<p>Important instructions below! Make sure you read the "How to use?" section!</p><h2>What is this?</h2><p>This is a collection of wildcards with extensive, customized lists of characters, clothing, location, pose, etc all logically nested together in order to automatically form a complete prompt. This includes character, clothing, and other loras, over 100 of them each tested and configured to have minimal impact on the style and composition of the image while retaining the lora's elements. All example images were generated with the same prompt/template or very similar. Just "__tt-all__" with a bit of flair like "4k wallpaper".</p><p>The original idea behind this system was to automatically generate myself a wallpaper each day or something, but it can be used for any purpose. Everything was tested to make sure it works well at 16:9, 1920x1080, but portrait or square ratios will only work better because wide ratio is the least supported by most models and loras.</p><p>Don't want to just automate the whole prompt? You can use individual elements as you like, "__tt-character__", tt-location, etc.</p><p>It's also nice to have a list of well configured characters or clothes to retrieve as needed with the auto complete extension like this:</p><img src="https://imagecache.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/9664b079-e942-434c-f4a3-c1425be84400/width=525/9664b079-e942-434c-f4a3-c1425be84400" /><p>You could also view this as more of a configuration list for a bunch of loras along with their trigger words, rather than just a set of wildcards.</p><h2>How to use?</h2><p>You will need 2 extensions first.</p><p>Note: It seems that the basic wildcard extension, the one given by civitai, conflicts with Dynamic Prompts, so remove that first if you have it. Dynamic Prompt extension makes it completely obsolete.</p><p>"<a target="_blank" rel="ugc" href="https://github.com/adieyal/sd-dynamic-prompts.git">Dynamic Prompts</a>" extension to use wildcards. Put the wildcards .txt files in the wildcards folder inside the extension folder.</p><p>"<a target="_blank" rel="ugc" href="https://github.com/hako-mikan/sd-webui-lora-block-weight.git">LoRA Block Weight</a>" extension to use custom lora weights. You will also need to add a custom preset that I used on most loras; add this line to the Weights Settings in the extension's menu then press Save Presets:</p><p>TT:1,0.2,0.2,0.2,0.5,0.5,0.5,0.5,1,1,1,1,0.5,0.5,0.5,0.5,0.5</p><p>I also highly recommend you get the "<a target="_blank" rel="ugc" href="https://github.com/DominikDoom/a1111-sd-webui-tagcomplete.git">Booru tag autocompletion</a>" extension, it's very helpful as it can autocomplete wildcard names and search through them for an entry.</p><p>And then the elephant in the room; if you want to use the collection as is you'll need the 100+ loras in it. I thought it would be a bit much to list all of them with links and expect people to download all of them that way (also that's more work and some are not available anymore), but I don't know about putting a huge archive of them on here on my own page either. For now I've included links to mega and google drive folders where you can either download whole folders or select the loras you want. Of course, you can also keep only the characters you like and add your own.</p><p>Some embeddings are also used and included in the archive.</p><p>Once set up, use wildcards in the prompt by putting 2 underscores around filenames, like "__tt-all__" for a complete prompt. All of my custom wildcards have the prefix "tt-" (some of the regular wildcards are used inside those, so they are also needed). Also, because wildcards are nested inside each others, you can't change the default wildcard wrapper unless you replace all of them.</p><h2>What checkpoint and settings to use?</h2><p>Use whatever you like. More realistic models probably won't work too well, but semi-realistic works fine, and anime models work best. My favorite is Kawaii25D, but it seems to have been deleted. I might reupload it if requested, but I don't know if that would be an issue. I don't really believe SD models belong to anyone, but eh. As for settings/negative prompt, you can use the examples as reference, but there's no reason your preferred settings wouldn't be better.</p><h2>Important Notes</h2><p>tt-all includes some basic quality tags. You can add some, or remove tt-quality from it if you want to manage quality tags only in prompt.</p><p>While there's no explicit sex acts included, there will be revealing clothes and nudity, so if you don't want to lewd certain characters, you'll have to remove them or something. For your convenience I've added "smol" and "nosmol" variants to all relevant wildcards, so you can use "__tt-all-nosmol__" or "__tt-character-named-nosmol__", etc. Separating them was a pain, and I've used the character's physique to do so because like 80% of anime characters are underage so that would be even more stupid. Basically Tatsumaki is smol, but most schoolgirls aren't.</p><p>If 2 or more characters are generated with a character lora in use, they will almost always be clones. This is not something I can do anything about and is an inherent characteristic of SD. Even without loras, characters will often look very similar. You can only try to prevent multiple characters from appearing; add 1girl, solo, etc. (not included in wildcards) Or use portrait ratio.</p><p>When testing and adding character loras, my objective was to add only the necessary descriptive tags for the character to be consistent enough, and retain as much flexibility as possible for other elements and varied image composition. You are probably aware that adding descriptive tags like eye color will heavily affect other elements, but you might not know the extent of it. For example specifying eye color doesn't just change other things' color, it also makes it much more likely for the character to face the viewer to show the eyes in the image, thus restricting composition. Ideally adding only the lora and character name is enough. As a result, you might sometimes get wrong eye color or something, but I think it's much better this way.</p><p>This collection is biased towards my preferences. My favorite characters and clothes appear more often, the tags I think work best are used, etc. Edit to your liking.</p><h2>Tips</h2><p>Want to add some creative spin to the image? Try adding something like " AND abstract:0.8" at the end of the prompt. This will add a layer of colorful abstract art and can cause some very interesting results that would never be generated normally, however it also makes nonsense like extra limbs happen more often. It's like blowing things up and hoping for something interesting to result from it, good for something more dynamic and action-oriented:</p><img src="https://imagecache.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/7dfe5524-fb1c-4bf0-1be5-b106f2325e00/width=525/7dfe5524-fb1c-4bf0-1be5-b106f2325e00" /><p>Speaking of extra limbs and other errors, they happen a lot more often in wide image ratios, and then if you add "lying" pose on top you end up with mangled bodies more often than not. I don't know of any way to prevent that other than not using wide ratio or removing lying pose from the list. Do as you like. You can try some negative embeds, but I've never seen them fix much and they all have more downsides than upsides in my opinion (more restrictive generations).</p>

Files with SHA256: eb90080e5e5e0463dd3f1042493a4bd5e9b27ebdfc4d7f241bc1c32fe9bb7c45