All previews don't use any other LoRAs. The Illustrious previews look mostly 3D because I was just starting to learn how Illustrious models work and there are no artist tags.<h1 id="commercial-use-of-this-model-is-not-allowed-zu7m165c3">❌💲 Commercial use of this model is not allowed</h1><h2 id="the-description-is-written-mainly-with-the-pony-v2-and-illustrious-v1-versions-in-mind-some-things-may-not-work-or-be-accurate-for-the-others.-58tz5jdhk">The description is written mainly with the Pony V2 and Illustrious V1 versions in mind, some things may not work or be accurate for the others.</h2><a target="_blank" rel="ugc" href="https://civarchive.com/models/835046/prof-worminkle-pdxl">Professor Worminkle has his own LoRA because this doesn't get his appearance right.</a>Yes, nearly all the base models this is trained for are SDXL family, but that's still too many to list in the title.I started working on the dataset for this model even before doing so for my <a target="_blank" rel="ugc" href="https://civarchive.com/models/540030/test-worms-team17-render-mix-sdxl">previous one</a>. I gathered over 800 images, mostly due to customisation, expressions, angles and stuff Worms hold."Why do Worms look so low-poly/angular here?" It's a cartoony 2005 game, and in the 2011 remaster they didn't modify the models at all.<h3 id="issues-o3234wmee">⚠ISSUES</h3><ul><li>Risk of wrong number of eyebrows/pupils.</li><li>Brown eyes tend to be blurry since their texture is in very low resolution. Eyes in any other colour should be slightly better quality.</li><li>Worms may get complete arms.</li><li>By default worms have very simple hands, there are some unusual hands with claws and grotesque fingernails, but worms may get fingernails unprompted.</li><li>Worms may get human ears (especially when they're generated with hair).</li><li>If you ask for a frowning worm, the pupils may get deformed because that's what happens in the actual game.</li><li>Worms' heads get severely deformed upside down. Worms' heads may also look very wrong from behind, with a chance of mouths and eyebrows on the back of head.</li><li>If you don't specify eyebrows to be black they may change colour. In rare cases, eyebrows may be transformed into stuff like antennae, horns or ossicones.</li><li>Ugly teeth, unprompted sharp teeth, small risk of double mouths, the worm may appear to have 2 tongues or a barely recognisable angular tongue.</li><li>Maws are usually bad. Worms don't have uvulas in this game.</li><li>The lower lip texture (seen in W4M) may be sharp.</li><li>I added a few sick worms, however this led to the unexpected side effect where if you prompt something to be green the worm's skin may turn green. Adding "green skin" to negative should help.</li><li>This may be just a personal opinion, but the area between eyes and snout may look obnoxiously sharp.</li><li>I included a bunch of images of worms holding weapons but those don't come out right.</li><li>The lidded eyes in-game use just one texture so there's not much variety with them.</li></ul>🤔 Things this model is NOT trained on but may be able to do (easily or brute-forced, it varies):<ul><li>Anthros (work best with Noob)</li><li>Hairstyles different from the in-game ones, can do blunt bangs despite the game not having them. The game doesn't have many hairstyles however so expect repetitive sharp bangs.</li><li>Non-canon eye colours</li><li>Eyelashes, eyeshadow (eyelashes may appear unprompted in some cases, but eyeshadow is difficult to get and it may look bad)</li><li>Sharp teeth (should be easy, may appear unprompted)</li><li>Tongue stuck out of mouth</li></ul>Dataset also doesn't contain any neck or body clothes (the game doesn't have those), try getting those at your own risk.<h3 id="dataset-27js0e3d4">🖼 DATASET</h3>I like how Worms look in Worms 4: Mayhem and Worms: Ultimate Mayhem (WUM is basically a remaster of W4M with some adjustments, Worms 3D campaign and DLCs), and in W4M I can freely control the camera and change game speed (thus most images come from there). Unfortunately there's no in-game functionality to remove elements with text, but I captioned them, maybe it will help in negative prompts.Obviously with a dataset of over 800 images tagging isn't fully consistent. I tagged images manually with replacement and mass removal assisted by Civitai and Notepad++. When AI randomly assigns genders to Worms and assumes they're Pоkemоn, can you really trust it with autotagging? 🌚<ul><li>Images are not edited in any way besides basic cropping.</li><li>Dataset consists entirely of game screenshots and does NOT contain fan-edited textures or fanmade items, all depicted assets are property of Team17.</li><li>A few pictures were taken in WUM with a mod, but the only sign of modding is the removal of the annoying sepia filter (it should be completely absent in the dataset) and removal of rim lighting (making it look a bit more like W4M).</li><li>Eyebrow related captions: floating black rounded eyebrows, eyebrowless</li><li>"Default worm" caption refers to worms with all of the following criteria: eyebrows, natural skin colour (healthy), no customisation, no Jetpack/Parachute selected.</li><li>Dataset does NOT contain teal alien worms out of pupil/eyebrow generation concerns. They're only present in Pony V1.</li><li>I did NOT caption most background elements. Base SDXL/Pony/Illustrious models might make derpy backgrounds mimicking the game.</li></ul>A few images showing some differences between W4M and WUM worm textures, and some insight on hair and moustache items (very specific details may not work well or at all if prompted, e.g. I can't get Professor Worminkle's correct appearance)<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/568f8ad8-4e13-4998-89ec-8fd8c707913e/width=525/568f8ad8-4e13-4998-89ec-8fd8c707913e.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/00bb7b75-1707-4146-8092-618b24993229/width=525/00bb7b75-1707-4146-8092-618b24993229.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/830aa5de-4e62-4455-aea1-e4afb89b5ba5/width=525/830aa5de-4e62-4455-aea1-e4afb89b5ba5.jpeg" />Negatives I tend to use: human, humanoid, anthro, lamia, furry, monster boy, legs, ears, human ears, animal ears, horns, antennae, single eyebrow, 3 eyebrows, 4 eyebrows, nose, animal nose, human nose, snout, nostrils, claws, fingernails, dimly lit worm, dim lightingNegatives I tend to use for anthros: amputee, disembodied, human, lamia, countershading, countershade torso, countershade face, fur, scales, human ears, ears, animal ears, horns, antennae, single eyebrow, 3 eyebrows, 4 eyebrows, nose, animal nose, human nose, snout, nostrils, claws, fingernails, toenails, digitigrade, humanoid feet, fake tail, anal tail

Worms Team17 Mayhem [multi model]

🗑 THIS LORA PRODUCES LOW QUALITY IMAGES, USE AT YOUR OWN RISK.Let this be an example on how NOT to train a NoobAI-XL V-Pred LoRA. I found a comment saying how they trained a NoobAI V-Pred LoRA, which apparently gave grey and blurry images... I have morbid curiosity and IDK when NoobAI will be one of the default options, if ever (I heard a rumor but no official confirmation), I went ahead and trained this anyway via the Civitai trainer with default settings, using the V-Pred-0.65S-Version (latest V-Pred version at the time of training). This costed me over 1000 Buzz, "enjoy" my sacrifice, I guess.Reused dataset from SD1.5, I didn't add any hand tags from Danbooru or E621 that would fit worms to avoid potential conflicts. Trained on December 4

Files with SHA256: 2ba5ebe4c98e8e0815630845d09bc9d8b950ab1da8db2ac75e9a02817ddec869