<h1 id="v15-to-v16-(illustrious-model)">V15 to V16 (Illustrious Model)</h1>So, like that, The WonderMix goes into SDXL ^^The 2.5D style is not completly there, but i'm close, very close to get it and i'm working on it. Right now it's the best i'm able to do with my limited ressources (mainly a very old gaming Laptop)<h3 id="prompt">Prompt</h3>here what i used for all the image:<ul><li>Postive : masterpiece, best quality, high quality at the start of the prompt</li><li>Negative : low quality, worst quality, bad anatomy, bad proportions, bad perspective, bad hands, bad feet, blurry, jpeg artifacts</li></ul>You can try other things, the model is very responsive and easy to use.<h3 id="generationupscale-settings">Generation/Upscale Settings</h3>First the generation Settings<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3cbcc250-c9cc-4df0-a989-cff8b0e8abc0/width=525/3cbcc250-c9cc-4df0-a989-cff8b0e8abc0.jpeg" />Then the Upscale Settings<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/481b25b7-d302-4e29-8233-52a389dc8b18/width=525/481b25b7-d302-4e29-8233-52a389dc8b18.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/c1774487-9498-4d47-9905-43d238f8590d/width=525/c1774487-9498-4d47-9905-43d238f8590d.jpeg" />This is what i used for all the showcase images.For those using ComfyUI, a simple workflow with an Hires pass using Nearest-Exact X1.5 with a 0.6 Denoising + an SD Ultimate upscale work well tooEnjoy :)<hr /><h2 id="edit-on-v14">Edit on V14</h2>I found that several Anime Charater are present in the model and don't need a LoRa. A list of what i found for now is below in the model details (if you find some i didn't ,tell me i will add to the list)[MAJ 15/06 : 42 Characters added to the list, most from Fate Serie, especially FGO and Touhou - Total : 50 Characters]<h1 id="the-v14-is-a-sd1.5-lcm-model.">The V14 is a SD1.5 LCM Model.</h1>After a big slump (merging, unhappy with the result, erasing it, taking a break.... rinse and repeat), i was finally able to release a new version.And with this one i think i really nailed the 2.5D feel.This version is, by far, the best i ever done : very flexible, stable and versatile.It respond very well to prompt, in fact even a bit too well.... So if you got a weird result, it's mostly because you were a prompt control freak (like me :p) and you need to remove some tags or change some.Give some freedom to this model and you'll get some amazing results.And, of course, it's NSFW/H compliant ^^Model and Lora used in the merge : <a target="_blank" rel="ugc" href="https://civarchive.com/collections/1506160">https://civarchive.com/collections/1506160</a>Don't hesitate to try them and to thumb up if you like. They deserve it!<h3 id="prompt">Prompt</h3>Simple : (masterpiece, best quality, highres) at the start and (worst quality, low quality:1.4) for Negative.For Negative Embeddings/TI..... just don't, especially thoses that are supposed to fix hands : they are completly messing them insteadAs for the rest.... most of them produce lesser quality. They especially reduce background details.I was talking about this version being flexible and versatile here why :if you want an Anime feel use (masterpiece, best quality, highres, anime coloring, flat color) (exemple : <a target="_blank" rel="ugc" href="https://civarchive.com/images/14211961">https://civarchive.com/images/14211961</a>)If you want a more semi-realistic feel use (masterpiece, best quality, highres, realistic, photorealistic) (exemple : <a target="_blank" rel="ugc" href="https://civarchive.com/images/14211979">https://civarchive.com/images/14211979</a>)<h3 id="generation-settings">Generation Settings</h3>Let an image do most of the talk ^^<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/2a3763c9-7ceb-4172-88e0-795942e4b0e9/width=525/2a3763c9-7ceb-4172-88e0-795942e4b0e9.jpeg" />Stay between 10 and 15 for Sampling Steps.Don't go over 15.And i mean it.... Even at 16, the same seed change a lot and it's getting messy.You can go a bit under 10, but i don't encourage it (the quality start to drop a bit too much)For the CFG Scale between 1 and 2.And the HiresFix, you can go down to 0.2 for Denoising Strenght, but don't go over 0.3.About the Steps, it depend of your Upscaler, so either stay with my settings (4xUltraSharp + 5 steps) or try your own settings with a different one<h3 id="characters-present-in-the-model.">Characters Present in the model.</h3>First i'll add, that you need to tweak some details for character (mostly breasts size or eyes color)Here the list of what i found so far :<ul><li>Tohsaka Rin, Artoria pendragon, Tamano, Mash Kyrielight, Scathach, Elizabeth Bathory, Shuten Douji, Minamoto no Raikou, partial Jeanne d'Arc (outfit), Illyasviel Von Einzbern from Fate Serie (Several outfits for some included)</li><li>Aqua and Megumin from Konosuba (outfit included)</li><li>Asuka Langley from Evangelion (Bodysuit included)</li><li>Makima from Chainsaw Man (missing the braid by default)</li><li>Partial Shiroko from Blue Archive (Missing the animal ears and the school uniform, but the Halo and everything else is here)</li><li>Mikasa Ackerman from Attack on Titan (uniform included)</li><li>Belfast from Azur Lane (only maid outfit tested)</li><li>Asuna from SAO (SAO Outfit Included)</li><li>Toga Himiko, Uraraka Ochako, partial Jirou Kyouka (missing the Ear Jack :( ), Asui Tsuyu from MHA (some Hero outfit are missing)</li><li>Konpaku Youmu, Hakurei Reimu, Remilia Scarlet, Flandre Scarlet, Izayoi Sakuya, Saigyouji Yuyuko, cirno, Komeiji Koishi, from Touhou (complet outfit included, meaning : ears, wings, accesorry and such.), Kochiya Sanae (wrong hair color and missing the frog hair ornament) and Kirisame Marisa's hat.... But only the hat XD</li><li>Noire, Purple Heart, Nepgear, Iris Heart, Tennouboshi Uzume from Neptunia Serie (this is quite a mess, some have complete outfit, other partial, and sometime it's a mix of different outfit XD.... But sill.... Planetune FTW :p )</li><li>Shihouin Yoruichi (missing ponytail by default) from Bleach</li><li>Yuzuku Yukari, Hatsune Miku, Megurine Luka, Kagamine Rin, MEIKO from Vocaloid (Only tested default outfit, they are complete)</li><li>Satonaka Chie, Amagi Yukiko from Persona 4 (they don't get her school uniform right, you need to work a bit to get them.... And yup, no Rise ;( )</li><li>Tifa lockhart, Aerith Gainsborough from FF7 (Outfits need some tweaks, and Aerith is missing her ponytail by default)</li><li>Kousaka Kirino from OreImo</li><li>Suzumiya Haruhi</li><li>Partial Tatsumaki from One Punch Man (couldn't get her outfit)</li></ul><h3 id="post-scriptum">Post Scriptum</h3>I don't think i can do better for The WonderMix.So, i'll enter "Maintenance Mode" for this model.Meaning : When new version of the models used in the merge create a reel improvement, i will release a new version too.Meanwhile : Enjoy this awesome version and don't hesitate to share your generation :)<hr /><h3 id="v1-to-v13">V1 to V13</h3>Basic SD1.5 settings : Euler a/Euler/DPM++2M @30 Steps and CFG7 with Hires fix with the same settins and upscale by 1.5For the positive and negative prompt, use the classics tags but avoid Negative Embeddings.

The WonderMix

Merged Models and loRa here : <a target="_blank" rel="ugc" href="https://civitai.com/collections/852928">https://civitai.com/collections/852928</a>Baked VAE : Blessed2DetailsThis version is more anime but still can produce "realistic" generation. Backgrounds are amazingly detailed ^^I came back to Blessed2 as VAE because, on this version, i found the colors from kl-f8-anime2 were too saturated.But if you want an even more Anime style, use it.Still not using any Negative Embeddings/TI,and the output is very stable (especiallu hands are not often a mess even with very demanding prompt using movement).Ofc, like always, you are free to use them :)No Fp32 for this one too, because i used only fp16 version for the merge

Files with SHA256: 2e5a2e46e027f7b148562277f35fb27a3940e76fbc6cf125f1f884762767bc18