 So you know how to use simple captioning, but you need something a bit more advanced. Not a problem. Real quick, a lot of what we're trying to do is emulate the workflow examples that we see from Civet AI. So if you'd like to see more examples of this stuff, you can find them if you follow the link in the pinned comment. First open Koya and under WD-14 captioning, click the folder and set it to your collection of images. Type in your basic prefix sentence. In my case, it's the keyword woman and green background, and then you add a post fix. Prefix is really only used if you just want to make sure that some tags are present regardless of mixing and auto caption. It's not necessary, but sometimes it helps. In my case, I'm just going to use realistic and 3D model. Also make sure to pay attention to your commas here. There should be a comma at the end of your prefix and at the beginning of your post fix. Now scroll down here and the only things you really need to worry about here are general and character threshold. Basically, the smaller these numbers, the more tags it will generate per image. So if you set it to zero, it will give you a bazillion different tags and it's going to be chaos. Point five is pretty good for more complex characters, but if you want more simple captions that are less strict and more flexible in the long run, we're just going to set these to point six. Press caption images and now the computer has gone through all of your images and came up with its own caption for each image. Now we're going to download a tag manager. If you follow the link in the description, click releases and download this zip file right here. Then drag the content to wherever you want. I personally prefer to keep it next to the Koya file. Once you have it, run the application and open the file with all of your images. Now from here, it will automatically load all your images and allow you to see the captions that the computer has generated for them. And you can see it will have done a fairly decent job describing each of them. For example, it picked up that this image was from the side view and that this one was from behind and it's done that for every single image. So now the only thing we really need to do is clean up the details on the right side contains all the tags used in all of your images combined. But the middle row contains the tags only for your selected image. These icons over here are just tools that you can use to adjust the tags of each of your images. Most of them are pretty self-explanatory like copy, move tag up, move tag down. But the most important ones are the X to delete a tag or the plus to add a new one. If you delete a tag like this, it will only remove it from the currently selected image. If you delete it on the right side, though, it will remove the tag from all images. So be careful when you're doing stuff over here. Now let's say that you wanted to replace the tag for hat with headwear, which is a bit more flexible across all of your images. Well, if you click this button, then under new tag, tell it what you want to replace it with. And now the tag hat has been replaced with headwear across all your images. Now one thing to keep in mind is the order of these tags is pretty important. Things at the top hold more weight than the things at the bottom, which is why our keyword is always first. And you'll probably notice that our basic prefix of woman green background is always at the top. And our post fix realistic and 3D modeling is always at the bottom. So it looks like everything is working pretty well. Okay, so let's talk about what should you be looking for when you're doing this process? Well, two things. First, you just want to check and remove false positives. For example, we're trying to train our AI on this female character. If there are words that should not be there, remove them. For example, if we found the word tiger in here and there clearly is no tiger in the picture, be sure to take it out. Likewise, if there is a tiger in the picture, and we have no tiger tag in the caption, the AI might misinterpret and assume that the tiger is actually part of our character. So remove things that shouldn't be there and add things that should. Now when you're doing this, and you get a tag that is just absolute nonsense, you probably want to remove it using the tools on the right side to get rid of the entire thing in all images entirely. So if you see a word like web address, then you can probably just cut that out from the whole set. The second main thing that you want to check for is repeating words. For example, if you find things like skirt and pleated skirt, just have skirt. Or if you have things like gloves and brown gloves, just have gloves. Here you can see I have one for high heels and high heel boots. We can go ahead and delete the boot version because we already have boots up here. So remove false positives and remove duplicates. Once you've done that, of course, make sure to save your work and you're done. Congratulations. Your data set is now ready to enter the training phase, which I'll show you how to do in the next video. But in the meanwhile, as always, I hope you have a fantastic day and I'll see you around.