Plugin: EmbyCredits, detect end credits and add auto skip.

[yocker 1943]

51 minutes ago, fbourg said:

Hi everyone,

I'm looking to test the latest beta version of the EmbyCredits plugin, but I cannot find the beta .dll file on GitHub.

Could someone point me in the right direction or share where to download the latest beta build? (Is it shared via PM, attached to a specific post in this thread, or hosted on another branch?)

Setup & Hardware Specs

CPU: Intel Core i7-14700K (Quick Sync Video active, no dedicated NVIDIA GPU)

RAM: 128 GB

Library size: ~61,000 episodes

Questions regarding settings:

1. Chromaprint triggering at the beginning of episodes

When trying to use Chromaprint (audio fingerprinting), it's misbehaving and matching way too early, near the start of the video:

What are the recommended settings or search offsets to force the audio scan to focus strictly on the last few minutes of the episode?

What are good values for the minimum match duration and confidence thresholds to prevent opening themes or incidental audio cues from triggering false positive markers?

2. OCR tuning & performance scaling

With 61,000 episodes, running OCR takes a massive amount of time:

Without a dedicated NVIDIA card (no CUDA), the OCR engine is relying purely on the CPU/RAM, while frame extraction can use Intel QSV.

With 20 cores / 28 threads on the 14700K and 128 GB of RAM, what are the recommended settings (worker threads, extraction FPS, temp directory configuration) to maximize throughput without bottlenecking the system?

Are there best practices to avoid the initial scan choking the server on such a large library?

Thanks in advance for your help and feedback!

Now that i'm sitting at my computer i can answer the other questions without having to fight my phones autocorrect.

1. Chromaprint is not very accurate and i generally advice against using it.
I'm planning some time in the future to try and merge it with blackframe detection to more accurately find the end credits.
It's a rather big operation to implement that so might be a good while before that is ready.
There are settings you can fine tune for the chroma detections, sadly the defaults are the best ones i have found so far.

I strongly advice using OCR when possible.

2. Only the PaddleOCR supports Nvidia cards. If using the Tesseract it's CPU only. The plugin it self can use GFX cards for extracting the images but in reality it's not really that much faster.
With an 14700K it shouldn't be to slow (I'm running it with a 7950X3D so not that much bigger a CPU).
It's not memory intensive so 128GB is WAY overkill.
If you insist on fine tuning and not using the presets already there then I suggest using Tesseract and the default settings but change:
1. Paraelle FFmpeg Sessions to 4 (If this bogs down the machine than lower it a bit)
2. FFmpeg Thread Limit: to 4 or higher. You can set it to 0 if you are comfortable with it being allowed to take all your CPU.
3. Hardware Acceleration Type: Set to Quick Sync. (If that fails set it to VAAPI, havn't tested if Quick Sync finally works with the newest Emby)
4: Parallel Batch Size: 4 (again lower this if you find it bogs down the machine.

The rest of the settings should be pretty self explanatory as they say if they are recommend to enable or disable.

A single high quality TV show episode takes about 20 seconds for me so... 65.000 episodes would be a bit over 2 weeks if running one at a time.  

If you have any questions don't hesitate to ask.
In case i'm not here to help then @Neminemmight be able to help as he is a big user of the plugin as well and always ready to help.

Edited Friday at 08:18 AM by yocker

[fbourg 3]

Thank you for your message. I haven't applied your suggested settings yet, but with the default ones, the scan stopped twice... Here is the log file.

embyserver.txt

[yocker 1943]

14 minutes ago, fbourg said:

Thank you for your message. I haven't applied your suggested settings yet, but with the default ones, the scan stopped twice... Here is the log file.

embyserver.txt 4.74 MB · 0 downloads

I'll take a quick look but sadly don't have much time as i'm going away for the weekend, sorry.
If i don't find anything now it will have top priority when i get back!

[yocker 1943]

@fbourg Looks like the files were changed mid detection and the detections failed because of that.
In fact the whole settings page looks to have soft crashed.

If the logs don't reveal anything more then I should be able to conquer something up to help with that. 

As said i'm going away for the weekend so will take a little while, sorry about that.
I promise it will have top priority once i'm back!

Edited Friday at 12:24 PM by yocker

[enterpr1se 3]

Hi @yocker,

First off, thanks so much for adding .strm file support earlier—it’s been super helpful! Really appreciate all your hard work on this awesome plugin.

I’d like to drop another quick feature suggestion if you're open to ideas:

Since local OCR and hash detection can sometimes be a bit hit-or-miss, there are times when a newly scanned item isn't on TheIntroDB yet. However, the data might get added there later on.

Would it be possible to add an option that checks for updated intro/credit timestamps from TheIntroDB during library scans, and automatically overwrites existing local data if a mismatch is found?

No rush at all, just thought it might be a handy toggle to have! Thanks again for considering it!

[fbourg 3]

On 9/18/2026 at 10:13 AM, yocker said:

Now that i'm sitting at my computer i can answer the other questions without having to fight my phones autocorrect.

1. Chromaprint is not very accurate and i generally advice against using it.
I'm planning some time in the future to try and merge it with blackframe detection to more accurately find the end credits.
It's a rather big operation to implement that so might be a good while before that is ready.
There are settings you can fine tune for the chroma detections, sadly the defaults are the best ones i have found so far.

I strongly advice using OCR when possible.

2. Only the PaddleOCR supports Nvidia cards. If using the Tesseract it's CPU only. The plugin it self can use GFX cards for extracting the images but in reality it's not really that much faster.
With an 14700K it shouldn't be to slow (I'm running it with a 7950X3D so not that much bigger a CPU).
It's not memory intensive so 128GB is WAY overkill.
If you insist on fine tuning and not using the presets already there then I suggest using Tesseract and the default settings but change:
1. Paraelle FFmpeg Sessions to 4 (If this bogs down the machine than lower it a bit)
2. FFmpeg Thread Limit: to 4 or higher. You can set it to 0 if you are comfortable with it being allowed to take all your CPU.
3. Hardware Acceleration Type: Set to Quick Sync. (If that fails set it to VAAPI, havn't tested if Quick Sync finally works with the newest Emby)
4: Parallel Batch Size: 4 (again lower this if you find it bogs down the machine.

The rest of the settings should be pretty self explanatory as they say if they are recommend to enable or disable.

A single high quality TV show episode takes about 20 seconds for me so... 65.000 episodes would be a bit over 2 weeks if running one at a time.  

If you have any questions don't hesitate to ask.
In case i'm not here to help then @Neminemmight be able to help as he is a big user of the plugin as well and always ready to help.

I really appreciate all your work and I can't wait to see the chromatic and black frame features merged. Would upgrading to an RTX 3060 or an RTX 4060 Ti significantly improve performance or speed up the process? Thanks in advance!

[yocker 1943]

50 minutes ago, fbourg said:

I really appreciate all your work and I can't wait to see the chromatic and black frame features merged. Would upgrading to an RTX 3060 or an RTX 4060 Ti significantly improve performance or speed up the process? Thanks in advance!

With your CPU? I would say not much.
I got a 4090 and using PaddleOCR and GPU acceleration in the plugin doesn't change the speed versus CPU only enough (for me) to warrant buying a new GFX card.. Specially not with the prices today!!

The detection can ofc. be made as fast as you are able to throw money after it.   

17 hours ago, enterpr1se said:

Hi @yocker,

First off, thanks so much for adding .strm file support earlier—it’s been super helpful! Really appreciate all your hard work on this awesome plugin.

I’d like to drop another quick feature suggestion if you're open to ideas:

Since local OCR and hash detection can sometimes be a bit hit-or-miss, there are times when a newly scanned item isn't on TheIntroDB yet. However, the data might get added there later on.

Would it be possible to add an option that checks for updated intro/credit timestamps from TheIntroDB during library scans, and automatically overwrites existing local data if a mismatch is found?

No rush at all, just thought it might be a handy toggle to have! Thanks again for considering it!

No problem.  

There's some big changes to chroma detection coming soon.
OCR should be very precise if you have set up the correct keywords.

The plugin can already be set to check TheIntroDB before doing any detections. If there are any end credits on TheIntroDB it then uses those.
For anything more than that you need TheIntroDB plugin.

Edited 20 hours ago by yocker

[Baenwort 125]

On 9/18/2026 at 2:23 AM, fbourg said:

Hi everyone,

I'm looking to test the latest beta version of the EmbyCredits plugin, but I cannot find the beta .dll file on GitHub.

Could someone point me in the right direction or share where to download the latest beta build? (Is it shared via PM, attached to a specific post in this thread, or hosted on another branch?)

Setup & Hardware Specs

CPU: Intel Core i7-14700K (Quick Sync Video active, no dedicated NVIDIA GPU)

RAM: 128 GB

Library size: ~61,000 episodes

Questions regarding settings:

1. Chromaprint triggering at the beginning of episodes

When trying to use Chromaprint (audio fingerprinting), it's misbehaving and matching way too early, near the start of the video:

What are the recommended settings or search offsets to force the audio scan to focus strictly on the last few minutes of the episode?

What are good values for the minimum match duration and confidence thresholds to prevent opening themes or incidental audio cues from triggering false positive markers?

2. OCR tuning & performance scaling

With 61,000 episodes, running OCR takes a massive amount of time:

Without a dedicated NVIDIA card (no CUDA), the OCR engine is relying purely on the CPU/RAM, while frame extraction can use Intel QSV.

With 20 cores / 28 threads on the 14700K and 128 GB of RAM, what are the recommended settings (worker threads, extraction FPS, temp directory configuration) to maximize throughput without bottlenecking the system?

Are there best practices to avoid the initial scan choking the server on such a large library?

Thanks in advance for your help and feedback!

I've got a 5950x and if you are willing to give it all the CPU by tweaking settings up it could be only a couple seconds per episode.

A easy episode takes 3-5 second and a long one seems to take 12-15 seconds.

I can't remember what all the defaults are but I know I turned the thread and CPU limiting ones to use as much CPU as possible (I used scheduling to have this only run afterhours when everyone was asleep). I turned up the parallel FFmpeg sessions to 8. I also increased the number of parallel batch size to as high as it took to max out the CPU usage (This is 16 right now but I think going as high as 24 still saw improvements). The last one I tweaked was Max Concurrent Files to accommodate your storage bandwidth as increasing this seems to help a little but no where near as much as the batch size and uncapping all the throttling settings.

I then set it to go while the family went on a week long vacation and it was done before we got back. Not sure on exact time length that it took. It only had ~24K episodes after TheIntroDB plugin was done.

The fastest is if TheIntroDB has results so make sure you have that working and run through it as it saves a lot of time if you run it first. The plug-in is able to skip any that have the credits detected when you do the second run with OCR running. Every episode that gets a hit will be done twenty times faster(or more) than OCRing.

 

 

[yocker 1943]

New beta in the catalog v2.8.9.0.

Changed:
1) Much more precise chromaprint/hash detection.
2) If chromaprint/hash detection doesn't find reliable consistent audio it will no longer add the time stamps and if enabled fallback to black frame detection.
2) Better blackframe detection. I advice to turn on blackframe fallback.
3) Added more settings for the Chromeprint/hash detection, default: on.
MultiPeer Consensus Refinement
Reject Repeating Intro/Bumper Matches
Use Boundary Search for Fine Grained Refinement
Use Subframe Interpolation
Explanations available in the settings.
4) Logging for "dry run with debug" has been improved.
5) The new settings have been added to the rule system.

Fixed:
1) Detection queue could sometimes be limited to a 1000 episodes.
2) Detection runs sometimes stopping without reason should be fixed (if i have correctly identified the problem).

Bugs:
1) When using Chromaprint detection the UI can make it look like the detections have temporary stalled.
This is not a bug per say but simply lack of data from ffmpeg to react on.

[fbourg 3]

Hi,

Thank you for this new beta release (v2.8.9.0) and all the improvements made to Chromaprint and Black Frame detection!

I would like to report an issue regarding the fallback behavior when using "Enable Hash with OCR Fallback".

When Chromaprint fails because it cannot find a consistent batch match across episodes (for instance on short format series like Kaeloo), the OCR fallback does not trigger. In the failed episodes summary, the error log displays:

All detection methods failed — Chromaprint Audio Fingerprint Detection: Chromaprint requires batch comparison across episodes; no batch result was available for this episode; OCR Detection: Chromaprint requires batch comparison across episodes; no batch result was available for this episode; BlackFrame: No black frame transition detected

As you can see, the entry prefixed with OCR Detection: simply duplicates the Chromaprint error message rather than actually attempting the OCR run via the Tesseract container.

Could you please look into whether a Chromaprint batch failure inadvertently bypasses or short-circuits the OCR fallback execution?

Thanks again for your continuous work on the plugin!

embyserver.txt

[fbourg 3]

7 hours ago, yocker said:

New beta in the catalog v2.8.9.0.

Changed:
1) Much more precise chromaprint/hash detection.
2) If chromaprint/hash detection doesn't find reliable consistent audio it will no longer add the time stamps and if enabled fallback to black frame detection.
2) Better blackframe detection. I advice to turn on blackframe fallback.
3) Added more settings for the Chromeprint/hash detection, default: on.
MultiPeer Consensus Refinement
Reject Repeating Intro/Bumper Matches
Use Boundary Search for Fine Grained Refinement
Use Subframe Interpolation
Explanations available in the settings.
4) Logging for "dry run with debug" has been improved.
5) The new settings have been added to the rule system.

Fixed:
1) Detection queue could sometimes be limited to a 1000 episodes.
2) Detection runs sometimes stopping without reason should be fixed (if i have correctly identified the problem).

Bugs:
1) When using Chromaprint detection the UI can make it look like the detections have temporary stalled.
This is not a bug per say but simply lack of data from ffmpeg to react on.

I've just sent you a PM regarding this.

[yocker 1943]

1 hour ago, fbourg said:

Hi,

Thank you for this new beta release (v2.8.9.0) and all the improvements made to Chromaprint and Black Frame detection!

I would like to report an issue regarding the fallback behavior when using "Enable Hash with OCR Fallback".

When Chromaprint fails because it cannot find a consistent batch match across episodes (for instance on short format series like Kaeloo), the OCR fallback does not trigger. In the failed episodes summary, the error log displays:

All detection methods failed — Chromaprint Audio Fingerprint Detection: Chromaprint requires batch comparison across episodes; no batch result was available for this episode; OCR Detection: Chromaprint requires batch comparison across episodes; no batch result was available for this episode; BlackFrame: No black frame transition detected

As you can see, the entry prefixed with OCR Detection: simply duplicates the Chromaprint error message rather than actually attempting the OCR run via the Tesseract container.

Could you please look into whether a Chromaprint batch failure inadvertently bypasses or short-circuits the OCR fallback execution?

Thanks again for your continuous work on the plugin!

embyserver.txt 2.55 MB · 0 downloads

Thank you, i will look in to it now.
Edit: Easy fix. Uploaded a new beta in the catalog v2.8.9.1

Edited 5 hours ago by yocker

[fbourg 3]

2 hours ago, yocker said:

Thank you, i will look in to it now.
Edit: Easy fix. Uploaded a new beta in the catalog v2.8.9.1

Thanks, Look at the remaining time... it's endless... it's in hash OCR fallback.

1 hour ago

Edited 3 hours ago by fbourg
Images + time

[enterpr1se 3]

Hi everyone,

I just found an AI OCR engine that supports CPU-only execution and is compatible with EmbyCredits. Has anyone tried it before?

Also, I noticed that pulling its Docker image returns a "404 Not Found" error. Has anyone else run into this issue?

[yocker 1943]

1 hour ago, fbourg said:

Thanks, Look at the remaining time... it's endless... it's in hash OCR fallback.

1 hour ago

Yes, it can take a long time depending on what system it runs on.

You can try and enable black frame fallback and use hash (Chromaprint) detection only.
That should speed things up.





I need to pull my self together and rename that option.

[yocker 1943]

8 minutes ago, enterpr1se said:

Hi everyone,

I just found an AI OCR engine that supports CPU-only execution and is compatible with EmbyCredits. Has anyone tried it before?

Also, I noticed that pulling its Docker image returns a "404 Not Found" error. Has anyone else run into this issue?

I don't really know tbh.
It was another user on this forum that made what looked to be a fork of EmbyCredits but claimed some other things i didn't really get.
I guess he deleted it all again.

Disclaimer: People are allowed to do what ever they want with this plugin, it is released under the MIT license so only thing required is that the copyright is kept.

[fbourg 3]

New problem ?

[fbourg 3]