A new tool - artificial intelligence to standardize actor images

[chef 3746]

2 hours ago, slevin7 said:

if you like, i could test this on ubuntu and macos. 

 

Cool. Yes, we need some cross platform tests.

 

[chef 3746]

Cool check this out

 it's still early but it's cool.

 

 

Edited September 7, 2022 by chef

[sydlexius 243]

22 hours ago, chef said:

Cool. Yes, we need some cross platform tests.

 

Maybe you can have a chat with @alucryd about how to package this up into a Docker container (I don't know much about how to create "compose" files)?

[sydlexius 243]

On an unrelated note, I sure wish there was a way to identify which actors get stuck with character art (especially of the animation-related kind), so I can go and replace it.  I've done my best on the metadata fetching side, but that hokey stuff still makes its way through!

[chef 3746]

Whoa! So busy. Two 'all nighters'! 

I built a web app, and it has an API. I have never done that before. It is a huge undertaking, but I did it. 

This is the web page that appears when you go the url. 

You authenticate with Emby. Then you are shown the last 10 results of the background standardizer.

Emby can request the images directly from the API, or this app can do a full replacement of the actor images in the file system.

You can choose to keep both files if you don't like the results, so there is a "Go Back" feature, but it does double the image count so you'll  have to have a large file system...

But once you confirm the results, this will do a clean up.

 

TODO:

• Choose to see processed actor list by movie/TV Show

 

Edited September 8, 2022 by chef

[chef 3746]

Some really exciting stuff here... it might not seem like it. But, it is.

So, I'll be able to put this out there.

I have now found the best way to show actors.

Next, I'll add the process button to process all.

It's so close to being done.

[slevin7 54]

Pretty amazing.

I am really looking forward to this 

[chef 3746]

 

Emby server discovery and login

[chef 3746]

Yeah, I'm experimenting with images, and I think after going back and fourth between liking it and not liking it...

I find that the darker image backgrounds look best. 

A lot of work for a simply... not so simple change.

            

             

 

              

 

         

Now that the entire web dashboard loads from inside the library... it should be very easy to run on linux.

 

 

[TheGru 38]

This is amazing!!

[pünktchen 1258]

Great stuff @chef! Can you also resize the images so they are always 2:3 please? I have many older images that have a strange aspect ratio. And if the AI allows it, zoom in images where you can see too much of the persons body like the first image in your last post. That would be the cherry on the cake.

[chef 3746]

5 minutes ago, pünktchen said:

Great stuff @chef! Can you also resize the images so they are always 2:3 please? I have many older images that have a strange aspect ratio. And if the AI allows it, zoom in images where you can see too much of the persons body like the first image in your last post. That would be the cherry on the cake.

I think it can do that yes.

Hey... Wanna try it?

[pünktchen 1258]

Of course!

[Cheesegeezer 3087]

On 18/09/2022 at 19:13, chef said:

Yeah, I'm experimenting with images, and I think after going back and fourth between liking it and not liking it...

I find that the darker image backgrounds look best. 

A lot of work for a simply... not so simple change.

            

             

 

              

 

         

Now that the entire web dashboard loads from inside the library... it should be very easy to run on linux.

 

 

Lol was just having a laugh at Sienna Miller in your basement torture chamber backdrop lol

[TheGru 38]

My Emby system is running in a Docker container on Unraid, I'd love to test this out and report back. Let me know what I may need.

[chef 3746]

9 minutes ago, TheGru said:

My Emby system is running in a Docker container on Unraid, I'd love to test this out and report back. Let me know what I may need.

Cool. Yes, Im not sure how to make a docker container yet.

 

[chef 3746]

46 minutes ago, Cheesegeezer said:

Lol was just having a laugh at Sienna Miller in your basement torture chamber backdrop lol

I know sooo funny. 

[TheGru 38]

16 hours ago, chef said:

Cool. Yes, Im not sure how to make a docker container yet.

 

Don't worry about that, I can spin up a virtual machine running whatever OS is needed to give this a spin against my Emby server container.

[nagetech 66]

I would definitely be interested in this as a plugin! 

Following this topic! 

[chef 3746]

Sure thing. Except that it's not an Emby plugin.

It actually creates a local web service provider that communicates with emby and then creates the images for emby.

The thing is massive. 

I actually just changed the code to save the overlay images for each actor using their tmdb ID.

Reason being, I can ship this tool with a bunch of actor image overlays already processed with background removal, and this will allow the user to build their new collection using their custom backgrounds faster.

I ran this on my own 10 000 actor images to create the cutouts.  It will be in a zip file, which can be used for people who don't have the GPU or processing power to crunch their own actor db.

It's just sorting the images now.

Obviously, it's an option you can use, but you can also crunch your own db if you want to. 

I'll see how quickly I can test these new changes.

BRB! 

 

[chef 3746]

Darn, swapping IDs is a massive undertaking. Especially since the only connect with emby is through the front end. The backend is a microservice that actually has no idea what emby is. It just looks for image urls to process.

So TMBD IDs will be for the future. 

Right now processing the person db will have to do. This makes it proprietary to emby anyway, which is the way I like it.

Brb! 

Edited September 21, 2022 by chef

[chef 3746]

I've got a lot to work on UI still.

 

These images don't upload to Emby yet. They are only created in a folder because there are still some images that don't  process well.

 

[chef 3746]

So, I'll make each image a button which will navigate to emby's actor details page where you can change the image form inside emby.

At that point, when the process is run again, the app will see the change for the image and update it.

 

Because.... that guy in the picture below. LOL!

[sydlexius 243]

8 minutes ago, chef said:

So, I'll make each image a button which will navigate to emby's actor details page where you can change the image form inside emby.

At that point, when the process is run again, the app will see the change for the image and update it.

 

Because.... that guy in the picture below. LOL!

At some point in time, are you going to add EXIF data to the modified images to indicate items such as name of the background file chosen, version of the AI tool that applied the modifications, as well as a timestamp of when the modifications were performed?  Additionally, some data about the model/person DB might be useful too (count of source images, for example).

Edited September 21, 2022 by sydlexius

[chef 3746]

37 minutes ago, sydlexius said:

At some point in time, are you going to add EXIF data to the modified images to indicate items such as name of the background file chosen, version of the AI tool that applied the modifications, as well as a timestamp of when the modifications were performed?  Additionally, some data about the model/person DB might be useful too (count of source images, for example).

Yeah, that is a good idea.

Right now when an image is processed it takes 10bits of the original image base64 data and encodes it to md5. That way I can know if the image has changed on the server, and if it needs to extract the overlay again. I only take 10bits because it is small enough to save, and long enough to know if the image has changed.

Exif data will also be a good idea.