Skip to main content

YouTube Chapter Markers From Active Sequence

This workflow analyzes the currently active Premiere Pro sequence and creates viewer-friendly YouTube chapter markers based on the spoken content that is actually used in the edit.

Instead of chaptering the raw source material or inventing timestamps by hand, you ask the agent to reconstruct the sequence-level narrative from the clips that are really audible in the timeline, then place sequence markers at meaningful section boundaries.

New to Automation Agent?

This recipe uses Automation Agent for Adobe Premiere. The ready-made Library script runs without a live MCP-agent session but still uses your configured Workflow AI provider for editorial judgment. The adaptable prompt uses the agent workflow via MCP. See Set Up Your AI for both setup paths.

Fastest Path: Run The Library Script​

Open Library > Examples > Publishing & Social and run Create YouTube Chapters From Active Sequence.

The script analyzes the transcripts of the sequence that is active when you continue from its setup dialog, creates chapter point markers, and can copy matching YouTube-style chapter text to the clipboard. The setup dialog lets you add optional video context or naming guidance, choose the chapter density, control whether the chapter text is copied, and optionally retain a detailed report or troubleshooting files. You can switch sequences while the dialog is open; the sequence active when you click OK becomes the target.

This is the quickest repeatable path for speech-led edits. The saved workflow already knows how to collect only the transcript ranges used in the edit, map them into sequence order, call Workflow AI with stable text IDs, validate the response, create and read back the markers, and derive the clipboard text from the same verified plan. It therefore usually takes less time and fewer agent tokens than asking a live agent to discover and build the complete workflow during every run. Provider speed and sequence length still affect the runtime.

You need the Python Runtime Pack and a configured Workflow AI provider. Codex CLI and Claude Code are the currently tested recommendations for this complete-context editorial task. Local models running through LM Studio have not yet been validated for it. The script sends the usable transcript text and your optional guidance to that provider.

This Library script is transcript-only

It does not inspect frames, title cards, B-roll, screen content, silent sections, or visual transitions that are not expressed in the transcript. Use the live-agent prompt below when those elements determine the chapter structure: a live agent can inspect a small number of targeted frames and adapt its analysis to what it finds.

Treat the result as a strong draft. Review every title, boundary, coverage warning, and YouTube compatibility note before publishing. Missing transcripts, unsupported timeline items, overlaps, and unusually short sections are reported instead of silently guessed.

After a successful run, the default keeps only the Premiere markers and optional clipboard text. Working files and the empty working folder are removed. Enable Keep detailed report to retain ::HOME::/automation-agent-sequence-chapters/Latest YouTube Chapters Report.md, or enable Keep troubleshooting files to retain the run artifacts in the same folder. Failed runs and retirement warnings may keep diagnostic files so the problem can be investigated.

This walkthrough shows the recommended repeatable path: analyze the active sequence, review the generated Premiere markers, refine them if needed, and copy the updated YouTube chapter text.

Use the Library workflow to propose, review, refine, and export YouTube chapter markers from an active Premiere sequence.

Earlier Live-Agent Walkthrough (Optional)​

The earlier walkthrough below documents the more open-ended live-agent route. It remains useful when the transcript alone cannot explain the structure and you want an agent to inspect selected visual evidence or adapt the approach to an unusual project. For normal repeatable, speech-led chaptering, start with the Library workflow above: its transcript collection, marker creation, and export steps are already fixed and verified.

Short walkthrough of the YouTube chapter markers workflow in Premiere.

When This Is Useful​

Use this workflow when:

  • you want YouTube chapters for an edited sequence, not for the raw source recordings
  • the final cut combines multiple clips and you want chapters based on what viewers actually hear and see
  • you want the chapter structure added directly as sequence markers in Premiere
  • you want chapter titles that reflect topic changes rather than arbitrary time slices

Typical examples:

  • talking-head videos with inserted B-roll
  • tutorials assembled from multiple screen-recording segments
  • interviews or explainers edited down from longer source material
  • review, demo, or presentation videos that move through several clear topics

What You Need For The Live Agent Prompt​

Before you run the prompt:

  1. Open the target sequence and make sure it is the active sequence in Premiere.
  2. Make sure Automation Agent is connected through your MCP workflow.
  3. Make sure the source clips used in the sequence have transcripts where possible, or at least enough transcript coverage for the agent to infer the structure reliably.
  4. If the sequence depends heavily on visuals rather than speech, expect the agent to inspect a small number of targeted still frames as part of the workflow.

Both paths only need to write sequence markers in Premiere. To protect the rest of your edit, open the Execution Permissions tab before running either one, enable Restrict write access in the Project section, then use + Add and choose Active Sequence Markers.

This adds write access for ::SEQ\ACTIVE::MARKER. The workflow can create chapter markers on the sequence captured for this run, but it cannot move clips, change track items, rename project items, or otherwise modify the project unless you add more write permissions.

The Library script also needs the installed Python Runtime Pack and filesystem access to its working folder under ::HOME::/automation-agent-sequence-chapters. Its Workflow AI call sends the usable transcript and optional guidance to your configured provider and may use network access, model tokens, and paid provider capacity according to that provider's configuration.

If the live agent exports transcripts or targeted still frames for analysis, also allow the narrow temporary folder it proposes. You do not need to grant broad filesystem access; review that path before the live run continues.

The Library script owns only markers whose final non-empty comment line is its exact versioned automation-agent:youtube-chapters tag. The useful reason and review notes therefore appear before the technical ID. On a rerun it creates and verifies the new marker set before retiring the previous owned set. It does not delete unrelated markers. If old owned markers cannot all be retired, the new verified markers remain and the report warns that both runs are visible.

Copy And Paste Prompt​

Paste this prompt into your MCP client exactly as-is:

Using Automation Agent in Premiere Pro, analyze the currently active Premiere Pro sequence and create meaningful YouTube chapter markers based on the spoken and, where useful, visual content that is actually used in the edit.

Follow these steps carefully:

1. Capture Premiere's transcript of the active Sequence.

   * Capture the currently active Sequence once, then check `HAS_TRANSCRIPT` on that Sequence ref.
   * Export that same Sequence ref once with `prExportTranscriptJson` and verify that the JSON contains usable timed words.
   * Treat this as the transcript shown in Premiere's Text panel. Its word times are already Sequence times.
   * Do not rebuild it from individual source clips or filter it by track mute, clip disabled state, or a chosen audio track.

2. Preserve Premiere's Sequence semantics.

   * Use Premiere's composed result for nested Sequences, Multicam selections, repeated placements, cuts, and trims.
   * Keep the relationship between transcript text and Sequence timecode unchanged, so chapter markers can be placed accurately.
   * If a different transcript scope is required, stop and ask the editor to activate or construct the Sequence whose Text-panel transcript represents that scope.

3. Understand the video content.

   * Read the combined timeline transcript as if it were the actual spoken content of the edited video.
   * Identify the main topics, sections, and changes in subject matter.
   * Look for natural chapter boundaries where the video moves from one content section to another.
   * Keep the temporal order in mind: understand not only what is being said, but also when each topic appears in the sequence.
   * Avoid creating chapters based on every edit or cut. Create chapters only when there is a meaningful change in content, purpose, or viewer interest.

4. Use visual analysis only when helpful.

   * Start with the transcript-based analysis.
   * If the transcript is sufficient to understand the video structure, do not export unnecessary still images.
   * If the transcript is incomplete, ambiguous, or does not fully explain what is happening visually, you may export selected still images from the sequence for visual inspection.
   * Visual analysis is especially useful for:

     * screen recordings,
     * software tutorials,
     * product demos,
     * slides or presentations,
     * before/after examples,
     * B-roll or montage sections,
     * long sections with little or no spoken content,
     * unclear transitions where the transcript alone does not reveal the actual content change.
   * Do not export frames continuously or at fixed short intervals by default.
   * Export still images only at strategically useful positions, such as:

     * near potential chapter boundaries,
     * at the beginning of long transcript gaps,
     * at the start of visually distinct sections,
     * or a few seconds before and after an unclear transition.
   * Use the visual information to refine chapter titles and marker positions, but keep the chapter structure focused on meaningful viewer-facing sections.

5. Create YouTube chapter candidates.

   * Based on the timeline-aware transcript and any useful visual observations, decide where YouTube chapter markers should be placed.
   * The first chapter should start at the beginning of the actual content, usually at or near 00:00 unless there is a clear intro before the main topic.
   * Each chapter should begin where a new meaningful section starts.
   * Prefer clear, helpful chapter titles that describe what the viewer will learn, see, or understand in that section.
   * Avoid vague titles like “Part 1”, “Section 2”, or “Introduction” unless they are genuinely appropriate.
   * Do not create too many chapters. Prefer fewer, high-quality chapters over many small ones.
   * Do not create chapters only because there is a cut, a new clip, or a short pause. The chapter must represent a meaningful change in content.
   * If a chapter boundary is semantically clear but its exact frame is slightly uncertain, use nearby clip boundaries, visible transitions, or title-card starts as refinement hints.
   * If a nearby cut clearly marks the start of the new section, prefer placing the chapter marker exactly on that cut instead of slightly before or after it.
   * Do not snap markers to cuts blindly. A cut is only a placement hint when it also matches a meaningful content transition.

6. Add the markers to the Premiere Pro sequence.

   * For every final chapter, create a sequence marker at the correct timeline position.
   * The marker must be a sequence marker, not a clip marker.
   * Place each marker exactly at the start of the corresponding chapter section.
   * The marker name must be the chapter title.
   * Do not include timestamps in the marker names.
   * Make sure the marker positions match the active sequence timeline.

7. Output the result in the chat.

   * After creating the markers, provide a list of the chapters in YouTube description format:
     `00:00 Chapter title`
   * Use the correct timestamps based on the sequence timeline.
   * Also briefly mention that the same chapters were added as sequence markers in Premiere Pro.
   * If visual still images were used, briefly mention that selected frames were inspected to refine the chapter structure.
   * If some clips had missing or incomplete transcripts, mention this limitation briefly.

Important quality guidelines:

* Base the chapters primarily on Premiere's active-Sequence transcript as shown in the Text panel.
* Do not reconstruct or filter that transcript from individual source clips.
* Keep the timeline order and timing accurate.
* Use visual still images only when they add meaningful information.
* Do not let visual analysis create unnecessary complexity or noise.
* Create chapters based on meaningful content changes, not merely on timeline edits.
* Chapter titles should be concise, specific, and useful for YouTube viewers.
* If the sequence contains very little spoken or visually interpretable content, create only the chapters that are justified by the available information.
* Before creating the final markers, reason through the chapter structure carefully and verify that each chapter boundary is useful.

What The Agent Will Do​

In practical terms, a good run of this workflow should:

  1. capture the active Sequence and export its Premiere transcript once
  2. verify that the exported JSON contains usable timed words
  3. use Premiere's existing Sequence-time transcript directly, including nested and Multicam edits
  4. identify the main viewer-facing topic changes and likely chapter boundaries
  5. use targeted still-frame inspection only where transcript context is not enough
  6. refine uncertain boundary placement using nearby cuts or visible transitions when helpful
  7. create sequence markers with concise chapter titles
  8. return the finished chapter list in YouTube timestamp format

Expected Result In Premiere​

The result should be:

  • a set of sequence markers on the active sequence, one per final chapter
  • chapter titles that describe the actual sections of the edited video
  • timestamps that match the sequence timeline rather than the raw source files
  • a chat output that you can paste directly into a YouTube description

This workflow is meant to produce a practical first pass that is already useful for publishing, while still allowing editorial judgment where chapter boundaries are subjective.

Prompt Variants​

You can adapt this workflow depending on the type of sequence and how much uncertainty the agent is likely to encounter.

1. Keep the workflow transcript-only unless ambiguity remains​

If you want the simplest and fastest run, add a stronger instruction that the agent should avoid visual frame export unless the transcript leaves a real ambiguity about the chapter structure.

That is useful when:

  • the sequence is mostly spoken-word content
  • transcript coverage is strong
  • visuals are supportive rather than structurally important

2. Encourage stronger use of visual verification​

If the sequence includes screen captures, slides, or visually driven transitions, add an instruction that the agent should inspect selected frames around likely chapter boundaries even when the transcript is partially usable.

That is useful when:

  • the main structure is shown visually rather than spoken explicitly
  • chapter titles should reflect what appears on screen
  • long silent or lightly narrated sections matter

3. Be stricter about snapping uncertain boundaries to cuts​

If your edit style often starts a new topic exactly on a cut, title card, or graphic change, you can strengthen the placement rule so the agent prefers cut-aligned markers whenever the semantic transition and the edit transition coincide.

That is useful when:

  • new chapters often begin with a new on-screen segment
  • your sequence uses clear editorial section breaks
  • you want timestamps that feel clean and intentional in both Premiere and YouTube

4. Use a deliberately different Sequence scope​

If the Text-panel transcript contains more or less material than you want to analyze, activate or construct a Sequence whose Premiere transcript represents that scope, then rerun the workflow. Track mute and clip disabled state are not transcript-scope controls for this workflow.

5. Give the agent outside context about the intended structure​

If the transcript alone is too vague, compressed, or context-poor, give the agent additional information about what the video is supposed to cover before it decides on the chapter structure.

That can include:

  • a script, outline, rundown, or presentation plan
  • a list of the expected topics or sections
  • a short explanation of the video's purpose and target audience
  • notes about which concepts or segments must definitely become separate chapters

That is useful when:

  • the transcript uses shorthand, vague references, or context-dependent language
  • you already know the intended chapter topics and mainly need accurate boundary placement
  • the agent does not need help understanding what the sections are, only where they begin

In those cases, the task becomes easier because the agent can use the outside context to interpret ambiguous transcript passages and focus more precisely on transition timing.

6. Tell the agent about structural chapter cues already present in the project​

If your sequence already contains editorial patterns that usually mark chapter starts, tell the agent about them explicitly and let it use them as constraints or strong hints.

That can include:

  • a title-card or graphic clip that appears at every chapter start
  • a specific video track that only contains chapter-intro animations
  • recurring text overlays, bumpers, or transitions that mark new sections
  • a known rule such as "only place chapter markers where this kind of cue appears"

That is useful when:

  • your project already encodes chapter boundaries visually or structurally
  • you want fewer false-positive chapter starts
  • the transcript suggests a topic change, but the exact placement should follow your editorial system

This can make the workflow much more precise, especially when the instruction is framed clearly. For example, you can tell the agent that chapter starts should only be considered valid when they coincide with one of those structural cues, or that those cues should be used to refine the exact marker placement after the semantic chapter boundaries have been identified.