a1111-sd-webui-tagcomplete

This release adds support for the new in-prompt hypernetwork and Lora notation used in the current webui version.

The old method of < will show Embeddings, Hypernetworks and Loras mixed together for convenience.
You can use <e:, <h: and <l: respectively as a prefix to limit completion to that type.
Filtering embeddings by version using v1/v2 as a prefix still works for both types, so <v1 or <e:v1.

Thanks to @Mulet-J for his first draft of the feature in #115!

The webui now uses proper sha256 hashes and has changed the way in which they are displayed. This release fixes the script to work with the new format.

Added:

• Ability to use model name in black/whitelist setting
  • Mixing names and hashes works too

Fixed:

• Both short (first 10 characters) and long form hash now work again in black/whitelists

Fixes the ? wiki links being shown for embeddings and wildcards as well, instead of just for tags like they were meant to.

Added:

1. New settings option for (de)activating the script based on a model hash black/whitelist. #14
2. New settings option for showing ? links next to the tags linking to their danbooru / e621 wiki pages. #109
   • Note: This is disabled by default since it links to an external site and the wiki pages more often than not contain NSFW examples of recent posts for that tag.

Fixed:

If the prompt was only one letter long and identical to the previous completion (e.g. due to accidentally deleting the first letter or changing your mind twice in a row), completion would not show. The script now correctly resets its internal state if the last letter is deleted.

This release includes:

• Some more fixes and adjustments for embedding loading not included in the previous release
• Small styling adjustments
• More utility functions moved to separate file for better code readability
• Results now use a proper object instead of array indices
  • This shouldn't change anything for now, but will make it easier to add more optional metadata in the future for various different completion types (if needed). Also massively helps with readability and is less error-prone.

Displays version metadata for embeddings & allows filtering for it by prefixing the search with v1 / v2.
This makes it easy to find matching embeddings for sd-v1.x or 2.x models respectively. Implements #97.

https://user-images.githubusercontent.com/34448969/210187734-c2e2292c-7649-4df6-b787-2b2195ab3a29.mp4

Note:

I wait for the webui to load the embeddings to then get their version info, but if the model doesn't load any (e.g. a v2 model, but the user has no v2 embeddings, or the same with v1), I can't test what version they are and default to v1 since it's more common.
I could theoretically check them manually, but that would mean embeddings being loaded twice which would increase startup time, so I compromised here.

Adds support for https://github.com/Klokinator/Umi-AI, which uses a yaml-based tagging system for dynamic wildcard template selection.
Completion takes previous Umi tags into account to only show possible candidates, which allows progressive filtering to close in on wanted templates.

See #84, #88, #90, #93.

Thanks to @ctwrs for the help.

Adds base functionality for autocompletion support in third party text areas added by other extensions or scripts.
Includes https://github.com/toshiaki1729/stable-diffusion-webui-dataset-tag-editor so far.
Adding new third party options is easily achieved by adding them in the thirdParty dictionary of the new _textAreas.js file, with their respective IDs or captions.
Feel free to open issues or pull requests for additional third party support.

Includes further fixes for #62 and #65.

Translations now need to be in a separate file to still be visible when typing the English tag counterpart in the new alias system.
See https://github.com/DominikDoom/a1111-sd-webui-tagcomplete#translations and the new config options related to it.

This new system is a bit more involved, but also allows for translating aliases, which wasn't possible before.

This release contains large updates to the default tag data snapshots, enabling the script to show both post count and search using aliases.

New

• Search using aliases (e.g. alias cat_eyes redirecting to the actual tag slit_pupils)
• Display post count for a given tag if available
• Highlight the matched part in the tag

Fixed

• After composing a word using an IME editor (e.g. for inserting Chinese or Japanese text on a Latin keyboard), the result list would instantly close after pressing enter to confirm the written word in the IME.

Changed

• Translation config options have been renamed to use "alias" instead, since it uses the same method in the code, but aliases are a superset of translations.
  • Translations can still be added like before and just add another alias for their tag.

• Fixes regression of editing tags in parentheses (#59)
• Forces cache workaround, which should lead to fewer cache-related issues

New

• UI and config option to switch whether commas should be added after tags on insertion (#49)
• Support for comments in wildcards (#52)

Fixed

• Fix parsing for words starting or ending with non-ASCII characters (#51)
  • Especially important when used to search translations

Fixes parenthesis parsing broken by a038664616b7a8258ca80a10c1c1ae3595ee68a9 on accident.

(1.9.1 was replaced since it contained a half-broken fix)

Contains a reworked tag matcher to allow editing tags in weighting brackets: (tag), [tag] and (tag:<number>).

Contains Linux bug fix for #46 as well as a fix for wildcards being loaded twice if the extension folder was also called wildcards.

Now supports autocompletion in multiline prompts, whereas before it would break for anything after the first line.

The script now works if you install it as an extension instead of in the root (#41).

Now also supports loading wildcards from extensions regardless of the extension name, including combining multiple wildcard sources.
This will work with every extension that has a wildcards folder in its root (#37).

Due to large amounts of separate wildcard files leading to bad performance issues with the previous method, they are now loaded on demand instead of preloaded.

Support for wildcards in the new extensions folder location (#35). Will only work for files in extensions\wildcards\wildcards and its subfolders at the moment, so you might need to rename the extension folder if you cloned it with another name.