Update README.md

README.md

@@ -442,10 +442,9 @@ With those way smarter resources out of the way, I'll try to produce a simple gu
    This is the largest part of Lora training. You will need to create a "dataset" of images to train with, along with corresponding text files containing descriptions for those images (tags in the case of anime).
 
    1. Find some images online representing the character/artstyle/concept you want to convey, possibly on sites such as [gelbooru](https://gelbooru.com/). You will need at least 10 images, I'd recommend at least 20, but more is almost always better.
-      * Optionally, you can get hundreds of them using [Grabber](https://github.com/Bionus/imgbrd-grabber/releases).
-      * If you want to do a character, I recommend filtering tags like this: `1girl solo character_name score:>10 -rating:explicit` (the explicit rating may include weird images, so it's fine to exclude them)
+      * Optionally, you can get hundreds of them using [Grabber](https://github.com/Bionus/imgbrd-grabber/releases). If you want to do a character, I recommend selecting gelbooru and pixiv, and filtering tags like this: `1girl solo character_name score:>10 -rating:explicit` (the explicit rating may include weird images, so it's fine to exclude them)
   
-   1. If you only have a few images you may tag them yourself, but it may be slow and inaccurate. If you're using photographs you should describe each one in detail using simple sentences.
+   1. Create your text files next to each image, with the same filename. If you only have a few images you may tag them yourself, but it may be slow and inaccurate. If you're using photographs you should describe each one in detail using simple sentences.
       * Optionally, add the [Tagger extension](https://github.com/toriato/stable-diffusion-webui-wd14-tagger) to your webui, through which you can automatically analyze all your training images and generate accurate anime tags for them. Instructions are as follows: First add and enable the extension, and restart your entire webui. Then go to the new **Tagger** tab, then *Batch from directory*, and select the folder with your images. Set the output name to `[name].txt` and the threshold at or above 0.2 (this is how closely each tag must match an image to be included). Then **Interrogate** and it will start generating your text files.
   
    1. Once your images and their text files are ready, put them all in a folder following this structure: A folder with your project name, containing at least 1 folder in the format `repetitions_name`, which each contain some images and their tags. Like this: