[FEAT] Intro + Credits Detection.. [HELP NEEDED]

[stevezau]

Feature: Add Intro, Credits, and Outro Detection

Summary

Add intro, credits, and outro detection to plex_generate_vid_previews using FFmpeg chromaprint audio fingerprinting (intros) and blackdetect/silencedetect filters (credits), then write markers to the Plex database's taggings table.

---

Research Findings

How Plex Does Detection Internally

Plex uses three detection methods (confirmed by the intro-skipper project wiki and Plex blog):

1. Chromaprint audio fingerprinting — an FFmpeg extension that compares audio across episodes in a season to find recurring segments (intro themes). This is the primary method for intro detection.
2. Black frame detection — FFmpeg's blackdetect filter identifies black/near-black frames marking transitions (credits start).
3. Silence detection — FFmpeg's silencedetect filter finds extended periods of silence that often precede credits.

Detection Parameters

• Intros: 15s–2min long, located within the first 25% of episode or first 10 minutes (whichever is smaller)
• Credits: 15s–5min long, located near the end of the video (typically last 25%)

How Plex Stores Markers

Plex stores markers in its SQLite database in the taggings table (confirmed by MarkerEditorForPlex source code). Key fields:

Field	Description
metadata_item_id	Plex episode/movie rating key
tag_id	Marker tag type (tag_type=12 in the tags table)
index	Ordering index for multiple markers
text	Marker type: 'intro', 'credits', or 'commercial'
time_offset	Start time in milliseconds
end_time_offset	End time in milliseconds
created_at	Timestamp
extra_data	JSON (PMS >=1.40) or legacy string with version and final flag

There is no public Plex HTTP API for creating markers. Both MarkerEditorForPlex and the now-EOL plex-credits-detect project write directly to the SQLite database. python-plexapi only has read-only Marker objects but no addMarker() method.

The project already has PLEX_CONFIG_FOLDER access, so the Plex database at PLEX_CONFIG_FOLDER/Plug-in Support/Databases/com.plexapp.plugins.library.db is accessible.

---

Architecture

flowchart TB
    subgraph detection [Detection Phase]
        A[Enumerate Library] --> B{Media Type?}
        B -->|TV Episode| C[Group by Season]
        B -->|Movie| D[Single File Analysis]
        C --> E["Chromaprint Fingerprint each episode"]
        E --> F["Compare Fingerprints across season"]
        F --> G["Identify Common Audio Segments"]
        G --> H[Intro Markers]
        C --> I["BlackDetect + SilenceDetect near end of each episode"]
        D --> I
        I --> J[Credits Markers]
    end
    subgraph writing [Marker Write Phase]
        H --> K["Write to Plex SQLite taggings table"]
        J --> K
        K --> L["Verify via python-plexapi read"]
    end

Loading

[aweigold]

I might be able to do this... however, there is no safe way that I can see doing this on windows with sqlite. My thoughts were to just document that and warn about that in the UI.

[stevezau]

I'd welcome a PR :)

When you say "no safe way that I can see doing this on windows with sqlite."

It is running in a container so why is windows an issue? The container should have sqlite? Or am i misunderstanding your comment ?

[aweigold]

I'd welcome a PR :)

When you say "no safe way that I can see doing this on windows with sqlite."

It is running in a container so why is windows an issue? The container should have sqlite? Or am i misunderstanding your comment ?

Windows Docker volume mounts don't support POSIX advisory file locking, which SQLite requires. This is a documented, unfixable limitation.

[aweigold]

Should be sane on non-windows.

Based on my research, I believe it is manageable to access the SQLite database safely without stopping Plex. Here's why:

The evidence is strong:

• Plex itself runs multiple processes (scanner, butler, main server) against the same DB concurrently — it's designed for this
• Plex uses WAL mode, which explicitly supports concurrent readers + 1 writer
• plex-credits-detect ran as a background daemon writing markers while Plex was running for its entire lifespan with zero reported corruption
• Plex has built-in SQLITE_BUSY retry logic ("Sleeping for 200ms to retry busy DB"), meaning it already expects external lock contention
• The taggings table is a simple table — no FTS indexes, no custom tokenizer involvement

The safe pattern is straightforward:

1. PRAGMA busy_timeout = 5000 on every connection
2. Open connection → single short transaction (INSERT + COMMIT) → close connection immediately
3. Never hold a connection open longer than necessary
4. Read-only connections for queries

[stevezau]

ah, didn't realise that on windows.

hmm, do we need sql access, it looks like the API's might work?
https://developer.plex.tv/pms/#tag/Library/operation/libraryMetadataPostMarker

POST /library/metadata/{ids}/marker — Create a marker (type, startTimeOffset, endTimeOffset, attributes)
PUT /library/metadata/{ids}/marker/{marker} — Edit a marker
DELETE /library/metadata/{ids}/marker/{marker} — Delete a marker
DELETE /library/sections/{sectionId}/markers/intro — Delete all intro markers in a section

[stevezau]

Also found this - https://github.com/danrahn/MarkerEditorForPlex

[aweigold]

I knew about that MarkerEditorForPlex app, but haven't deployed it, I have half a petabyte of content and growing, and it looked manual, so a no go for me. 😆

I looked into the REST API, and it returns 400 Bad Request for type=credits, type=credit, and type=intro. Only type=bookmark works. This matches the community reports.

The Plex REST marker API only supports bookmarks. We must use direct SQLite writes for intro/credits markers unfortunately. Really wish Plex exposed this in the API as it's the clear way to go. 😡

[stevezau]

yes, similar size for me. I shared that other project as i figured we might be able to reuse or claud to learn from how it works.

Ah ok, damn. Plex is a little frustrating like this.

Well, i think it's fair to exclude windows given the constraint, but i guess we should check on startup, and provide a warning to the user and disable credits+intro detection... if the path mapping is on windows?

[aweigold]

Since you are active online, once I have a plan from claude, and review it, I'll paste it here for feedback. Mostly telling it not to be lazy and not defer the intro detection at the moment. 😆

[stevezau]

no worries, happy to jump on a discord if you want? might be faster to communicate vs flooding this issue?