurate. After slicing, the first and last frames of the clips were extracted and compared against the transcription to confirm that the discrepancy between timestamps and on-screen embedded subtitles was within 1 second.

## Results

| Metric | Manual Production | Talos |
| --- | --- | --- |
| Total Time | Over 1 hour (estimated) | About 6 minutes |
| Full Subtitles | Transcribe 17 minutes of footage sentence by sentence by ear | 578 subtitle entries automatically saved to disk, second-level timestamps |
| Segment Location | Repeatedly drag the progress bar to audition | 11-segment index table delivered in one go |
| Clip Output | Manually export segment by segment in editing software to try | 2 clips, with visuals and audio both present, 1080p |
| Output Format | Manually typed text | Transcription md plus SRT, ready to import into editing software |
| Reusability | Re-transcribe each video from scratch | Save the prompt, replace the video, and the process stays the same |

In about 6 minutes, subtitle extraction, segment location, and slicing for 17 minutes of footage were completed, yielding 578 timestamped subtitle entries, one SRT file, and 2 ready-to-use finished clips.

## Business Impact

- **Per single edit**: material preparation shifted from the most time-consuming step to the most time-saving one, allowing more time to be spent on segment selection and conceptual framing.
- **Material reuse**: different themes within the same long video can be located and turned into clips separately, enabling one piece of footage to support multi-platform distribution.
- **Timestamp quality**: second-level timestamps provide a unified reference for clipping, subtitle alignment, and secondary creation scripts, reducing back-and-forth rework.
- **Format compatibility**: SRT and clips encoded in standard formats can be used directly in editing software without extra transcoding.

## Replicable Insights

1. **Write precision requirements into the prompt**: explicitly state that timestamps must be precise to the second, no estimates, to get time points that are directly usable.
2. **Get the table before slicing**: first request a segment index table for confirmation to avoid cutting the wrong segments after the fact.
3. **Explicitly require visuals and sound**: clearly state that both visuals and audio must be included to avoid receiving silent clips.
4. **Save the prompt for reuse**: to use with a different video, simply replace the link, duration, and target theme while keeping the process unchanged.

## Frequently Asked Questions

**Q: The subtitles are auto-generated — will recognition errors affect slicing?**

A: Slicing itself is not affected. The timeline comes from the platform's subtitle track; recognition errors only impact some of the text. This time, obvious errors were flagged separately in the transcription, and when quoting original lines, the transcription text was taken as authoritative.

**Q: Can the cut clips be imported directly into editing software?**

A: Yes. The clips are 1920×1080, H.264 video with AAC stereo audio, and the SRT can also be directly imported, with both timelines matching up.

**Q: What happens if the reported time points fall between two keyframes?**

A: The starting point will be affected by keyframe alignment and may be slightly earlier than requested. Talos will either re-encode with precise trimming or fall back to recording from the player, and will explain any deviation upon delivery.

**Q: Can this method be used for other videos?**

A: Yes. The workflow has only two steps: extract timestamped subtitles, then slice by time points. To switch to another video, simply replace the link, duration, and the target theme you are looking for.