UPDATE: Following the common practice, the most recent version is trained on NAI with clip skip 2The advantage of my models is clearly the ability to put multiple characters in a same image(I was considering training a full model with the same dataset as well, but the LoHa seems to be reasonably good at this point The other models are otherwise trained on <a target="_blank" rel="ugc" href="https://huggingface.co/JosephusCheung/ACertainty/tree/main">ACertainty</a> Please check the version notes for further details including the clip skip to useNormally you should be able to use the lycoris models on a lot of different checkpoints. In case of incompatibility, you can lower lycoris weight or better, merge your base model with AC/nai.More checkpoints can be found on the associated hugging face repository <a target="_blank" rel="ugc" href="https://huggingface.co/alea31415/oshinoko">https://huggingface.co/alea31415/oshinoko</a>Screenshots and fan arts are respectively tagged with aniscreen and fanart. Using aniscreen gives both stronger anime style and better resemblance. Using aniscreen on style models may cause weird effects.You may want to use ; to separate characters (see the prompts of the example images)<h2 id="heading-2701">Characters</h2>EP01<ul><li>HoshinoAi (add hair ornament if you want)</li><li>AmemiyaGoro</li><li>TendojiSarina</li><li>Aquamarine baby</li><li>Aquamarine child</li><li>Aquamarine</li><li>Ruby baby</li><li>Ruby child</li><li>Ruby (add red ribbon if you want)</li><li>ArimaKana child</li><li>ArimaKana (I suppose you should add some hat)</li><li>SaitoIchigo</li><li>Miyako</li><li>GotandaTaishi</li><li>KurokawaAkane</li></ul>Main improvement of EP02<ul><li>Aquamarine</li><li>Ruby</li></ul>Added in EP04<ul><li>NarushimaMelt</li><li>KaburagiMasaya</li><li>KichijoujiYoriko</li><li>ShiranuiFrill</li><li>KotobukiMinami</li></ul>Added at the end of the season<ul><li>MemCho</li></ul><ul><li>KumanoNobuyuki</li><li>SumiYuki</li><li>MorimotoKengo</li><li>Pieyon</li></ul>Comments on the most recent model:Pieyon is not learned very well as can be seen from the example images, showing this is a more difficult task than learning the appearance of the charactersThe star-shaped eyes are learned pretty well in the newest model, probably too well that it sometimes wants to put it on other characters, especially when we do inpaintingYou may want to avoid this by putting related keywords in negativeI suppose this suggests a sort of character blendingSomeone reported to me the EP04 version has more important character blending compared to the EP02 versionI am not sure how it would be the case for this new one as I do not have an objective way to judge thisHopefully it works fine for you<h2 id="heading-2702">Dataset composition</h2>fanart 2432screenshots<ul><li>EP 01 1268</li><li>EP 02 357</li><li>EP 03,04 777</li><li>EP 05-11 2582</li></ul>regularization ~35K--I don't need support or credit, but I would be glad to know that you are using the models I trained and find it useful.Moreover, I would like to advocate for more franchise models.You can take a look at my workflow <a target="_blank" rel="ugc" href="https://github.com/cyber-meow/anime_screenshot_pipeline">https://github.com/cyber-meow/anime_screenshot_pipeline</a> if you are interested. You may also check this nice page <a target="_blank" rel="ugc" href="https://civarchive.com/articles/262/directory-of-girlpacks-character-packs">https://civarchive.com/articles/262/directory-of-girlpacks-character-packs</a> for a collection of such models,

OshiNoKo-characters-LoHa/LoCon/FullCkpt 推しの子 || Hoshino Ai / Hoshino Aquamarine / Hoshino Ruby / Arima Kana / Saito Miyako / Kurokawa Akane / Kotobuki Minami / Shiranui Frill / Sumi Yuki / ...

Trained on clip skip 1.Up to this point, Aqua, Ruby, Kana, and Akane are only trained with limited fan arts, so the quality of these characters will be largely improved later after more episodes come out.The trained LoHa provides a smaller alternative. However, the general rule seems to indicate that the smaller the network is the easier you get style conflict. Therefore, you may want to avoid using aniscreen on any model except for ACertainty. Fanart should be fine.After some test the model seems to work well but I don't plan to make new example images, so I only updated the last four images for some comparisons. As for the full checkpoint it seems to get the anime outfit right more frequently.Please also adjust weight and prompt accordingly for overfitted characters. Details such as star pupils would be the first thing you lose when you switch base model. It turns out you can still get it right with aniscreen on ACertainty.Other checkpoints can be found at <a target="_blank" rel="ugc" href="https://huggingface.co/alea31415/oshinoko/tree/main/checkpoints_0414ep01/loha">https://huggingface.co/alea31415/oshinoko/tree/main/checkpoints_0414ep01/loha</a>

Files with SHA256: 0a5e6d4dc7437c5c32ecaac4d1adfdc8b7c940cfd7ad3584e6be1172594bbc3e