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

Breaking a long video into short clips that can be distributed individually is the most common content repurposing action for self-media creators. For a 17-minute video, manually transcribing the full subtitles, locating the segments discussing a specific topic, and finally cutting out the finished clips based on timestamps usually takes over 1 hour (estimated).

Using Talos this time, it took about 6 minutes to produce a full transcript with timestamps from 578 subtitle entries, an SRT file, and 2 finished clips with both video and audio.

The difficulties were concentrated in three areas. First, locating the segments about content planning and how titles are decided within the 17-minute footage by dragging the progress bar and repeatedly listening easily leads to losing your place. Second, the transcript timestamps need to be accurate to the second, but estimated time points do not align in editing software. Third, when slicing by timestamps, the start point can be affected by video keyframes, potentially causing a misaligned cut or resulting in a file that has only video but no audio.

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

Step 1: Write Clear Requirements Upfront

The core of the prompt is to clearly state the source material, expected outputs, time precision, and delivery cadence once. The complete prompt used in the first round:

I need to cut a few usable clips from a long video.

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

This is "150 Videos in One Year? A 2024 Workflow Sharing by Yi Film Hurricane," 17 minutes long, discussing their content workflow.
The content I'm looking for: the segments about content planning and how they decide on titles.

Please:
1. Open the video, fully extract the subtitles, compile them into a transcript with timestamps, and save it to the current session directory.
2. Find all segments related to content planning and titles from the transcript, and list them in a table: start time, end time, what this segment is about, and one direct quote from it.
3. Do not cut anything in this round; just show me the table.

Requirements:
- Timestamps must be accurate to the second; do not estimate.
- If no relevant content is found, simply state that there is none; do not fabricate.

This prompt contains four key elements:

• Clear source material specification: Provided the video link, show name, duration, and content theme, saving a round of follow-up questions.
• Clear output and storage location: The subtitles needed to be compiled into a timestamped transcript file and specified to be saved in the session directory.
• Clear time precision and search method: Required timestamps accurate to the second, no estimation, and organized into a table by topic.
• Clear delivery cadence: No cutting in this round; confirm the table first before moving to the next step.

Step 2: Extract Subtitles and Compile Transcript

After receiving the prompt, Talos executed six stages:

1. Open the video page: Read the platform's subtitle track, confirming the total number of subtitle entries and coverage range.
2. Expand each subtitle entry: Extract the start time of each subtitle minute by minute, without interpolation or estimation.
3. Compile into a transcript: Write the 578 subtitle entries sequentially into a timestamped file, preserving auto-generated subtitle recognition errors and annotating them separately.
4. Generate SRT: Use the same subtitle data to provide start and end times for each entry.
5. Verify the timeline: Count entries one by one, confirming times increased monotonically with no duplicate entries.
6. Organize segment index: Read through the transcript, compiling the 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: a transcript, an SRT, and a segment index table.

Step 3: Slice by Timestamps

After confirming the segment index table, the second round reported only two timestamps:

Cut the following segments I selected: 3.46-4.10; 4.38-4.43

Requirements:
1. Save each segment as a separate video file, named Clip_SequenceNumber_StartTime.mp4, in the current session directory.
2. Include both video and audio; do not output video-only clips.
3. After cutting, provide a list: filename, duration, and one or two corresponding original sentences from the transcript.

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

File	Duration	Content
Clip_1_3m46s.mp4	24.0 seconds	3:46-4:10, discussing why some people can tell the same story in an engaging way, and packaging content into a shell
Clip_2_4m38s.mp4	5.2 seconds	4:38-4:43, performance data of the moiré video

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

Two deviations during the process were handled separately. For the first segment, after aligning to keyframes, the start point fell at 3:45, 1 second earlier than requested; Talos used ffmpeg to trim precisely back to 3:46. For the second segment, keyframe-based cutting could not yield a 5-second clip starting exactly from 4:38, so it was captured in real-time from the player, achieving accurate start and end points. After slicing, the first and last frames of the clips were extracted and compared against the transcript to confirm the timestamp error against the on-screen embedded subtitles was within 1 second.

Metric	Manual Production	Talos
Total time spent	Over 1 hour (estimated)	About 6 minutes
Full video subtitles	Sentence-by-sentence transcription of 17-minute footage	578 subtitle entries saved automatically, second-level timestamps
Segment location	Repeatedly dragging the progress bar to listen	11 entries in the index table delivered at once
Clip output	Manually exporting in editing software segment by segment	2 clips, with both video and audio, 1080p
Output format	Manually typed text	Transcript as markdown plus SRT, ready for direct import into editing software
Reusability	Re-transcribe every new video	Prompt saved; just swap the video link

In about 6 minutes, the 17-minute footage's subtitle extraction, segment location, and slicing were completed, producing 578 timestamped subtitle entries, one SRT file, and 2 ready-to-use finished clips.

• Single editing task: Material preparation goes from the most time-consuming step to the most time-saving, allowing more time for segment selection and creative direction.
• Material repurposing: Different topics within the same long video can be located and clipped separately, supporting distribution across multiple platforms from one source.
• Timestamp quality: Second-level timestamps provide unified coordinates for slicing, caption alignment, and derivative scriptwriting, reducing back-and-forth rework.
• Format compatibility: SRT and conventionally encoded clips can be imported directly into editing software without additional transcoding.

1. Write accuracy requirements into the prompt: Clearly specify timestamps accurate to the second, no estimation, to get directly usable time points.
2. Produce the table before slicing: Obtain a segment index table for confirmation first, to avoid cutting unwanted segments.
3. Explicitly specify video and audio: Clearly state that both are required to avoid receiving silent clips.
4. Save the prompt for reuse: For a different video, simply replace the link, duration, and topic; the workflow remains unchanged.

Q: The subtitles are auto-generated; do recognition errors affect slicing?

A: They do not affect the slicing itself. The timeline comes from the platform's subtitle track; recognition errors only impact specific words. In this case, obvious errors in the transcript were annotated separately, and any direct quotes were based on the transcript as the source.

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; the SRT can also be imported directly, and the timelines align.

Q: What if the specified time point falls between two keyframes?

A: The start point can be affected by keyframe alignment, potentially starting slightly earlier than requested. Talos will either re-encode to trim precisely, or capture from the player instead, and will explain any deviation upon delivery.

Q: Can this method be used on other videos?

A: Yes. The process is just two steps: extract subtitles with timestamps, and slice by time points. For a different video, simply replace the link, duration, and topic you are searching for.