Has Anyone Integrated AI With Emby? Looking for Real-World Tools, Setups & Dev Plans

[dexus 31]

I am not going to get into the Ai vs no Ai argument here.. Truth is there is too many facets to that discussion.  Me personally see it as a tool as any other, and use it both professionally and privately.  If anyone wishes to not use it, good on you. I won't try to convince you.

That said.. I have made myself a tool that I use to get Norwegian subtitles when none exist, based on existing subs in other languages (any language in fact). I DOES use AI, because a good LLM is needed to get great translation, as it understands context and slang etc, making the subtitles feel normal and with a good flow. In my opinion, it makes virtually flawless subtitles.  It is very easy to use, and you can utilize both API calls to external LLM's or local if you have hefty hardware.. but to be honest, API calls to a very capable LLM is the way to go as of now, if you want flawless subs.  I use DeepSeek API, and yes.. I know.. it's chinese.  I don't care.  It is by far the cheapest option, and costs virtually nothing to use compared to other options.. And ALL i send them is subtitles to be translated, nothing else.   If you would like to check the project out, it is here: https://github.com/dexusno/Translate_Subs

It is of course fully open source with an MIT license so you can do with it what you want, and is python based.. with wrappers for powershell and bash (debian / ubuntu). AND of course you can translate to ANY language of your choice, from any language that is available.

To be clear, I do respect everyone that does not want to use AI, but the truth is, we can't put the cat back into the bag, it's here to stay.  And I for one, will use the tool, rather than ignore its existence.   Still, everyone must decide for themselves.  I also believe that AI is not needed in the base Emby code, but plugins are always welcome.. That way everyone has a choice.

Edited 22 hours ago by dexus

[dexus 31]

On 2/4/2026 at 3:48 AM, yocker said:

I would rather have an AI create subtitles than painstakingly making them my self.
In fact i often have to using Bazarr and whisper because subtitles in Danish can be very hard to find.

Hey, my Danish neighbor! check out my translation project using AI here: https://github.com/dexusno/Translate_Subs It works like a charm, and DeepSeek API is so cheap it's ridicolous.

[BillOatman 597]

2 hours ago, dexus said:

I am not going to get into the Ai vs no Ai argument here.. Truth is there is too many facets to that discussion.  Me personally see it as a tool as any other, and use it both professionally and privately.  If anyone wishes to not use it, good on you. I won't try to convince you.

That said.. I have made myself a tool that I use to get Norwegian subtitles when none exist, based on existing subs in other languages (any language in fact). I DOES use AI, because a good LLM is needed to get great translation, as it understands context and slang etc, making the subtitles feel normal and with a good flow. In my opinion, it makes virtually flawless subtitles.  It is very easy to use, and you can utilize both API calls to external LLM's or local if you have hefty hardware.. but to be honest, API calls to a very capable LLM is the way to go as of now, if you want flawless subs.  I use DeepSeek API, and yes.. I know.. it's chinese.  I don't care.  It is by far the cheapest option, and costs virtually nothing to use compared to other options.. And ALL i send them is subtitles to be translated, nothing else.   If you would like to check the project out, it is here: https://github.com/dexusno/Translate_Subs

It is of course fully open source with an MIT license so you can do with it what you want, and is python based.. with wrappers for powershell and bash (debian / ubuntu). AND of course you can translate to ANY language of your choice, from any language that is available.

To be clear, I do respect everyone that does not want to use AI, but the truth is, we can't put the cat back into the bag, it's here to stay.  And I for one, will use the tool, rather than ignore its existence.   Still, everyone must decide for themselves.  I also believe that AI is not needed in the base Emby code, but plugins are always welcome.. That way everyone has a choice.

Nice!  I wrote this one that creates subs from the audio that uses Whisper

SubtitleCreator

[dexus 31]

1 minute ago, BillOatman said:

Nice!  I wrote this one that creates subs from the audio that uses Whisper

SubtitleCreator

LOL!  I am actually working on one that does that as well, but also uses api calls to an LLM to fine-tune the subs, and keeping them in the "netflix rules" for subs, correcting grammar, punctuation and shortening sentences that are longer than the standard etc.  Refinement stage.

In the translation project there is alot more to it than meets the eye.. it uses quite advanced processes to ensure that we have bulletproof timing and that the llm doesn't f*ck it up.

[Neminem 1626]

@dexus

Nice more options are always nice

I have used this for 1 or 2 years now.

lingarr-translate/lingarr: Lingarr is an application that supports both local and SaaS translation services to translate subtitle files into a specified target language. With automated translation options, Lingarr simplifies translating subtitles.

With a combination of Bazarr and Whisper for those pesky not found subs in any language.

It runs in docker, so that might not be for all.

But using that with libretranslate.

This is mostly because I strip out all embedded subs from my media.

And only use .srt subs. 

Edited 19 hours ago by Neminem
adding

[yocker 1427]

5 hours ago, dexus said:

Hey, my Danish neighbor! check out my translation project using AI here: https://github.com/dexusno/Translate_Subs It works like a charm, and DeepSeek API is so cheap it's ridicolous.

Hej der!  

For those i can't find i just use whisper, its free. 
But nice project.

[dexus 31]

13 hours ago, yocker said:

Hej der!  

For those i can't find i just use whisper, its free. 
But nice project.

Hej! I actually made a project to use whisper as well..  But as I am a bit of a perfectionist, I was not very happy with the pure whisper output.. so I made it a 5 step process, where whisper first transcribes, then some smart logic and 2 passes to a LLM gets the subs to where I would like them.. near perfect.  The project is here: https://github.com/dexusno/Transcribe-Subs

You can read about the inner workings of the project in the Readme.md on git..  It was fun to develop.

[dexus 31]

17 hours ago, Neminem said:

@dexus

Nice more options are always nice

I have used this for 1 or 2 years now.

lingarr-translate/lingarr: Lingarr is an application that supports both local and SaaS translation services to translate subtitle files into a specified target language. With automated translation options, Lingarr simplifies translating subtitles.

With a combination of Bazarr and Whisper for those pesky not found subs in any language.

It runs in docker, so that might not be for all.

But using that with libretranslate.

This is mostly because I strip out all embedded subs from my media.

And only use .srt subs. 

I actually tried lingarr based translation, but wasn't satisfied with the quality.  Also had a project where i used NLLB-200, the translating engine from Meta, which translates between all the FLORES languages, and actually trained it on norwegian and english subs to get it better at subtitle translation, but in the end, even if it was pretty good, my newest version using a mix of smart scripts and LLM processing is much better.. nearly flawless.  Libretranslate was a fallback option in the NLLB-200 project, and it also had a web-interface with drag and drop etc. It is here: https://github.com/dexusno/srtxlate  And if you want to use local models only, probably the best i have seen.  But as deepseek api is so incredibly cheap, and also produce so much better results, I use the new project instead.  I mean, you can literally translate hundreds of episodes for under 5 usd in api calls.. 

My newest project for files where subs are not available at all, uses a 5 step process using whisper v3 large in conjunction with local smart scripts and LLM calls to extract audio, transcribe, punctuate correctly, clean up and catch hallucinations by whisper using professional grade subtitle standards as a framework.. It works very well. https://github.com/dexusno/Transcribe-Subs

[yocker 1427]

16 minutes ago, dexus said:

Hej! I actually made a project to use whisper as well..  But as I am a bit of a perfectionist, I was not very happy with the pure whisper output.. so I made it a 5 step process, where whisper first transcribes, then some smart logic and 2 passes to a LLM gets the subs to where I would like them.. near perfect.  The project is here: https://github.com/dexusno/Transcribe-Subs

You can read about the inner workings of the project in the Readme.md on git..  It was fun to develop.

Sorry i meant to say i use "Bazarr".
99% of the stuff i find have english subtitles, no problem to translate with Bazarr for that 1%.  
Also the only place where i and my users mostly use subtitles is for anime, never translate them to danish.. I don't know, just feels wrong watching those with danish subtitles for some strange reason. Looks strange watching "ILD KUGLE" being written every time someone attacks some one else.  

[dexus 31]

22 minutes ago, yocker said:

Sorry i meant to say i use "Bazarr".
99% of the stuff i find have english subtitles, no problem to translate with Bazarr for that 1%.  
Also the only place where i and my users mostly use subtitles is for anime, never translate them to danish.. I don't know, just feels wrong watching those with danish subtitles for some strange reason. Looks strange watching "ILD KUGLE" being written every time someone attacks some one else.  

Lol! I see what you mean with "Ild Kugle!" that would be just wrong..  I just feel that Bazarr, even if it translated pretty good, it still feels unnatural often.. like machine translating sentences like a book, not like a spoken dialog.  like if the original text is "Hi, do you want coffee?  No, I'm good" would translate to "Hej, vil du have kaffe?,  Nej, Jeg er god" and so on.. It does not understand slang or spoken dialog as well as an llm..