hollowstrawberry
/

stable-diffusion-guide

@@ -489,7 +489,7 @@ With those way smarter resources out of the way, I'll try to produce a simple gu
    The most important thing for characters and concepts is the tags. You want a dataset with varying poses and scenarios, but if they're tagged incorrectly it's not gonna work.
    When training a character or concept you should place an **activation tag** at the start of every text file, and set `keep_tokens` to 1 before training. An activation tag is how we'll invoke your Lora to work. Having done that, **some people reccommend** removing or "pruning" tags that are intrinsic to your character or concept, such as hair color and eye color. For example, if a character always has cat ears, you would remove tags such as `animal ears, animal ear fluff, cat ears`, etc. This way they become "absorbed" by your activation tag, and makes your Lora easier to use, but less flexible.
-   * You can use the [Tag Editor extension](https://github.com/toshiaki1729/stable-diffusion-webui-dataset-tag-editor) to add an activation tag. Instructions are as follows: After adding the extension and restarting your webui, go to the new **Dataset Tag Editor** tab then *Batch Edit Captions*. Write your activation tag under *Edit Tags* and turn on *Prepend additional tags*. Then apply your changes, scroll up and save your changes.
    This "absorption" of details not provided by tags is also how Loras work at all, by representing things normally imperceptible or hard to describe like faces, accessories, brushstrokes, etc. If you desire you may also prune redundant clothing tags, such as removing "red tie" but keeping "tie". You can even define an additional activation tag for different outfits your character may wear regularly, eg. character-default, character-bikini, etc. But there's more than one way to do it. In any case, with the correct usage of tags, your character should easily be able to change clothes.

    The most important thing for characters and concepts is the tags. You want a dataset with varying poses and scenarios, but if they're tagged incorrectly it's not gonna work.
    When training a character or concept you should place an **activation tag** at the start of every text file, and set `keep_tokens` to 1 before training. An activation tag is how we'll invoke your Lora to work. Having done that, **some people reccommend** removing or "pruning" tags that are intrinsic to your character or concept, such as hair color and eye color. For example, if a character always has cat ears, you would remove tags such as `animal ears, animal ear fluff, cat ears`, etc. This way they become "absorbed" by your activation tag, and makes your Lora easier to use, but less flexible.
+   * You can use the [Tag Editor extension](https://github.com/toshiaki1729/stable-diffusion-webui-dataset-tag-editor) to add an activation tag to all files at the same time. Instructions are as follows: After adding the extension and restarting your webui, go to the new **Dataset Tag Editor** tab then *Batch Edit Captions*. Turn off "Show only the tags...", turn on "Prepend additional tags", then add your activation tag inside the *Edit Tags* text box. Then apply your changes, scroll up and save your changes. This will add a new tag at the beginning of every text file.
    This "absorption" of details not provided by tags is also how Loras work at all, by representing things normally imperceptible or hard to describe like faces, accessories, brushstrokes, etc. If you desire you may also prune redundant clothing tags, such as removing "red tie" but keeping "tie". You can even define an additional activation tag for different outfits your character may wear regularly, eg. character-default, character-bikini, etc. But there's more than one way to do it. In any case, with the correct usage of tags, your character should easily be able to change clothes.